Intravital and High-Content Multiplex Imaging of the Immune System

Abstract

Highly motile and functionally diverse immune cells orchestrate effective immune responses through complex and dynamic cooperative behavior. Multi-photon intravital microscopy presents a unique and powerful tool to study the coordinated action of immune cell interactions in situ. Here, we review the current state of intravital microscopy in deepening our understanding of the immune system and discuss its fundamental limitations. In addition, we draw insights from recent technical advances in multiplex static tissue imaging methods and propose an approach that could enable simultaneous visualization of cellular dynamics, deep phenotyping and transcriptional states through a new type of correlative microscopy that combines these imaging technologies with advances in complex data analysis.

The Immune System as a Biological System Shaped by Dynamic Cellular Organization

Complex multicellularity emerges with the development of molecular adhesion, intercellular signaling pathways and defined differentiation programs in simple organisms [1]. Individual cells no longer operate merely as a community of uniform cell aggregates, but as a diverse collective of specialized cell types that can interpret local positioning and receive input from neighbors to perform complex biological functions. This dynamic ‘feedback loop’ amongst interacting cells—often involving complex layers of interplay between signaling pathways, protein expression and transcriptional regulation—enabled formation of tissues supporting such specialized functions as locomotion, nutrient absorption, and host defense. Comprehensive understanding of biology thus requires a systems level approach that simultaneously interrogates how diverse components interact across spatio-temporal scales to ultimately shape structure, organization and function.

While widespread changes in cellular organization are typical of embryonic development and the pubertal period, in adult organisms, the immune system almost uniquely represents a continuously dynamic, complex system characterized by constantly changing positions of its diverse components. Such distinctive behavior allows the immune system to adapt to specific challenges encountered when mediating exchange with the external milieu as well as during internal homeostatic regulation across tissues and organs of the body.

At least two unique challenges shape the behavior and property of immune cells. First, the fast replication kinetics of pathogens demand effective responses capable of acute sensing of large surface areas, precise relay of information and timely eradication of pathogenic threat. Thus, unlike their epithelial or stromal counterparts, many immune cells are highly motile and can navigate through dense tissue environments. Indeed, when intravital microscopy became available for deep tissue imaging, it revealed in detail the dynamic components of immune cell cooperation: the coordinated swarming of neutrophils in tissue lesions [2, 3]; high rate scanning of passing lymphocytes by strategically positioned antigen-presenting cells [4, 5]; migration of sentinel dendritic cells across tissues to deliver critical priming signals [6, 7] are just a few examples. Second, the unpredictable nature of external threats—from simple viruses to large parasitic worms—dictates a need for diverse arsenals to enable effective host defense. For instance, the functional diversity and plasticity of CD4+ T cells embody an evolved solution to such challenges.

With the advent of high parameter flow cytometry and single-cell sequencing technologies, the striking extent of such diversity has become increasingly apparent. However, despite collection of many single cell datasets using these powerful methods, how these different cellular components engage in the complex sequence of dynamic events that determine cell fate specification and shape specific immune outcome, and how these events transpire in a spatially-organized manner within lymphoid and parenchymal tissues, remain fundamental yet unresolved questions in immunology. Here, we provide an overview (including some recent case studies) of how multi-photon intravital microscopy (MP-IVM) has become integral to acquiring a greater understanding of the immune system. We then pivot to a discussion of emerging highly multiplex static imaging methods and propose that combination with MP-IVM yields an especially powerful approach for immune analysis.

Intravital Microscopy at the Forefront of Understanding Immune Cell Interaction: Current State and Limitations

The highly migratory nature of immune cells made investigating their dynamic movement in situ a critical step in understanding how the immune system orchestrates a functional versus dysfunctional response. Some of the earliest efforts to study the immune system—as long ago as the 19^th century—involved brightfield light microscopy of thin, translucent animal tissues to study immune cell migration. However, until the introduction of multi-photon microscopy into immunological research in the early 2000s [8–12], most attempts to study immune cell migration involved experiments utilizing in vitro 2-dimensional (2D) cell culture, 3-dimensional (3D) collagen gel systems, or brightfield video visualization of the emigration of blood-borne cells into nearby tissue [13].

Today, MP-IVM is the mainstay technique for live tissue imaging and tracking of cell dynamics in situ. MP-IVM enables two key types of data to be gathered (Figure 1A). First, dynamic cell behavior and the complex interplay between different fluorescently labeled cell populations can be visualized and examined deep (several hundred microns) within a diversity of tissues and organs (summarized in [14]). Second, with appropriate fluorescent indicators, live signal transduction events may also be recorded as cells actively engage in interactions or receive soluble stimuli. Prominent examples include use of calcium flux signaling [15–18] and NFAT nuclear localization [19–21] as putative indicators of antigen-specific T cell receptor (TCR) engagement, or deployment of STAT1 and Gamma interferon Activation Sites (GAS) fluorescent biosensors to visualize IFN-γ induced signaling in tumor cells [22, 23]. Further, multi-photon excitation is also capable of illuminating autofluorescence and second-harmonic generation, thus enabling real-time visualization of cellular metabolism (e.g., metabolic substrates including NADH and FAD) as well as tissue structural components (e.g., collagen) respectively. Such metabolic imaging is typically achieved through fluorescence lifetime imaging (reviewed further in [24]). Together, MP-IVM allows in situ functional activity of immune cells to be understood in the context of their spatio-temporal dynamics in a manner that other methods cannot achieve.

