Histo-Cytometry: in situ multiplex cell phenotyping, quantification, and spatial analysis applied to dendritic cell subset micro-anatomy in lymph nodes

Summary

Flow cytometry allows highly quantitative analysis of complex dissociated populations at the cost of neglecting their tissue localization. In contrast, conventional microscopy methods provide spatial information, but visualization and quantification of cellular subsets defined by complex phenotypic marker combinations is challenging. Here we describe an analytical microscopy method, "Histo-Cytometry," for visualizing and quantifying phenotypically complex cell populations directly in tissue sections. This technology is based on multiplexed antibody staining, tiled high-resolution confocal microscopy, voxel gating, volumetric cell rendering, and quantitative analysis. We have tested this technology on various innate and adaptive immune populations in murine lymph nodes (LN) and were able to identify complex cellular subsets and phenotypes, achieving quantitatively similar results to flow cytometry, while also gathering cellular positional information. Here, we employ Histo-Cytometry to describe the spatial segregation of resident and migratory dendritic cell subsets into specialized micro-anatomical domains, suggesting an unexpected LN demarcation into discrete functional compartments.

Introduction

Decades of research have revealed the exceptional diversity of hematopoietic cell populations that comprise the innate and adaptive immune systems (Germain, 2004). Much of our current understanding of this heterogeneity comes from the application of two key technical advances, monoclonal antibodies (Kohler and Milstein, 1975) and flow cytometry (Perfetto et al., 2004). Cell types initially believed to represent a single lineage are now understood to comprise many distinct differentiated subpopulations with divergent functions in immunity. The definition of various distinct cell types is now typically achieved using highly multiplexed flow cytometric analysis for up to 17 parameters, with the newly developed mass spectrometry-based CyTOF method allowing more than 40 parameters to be studied at once (Bendall et al., 2011; Newell et al., 2012).

Contemporaneous with progress in dissecting the immune system’s components as isolated cells, optical imaging has revealed specialized anatomical localization of distinct cellular subsets in the steady-state or during immune responses, for example, the re-positioning of activated B cells at the T-B border during the development of T-dependent humoral immunity (Ansel et al., 1999; Garside et al., 1998; Reif et al., 2002). More recently, live intravital imaging has added information on the dynamic behavior of immune cells within various secondary lymphoid organs and peripheral sites (Germain et al., 2006; Sumen et al., 2004).

The crucial role played by tissue anatomy and cellular positioning in the development of effective immune responses emphasized by these recent microscopy-based experiments raises a key issue, especially as regards non-human primates or humans where the choice of analytic methods is more limited than with mice. The cells imaged in either static or dynamic modes by available methods are typically identified by one or a very few markers, in striking contrast to how most immunological studies are performed using flow cytometric methods. This precludes relating the spatial insights that can be obtained from optical imaging with the dense and precise phenotypic data derived from flow analysis. Yet only a combination of the two approaches can provide the field with optimal insight into how the immune system is organized and operates in health and disease.

Dendritic cells (DC) are a prime example of a cell type for which a method that can combine these two technologies would be of particular value (Chow et al., 2011). DC are critically involved in detecting, sampling, and processing information from invading pathogens and regulating the activation, differentiation, and expansion of adaptive CD4⁺ and CD8⁺ T cells (Heath and Carbone, 2009). DC, often characterized simply by co-expression of major histocompatibility complex class II molecules (MHC-II) and CD11c, are in reality a highly heterogeneous cellular population composed of distinct subsets with variable expression patterns of specific lectins, Toll-like receptors, inflammatory cytokines, and co-stimulatory molecules. These distinguishable subpopulations of DC have been reported to play specialized roles in sensing various infections, and to induce activation and differentiation of distinct types of effector CD8⁺ and CD4⁺ T cells (Edwards et al., 2003; Heath and Carbone, 2009; Helft et al., 2010; Kawai and Akira, 2011; Sancho et al., 2009; Shortman and Heath, 2010). As a prime example of subset complexity within tissues, murine skin draining lymph nodes (dLN) typically contain conventional CD11c^HIGHMHC-II^INT (intermediate) lymphoid-tissue resident DC (composed of CD8⁺ and CD11b⁺ subsets) and CD11c^INTMHC-II^HIGH peripheral tissue-derived migratory DC (composed of CD207⁺CD103⁺ dermal DC (dDC), CD11b⁺CD207⁻CD103⁻ dDC, and CD207⁺CD103⁻ Langerhans cells (LC)), as well as B220⁺ plasmacytoid DC (Heath and Carbone, 2009; Helft et al., 2010; Villadangos and Schnorrer, 2007).

Considering that DC subset markers are not exclusively expressed by one or another subpopulation or even DC in general, imaging analysis of subset specific localization differences has been challenging. Nevertheless, by analyzing fluorescently labeled cells after fluorophore/irritant skin painting of Langerin (CD207) reporter animals (murine-promoter Langerin-eGFP mouse), it has been shown that migratory CD207⁻ dDC and CD207⁺ DC (a combination of LC and CD103⁺ dDC), subsets with known functional differences, localize into highly discrete areas of the skin dLNs (Kissenpfennig et al., 2005; Klechevsky et al., 2008). The specific distribution of functionally distinct LC and CD103⁺ dDC is currently unknown (Igyarto et al., 2011; Kaplan, 2010). Most other reports on the visualization of lymphoid tissue-resident DC subsets have been mainly focused on the spleen, an organ with greatly reduced subset heterogeneity. Such splenic imaging has documented differential resident DC subset localization to discrete spatial micro-compartments (Dudziak et al., 2007; Idoyaga et al., 2009; Qiu et al., 2009). Importantly, location-specific differences in antigen (Ag) sampling, infectious spread, and even inflammatory cytokine production by these splenic DC subsets have been observed (Edelson et al., 2011; Qiu et al., 2009; Rothfuchs et al., 2009). Together, these studies suggest that lymphoid tissues may contain functionally specialized micro-compartments that are specifically geared towards generation of distinct immune responses.

