Imaging mass cytometry

The work flow of imaging mass cytometry is shown in Figure 1—figure supplement 2 and explained in detail below.

Conjugation of antibodies with lanthanide metals. Lanthanide metal-conjugated antibodies were either obtained from Fluidigm (Markham, Ontario, Canada) or conjugated at SickKids-UHN Flow and Mass Cytometry Facility (Toronto, Ontario, Canada), using the MaxPar X8 labelling kit from Fluidigm (catalogue number 201169B) as previously described (Han et al., 2018). Briefly, a purified carrier-free antibody was partially reduced with TCEP buffer (Fluidigm, catalogue number 77720) at 37°C. The reduced antibody was then incubated with an excess of metal-loaded MaxPar X8 polymer for 90 min at 37°C. The metal-labeled antibody was then recovered using a 50 kDa size exclusion filter. The percent yield of metal-conjugated antibody was determined by measuring the absorbance of the conjugate at 280 nm. The recovery of our metal-conjugated antibodies was 69–78%. Antibody stabilizer was then added to the metal-conjugated antibodies before long-term storage at 4°C.

Staining for Imaging Mass Cytometry. On the day of staining, the slides were brought to room temperature and rehydrated with 0.05% PBS-Tween in a humidified chamber for 20 min at room temperature. Non-specific protein binding was blocked by incubation with 10% normal goat serum for 20 min at room temperature followed by incubation with blocking solution (ThermoScientific Superblock Blocking Buffer in PBS) for 45 min at room temperature. A cocktail of primary antibodies, diluted in 0.5% BSA, was applied overnight at 4°C at the dilutions indicated in Key Resouces Table. The following day, slides were first washed with 0.05% PBS-Tween and then with PBS, followed by incubation with Iridium-conjugated intercalator (Fluidigm, catalogue number 201192B), diluted 1:2000 in 0.5% BSA for 30 min at room temperature. Lastly, slides were dipped in water (Invitrogen ultrapure distilled water), air dried and stored at room temperature until they were ablated.

Identification of region of interest (ROI) for laser ablation. Two serial sections each stained for either IF or IMC, were used. Based on IF staining with an antibody specific for proteolipid protein (PLP) (to visualize myelin) and DAPI (to visualize nuclei), ROIs were selected for ablation to capture the regions of interest for this study.

High-spatial resolution laser ablation of tissue sections. Tissue sections were analyzed by IMC, which couples laser ablation techniques and CyTOF mass spectrometry (Bandura et al., 2009) (Cytof software v6.7). Briefly, a UV laser beam (λ = 219 nm) with a 1µmx1µm spot size is used to ablate the tissue. The laser rasters across the tissue at a rate of 200 Hz (200 pixels/s) with the requisitie energy to fully remove the tissue within the selected region of interest. The time needed to analyze each image of 1 mm², using the methodology described in this manuscript, is approximately 1 hr and 45 min. The ablated tissue is then carried by a stream of inert helium and argon gas into the Helios (a CyTOF system) where the material is completely ionized in the inductively coupled plasma. The ionized material then passes through high pass ion optics to remove ions with a mass less than 75amu before the ions enter the time of flight detector where they are separated based on their mass (Bodenmiller et al., 2012; Bendall et al., 2011).

Data analysis and image visualization. Images of each mass channel were reconstructed by plotting the laser shot signals in the same order they were recorded, line scan by line scan, generating pseudo-colored intensity maps of each mass channel. These data were examined using MCD Viewer (V.1.0.560, Fluidigm). For qualitative assessments, images remained at the automatic threshold generated by MCD Viewer, based on the on the 98th percentile of signal. For further analysis, data were exported from MCD Viewer as tiff files, and each channel was run through an individual analysis pipeline in CellProfiler (Carpenter et al., 2006; Jones et al., 2008) (V3.185) in order to despeckle the image. Composite images were created for each ROI using Image J (V1.52a), and any changes to the brightness or contrast of a given marker were consistent across ROIs.

Calculation of the limit of detection. MCD Viewer was used to export text files acquired with the Hyperion IMC instrument (Fluidigm Inc, Markham ON), which were then converted to 32-bit single-channel TIFF images. The polygon tool within ImageJ 1.15 s was used to manually outline the ROI (white matter of control, normal-appearing white matter) or subROI (periplaque white matter, lesion edge, lesion core), manually identified on the bases of PLP, HLA and Iridium-intercalator signals. Gray matter was excluded from subsequent analysis. Each image was despeckled in Definiens Developer XD 2.7 (Definiens Inc, Munich, Germany), using a 2D gray-level morphological opening filter with kernel radius of 1. In addition, the intensities of each marker were normalized using a modified z score approach, in which the intensity of each pixel is divided by the sum of (mean intensity of the image plus 3 times the standard deviation of the pixels in the image) I_zs = I/(μ_Im+3*σ _Im). This normalization approach has been previously used (Ellingson et al., 2012) and we found that it allows for a reliable comparison between IMC markers across different channels, with per-marker comparisons holding robustly across a 16-fold antibody dilution series (data not shown).

Single-cell segmentation. In order to define cells, we used a customized segmentation algorithm that took into account both the presence of nuclear DNA Iridium-intercalator as well as a set of markers of interest (see example in Figure 7—figure supplement 1). In brief, a Gaussian blur was applied to the DNA signal and the resulting blurred image was segmented to identify nuclear content (Figure 7—figure supplement 1a). Segmentation around the nuclei was expanded to simulate the cytoplasm, corresponding to individual cell areas, using a combination of threshold and watershed filters (Figure 7—figure supplement 1b). Next we interrogated the segmented image for the presence of specific markers or combinations of markers that are either biologically co-expressed, or whose expression is mutually exclusive, according to the combination of markers indicated in Table 2. If a nucleated cell was positive for a marker or a combination of markers (see example for CD3 in Figure 7—figure supplement 1c), the marker(s) signal was used to refine the initial nuclear segmentation. Nucleated cells that were not positive for any of the markers used, were segmented purely based on DNA signal and expanded to simulate the cytoplasmic area around the nucleus.