(Key Figure) Figure 1. Key data derived from intravital microscopy and highly multiplexed static tissue microscopy.

(A) Intravital microscopy enables collection of time-lapsed recording of cellular migration, signal transduction events and cell-cell interactions in situ. Broken lines indicate a short history of the cells’ movement.

(B) Deep phenotyping capability of highly multiplexed static tissue imaging achieved through iterative staining (cyclic immunostaining). Each iteration consists of immunostaining with 4–7 antibody marker, imaging and fluorophore inactivation. Complete dataset is then constructed computationally with images from each iteration to generate highly multiplexed view of the cells in tissue. Abbreviation: cDC1, cDC2: conventional dendritic cell, type 1 or 2.

The unique advantages and the shortcomings of each technique are denoted in filled and open circles, respectively.

While multi-photon excitation enables deep tissue intravital imaging, the point-scanning nature of MP-IVM nevertheless limits the imaging speed at which fast cellular motion can be captured, such as of cells or bacteria traveling in vessels under high flow pressure. To this end, spinning-disk confocal microscopy is typically employed to record activity at high frame-rate, and this technique has been especially useful for imaging the liver [25, 26] and kidney [27]. The disadvantage of spinning-disk microscopy is the limited tissue penetration of single-photon excitation, making its use highly dependent on the specific surgical preparation of the tissues to expose the regions of interest. Light-sheet microscopy has also gained popularity for fast-scanning and low intensity illumination (less phototoxicity) applications. However, the orthogonal arrangement of its dual objectives still imposes physical constraint on the size of specimens that can be examined, and is so far limited to relatively transparent and small organisms, such as zebrafish, Drosophila and the developing mouse embryo. Advances in lattice light-sheet microscopy have recently permitted migration of macrophages and cancer cells within zebrafish embryo to be recorded at super-resolution detail [28].

An exciting development in light-field microscopy (LFM) is worth mentioning here: a novel framework of digital adaptive optics scanning light field tomography (DAOSLIMIT) has enabled intravital imaging of mammalian tissues at an unprecedented level of spatial and temporal resolution [29]. Unlike many of the microscopy methods described thus far, LFM utilizes a microlens array to capture multiple perspectives of the specimen at once, with computational methods applied post-acquisition to deconvolve and reconstruct the 3D image. Through this cutting-edge approach, high-resolution cellular processes of neutrophil as well as tumor cell migration in the mouse liver could be captured, revealing details such as the shedding of migrasome-like vesicular structures during contact with blood vessels [29]. These details can provide invaluable insights into how long-range communication occurs between immune cells.

Recent insights gained from intravital imaging of the immune system

Beyond the seminal findings of early MP-IVM studies summarized in several reviews [14, 30, 31], a number of recent applications of this method have substantially expanded our understanding of immune cell dynamic behavior and the relationship of these dynamics to physiology and pathology.

Immune surveillance in tissues

Immune surveillance in tissues constitutes the first line of responses mounted against tissue injury, pathogen entry, or malignant tumor outgrowth. The versatility of intravital microscopy and its importance in uncovering the dynamic mechanisms that underlie immune surveillance can be illustrated in the following studies.

The skin and mucosal surface form a protective barrier against external environment and are populated by a variety of immune cell populations. Intravital imaging had revealed that neutrophil swarming at focal sites of tissue damage results in a self-amplifying inflammatory response that isolates the damage/infection, but also disrupts the local tissue architecture [3]. These findings raised the question of whether cell death during regular cellular turnover in tissues would also attract such neutrophil-driven inflammatory damage. Again using MP-IVM, a recent study described a regulatory role of tissue-resident macrophages, showing that these myeloid cells sense individual parenchymal cell death and rapidly sequester (“cloak”) the corpse, preventing initiation of the molecular cascade that drives neutrophil swarming with its attendant tissue disruption [32]. This mechanism prevents gradual tissue deterioration over time that would occur if inflammatory responses to episodic cell death were not prevented. A similar role has also been reported for lung alveolar macrophages, with intravital imaging revealing the unexpectedly dynamic crawling behavior of these macrophages and their constant clearance of inhaled bacterial pathogens, preventing neutrophil recruitment and the resulting inflammation [33]. Further extending the role of macrophages in maintaining tissue homeostasis, a recent study showed that free-floating peritoneal GATA6+ macrophages rapidly aggregate around focal injuries and physically seal the lesions to promote tissue repair [34].

During inflammation, an intrinsic feedback mechanism that limits excessive swarming of self-amplifying neutrophil recruitment has also been revealed through MP-IVM [35]. While the self-amplifying process is driven by GPCR sensing of leukotriene B4, GPCR kinases (GRKs) expressed by neutrophils mediate a negative feedback loop that promotes GPCR desensitization [35]. Neutrophils deficient in GRK2 engage in persistent swarming behavior that leads to excessive accumulation of neutrophils over time while also interfering with the efficiency of bacterial clearance. Intravital microscopy thus helped reveal the regulatory dynamics that carefully balances acute inflammatory responses, pathogen constraint, and unrestrained autoimmune damage.