These findings highlight the need for robust identification, visualization, and quantification of complex cellular populations directly in their microenvironment. Such methodology would in principle require staining with and visualization of multiple phenotype-specific antibodies, together with analysis of entire tissue cross-sections for cells possessing the relevant marker combinations most often defined by flow cytometry. Although most commercially available microscopes are technically able to discriminate numerous spectrally-distinct fluorophores and image large cross-sectional areas at high resolution through motorized image tiling (Conchello and Lichtman, 2005; Garini et al., 2006), there are virtually no reports examining cell subpopulations identified by expression of multiple markers directly in situ.

By combining multi-parameter, high-resolution 3D confocal imaging, robust spillover and deconvolution correction, accurate identification and 3D reconstruction of specific cells of interest (COI) in LN cross-sections, and graphical/positional plotting, we have been able to develop a novel analytical pipeline, "Histo-Cytometry," that allows visualization, quantification, and positional analysis of complex cellular populations and phenotypes directly in tissue sections from experimental animals and humans, using instruments and analytic tools available in most imaging core laboratories. Here, we describe the method, validating the approach by visualizing and identifying innate and adaptive immune cell subsets in LN sections and by tracking phenotypic changes occurring in Ag-specific CD4 and CD8 T cells during activation. As an example of the utility of this new method, we utilize Histo-Cytometry to conduct a thorough analysis of the spatial localization of highly heterogeneous conventional DC subsets in unperturbed steady-state LNs. We report anatomical segregation of distinct resident and migratory DC subsets, suggesting a highly regulated, fine-grained organization of LNs into functionally discrete spatial micro-domains whose existence has important implications for understanding how polarized immune responses develop and for designing vaccine strategies in which delivery of material to the right DC subset is crucial.

Results

Histo-Cytometry: Methodology

To gain positional and quantitative information on complex cellular subsets/phenotypes (defined by multiple markers) directly in tissue sections we have devised a general Histo-Cytometry workflow (Fig. 1). First, the cross-sections are stained with large panels of phenotypically appropriate antibodies, 6–8 colors at present, although in principle the method can be applied to more colors with appropriate adjustments to the protocol (Fig. 1[1]). Because of the densely packed nature of tissues, very high-resolution imaging and accurate signal 3D allocation is needed for optimal signal to cell assignment (Scriven et al., 2008). However, a practical balance must be achieved between the highest-possible quality image resolution, the overall imaging volume, and imaging duration. Limitations to super-resolution microscopy in this context make conventional diffraction-limited confocal microscopy the most practical choice (Hell, 2009). With this latter imaging technology, accurate signal allocation of multiple different fluorophores in all three axes (x,y,z) can be achieved using high numerical aperture objectives (Wright and Wright, 2002). In addition, deconvolution algorithms that mathematically reverse optical distortions can be applied after initial data acquisition to improve image quality (Conchello and Lichtman, 2005; Wallace et al., 2001). Therefore in our processing pipeline, the stained tissues are imaged at high optical resolution with conventional tiling confocal microscopes capable of discriminating multiple spectrally-distinct fluorophores (Fig. 1[Step 1]), with the optical distortions minimized by image deconvolution after data acquisition (Fig. 1[Step 2]).

Figure 1. Histo-Cytometry Workflow Pipeline.

Multi-parameter confocal images of tissue sections are taken with a tiling confocal microscope (1). Images are then deconvolved (2) and compensated for fluorophore spillover (3). Voxels exhibiting specified combinations of signals in the original channels (above/below designated thresholds) are used to create a new masking channel (4), which is then used to gate/mask all other parameters of interest (5). 3D COI surfaces are constructed based on the gated signal expressed by the COI through use of semi-automatic volumetric rendering and segmentation (6). COI surface statistics are exported for quantitative analysis and phenotypic gating (7), and the identified gate thresholds are used for quantitative visualization (8). Bars represent 20 µm, unless otherwise stated.

As in flow cytometry, highly multiplex staining using fluorophores with overlapping emission spectra can generate substantial spillover artifacts and thus requires robust compensation correction. While some microscopy systems can collect hyper-spectral data and remove fluorophore cross-talk through spectral unmixing, it is more practical to collect single stain control images and calculate the relative spillover of each fluorophore into the other detectors. We mathematically apply simple compensation correction algorithms to every imaged voxel after data acquisition in a manner akin to flow cytometry (Roederer, 2002), which results in robust removal of the spillover signal (Fig. 1[Step 3], S1A).

Due to the lack of molecular level resolution, diffraction-limited confocal imaging does not spatially separate neighboring fluorescent molecules, instead colocalizing them to the same voxel (volumetric pixel). This makes correct signal to cell allocation potentially problematic with densely packed cells sharing surface molecule expression (Scriven et al., 2008). However, one benefit of such spatial photon colocalization is that it allows software-based identification and computational isolation of voxels displaying specific combinations of markers. Thus, by selecting (“gating”) voxels positive for fluorescent signals from COI-specific antibodies and negative for irrelevant or inappropriate cell markers, we can identify discrete image regions corresponding to cells of interest (COI) and not to contaminating noise or irrelevant cells (Fig. 1[Step 4]), a process highly similar to flow cytometric Boolean gating and the use of “dump” channels for exclusion of irrelevant cell populations (Roederer, 2002). This is accomplished by creation of a new COI-specific binary channel based on the selected fluorescence thresholds identified by visual determination of positive/negative signal cutoffs and with the assistance of automated thresholding algorithms (Fig. S1B,C). The COI binary channel is then used to gate/mask all other relevant parameters, and this permits qualitative visualization of distinct cellular subsets/phenotypes, as the visual image complexity is reduced only to the remaining COIs (Fig. 1[Step 5], S1D). Finally, to achieve flow cytometry–like cellular phenotypic profiling and quantification based on imaging data, 3D volumetric surface objects corresponding to individual cells are created (Fig. 1[Step 6]). Here, we utilize the semi-automatic surface rendering and watershed-based segmentation algorithms currently implemented in the Imaris imaging software, although similar tools available in other imaging platforms can be used for this purpose (Meijering et al., 2009; Wahlby et al., 2004).

