Research ArticleCANCER

MIBI-TOF: A multiplexed imaging platform relates cellular phenotypes and tissue structure

See allHide authors and affiliations

Science Advances  09 Oct 2019:
Vol. 5, no. 10, eaax5851
DOI: 10.1126/sciadv.aax5851
  • Fig. 1 MIBI-TOF is designed for high-throughput, highly multiplexed tissue imaging.

    (A) MIBI-TOF experimental procedure. A tissue section is stained with a mix of antibodies, each labeled with a unique metal isotope. To detect the isotopes, the tissue is rastered, pixel by pixel, by primary charged ions. Secondary ions, released from the tissue, are measured by time-of-flight mass spectrometry (TOF-MS) to generate an N-dimensional image. (B) MIBI-TOF instrumentation. MIBI-TOF is composed of several key parts: the sample stage, where slides are loaded into the instrument; a primary ion source, which shoots charged particles at the specimen, releasing secondary ions; an electrostatic analyzer (ESA), which filters secondary ions; a TOF mass spectrometer, which measures element abundance; and a raster scanner, which rasters the primary ion beam across the selected FOV in a stepwise pattern. (C) An 800 μm × 800 μm FOV of breast carcinoma imaged by MIBI-TOF. Inset depicts a 100 μm × 100 μm region, which can be imaged by NanoSIMS 50 L. (D to J) Comparison of various parameters for NanoSIMS 50 L and MIBI-TOF. Purple indicates more favorable values. (D) FOV sizes. (E) Current density (current per spot size). (F) FOVs (100 μm × 100 μm) were stained and visualized by NanoSIMS (top, 17 min) and MIBI-TOF (bottom, 3 min). Counterstain for the top image was hematoxylin and for the bottom image was carbon. (G) Scan times for the FOVs shown in (F). (H) Schematic depicting the mass range (x axis) measured by NanoSIMS and MIBI-TOF. Whereas the NanoSIMS uses a magnetic sector that inherently limits the number of detectable masses to the number of detectors (less than seven), MIBI-TOF acquires the entire spectrum from 1 to 240 m/z. (I) Sensitivity, measured by scanning a molybdenum standard. (J) Mass resolution (mass divided by full width of the peak at half mass). (K) FFPE human hippocampus was stained with a panel of 40 antibodies, and a 500 μm x 500 μm FOV was imaged by MIBI-TOF. The mass spectrum in the range of 165 to 175 m/z, summed over the entire image, is shown. (L) Energy filtering by MIBI-TOF. The primary ion beam hits the sample, releasing both elemental isotopes and polyatomic adducts, which may overlap in mass. Polyatomics are energy-filtered by the ESA, which prevents them from entering the TOF. Einzel lenses in the secondary ion optics are omitted for clarity. (M) Multiplexed staining for dsDNA, VGLUT2, CD298, glial fibrillary acidic protein (GFAP), EIF4A2, IBA1, and myelin basic protein (MBP), corresponding to the spectrum in (K). (N) A 200 μm × 200 μm FOV of human gastrointestinal (GI) tract was stained with a 20-plex panel and scanned with MIBI-TOF. Left: Secondary electron density (SED) image. Middle: Antibody staining for dsDNA (blue), Epcam (green), and vimentin (red). Right: Overlay of antibody stains with SED image. (O) A tissue microarray of biopsies from 41 TNBC patients was stained with a panel of 36 antibodies. FOVs (800 μm × 800 μm) were measured continuously, 24 hours a day for 12 consecutive days, with no tuning of the machine. Left: For each biopsy, the intensities, summed over the entire FOV, for dsDNA, β-catenin, and CD56 as a function of acquisition time are shown. Right: Staining for dsDNA for a 150 μm x 150 μm inset for the first and last patients measured.

  • Fig. 2 MIBI-TOF can sensitively image across the elemental spectrum with a quantitative dynamic range of > 100,000.