On a separate front, solid tumors can sporadically form along epithelial layers. MP-IVM revealed that tissue-resident CD8+ memory T cells (T_RM) that populate the skin epidermal layer dynamically survey the local microenvironment, and the presence of these T cells correlated with reduced formation of subclinical tumor lesions [36]. Furthermore, intravital microscopy demonstrated that mucosal and skin T_RMs proliferate in situ upon antigen stimulation [37, 38], thereby revealing the autonomous surveillance capability of this non-circulating memory lymphocyte subset that inhabits peripheral tissues for extended times. Another recent study found distinct patterns of productive TCR signaling amongst tumor-infiltrating regulatory T cells (Treg) and CD4+ T helper cells [21]. Instead of forming stable contact with dendritic cells (DCs), Tregs mediate their suppressive function through transient, unstable contacts with DCs in a CTLA-4 self-regulatory feedback loop. Given the increasing usage of CTLA-4 immunotherapy in cancer patients, understanding the dynamic interactions of tumor-infiltrating T cell subsets thus uncovers the intricate balances required to achieve effective treatment.

Beyond skin and mucosal surfaces, the central nervous system (CNS) is yet another environment that demands tight regulation of immune cell activity. Although meningeal lymphocytes have been detected at the CNS border, they were initially thought to have derived from systemic circulation. A recent report utilizing MP-IVM showed that meningeal B cells found at the CNS border are instead derived from calvarial bone marrow [39], indicating the presence of an immune privilege niche that harbors a private source of meningeal B cells. These B cells would likely to be depleted of local self-antigen auto-reactive clones, thus allowing them to perform immune surveillance at the CNS while reducing the risk of causing autoimmune pathology.

Migratory mechanics of immune cells in lymphoid tissues

Migratory mechanics of immune cells in lymphoid tissues constitute another domain in which IVM has been especially valuable. Although lymphocyte migration in the lymph nodes has been extensively characterized over the past decade, their movement within the spleen with a very different tissue architecture has been less well studied, in part due to technical limitations in employing standard MP-IVM methods as well as the abundance of highly autofluorescent red blood cells. Through the use of an improved technique, a recent study [40] uncovered a unidirectional perivascular migration pattern of T cells into splenic T cell zones through the marginal zone bridging channels [41]. Splenic T cells in the red pulp sinuses latched on to perivascular stromal cells in a GPCR-dependent manner, while the one-way migration is mediated by CCR7-dependent chemoattraction. Interestingly, such crucial bridging channels collapse when using explanted spleen, thus further emphasizing the importance of live intravital tissue imaging approach under certain contexts.

Limitations of intravital microscopy

Despite the valuable data it provides, MP-IVM (or light-field microscopy) approaches have some major limitations. The low multiplexing capacity of the existing technology, in which only ~3–5 populations with distinct fluorescent signatures can be realistically visualized per experiment, is a considerable downside. Visualization of cells expressing specific protein markers is highly dependent on the availability of fluorescent transgenic reporters. Additionally, many existing reporter cell lines and transgenic animal strains express the same set of fluorescent proteins (commonly GFP, YFP or RFP variants) that cannot be spectrally resolved in a “mix and match” fashion. Considering the diversity of immune cell subpopulations that have been uncovered, practical attempts to visualize deep phenotypic variations of immune cells would require creation of a very large number of new reporter animals expressing unique sets of fluorescent proteins. Consequently, dynamic in vivo imaging often demands researchers to choose between breadth (visualizing multiple major populations) and depth (visualizing a specific subset at the expense of other major cell types). This limits the capacity of MP-IVM to address many critical questions that require fine grained phenotypic dissection of cell types, as opposed to reducing highly heterogeneous subpopulations of immune cells (e.g. CD4+ T helper subsets) into an averaged representation that can bias interpretation.

Future improvements in new fluorescent proteins, optics, detector capacity, scanning speed will be instrumental toward advancing the versatility and utility of dynamic tissue imaging. However, we believe that the extensive and exciting developments that have taken place in the field of static tissue imaging are highly connected to improving the information to be obtained by live imaging. While lacking the temporal dimension of dynamic IVM, the extensive multiplexing capacity of these emerging techniques strongly complement the limitations of MP-IVM.

Deep Phenotyping Using High Content Static Tissue Imaging

Immunofluorescence (IF) microscopy delivers the much-needed, complementary spatial information that tissue dissociation-based methods (e.g. flow cytometry, single-cell RNA sequencing) lack. However, IF microscopy is complicated by its own unique set of technical challenges and until recently, high-content, single-cell level quantitative analysis of tissue samples was not a practical option. Owing to the instrument design, the nature of the specimens used, preparation techniques and detection methods, IF microscopy has been traditionally constrained by: 1) low plex detection capability; 2) population under-sampling due to thin cross-section of the tissue specimens; 3) low throughput; 4) lack of standardized single-cell quantification calibration and methods; and 5) lack of robust quantitative algorithms that can unravel the complex spatial relationship between diverse cell types in tissue space.