It is important to note that current segmentation algorithms can suffer from poor separation of very dense, phenotypically homogeneous cellular clusters into individual cell objects, especially when visualizing surface markers. Advanced image segmentation algorithms are under investigation and will undoubtedly improve 3D object discrimination in the future (Indhumathi et al., 2011; Meijering et al., 2009). Here we describe one such approach specifically applicable to major immune populations of interest, small resting lymphocytes. For many important questions, however, it is sufficient to analyze tissue sections for relatively non-clustered cells. In turn, these COI-based objects have various quantitative information associated with them, including (X,Y,Z) position and average voxel fluorescence intensities for the imaged parameters. Through graphical plotting of these parameters, one can achieve quantitative phenotypic subset analysis (akin to flow cytometry) with the added bonus of subset-specific positional characterization within the tissue (Fig. 1[Steps 7,8]).

Basic Immune Subset Discrimination

We first tested the ability of this Histo-Cytometry method to accurately identify and quantify major immune subsets in sections from cell-subpopulation rich LN. For this purpose, we created bone marrow (BM) chimeric animals constructed by transfer of 1:99, 5:95, or 10:90 mixtures of CD45.2⁺ BM and CD45.1⁺ donor BM into irradiated CD45.1⁺ recipients, providing a useful test case in which the difficulty of identifying cells of the same phenotype in close apposition to one another was largely avoided. Six weeks after reconstitution, we prepared sections from LN of these animals and applied Histo-Cytometry in an attempt to localize and quantify different CD45.2⁺ cell subsets (Fig. 2A). In this scenario, the CD45.2⁺ COI would be represented in all hematopoietic cellular lineages, with distinct populations reconstituting with slightly different frequencies in each chimeric animal. These cells would not be directly adjacent to one another in very high densities as in a normal LN, but would still be surrounded by numerous irrelevant CD45.1⁺ cells, and would be located in their presumably normal micro-anatomical compartments. The quantitative data obtained from microscopic imaging were compared with results from more conventional flow cytometric analysis of cells from dissociated contralateral LNs of the same animals (Fig. 2A). We stained the LN sections with a panel of seven spectrally compatible antibodies that would allow discrimination of the CD45.2⁺ specific populations from the 90–99% of surrounding neighboring CD45.1⁺ cells, as well as permit identification of specific leukocyte subsets. We then imaged whole LN sections as well as single stained controls via confocal tiling microscopy, deconvolved the data, and compensated for fluorophore spillover (Fig. 2B and data not shown).

Figure 2. Basic Immune Subset Discrimination.

Irradiated CD45.1⁺ recipients were injected with a 1:99, 5:95, or 10:90 mixture of CD45.2⁺ to CD45.1⁺ donor BM and were allowed to reconstitute for 6 weeks, after which contra-lateral inguinal LN were taken for comparative analysis by flow cytometry and Histo-Cytometry (A). LN sections were stained with a panel of indicated antibodies and imaged (B). CD45.2⁺CD45.1⁻ voxels were used to create a masking channel that was utilized to further gate/mask all other imaged parameters, with the gated CD45.2 signal used for creating 3D surface renderings of CD45.2⁺ cells (C). Surface statistics were exported and plotted for identification of distinct immune subsets and compared to results obtained by flow cytometry (D). X and Y positions of CD45.2⁺ surfaces were plotted for each subset (D). Relative frequencies of identified subsets in the 1–10% chimeric animals, with (left) or without (right) inclusion of CD11c⁺ and unclassified events, were determined and compared to flow cytometry-based quantification (E). (Representative of three independent imaging quantifications)

To minimize signal contamination from closely apposed irrelevant cells, we utilized a CD45.1 "dumping" technique, similar to that used in flow cytometry (Perfetto et al., 2004), and selectively gated on CD45.2⁺CD45.1⁻ voxels to create a new binary colocalization channel specifically representing only the CD45.2⁺ COI (Fig. 2C). This channel was then used to further mask/gate other parameters of interest, effectively removing unwanted signals from neighboring cells and allowing for clear visualization of individual CD45.2⁺ cells belonging to one or another lymphoid lineage (Fig. 2C). We next volumetrically rendered/segmented the gated CD45.2 signal, thus creating 3D surface objects representing the CD45.2⁺ COI (Fig. 2C). These surfaces provided quantitative statistics, which were exported to Excel for further analysis. We analyzed the mean voxel intensities for the different gated parameters inside each surface instead of the total sum of voxel intensities, as the intensity sum was highly dependent on the exact object volume. Plotting of these data as 2D histograms revealed the capacity of the method to provide cellular subset discrimination of CD11c⁺ cells, B cells (B220⁺CD3⁻CD11c⁻), and CD4⁺ and CD8⁺ T cells (CD3⁺B220⁻CD11c⁻) with patterns that were highly similar to the data derived from conventional flow cytometric analysis of dissociated cells (Fig. 2D and 2C). Moreover, the X,Y positional information corresponding to the volume-rendered individual cell surfaces allowed us to directly visualize the spatial localization of gated subsets within the LN. We observed clear-cut localization of computed B cell surfaces to structures corresponding to the B cell follicles, of CD4⁺ and CD8⁺ T cell surfaces to the T cell zone, and of CD11c⁺ surfaces distributed in the paracortical T cell zone, and with somewhat higher densities, in the interfollicular and lymphatic regions (Fig. 2D). The cell surface subset spatial positioning was thus clearly reflective of the normal biological localization typically observed for these cell types using more conventional one and two color immuno-histochemical imaging (Fig. 2B).

Quantitative subset discrimination also allowed ratiometric comparisons to the results generated by flow cytometric analysis of cells from the contra-lateral LN. Comparison of the relative proportions of B cells, CD4⁺ T cells, and CD8⁺ T cells in chimeras with different CD45.2/45.1 frequencies demonstrated a near 1:1 correspondence between flow cytometry and Histo-Cytometry quantification methods, suggesting that Histo-Cytometry can be used for highly quantitative analysis of cell subset composition and distribution in tissue sections at least for cells of this simple morphology (Fig. 2E). However, there were also clear differences between these analyses. Histo-Cytometry yielded markedly higher percentages of CD11c⁺ cells relative to flow cytometry. As CD11c⁺ DC are cells of very high morphological complexity, the use of semi-automatic segmentation algorithms resulted in cellular over-segmentation of 6–17% of CD11c⁺ events, depending on the analyzed image (data not shown), thus yielding multiple apparent cells from what is in reality a single cell. We thus manually curated the cellular surfaces to exclude over-segmentation artifacts. Even with this adjustment, the frequency of CD11c⁺ cells identified via Histo-Cytometry greatly exceeded that identified via flow cytometry (Figure 2E). As DC are known to tightly bind stromal elements of the LN and are difficult to extract with standard tissue disruption methods (Vremec et al., 1992), these cells are likely to be heavily underrepresented in flow cytometric analysis of single cell suspensions. For this reason, Histo-Cytometry may therefore be providing a more accurate and complete picture of DC frequency and location in the absence of extraction artifacts. A number of surfaces could not be clearly allocated to a distinct lineage, potentially due to decreased fluorescence separation in confocal microscopy-based analyses as compared to flow cytometry or to noise contamination derived from spatially proximal cells (Fig. 2E); these results reflect the lower signal to noise characteristics of cell imaging in thin tissue sections vs. total fluorescence capture from individual cells in a flow stream.