    (A) Human FFPE tonsil sections were stained with a single CD45 antibody clone conjugated to 16 different elemental reporters, and the amount of metal detected by MIBI-TOF was assessed by comparing to values attained using normalized, solution-phase inductively coupled plasma mass spectrometry (ICP-MS). For each element, the number of antibodies needed for detection (assuming ~100 metals per antibody) is shown. The corresponding number is displayed for 159Tb, the most sensitive channel on CyTOF, which is, on average, 15-fold less sensitive. (B and C) FFPE human breast carcinoma was stained with a panel of 36 antibodies, and a 500 μm × 500 μm FOV was imaged by MIBI-TOF. (B) Mass spectrum, summed over the entire image, at various scales. (C) Multiplexed staining for dsDNA, pan-keratin, and CD45, corresponding to the spectrum in (B). White box marks inset shown in (D). (D) Zoom-in on (C), depicting staining for dsDNA, CD3, and FoxP3, ranging over three orders of magnitude in intensity. (E) For all channels, their intensities, summed over all pixels in the image from (C), over five orders of magnitude, are shown. (F) Measurements of counts per second for a gallium arsenate probe (y axis) for a range of extraction currents (x axis). Red boxes denote limits of linear range, which spans five orders of magnitude. Inset: Gallium isotopic ratio (69Ga/71Ga) measured by MIBI-TOF (black) matches the known ratio (1.5, red) and is constant over three orders of magnitude before saturating the counts system. (G) FFPE human breast carcinoma was stained with the same panel as in (B) and two 500 μm × 500 μm FOVs were imaged by MIBI-TOF. Top left image depicts the entire region. Surrounding images highlight expression of distinct markers in specific regions.

  • Fig. 3 MIBI-TOF enables imaging of metal-tagged antibodies simultaneously with histochemical stains and endogenous biological elements.

    Tissue sections from tonsil (left), spleen (center), and GI (right) depicting simultaneous measurements of natural elements (12C, 31P, and 56Fe), metal-tagged antibodies (CD20, dsDNA, CD68, HO1, and Epcam), and counterstains (uranyl acetate). Arrows denote CD68+ HO1+ macrophages with high 56Fe.

  • Fig. 4 MIBI-TOF enables multiscale, high-dimensional rescanning for customizable tissue pathology workflows and three-dimensional tissue reconstruction.

    (A) A 200 μm × 200 μm melanoma biopsy (inset) was scanned four times at resolutions ranging from 260 to 1000 nm (y axis). The scan times for the different resolutions are shown (x axis). (B) Zoom-in on the inset in (A), showing staining for dsDNA (top) or lamin A/C (bottom) for the survey scan (left) and high-resolution scan (right). (C) Zoom-in on the inset in (B), showing staining for dsDNA (top) or lamin A/C (middle), and their overlay (bottom) for the different scans. Increasing detail of lamin A/C becomes available at higher resolutions. (D) Three FFPE breast carcinomas were stained with a 36-plex panel in May 2017, and 1-mm cores were visualized by MIBI-TOF days after staining (insets), or a year later (large images), with negligible deterioration in signal. (E) Second FOV from (D) was divided into a 10 × 10 grid. CD20 counts for each of the resulting 100 tiles in the images from 2017 (x axis) and 2018 (y axis) are shown. (F) Three-dimensional reconstruction of a 15-layer depth profile of archival 4-μm-thick FFPE melanoma tissue stained with a 20-plex panel. (G) Snapshots of selected depths. White and yellow arrows denote nuclei (dsDNA, green) that either appear or disappear, respectively, with depth, with corresponding disappearance or appearance of cytosol (vimentin, red) and membrane (Na-K ATPase, blue). (H) Images showing select marker expression in a colorectal tumor stained with a 34-plex panel. HH3, histone H3; PanCK, pan-keratin; Ecad, E-cadherin; Vim, vimentin. A select area is expanded in the bottom row, with cell segmentation outlines overlaid. Green and cyan arrows mark cells with high and low NC ratios of β-catenin, respectively. (I) NC ratio of β-catenin was calculated in all β-catenin+ tumor cells (PanCK+ or ECAD+). Cells within the highest and lowest quintile of β-catenin NC values were gated. (J) Expression of Ki-67, IDO1, HIF1α, VIM, PanCK, and HLA1 was compared between β-catenin NChi and NClow tumor cells in vertical scatterplots. Asterisks denote significant differences in mean cell signal intensity in the populations, ****P < 0.0001; ns, not significant.

  • Fig. 5 Intrapatient heterogeneity in tumor phenotypes is associated with spatial location.

    (A) Diagram of microscope-like analytical workflow, enabled by MIBI-TOF. (B) Archival breast cancer tissue section was stained with 36 antibodies, and a 0.8 cm × 1.04 cm FOV was visualized using MIBI-TOF at 1.5-μm resolution. The full image was constructed from 520 400-μm FOVs, tiled 26 × 20. Eight regions (white squares) were subsequently rescanned at 500-nm resolution. (C) Expression of four proteins across three regions shows regional heterogeneity in expression. Colored arrows mark cells belonging to specific tumor clusters in (D). (D) Tumor cells from all regions were clustered by protein expression. Expression values for each protein are scaled from zero to one. (E) tSNE embedding for tumor cells. Cells are colored by either phenotype as in (D) (left) or region (right). (F) Pseudo-coloring of all regions. Tumor cells are color-coded as in (D). Other cell types are colored gray. (G) For each tumor region (x axis), the number of cells from each tumor phenotype is shown (y axis). (H) Pearson correlations of pairwise comparisons of tumor cell composition in different regions (y axis) are shown as a function of the distance between the regions (x axis). Overall correlation indicates that close regions are more similar in composition of tumor phenotypes. (I) Cumulative distribution functions depicting the fraction of cores (y axis) with up to a number of cells from a given tumor phenotype (x axis) for the CK+ Ki-67+ phenotype (left) and CK+ HLA-DR+ Ki-67+ phenotype (right). Observed distributions (cyan and pink, respectively) are compared with the Poisson distribution expected from random sampling using the Kolmogorov-Smirnov statistic and FDR-corrected for multiple hypothesis testing across all tumor and immune phenotypes.