Circumventing the fundamental limits of fluorescence spillover:

A critical limitation in modern IF microscopy is the relatively small number of fluorescence channels that can be recorded from each sample. One major reason is the fluorescence spillover between fluorophores that fundamentally limits multi-parameter detection without spectral compensation – routine for flow cytometry but hardly a common practice in IF microscopy. Furthermore, many commercial fluorescence microscopes have only a small number of fixed detectors (typically 4–5) and band pass filter sets configured to capture fluorescence emission from a particular set of fluorophore combination. More elaborate staining method such as OPAL [42] that utilizes proprietary fluorophores can detect up to 7 colors per sample, but even this is underwhelming compared to a standard flow cytometer equipped with 10–18 detectors.

Nevertheless, use of spectral array detectors as tunable band pass filters in a confocal microscope, when combined with carefully designed fluorophore panel and single-color spillover compensation, can permit sequential imaging of 7 to even as many as 14 colors without resorting to recording hundreds of spectral channels [43], although the signal-to-noise output becomes increasingly limited at the high end of this range by the need to discard more and more signal that spills into adjacent channels.

An alternative to fluorescence-based detection is found through the use of heavy metal ion-tagged antibodies. When combined with mass spectrometry, metal isotope tags can be detected based on their distinct time-of-flight profile, thus overcoming major spectral overlap issues intrinsic to fluorophores [44]. Initially deployed in combination with flow cytometry (termed mass cytometry [44]), it was subsequently adopted to laser/ion beam-based imaging platform to achieve high-content immunophenotyping visualization [45, 46]. These imaging platforms have recently been deployed to define spatial organization of tumor-immune interactions in human cancer samples [46–48]. However, these methods still require very expensive, hard to maintain instruments and are constrained by the availability of high-quality metal ion conjugated antibodies. Increasing commercialization of mass cytometry imaging platforms will certainly reduce these limitations going forward.

Dye cycling constitutes another alternative to overcome the limits of fluorescence spillover (Figure 1B). It can be achieved through photobleaching (MELC [49]), chemical inactivation (MxIF [50]; t-CyCIF [51]; IBEX [52]) or antibody stripping [53]. In principle, cyclic immunostaining allows unlimited number of markers to be visualized through multiple iterations of staining-imaging-stripping process. However, the stability of tissue and epitope quality over many cycles of light or chemical treatment may limit maximum cycle count. Though its practical ceiling has yet to be determined, >65-plex staining has been reported [52]. Additionally, potential distortion of tissue from photo/chemical treatments demands post-processing that accurately registers protein signal in images from the sequential cycles. Use of robotics (usually custom made) to automate the dye cycling process can also add to the cost of instruments.

An interesting variation of cyclic immunostaining approach involves the use of oligonucleotide-tagged antibodies – single-stranded DNA oligomers that carry unique oligonucleotide barcode assigned to each protein marker that can be subsequently visualized when bound to fluorophore-conjugated complementary strands. This approach has been employed successfully to achieve detection of ~60 markers in spleens from a mouse lupus model [54] and in human colorectal cancer samples [55]. The use of DNA oligomer tags is attractive for two additional reasons: 1) potential compatibility with RNA probe staining (discussed below); and 2) branched assembly of the DNA scaffold allows in situ signal amplification by increasing binding sites available for the fluorescent strands [56] – a much needed improvement for detection of weakly expressed protein in tissues [57]. Nonetheless, fully developed systems for utilization of this approach are commercial systems with limited panels of proprietary barcodes on pre-selected antibodies. This may change rapidly with the increasing availability of barcoded antibodies for CITE-seq studies [58] that can be re-purposed for tissue staining and whose oligo tags are openly available, allowing easy custom design of multiplex complementary oligo-fluor conjugates for detection [52].

Toward Whole Tissue Sampling and Quantitative 3D Tissue Microscopy:

Tissue imaging is conventionally performed on thin sections cut using a cryotome or microtome. While adequate for general visualization of cell and protein location, thin cross-sections severely under-sample cell populations of interest and can introduce biases in quantitative analysis. Furthermore, dynamic immune cell activity often takes place in anatomical substructures, such as the vasculature, germinal centers, and inflamed tissue lesions that are 3D in nature. Depending on how a section is cut, many important features may be missed or disproportionately represented (Figure 2).

Figure 2. Towards whole tissue sampling and quantitative 3D tissue microscopy.

Advances in optical tissue clearing have enabled imaging of cleared whole tissue (>1mm thickness) (left column). Major advantages of whole tissue sampling include complete visualization of 3D spatial relationship of cells with intact microanatomical structures and tissue vasculatures, as well as the full sampling of low frequency cell types. Due to the large tissue size, image resolution is often compromised in favor of faster acquisition speed, and may complicate quantitative analysis at single-cell level. Conversely, acquisition at lower speed (higher image resolution) leads to excessive photobleaching and unrealistically long acquisition time. Large tissues are also less amenable to iterative immunostaining methods due to antibody penetration issues and incomplete fluorophore inactivation. At the other end of the spectrum, thin cryosections (~5–20μm) (right column) benefit from established iterative and high-content multiplexing techniques (>60 parameters) together with highly developed quantitative analysis pipelines. Nonetheless, the thin cross-section often represents only a sparse sampling of total cell population in the tissues, preventing comprehensive view of the cellular spatial relationship in 3D spaces. In practice, optically cleared thick tissue slices (~100–300μm) (middle column) sectioned using vibratome represents an acceptable compromise that delivers adequate sampling, performance and potential for quantitative analysis.