Besides analysis of spatially separated CD45.2⁺ cellular events, complete quantitative analysis of all constituent cells within an imaged tissue section is also highly desirable. Membrane staining alone does not currently allow for direct segmentation of tightly juxtaposed cells with shared surface phenotype. However, cellular nuclei are relatively well separated spatially, and we found that it was possible to segment most small lymphocytes based on nuclear staining despite close packing (Figure S2A). Resting T and B lymphocytes have relatively small cytoplasmic volumes and the majority of the cellular membranes lie in close nuclear proximity, providing sufficient signal overlap between membrane staining and the nuclear dye signal to permit individual cell identification, subset gating, and quantification (Figure S2B). By excluding poorly segmented events through volumetric and nuclear MFI gating, and looking at CD45.1⁺ and CD45.2⁺ events, we were able to robustly identify all B cell and CD4/8 T cell sub-populations within the imaged LNs despite dense cell packing. The cellular frequencies closely matched the membrane-based Histo-Cytometry data for CD45.2⁺ events and flow cytometry data for both CD45.1⁺ and CD45.2⁺ populations (Fig. S2C,D). This specialized segmentation method relies on close membrane/nucleus juxtaposition, making it suitable strictly for quantitative analysis of resting lymphocytes and not for blasting or morphological complex cell populations.

Phenotypic Profiling of T cell Activation

We then examined Histo-Cytometry for its capacity to detect well-characterized stimulus-associated phenotypic changes in defined cellular subsets. Upon sensing cognate antigens during infection or immunization, specific CD8⁺ and CD4⁺ T cells undergo rapid activation that is associated with changes in expression of distinct surface markers and cellular proliferation that is associated with changes in cell cycle proteins (Smith-Garvin et al., 2009). Using the Histo-Cytometry pipeline described above, we attempted to visualize these phenotypic changes in responding Ag-specific T cells directly in tissue sections and to determine the spatial distribution of T cell activation with respect to the location of cognate Ag within the LN. For this purpose, we adoptively transferred CD45.2⁺ OVA-specific T cell receptor (TCR) transgenic CD8⁺ (OT-I) and CD4⁺ (OT-II) T cells, along with irrelevant TCR transgenic CD4⁺ T cells (SMARTA), into CD45.1⁺ congenic recipient mice (Fig. 3A). To separate the transferred CD45.2⁺ cells into the correct transgenic populations, we utilized genetically labeled OT-I GFP cells and ex vivo Cell Tracker Blue (CTB) labeled SMARTA cells. One day later, mice were subcutaneously immunized in two contralateral flanks with CpG adjuvant and OVA-conjugated fluorescent microspheres, to allow for precise tracking of Ag localization in relation to cellular activation. Contralateral dLN were then harvested at different times post immunization for comparative Histo-Cytometry and flow cytometry analyses (Fig. 3A).

Figure 3. Phenotypic Profiling of T cell Activation.

1.5×10⁶ CD45.2⁺CD44^low OT-II, OT-I.GFP and CTB-labeled SMARTA were adoptively transferred into CD45.1⁺ recipients, which were immunized s.c. one day later with OVA-conjugated beads and CpG. Contra-lateral dLN were taken for comparative analysis by flow cytometry and Histo-Cytometry at indicated time-points (A). A CD45.2⁺CD45.1⁻ masking channel was used to gate all other parameters and the gated CD45.2 signal was used to create cell surfaces. CD45.2⁺ surface statistics were plotted for identification of OT-I, OT-II, and SMARTA populations, and were visually validated and compared to flow cytometry (B). CD45.2⁺ surfaces were phenotypically subsetted by fluorescence level s of gated CD69 and Ki-67, visually validated and compared to flow cytometry (C). Percentage of CD69⁺ and Ki-67⁺ T cell populations, as well as OT-I/II fold expansion determined by normalization to non-dividing SMARTA cells, were quantified (D). Minimum distances of OT-I/II surfaces to OVA-beads at the 38hr time-point were calculated and compared for the CD69^+/− populations (left) and plotted as a 2D contour graph for OT-II (E). N=3 for each time-point.[Ron, please indicate the meaning of the error bars in panel D]

We utilized the CD45.2⁺CD45.1⁻ voxels to selectively gate/mask all of the parameters of interest to focus our analysis specifically on the transferred T cells. Gated CD45.2 signal was then used to generate surfaces for contour plot analysis and directly compared to flow cytometric datasets (Fig. 3B, 3C). Both methods allowed clear discrimination and gating of CD4⁻GFP⁺ (CD8⁺) events representing OT-I CD8⁺ T cells, of CD4⁺GFP⁻CTB⁻ OT-II CD4⁺ T cells, and of CD4⁺GFP⁻CTB⁺ SMARTA CD4⁺ T cells (Fig. 3B). Because the SMARTA CD4⁺ T cells were not specific for any antigen present in this experimental set-up, they did not proliferate and did not lose CTB fluorescence over time (Fig. S3A and data not shown). Later in the immune response, some of the Ag-specific T cells formed dense cellular aggregates directly around areas with Ag-coated beads (Fig. S3B). Certain very tight clusters proved difficult to accurately segment into individual objects as noted above and thus were excluded from the quantitative analysis by volumetric filtering/gating.