Gating strategy for quantitative analysis of T cell, B cell, macrophage and microglial cell subsets. Segmented cell exports of raw and normalized marker intensities for all channels in each region of interest were exported as a single csv file. The per-cell mean intensities of each marker combination, (see marker list in Table 2), were linearly rescaled for visualization purposes. 2D log-log biaxial scatterplots of these intensities were generated in Python (V3.6.8) using matplotlib (V3.0.3). A positive- and negative-gating strategy was applied to establish thresholds that identify particular cell types. Quadrants were set on pathologist-verified positive cells. In brief, ROI was examined in Definiens Developer XD 2.7. Cells were manually annotated by a pathologist, based on the expression of a biologically relevant set of markers to identify cells in each class of interest as defined below (see examples of manual selection to identify myeloid cells in Figure 7—figure supplement 2; to identify T cells in Figure 7—figure supplement 3; to identify B cells in Figure 7—figure supplement 4). These identified positive cells were superimposed to the 2D log-log scatterplots to definitively establish gates that would capture the appropriate positivity range of each cell phenotype as shown in Figure 7—figure supplement 5.

For T cells: All nucleated cells expressing Igκ, Igλ, IgM, CD68 and HLA were eliminated, as these markers are not expressed on T cells. Gates were established for CD3 and CD45 based on a 2D log-log scatterplot of these markers. Following the identification of T-lineage cells, the same procedure was performed for CD3 vs CD4 and CD3 vs CD8, resulting in the identification of two subpopulations: CD4⁺ T cells and CD8⁺ T cells. Thresholds for Ki67 (a marker of proliferation) and NFAT1 (a marker of activation) were established based on manually annotated CD3⁺KI67⁺ and CD3⁺NFAT1⁺ cells, as described above. All cell populations were validated by manual annotation as described above.

For B Cells: All nucleated cells expressing CD3, CD4, CD8 and CD68 were eliminated as these markers are not expressed on B cells. B cells were further identified by CD45 above the same threshold set for T cells. Scatterplot comparison for Igκ and Igλ intensity identified Igκ+ and Igλ⁺ single-positive populations. Igκ⁺Igλ⁺ double positive cells were eliminated as artifactual, since the two allelic variants cannot co-exist on a given cell. Within Igκ+ or Igλ⁺ B-lineage cells, we compared IgM to CD38 to determine the relative abundance of IgM⁺ or CD38⁺ cell subpopulations. All cell populations were validated by manual annotation as described above.

For macrophages and microglia: All nucleated cells expressing CD3, CD4, CD8, Igκ/λ and IgM were eliminated as these markers are not expressed on macrophages and microglia. Discrimination of the remaining cells was visualized in a scatterplot for TMEM119 and CD45. The threshold for TMEM119 positive signal was determined by comparison to TMEM119⁺ microglial cells that were identified by manual observation, relative to other cell types. The threshold for CD45 high or low signal was determined by the comparison to manually identified TMEM^- macrophages. Manually identified microglial cells were used to establish the lower limit of the CD45 quadrants. Cells that were low for both TMEM119 and CD45 were labeled ‘other’ and ignored from subsequent analysis. These latter cells, likely correspond to astrocytes, oligodendrocytes and other cell types. Both TMEM119⁺CD45^low/+ microglial cells and TMEM^-CD45^high macrophages were further evaluated for HLA, CD68 and PLP (depicted as scatterplots), to differentiate microglia or macrophages that are either resting or activated/phagocytic/demyelinating. All cell populations were validated by manual annotation as described above.

Generation of cell density map. The gating strategy described above was confirmed by plotting the appropriately gated cell types for major lineage markers (see examples in Figure 7—figure supplement 6). Note that Igκ>Igλ in Figure 7—figure supplement 6a consistent with over-representation of κ⁺ B cells in humans (Koulieris et al., 2012). Following this confirmation, the density of all relevant cell subtypes was computed within each biological region of interest. A heat map, generated using Seaborn (V0.9.0), displayed the cell counts per mm² of tissue.

Generation of distance map. To assess the location of identified cells relative to blood vessels, collagen⁺ perivascular regions of >10 µm diameter and >800 µm² area were segmented, and the distance between cells of interest and the border of these perivascular regions was calculated. Similarly to the cellular density calculations, average vessel distances corresponding to the mean of the per-cell vessel distance values were computed, expressed in µm, and presented as a distance heat map. Some regions did not contain any cells of a particular type, leading to undefined values for those particular regions and cell type combinations (presented as ‘n/a’, not applicable).

Potential of Heat-diffusion Affinity-based Transition Embedding (PHATE) mapping. To study the dynamics of T cell phenotypes, Potential of Heat-diffusion Affinity-based Transition Embedding (PHATE) mapping was performed in R (Moon, 2019). Relevant parameters were selected for analysis of T cells (CD3, CD4, CD8, CD38, CD45, HLA, Ki67, NFAT1, distance to blood vessels). Raw, mean single-cell marker intensity values extracted from segmented IMC images as well as measured distances to blood vessels were subjected to log10 transformation followed by Z-score normalization. These normalized single-cell parameter values along with metadata indexing the class and lesion type of residence for each cell served as input for PHATE mapping, as detailed in the online user guide. The PHATE algorithm was executed with k = 100 and other parameters left as their default specifications. Plots were generated using the R ggplot2 package (Hadley, 2016).