Recent advances in optical clearing techniques offer a possible path forward that enables collection of larger scale, whole tissue and even small organism information by minimizing light scattering through chemical clearing processes. Multiple variations of optical clearing methods have been published, with each designed and optimized for specific applications (reviewed in [59]). Notably, the major implication for biological research is that—for the first time—it becomes possible to envisage whole-tissue quantitative histo-cytometry that can yield detailed 3D “atlases” that systematically catalog diverse (immune) cell subsets and their spatial relationships within tissue microanatomical niches. Sparse cell types can be readily identified, including visualization of T cells at their normal rare frequency in the polyclonal repertoire [60] or rare single metastatic tumor cells and their accessibility to drug treatment as demonstrated in a mouse breast cancer model [61]. Nonetheless, as is usual for nascent technologies, technical challenges abound for whole-tissue staining and imaging. Boxes 1 and 2 detail the challenges and advances being employed to address these issues. In practice, optically cleared thick tissue slices (100–300μm) cut using vibratome represent an acceptable compromise for both adequate sampling as well as thorough quantitative analysis (Figure 2).

BOX 1: Challenges in Single-Cell Quantitative Microscopy.

Robust and accurate single-cell quantitative analysis is essential for deriving morphological, spatial, phenotypic and transcriptional information from complex imaging datasets. Nevertheless, improvement is needed to tackle the many image processing and analysis challenges that are unique to such techniques.

Cell Segmentation and Object Classification:

Most existing cell segmentation algorithms are tailored toward detecting spherical objects (e.g., nuclei or round cells) rather than irregularly shaped cells (e.g., stromal cells and dendritic-shaped cells). The latter remains a difficult object classification challenge. A number of cell segmentation and annotation tools have been created over the years that employ various detection and segmentation algorithms, with each offering varying level of features, long-term software updates, and integration with commonly used image processing software e.g., ImageJ (reviewed in [69]). However, given the vast differences in image quality influenced by instrument configuration and user settings, as well as different tissue types subjected to widely variable tissue processing strategies, there is currently no general solution for cell segmentation in images.

Touching membranes:

The benefit of tissue microscopy—preserving cellular positioning within a native tissue organization—also leads to a critical problem: the fundamental limits in the resolving power of microscope optics prevents touching membrane signals from being clearly distinguished. While some efforts have been made to “compensate” for the fluorescence intensity of segmented adjacent cells based on the proportion of region shared between touching partners [54, 70], this remains a problematic issue unique to tissue imaging-based techniques.

Spatial Analysis:

For image quantitative analysis of single cells in tissues, early implementations of histo-cytometry have relied on exporting datasets into FCS format that can be read using flow cytometry analysis software [43, 71, 72]. As the number of parameters detected by multiplex imaging increases, high dimensional analytic methods are required to fully extract the information present in the dataset. There are presently only a few available spatial analysis algorithms and software tools that can quantify spatial interactions and cell neighborhood information in high dimensional imaging datasets [54, 73].

Finally, ease of use in data processing and analysis pipeline is one of the major factors that underpins the widespread acceptance of many single-cell methods. The recent proliferation of scientific computing tools and the development of open-source software ecosystems by flourishing communities of image analysis developers in biological sciences is a major step towards this direction.

BOX 2: Challenges and Advances in Whole-Tissue Imaging.

Uniformity in Immunostaining:

For thick tissue samples, fixative cross-linking and tissue extracellular matrix (e.g., collagen) can hamper antibody penetration and prevent uniform staining across tissue depth. Staining buffers often include mild detergents (e.g., Triton X-100) to facilitate antibody penetration, but over-treatment can result in loss of protein epitopes and spatial redistribution of GPI-anchored membrane proteins.

Advances:

• Searches for new detergent candidates, e.g., zwitterionic surfactant CHAPS for thick human tissue staining [74].

• Use of antibody fragments (e.g., Fab, scFv, nanobody) with substantially lower molecular weight to improve deep tissue staining [75, 76].

Slow scanning speed:

Using traditional laser point-scanning confocal microscope for the collection of high-resolution, thick whole tissue with increasing number of sequential scans to accommodate for more antibody markers can substantially raise the total imaging time (i.e., many hours to days) for a single sample.

Advances:

• Use of light-sheet microscopy that scans an oblique plane of the sample can improve acquisition speed by orders of magnitude. Future iterations of light-sheet microscopy design, such as single-objective light-sheet microscope would enable light-sheet scanning on samples laid against a flat surface (e.g., microscope slide).

• Computational approaches such as deep learning-based image restoration methods have been used to remove Poisson shot noise arising from low light illumination, high speed scanning aimed at reducing photobleaching and phototoxicity to live cells [77]. The rigor and accuracy of output from these nascent machine learning techniques, however, are not yet fully determined (reviewed in [78]).