To track the phenotypic changes associated with T cell activation and proliferation, we stained for CD69, an early marker of T cell activation, and Ki-67, which is specifically expressed in the G1-M phases of the cell cycle (Fig. 3C, S3B). Both flow and Histo-Cytometry analyses clearly identified expression of CD69, but not Ki-67, early after immunization uniquely on Ag-specific T cell populations (Fig. 3C, S3C). Three days after immunization, some of the Ag-specific OT-I and OT-II T cells remained CD69⁺ but a large fraction now showed Ki-67 expression, suggesting that the cells were actively cycling (Fig. 3C, 3D, S3C). Quantitative comparisons of flow and Histo-Cytometry datasets revealed similar kinetic trends of OT-I and OT-II cellular expansion, as well as of CD69 and Ki-67 expression with respect to the control non-dividing SMARTA cells (Fig. 3D). Nonetheless, there were some quantitative differences observed using the two methods. By quantifying the increase of Ag-specific T cells in relation to the predominantly stable Smarta cells (Fig. S3A), we found a somewhat reduced expansion of Ag-specific T cells as measured by flow cytometry (Fig. 3D). This potentially reflects poor tissue extraction of highly adhesive, recently activated T cells even with collagenase digestion (Jabbari et al., 2006). Consistent with these findings, we also observed a higher frequency of Ki-67⁺ OT-I CD8⁺ T cells by Histo-Cytometry (Fig. 3C). Finally, there was a reduced frequency of CD69⁺ cells observed by Histo-Cytometry, likely due to the lower signal detection and separation for this parameter by microscopic imaging as compared to flow cytometry (Fig. 3C, 3D).

To ascertain the spatial distribution of activated T cells in relation to their cognate Ag, we obtained the positional coordinates for T cells and OVA-coated fluorescent microspheres. The latter were clearly visible in dLN at 14hr and substantially increased in number at later time points (Fig S3B, not shown). We assumed these Ag-bearing particles were localized in relevant antigen-presenting cell populations within the LN and therefore calculated the closest distances between T cells and microspheres. This analysis revealed a statistically significant relationship between CD69⁺ expression and distance from Ag, with the majority of CD69⁺ cells located within 100µm of the Ag at 38hr (Fig. 3E). This relationship was somewhat underestimated due to cluster (CD69⁺ cells) exclusion and also became less apparent at 72hr time points, as the number of Ag⁺ cells continued to increase (not shown) and the responding T cells regained high levels of motility (Henrickson and von Andrian, 2007). These data demonstrate that Histo-Cytometry can quantitatively detect cellular phenotypic changes and the spatial positioning of specific activated cell subsets in relationship to distinct points of interest in complex tissues.

DC Subset Visualization

Having tested our new method first by identifying basic immune subsets and then cellular phenotypic changes in response to immunization, we applied Histo-Cytometry to examine in detail the spatial positioning of the various DC subsets in skin dLN. To this end, we first co-stained inguinal LN sections from C57BL/6 animals with a panel of antibodies against CD11c, MHC-II, CD3, B220, CD8, and CD11b, together with antibody to Lyve-1 to mark the lymphatic endothelium (Fig. 4A). Next, we restricted the image analysis to basic DC subsets by specifically gating CD8 and CD11b signals within CD11c⁺MHC-II⁺CD3⁻B220⁻ voxels (Fig. 4B). The gated CD8 and CD11b signals largely localized to distinct cells (resident CD8⁺, resident or migratory CD11b⁺ DC) that were dispersed throughout the LN paracortex and medullary/lymphatic regions with a sparse presence in the B cell follicles, as expected. Interestingly, the CD8⁺ DC displayed a more centralized localization in the T cell zone, as compared to more peripherally distributed CD11b⁺ DC (Fig. 4B, 4C). We were concerned that even though B220⁺CD3⁺ cells were excluded from image analysis, the high density of CD8⁺ T cells in the paracortical T cell zone (Fig. 4A) could have introduced elevated level s of non-specific CD8 signal into the DC-gated CD8 channel, thus potentially skewing data interpretation. We therefore imaged and analyzed DC subset distribution in LN from BATF3.HET and BATF3.KO animals (Fig. 5A). As BATF3.KO mice are largely devoid of CD8⁺ resident DC but contain normal numbers of CD8⁺ T cells, any artifacts from mis-assignment of CD8 signals from T cells to other cell volumes would be evident (Edelson et al., 2010; Hildner et al., 2008). We first restricted image analysis to conventional DC by specifically gating all parameters of interest within CD11c⁺MHC-II⁺CD3⁻B220⁻ voxels. Similar to C57BL/6 mice, BATF3.HET LN displayed an abundance of DC throughout the LN paracortex and not in the B cell follicles, as ascertained by visualization of the gated CD11c signal (Fig. S4B). We observed a clear enrichment of CD8⁺ DC in the LN center and a denser distribution of CD11b+ DC in the outer cortex and lymphatic/medullary regions (Fig. 5B). In contrast, BATF3.KO animals displayed little to no CD11c⁺MHC-II⁺CD3⁻B220⁻ gated CD8 signal, while maintaining a distribution of the gated CD11b signal similar to BATF3.HET LN, thus validating the observed CD8⁺ DC distribution in wild-type animal LN (Fig. 5B).

Figure 4. DC Subset Visualization.

Inguinal LN sections from C57BL/6 mice were stained with the indicated antibodies and imaged (A). CD11c⁺MHC-II⁺CD3⁻B220⁻ voxel gating allows visualization of CD11b⁺ and CD8⁺ DC (B). Original (non-gated) CD8, CD11c, MHC-II, and Lyve-1 signals provided a clear separation of the LN into discrete zones (C, left): lymphatic/medullary, inter-follicular (IFZ), T zone, and B zone, which were overlaid onto the DC-gated CD8 and CD11b image (C, right). Representative of at least three independent imaging sessions.

Figure 5. Resident DC Subset Visualization and Quantification.

Inguinal BATF3.HET and BATF3.KO LN sections were stained with the indicated antibodies and imaged (A, BATF3.HET presented). CD11c⁺MHC-II⁺CD3⁻B220⁻ voxel gated CD11b and CD8 signals are displayed for BATF3.HET and BATF3.KO LN sections (B). DC surface statistics were used for identification and spatial visualization of resident CD8⁺ and CD11b⁺ DC subsets, with confirmation by flow cytometry for CD3⁻NK1.1⁻CD19⁻CD11c⁺MHC⁺ gated DC subsets (C). Frequencies of CD11c^HIGHMHC-II^INT resident CD8⁺ and CD11b⁺ DC were quantified and compared to results obtained by flow cytometry (D). Percentage of cells localized to the indicated zones within the indicated DC subsets was quantified (E). Representative of at least two independent experiments, N = 3.