  • Fig. 6 Structured immune composition and organization across tumor regions.

    (A) Immune cells from all regions were clustered by protein expression. Expression values for each protein are scaled from zero to one. (B) Pseudo-coloring of all regions. Immune cells are color-coded as in (A). Other cell types are colored gray. (C and D) tSNE embedding for immune cells. Cells are colored by either immune lineage (C) or region (D). (E) Immune composition normalized from zero to one (y axis) in all regions (x axis). Regions are ordered by percentage of CD4+ T cells (magenta). (F) For five immune populations (x axis), their presence or absence across a cohort of 41 TNBC patients (26) (y axis, patients numbered 1 to 41) and in the eight regions from a single patient depicted in this study (orange rectangles, R1 to R8) is shown. (G) Tumor-immune mixing scores were calculated for a cohort of 41 TNBC patients and the eight regions from a single patient depicted in this study. The probability density function of tumor-immune mixing across patients (blue) and across regions (orange) is shown. Seven of eight regions are defined as compartmentalized, according to a previously defined threshold (26). (H) Same as (G), for the log ratio of PD-L1+ tumor and immune cells. In all regions, PD-L1 is expressed predominantly on immune cells. (I) Same as (G), for the log ratio of PD-1+CD8+ and PD-1+CD4+ T cells. In all regions, PD-1 is expressed predominantly on CD4+ T cells. (J) Heatmaps depicting spatial enrichment z scores between pairs of proteins in regions R1 (top) and R4 (bottom). Green squares denote enriched spatial proximity of myeloid-derived suppressor–like cells (MDSC-like) near the tumor and of lymphocyte-rich regions. (K) Left: Color overlays of PD-L1 (cyan), IDO (magenta), and cell segmentation (white) in regions R1 (top) and R4 (bottom). Tumor-immune border as identified by an automatic pipeline is highlighted in yellow. Right: Zoom-in on boxed region in left panel. Immunoregulatory proteins PD-L1 and IDO are predominantly expressed by immune cells close to the tumor-immune border. (L) Pseudo-coloring of immune cells as in (B) for the zoomed-in regions in (K), showing a concentric organization of tumor cells, surrounded by MDSC-like and then lymphocyte-rich regions.

Supplementary Materials

  • Supplementary material for this article is available at http://advances.sciencemag.org/cgi/content/full/5/10/eaax5851/DC1

    Fig. S1. MIBI-TOF is structurally designed for high-throughput, highly multiplexed imaging with high sensitivity and low noise.

    Fig. S2. Thirty-six–plex staining using MIBI-TOF.

    Fig. S3. MIBI-TOF can sensitively image across the elemental spectrum and allows customized ion dose/area to detect antigens with varying abundances.

    Fig. S4. Evaluation of observed and expected regional distribution of tumor and immune subtypes within a single patient.

    Table S1. Sensitivity measurements.

    Table S2. Imaging conditions.

    Table S3. Tissues and antibodies.

    Movie S1. Depth profile of melanoma.

  • Supplementary Materials

    The PDF file includes:

    • Fig. S1. MIBI-TOF is structurally designed for high-throughput, highly multiplexed imaging with high sensitivity and low noise.
    • Fig. S2. Thirty-six–plex staining using MIBI-TOF.
    • Fig. S3. MIBI-TOF can sensitively image across the elemental spectrum and allows customized ion dose/area to detect antigens with varying abundances.
    • Fig. S4. Evaluation of observed and expected regional distribution of tumor and immune subtypes within a single patient.

    Download PDF

    Other Supplementary Material for this manuscript includes the following:

    • Table S1 (Microsoft Excel format). Sensitivity measurements.
    • Table S2 (Microsoft Excel format). Imaging conditions.
    • Table S3 (Microsoft Excel format). Tissues and antibodies.
    • Movie S1 (.mov format). Depth profile of melanoma.

    Files in this Data Supplement:

Stay Connected to Science Advances

Navigate This Article