Photobleaching:

Repeated confocal scanning of the same area along z-axis for optical sectioning of thick samples can cause photobleaching and distort fluorescence intensity measurements. This can be further exacerbated during high-resolution scanning with higher pixel dwell time (slower scanning speed) due to longer light exposure.

Advances:

• Development of bright and photo-resistant dyes [79].

• Improvement in microscope optics that increases detection sensitivity with less laser excitation.

• Use of oblique plane scanning light-sheet microscopy rather than the rasterized point scanning confocal microscopy can substantially reduce light exposure.

Objective Working Distance:

The design of high numerical aperture (NA), high magnification objective lens is associated with lower working distance that limits z-axis coverage to no more than a few hundred microns depth. For cleared, thick specimen imaging, lower NA and magnification objective is often used at the expense of its resolving power.

Advances:

• Specialized objective designed with correction collars that allow adjustments to match the refractive index of the clearing/mounting medium can deliver good NA, magnification and large working distance.

• Computational 3D reconstruction of overlapping tiles of images taken from different orientation of the specimens [76], or from specimens physically sectioned into smaller slices.

Reading transcriptomic signatures in tissue sections:

While single-cell sequencing has revealed a central role of tissue environment in shaping the heterogeneity of cellular transcriptional states, the fine-grained manner in which these tissue-specific differentiation processes are controlled has not been addressed because extraction of cells for analysis also prevents locating them in the anatomical context of the tissue. To this end, several methods have been devised to perform transcriptomic profiling that also contain some level of spatial information (detailed further in Box 3). Such ‘spatial transcriptomic’ approach has allowed spatial mapping of the transcriptional states of tumor and stromal cells within pancreatic ductal adenocarcinoma [62]. Another recent study combined single-cell RNA sequencing, spatial transcriptomics and multiplexed ion beam imaging to identify tumor-specific keratinocyte populations and their spatial interaction with diverse immune cell populations [63]. Although the use of these new spatial profiling techniques in studying the immune system is still limited, such promising demonstrations suggest that detailed spatial mapping of immune cell gene expression is attainable in the near future.

BOX 3: Linking of spatial, protein and transcriptional network information within tissue context.

With the proliferation of single-cell sequencing datasets, assigning spatial relationship between cells types has become a new frontier towards situating cellular transcriptional states within spatial contexts. Multiple approaches have been employed to extract high-content RNA information from tissue sections.

Microarray-based spatial transcriptomics:

This method involves specially designed microscope slides embedded with barcoded bead arrays that can capture mRNA when tissue sections are adhered on the slide surface. Tissue digestion would then leave the capture mRNA tethered to the beads to be sequenced with conventional RNA sequencing method. Because the bead barcodes correspond to specific microscopic regions on the slide, the transcriptomic data can be mapped to their corresponding regions for spatial reconstruction [80, 81]. Current commercial versions of ‘spatial transcriptomics’ have relatively low spatial resolution (~100μm diameter for each region), but a recent improvement of this technique has achieved subcellular resolution [82], potentially allowing the transcriptomic profiles of densely packed cells in tissues to be probed.

Single-molecule fluorescent in situ hybridization (smFISH):

One approach, MERFISH, combines smFISH with barcoded RNA library to reveal 100s to 1,000s of RNA transcript species without tissue digestion [83]. This method has been successfully deployed to study transcriptional changes during animal social behavior in up to a million cells among mouse neuronal populations in the hypothalamus [84]. Because individual RNA molecules are detected, smFISH-based approach also enables absolute counting of the transcript copy number, although distinguishing highly crowded single RNA spots remains a challenging feat. Combination with expansion microscopy [85] can alleviate the optical crowding issue [86], whereas expanding the barcode “palette” using an improved decryption algorithm has achieved a resolution of detecting 1,000s to 10,000s of mRNA species [87]. To enhance signal-to-noise ratio in tissue samples, variants involving in situ sequencing/amplification steps have also been developed to visualize RNA molecules in thick tissue slices [56, 88].

Co-detection of protein expression and transcriptional states in tissues

Co-detection of protein expression and transcriptional states in tissues remains a challenging task because proteinase digestion is typically employed in FISH staining to facilitate unmasking of RNA probe binding sites. As expected, this comes at a cost of destroying protein epitopes. The use of oligonucleotide-tagged antibody may render feasible co-staining of both proteins and RNAs, as DNA-based probe can still be preserved after extensive enzymatic digestion of the tissue proteins, including the antibody stems [86]. Although current co-detection methods are still limited in scope, the practical considerations to combine both DNA oligo-tagged antibody and RNA probe staining are sound in principle.

Perspective: Towards an Integrative, Multi-Dimensional View of the Immune System Using a New Type of Correlative Microscopy

In the preceding sections, we have discussed various emerging methods and technologies now in use, or being developed, to classify cell types and situate them within 2D tissue sections and 3D whole tissue contexts. While these techniques will eventually enable detailed mapping of spatial, phenotypic and functional states, static imaging alone cannot infer the dynamic behavior and cell-cell interactions in tissues that MP-IVM can provide nor can RNA studies truly address function that resides in proteins, their post-translational modifications, and their changing cellular localization in response to external signals. An integrative understanding of how these components interact in vivo is critical towards illuminating how the intersection of intercellular signaling, transcriptional changes and cellular functions collectively drives immune outcome.