We then created DC volumetric surfaces, based on the gated CD11c signal, and analyzed DC populations via contour plot analysis and direct comparison to flow cytometry data. Both analyses allowed discrimination of CD11c^HIGHMHC-II^INT resident and CD11c^INTMHC-II^HIGH migratory DC populations (Fig. 5C), albeit the resolution of subpopulations by flow cytometry was noticeably clearer than that achieved with Histo-Cytometry for the reasons noted above. The threshold cutoffs for the gating were validated by examining CD11c/MHC-II expression level s on migratory CD207⁺ DC (Fig. S4A). Resident DC analysis revealed distinct CD11b⁺ and CD8⁺ subset separation by both analytical methods, with the CD8⁺ sub-population being clearly absent in the BATF3.KO animals (Fig. 5C). Moreover, we observed highly comparable quantification of resident DC subset composition via flow and Histo-Cytometry, with BATF3.KO animals having reduced percentages of CD8⁺ DC and increased percentages of CD11b⁺ DC (Fig. 5D). Importantly, Histo-Cytometry allowed us to quantitatively examine subset-specific localization differences (Fig. 5C,E). By identifying specific zones in the LNs (Fig. S4B), we observed predominant localization of resident CD8⁺ DC to the T cell zone, albeit some cells with this phenotype were also present in other regions, and the principal localization of resident CD11b⁺ DC to the lymphatic/medullary regions with a relatively minor presence of these DC elsewhere (Fig. 5E, S4B). These differences were reproduced in normal B6 mice (not shown), in CD11c-YFP reporter mice for inguinal, popliteal, and auricular LNs (Fig. S4C–E), and by utilizing the CD205 discriminatory marker (instead of CD8) for identifying CD8⁺ resident DC (Fig. S4F). Minor morphological over-segmentation artifacts did not affect the regionalized distribution of DC subsets derived from this analysis (data not shown).

Interestingly, migratory DC showed a mixed distribution between the interfollicular and T cell zones, and were largely absent from the medullary/lymphatic zones (Fig. 5F), which is consistent with the recent observations of localized DC immigration into the LN parenchyma through the interfollicular regions (Braun et al., 2011; Schumann et al., 2010). To further discern whether specific localization differences exist for different migratory DC subsets, we first attempted to co-stain LN sections with CD11b, CD207, and CD103. Unfortunately, in our hands neither the 2E7 or M290 CD103 antibody clones nor a polyclonal antibody to this marker yielded reproducible and specific staining consistent with the CD207 signal distribution (data not shown). Instead, we turned to a recently described langerin (CD207) YFP reporter model that is driven by the human and not the mouse langerin promoter, in which only the CD207⁺ Langerhans cells (LC) and not the CD103⁺CD207⁺ dermal DC (dDC) express the YFP protein (Hattori et al., 2011; Kaplan, 2010; Kaplan et al., 2007). We reasoned that additional staining of LN sections from these animals with the CD207 antibody would reveal two populations of cells, CD207⁺YFP⁺ LC and CD207⁺YFP⁻ CD103⁺ dDC. Consistent with this prediction, analysis of CD11c⁺MHC-II⁺ gated CD11b, CD207, and YFP channels revealed a highly variegated distribution of CD11b signal, with dense staining in the lymphatic/medullary zone (largely resident CD11b⁺ DC) and in the interfollicular zone (presumably migratory CD11b⁺ dDC), and distinct localization of CD207⁺YFP⁺ and CD207⁺YFP⁻ signals mainly deep in the T cell zone (Fig. 6A). Volumetric DC rendering and Histo-Cytometry analysis of CD11c^INTMHC-II^HIGH migratory DC revealed clear-cut separation of CD11b⁺CD207⁻ dDC, CD207⁺YFP⁺ LC, and CD207⁺YFP⁻ dDC populations (Fig. 6B). As expected, CD207 was absent from the resident CD11b^+/− DC populations. Flow cytometric evaluation of these populations (with EpCAM used instead of CD207 for the analysis) revealed a highly similar migratory DC subset composition and quantitative result (Fig. 6B, 6C). Importantly, Histo-Cytometry allowed us to analyze subset-specific localization differences. Consistent with the data presented above, the resident CD11b⁻ and CD11b⁺ DC differentially distributed between the T cell zone and the lymphatic/medullary regions, respectively, as analyzed by the distance to LN lobe center (Fig. 6B, 6D). Interestingly, the migratory DC subsets also were differentially distributed, with the CD11b⁺ dDC located far away from the LN center, in or near the interfollicular zones (peripheral paracortex). On the other hand, we did not observe major localization differences between the CD103⁺ dDC and the LC, with the large majority of cells located deep inside the LN paracortex (Fig. 6B, 6D). We consistently found these spatial distribution differences between the CD11b⁺ dDC and CD207⁺ migratory DC populations in non-transgenic animals, in CD11c-YFP reporter mice, and after chemical-induced skin irritation, although the regionalized localization was somewhat less pronounced in this case (Fig. S5). Together these data demonstrate a previously unappreciated diverse micro-anatomical positioning of distinct resident and migratory DC subsets in steady-state LN (Fig. 7).

Figure 6. Migratory DC Subset Visualization.

Inguinal LN sections from (human-promoter) Langerin-Cre x YFP-flox reporter animals were stained with the indicated antibodies and imaged (A). Resident and migratory DC subsets were identified and spatially visualized by Histo-Cytometry (LN outline added for clarity), with confirmation by flow cytometric evaluation of CD3⁻NK1.1⁻CD19⁻CD11c⁺MHC⁺ gated DC subsets (B). Migratory DC subset compositions derived by both analytical methods were quantitatively compared (C). Distance to the LN lobe center for individual cells of distinct DC subsets were compared (D, left), and the average distance for the different populations in three individual LNs was quantified (D, right). N = 3.

Figure 7. Micro-Anatomical Separation of Distinct DC Subsets.

Immunofluorescence microscopy-based (left) and cartoon (right) models of the distribution of different resident and migratory DC subsets among discrete LN micro-compartments are presented. Staining for the stromal marker ER-TR7 allows for direct visualization of LN compartmentalization into discrete micro-domains, with clear T cell zone demarcation (indicated by the CD8 stain) from the interfollicular and lymphatic/medullary regions (left). The microscopy image was mirrored and the afferent lymphatic vessels added for enhanced model clarity.

Discussion

The capacity to determine both individual cellular identity and precise tissue distribution of multiple populations within complex tissues and organs is a long-standing goal in all fields of cellular biology. However, it can be highly problematic to clearly identify and visualize complex cellular populations defined by several non-exclusive markers in highly heterogeneous and densely packed biological tissues. Histo-Cytometry permits analytical processing of tissue section images for direct visualization and quantitative gating of cellular subsets and phenotypes defined by multi-parametric stained marker combinations. We validated our technique in large cross-sections of mouse LNs by identifying and analyzing the distribution of various leukocytes. We were able to spatially separate basic immune populations, including CD4⁺ and CD8⁺ T cells, B cells and DC, visualize the early stages of Ag-specific T cell activation and proliferation with respect to the localization of cognate Ag, and decipher the intranodal spatial anatomy of all major conventional DC subsets. For readily isolated cells, Histo-Cytometry achieved cell phenotyping and enumeration results quantitatively similar to flow cytometry, the “gold standard” for cell subset/phenotype identification and enumeration, while surpassing flow results with respect to cells that are difficult to extract quantitatively from tissue. Thus this new technology can be used in situations requiring quantitative cellular analysis of tissue sections, while also adding spatial information not available using flow methods.

Using this technique, we extend previous studies that suggested a non-random distribution of some DC subsets within skin dLN. To our knowledge this is the first extensive visualization and image-based quantification of multiple DC subsets obtained using the classical marker definitions established by flow cytometry. Our data reveal a clear subset regionalization to discrete anatomical sub-compartments in murine LNs (Fig 7). It is important to note that each of the delimited LN regions was not occupied exclusively by a specific DC subset and any given subset did not reside exclusively in one region; rather, the indicated subsets showed heavily skewed distributions with predominance in the indicated locations. Most previous studies addressing this issue have described the localized distribution of a limited number of distinct resident DC in the spleen, an analysis that has been challenging to achieve in LNs, where multiple migratory and resident subsets with overlapping markers are present (Dudziak et al., 2007; Qiu et al., 2009). We found that the resident CD8⁺ and CD11b⁺ DC subsets largely segregate to the T cell zone or the lymphatic/medullary zone, respectively. This is highly reminiscent of differential localization of splenic CD205⁺ (CD8⁺) and 33D1⁺ (CD11b⁺) resident DC to the T cell zone and the red pulp (respectively), the spatial separation of CD205 and CD11b signal observed in the LN, and the localization of Neclin-like protein 2⁺ cells (CD8⁺ DC) to the T cell zones in human and murine spleens (Dudziak et al., 2007; Galibert et al., 2005; Sixt et al., 2005). We further found that that the migratory DC subsets also segregate into distinct anatomical zones, with the CD11b⁺ dDC largely inhabiting interfollicular and outer paracortical regions, and with the LC and CD103⁺ dDC accumulating deeper in the T cell zones. These latter findings are similar to what has been previously reported in a skin irritation and fluorophore painting model (Kissenpfennig et al., 2005), with the addition that we were able to achieve specific localization of all currently known migratory DC subsets, including the resolution of the CD207⁺ CD103⁺ dDC from the LC. In addition, Histo-Cytometry permitted observation of migratory DC localization in both steady state and inflammatory LNs.

The observed differences in micro-anatomical distribution suggest that different DC subsets preferentially respond to spatially distinct cues, which may include ligands for chemokine receptors, integrins, and/or lectins, that guide them to specialized locations within lymphoid tissues. It is likely that stromal cells contribute to this regionalization of DC subsets by providing population–specific cues. Differential stromal cell compositions and structures in LN have been previously described (Katakai et al., 2004; Mueller and Germain, 2009), with the stromal cell-enriched cortical ridge likely playing a critical role for micro-anatomical compartmentalization (Fig. 7). Overall, these results point to a highly complex and regulated LN structural organization and suggest that the segregation of innate sensory DC subsets within lymphoid tissues can create sub-compartments with distinct properties with respect to Ag access/MHC presentation, TLR-induced maturation, and inflammatory cytokine production, thus together leading to regional specialization in T cell activation and effector differentiation. In accord with the findings reported here, preliminary studies in our laboratory have revealed differential Ag uptake by DC subsets that match predictions based on these newly described localizations. These studies also show specialized zones of initial CD8⁺/4⁺ T cell migration arrest and activation that both fit with these location-associated antigen acquisition data and previously described DC subset specialization in MHCI/II processing machinery (manuscripts in preparation)( Heath and Carbone, 2009; Shortman and Heath, 2010).

There are currently several Histo-Cytometry pipeline issues that demand special attention. To obtain highly accurate Histo-Cytometry data, it is important to utilize high magnification and high NA objectives, for improved lateral/axial resolution, and accurate cellular segmentation. It may be possible to further decrease out of plane noise contamination through background subtraction algorithms (Zinchuk et al., 2011). Second, due to lack of signal separation between apposed cells in highly aggregated clusters, the current 3D segmentation algorithms do not allow for precise separation of phenotypically identical cellular objects based solely on membrane staining. As a first step in overcoming this limitation, we describe here an imaging routine for quantitatively analyzing densely packed resting T and B lymphocytes. By staining for and segmenting based on nuclei, we were able to achieve accurate Histo-Cytometric quantitative profiling of the majority of lymphocytes in entire tissue sections from resting LN.