Especially important for the highly motile immune cells are live imaging techniques that can map dynamic behavior and cell-cell interactions as well. However, it should be evident that these various methods are typically used in isolation and as such, are difficult to integrate because the data are collected from different samples and at distinct times in the evolution of a biological process. For the immune system in particular, how the diverse, complex and dynamic components come together, mechanistically, to drive specific outcome in tissue environment is a fundamental question that remains unresolved. In other words, how can we piece together the disparate views that have been revealed to us by various methods to form a more coherent, unified understanding of immunity?

A fundamental limitation of MP-IVM, as discussed earlier, is its inadequacy in capturing the phenotypic diversity of immune cell subsets during dynamic in vivo imaging. We propose that a correlative approach that combines dynamic MP-IVM with the static multiplexed 3D tissue imaging can help provide an integrative view that fuses observation of cellular dynamics with deep measurement of the phenotypic and functional states of the tracked cells (Figure 3). This approach is conceptually similar to correlative light and electron microscopy (CLEM), in which IF microscopy is first used to capture immunostaining signals (protein markers), followed by subsequent tissue processing for scanning electron microscope imaging to delineate ultrastructural features (reviewed in [64] and [65]) of the specific cells identified in the IF step. The unique advantage of this approach is that it provides a composite view of the same cells that can reveal both structural and protein localization information, which otherwise would not have been possible to obtain.

Figure 3. A correlative approach to integrate dynamic intravital microscopy with static 3D tissue imaging.

Correlative intravital and static immunofluorescence microscopy provides a means for deep phenotyping of cells tracked using intravital microscopy, which typically carries only limited information of the cells’ subset phenotypes and functional states. Intravital microscopy is first performed to collect dynamic information that entails cellular migration, interaction history and signal transduction events (left column). Tissue is then preserved through fixation and subjected to subsequent rounds of immunostaining to reveal additional protein markers (right column, color code denotes different cellular subsets identified via additional markers). Fixative injection “freezes” cellular movement and spatially encodes their identities, which can then be retrieved through post-acquisition computational methods (i.e., 3D image registration) that spatially align the images from both imaging modalities. High-dimensional phenotypic and spatial analysis (lower right panels) are then merged with cellular dynamics information (lower left panels) to form an integrated view of immune cell activity across time and space (center panel).

Correlative MP-IVM combined with static 3D immunofluorescence microscopy thus represents an intriguing approach to expand cell phenotypic granularity via immunostaining while also preserving temporal information: dynamic movements are first recorded with MP-IVM, followed by fixation of the tissue to arrest cellular movements (which allows cellular identities to be spatially encoded for later retrieval). The fixed tissue can then be subjected to subsequent rounds of immunostaining and RNA staining to add deep phenotypic information.

Indeed, such approach had been attempted before, to visualize immunostained TCR clustering at the apical dendrite projections of dermal γδ T cells [66]. However, it is only recently that technical advances have enabled the systematic application of this method. We anticipate that correlative MP-IVM, when combined with the latest advances in static highly multiplexed and 3D tissue clearing techniques, will be capable of vastly expanding the number of detectable phenotypic markers, as well as having the potential to incorporate transcriptional profiling through mRNA FISH staining.

Breakthroughs in adaptive light-sheet microscopy have enabled long-term whole-organism recording of mouse embryogenesis [67] and neurogenesis [68], where the distinct cell fates that contribute to different germ layers can be followed at single-cell level. These techniques employed multiple transgenic fluorescent reporters to track specific cell types. As discussed earlier, current technological constraints limit the number of fluorescent populations that can be visualized simultaneously. It is conceivable that a correlative approach that combines whole-embryo imaging with immunofluorescence staining at experimental endpoint could provide valuable information regarding protein/mRNA distribution, cell-cell interactions and the distinct state of individual cells.

In practical terms, an obvious benefit of this approach is that existing conventional hardware (commercial multi-photon, confocal and light-sheet microscopes) can be readily utilized without necessitating the construction of customized instrument, or even demanding an entirely new microscope design. Instead, challenges will lie in achieving high quality staining and tissue clearing, as well as developing robust computational methods that can efficiently register complex image datasets from both imaging modalities. The ability to perform single-cell level quantitative analysis of 3D tissues underpins the utility of this approach. While a fully operational version of this new correlative method has not yet been fully achieved, we have recently designed and tested many of the required experimental and computational components, and a working prototype of the entire method is beginning to emerge. These promising first steps suggest that this approach is likely to become a valuable addition to the tool set for examining immune cells or other dynamic cell processes in tissues.

Concluding Remarks

We are now at a critical juncture where data derived from different sources—intravital microscopy, high throughput spatial and flow cytometry, genomics and transcriptomics—each presents a unique slice of information and need to be integrated in a “multi-omics” fashion to understand crucial emergent properties of the immune and other biological systems. In an era of single-cell transcriptomics and data-driven research, intravital imaging techniques are poised to deliver the missing spatio-temporal context.