Several other platforms have been specifically designed for identification and phenotype analysis of cells directly in tissue sections (Ecker and Steiner, 2004; Grierson et al., 2005; Harnett, 2007; McGrath et al., 2011). However, these systems typically require specialized microscope systems and software, and do not allow for spillover correction, deconvolution, and voxel restriction. Some of these existing approaches choose to limit axial resolution in an attempt to maximize signal collection, and are restricted to 2D cellular analysis, thus not permitting accurate volumetric image interrogation of complex tissues and multiple diverse cell types. An alternative approach to simultaneous multiplexed antibody labeling, called multi-epitope-ligand cartography (MELC), has been demonstrated to allow robust sampling of hundreds of different antibody probes labeled with the same fluorophore by utilizing sequential staining, imaging, and fluorophore bleaching (Schubert et al., 2006). While being a powerful method for “toponome” mapping, MELC is limited by the need for special robotic instrumentation, very long imaging runs, and binary signal compression (Schubert et al., 2006). Histo-Cytometry, on the other hand, is an analytical pipeline that utilizes widely available microscope equipment and image analysis platforms along with commonly available fluorophores and combinations of antibodies already in use for conventional flow cytometry, and thus can be rapidly integrated into current laboratory workflow.

Both flow cytometry and Histo-Cytometry have specific limitations and advantages. Flow limits analysis to single cell suspensions of readily extractable cells, but allows for rapid throughput and higher signal separation. On the other hand, Histo-Cytometry is currently limited by somewhat lower signal detection, potential noise contamination, and can be problematic for separating tightly aggregated cellular clusters of identical cells of irregular shape, but provides the ability to examine and quantify cellular subsets/phenotypes directly within tissues, thus adding an unprecedented layer of positional information to phenotypic analyses. These techniques are thus best applied as complimentary approaches, together serving as both investigation and validation methods. Importantly, Histo-Cytometry can be directly applied to highly precious non-human primate and human samples that cannot be subjected to transgenic reporter-based or large-scale flow cytometric analyses, and we have recently applied elements of Histo-Cytometry to the in situ detection and spatial characterization of T follicular helper cells during SIV infection in spleen and LN tissues of rhesus macaques (Petrovas et. Al., JCI in press). It is also adaptable to the detection of such additional types of information as the cellular content and subcellular localization of specific phospho-proteins, cytokines and/or the relationship of extracellular matrix components to distinct cellular subsets. We thus believe that Histo-Cytometry offers a novel capacity to investigate the relationships between cellular subsets and their anatomical distribution, location-based function, differentiation / activation state, and intercellular interactions across different biological systems of interest and in different species.

Experimental Procedures

Mice

CD45.1⁺ C57BL/6, CD11c-YFP (Lindquist et al., 2004) and C57BL/6 OT-II TCR transgenic mice (Barnden et al., 1998) were obtained from Taconic Laboratories. OT-I CD8⁺ T cell transgenic mice (Hogquist et al., 1994) were obtained from Taconic and crossed to C57BL/6 ubiquitin-GFP mice (Schaefer et al., 2001) (Jackson Laboratories). SMARTA CD4⁺ T cell transgenic animals (Oxenius et al., 1998) were provided by Ethan Shevach (National Institutes of Health, Bethesda, MD). BATF3.KO mice on the 129 background (Hildner et al., 2008) were provided by Kenneth Murphy (University of Washington, St. Louis, Missouri, US) and backcrossed to C57BL/6 for 2 generations. BATF3.HET littermates were used as controls. LN from (human-promoter) Langerin-Cre x YFP-flox mice were provided by Daniel H. Kaplan (Hattori et al., 2011; Kaplan et al., 2007). All mice were maintained in specific-pathogen-free conditions at an Association for Assessment and Accreditation of Laboratory Animal Care-accredited animal facility at the NIAID. All procedures were approved by the NIAID Animal Care and Use Committee (National Institutes of Health, Bethesda, MD).

Microscopy and Histo-Cytometry

Detailed methods for sample preparation, imaging, and data processing are available as supplementary Materials and Methods. In brief, fixed LN sections were stained with panels of excitation/emission wavelength compatible fluorophore-conjugated antibodies and imaged with tiling confocal microscopy. Images were then processed for analysis with the Histo-Cytometry pipe-line, as detailed in the supplement.

BM Chimeras, Adoptive Transfers, and Immunization

CD45.1⁺CD45.2⁻ animals were exposed twice to 600 rad of gamma irradiation from a cesium source separated by a 3hr rest period. 5×10⁶ total donor BM cells, spiked with 1, 5, or 10% of CD45.2+CD45.1-cells, were injected intraevlously (i.v.) the same day. Mice were rested and given Neomycin or Sulfatrim water afterwards for 4 weeks. Tissues from chimeric animals were harvested for flow cytometry and Histo-Cytometry analyses six weeks after irradiation.

Naïve (CD44^low) OT-I, OT-II and SMARTA T cells were purified from LN and spleen tissues by Miltenyi negative selection T cell isolation kits (Miltenyi Biotec) in combination with 1:10,000 CD44-bio antibody. SMARTA T cells were labeled with CMF2HC (4-chloromethyl-6,8-difluoro-7-hydroxycoumarin) CTB at 100µM concentration and 1:10 pluronic polyol:dye ratio (Invitrogen) for 15 minutes at 37 Celsius and washed with 10% fetal calf serum RPMI media. 1.5×10⁶ CD45.2⁺ OT-I.GFP, OT-II and SMARTA.CTB were adoptively transferred i.v. into CD45.1+ recipient mice. One day later, animals were immunized with 25 µg of CpG 1668 and ~10⁶–10⁷ blue-green fluorescent FluoSpheres polystyrene microspheres (Invitrogen) covalently conjugated with OVA protein (Sigma-Aldrich). Inguinal dLNs were harvested at various times after immunization. In some animals, ear pinnae were irritated by skin application of 10µl of 50:50 (V:V) of dibutyl phthalate:acetone, with dLNs harvested 3 days later.

Cell Isolation and Flow Cytometry

Inguinal LNs were harvested and treated with 400 U/ml collagenase D (Roche Applied Science) solution for 25 min at 37°C. EDTA (10 mM) was added for 1 min and quenched with PBS wash. Single cell suspensions were stained with the indicated antibodies, acquired on a LSR-II flow cytometer (BD Biosciences), and analyzed using FlowJo software (TreeStar Inc.).

Statistical Analysis

The statistical significance of differences in mean values was analyzed by a two-tailed Student's t test. ***p < 0.0001, **p < 0.005, *p < 0.05