Concurrently, emerging technical advances in static tissue imaging have transformed the role of fluorescence microscopy that now promises construction of high-dimensional single-cell 3D “atlases” in which cellular interaction network can be spatially mapped in situ. We anticipate that combination of these new imaging techniques with MP-IVM through a correlative method can fuse measurement of cell dynamics and live signal transduction events with the functional and transcriptional states of the recorded cells. Although far from a trivial process (see Outstanding Questions), we anticipate that when approached correctly, such integrative effort will be not just an incremental advance, but a transformative one.

OUTSTANDING QUESTIONS BOX.

• To what extent can the spatial resolution, field-of-view, imaging depth and scan speed of MP-IVM be further improved through new optical and/or microscope designs, computational methods (e.g., machine learning-based denoising algorithms), adaptive optics and so on?

• How can the design and genetic engineering of new fluorescent reporters that enable live visualization of protein expression and signal transduction pathways be expedited?

• Development of computational methods that can robustly extract single-cell information from protein and mRNA staining for quantitative analysis remains a challenging task. What are the potential and emerging solutions that can adequately address these challenges?

• For intravital imaging, tracking of cell movements, especially amongst densely clustering cells, is still a very difficult task. What future improvements in cell tracking algorithms can we anticipate to accurately analyze cell movement in an automated fashion?

• Given the integrative nature of correlative microscopy, what kind of control/reference data will need to be collected and what normalization/processing steps will be required across samples/experiments for robust quantitative analysis of tissue microscopy datasets and to enhance reproducibility?

Highlights.

Immune responses are shaped by complex sequences of dynamic interactions between a multitude of functionally diverse immune cell types

Multi-photon intravital microscopy serves as a powerful tool to study immune cell dynamics, but is fundamentally constrained by low plex visualization

Emerging advances in static tissue microscopy deliver high parameter 3D tissue imaging capability, enabling deep immunophenotyping of immune cell subsets within spatial contexts

A new type of correlative microscopy that combines dynamic intravital microscopy and high-content static 3D tissue imaging may enable deep phenotypic visualization of immune cell interactions at unprecedented scale

GLOSSARY

Expansion microscopy

A method that utilizes chemical process to physically expands tissues isotropically by embedding tissues in a dense, cross-linking network of swellable hydrogel. During tissue expansion, biomolecules are pulled apart but retain their relative spacing with one another, and enabling super-resolution detection of these molecules.

Fluorescence lifetime imaging microscopy (FLIM)

An imaging technique that exploits the exponential decay rate of fluorophores. The fluorescence lifetime is the average time that a fluorophore remains in its excited state. Endogenous fluorophores such as metabolic substrates exhibit distinct lifetime that allows their detection through autofluorescence imaging.

Fluorescence spillover

Fluorochromes can exhibit broad emission spectra that can be detected in channels assigned to detect other fluorochromes.

GPCR

G-protein-coupled receptor is a large family of surface receptors expressed by immune cells for sensing of chemokines and chemoattractants, which are integral for immune cell migration in tissues.

Light-field microscopy

A type of microscopy that captures entire volume instantaneously using microlens array to generate perspective views of the specimen, after which deconvolution is applied computationally to reconstruct the 3D image volume. This method enables high-speed acquisition of the samples with low light excitation.

Light-sheet microscopy

A type of fluorescence microscopy that illuminates with a sheet of light beam to illuminate entire plane of the specimen at once. A detection objective is placed orthogonally to the illumination objective to collect fluorescence signal from the illuminated specimen plane, enabling acquisition at high-speed.

Marginal zone

The interfacing region that separates the red pulp and the lymphocyte-rich white pulp in the spleen.

Multi-photon microscopy

A fluorescence microscopy technique that utilizes simultaneous excitation of two (or more) photons to generate fluorescence signal. The longer wavelength used (multiples of single photon excitation) permits light penetration into deep tissue and reduces photodamage, enabling live tissue imaging.

NFAT

Nuclear factor of activated T-cells is a family of transcription factors expressed in many immune cell types. During T cell receptor (TCR) ligation, calcium influx induces a signaling cascade that promotes dephosphorylation and nuclear translocation of NFAT to regulate immune gene expression.

Registration

Image alignment process where two or more image datasets (usually taken at different times, in a different orientation or with different instruments) are transformed into a single coordinate system.

Second-harmonic generation

A non-linear optical process that generates signal at half the wavelength of incident light when passing through materials with a non-centrosymmetric crystalline structure such as collagen and myosin. It is especially useful for visualizing extracellular matrix and muscle fibers.

Single-molecule fluorescent in situ hybridization (smFISH)

A fluorescence microscopy method that allows detection of individual RNA molecules through the use of fluorophore-conjugated complementary nucleic acid probes. Single molecule detection is typically achieved via some form of signal amplification scheme to enhance detection efficiency.

Spectral compensation

A mathematical correction of fluorescence spillover between spectral overlapping fluorochromes in detector channels.

Spinning-disk confocal microscopy

A type of confocal microscopy that instead of a single pinhole, utilizes 1000’s of pre-arranged pinholes in a spiral pattern on an opaque disk, which rotates at high speed during image capture to significantly improve acquisition speed over point-scanning confocal microscope.

T_RM

Resident memory T cells are a memory lymphocyte population that resides in tissues without recirculating.