Adjacent Cell Marker Lateral Spillover Compensation and Reinforcement for Multiplexed Images

Abstract

Multiplex imaging technologies are now routinely capable of measuring more than 40 antibody-labeled parameters in single cells. However, lateral spillage of signals in densely packed tissues presents an obstacle to the assignment of high-dimensional spatial features to individual cells for accurate cell-type annotation. We devised a method to correct for lateral spillage of cell surface markers between adjacent cells termed REinforcement Dynamic Spillover EliminAtion (REDSEA). The use of REDSEA decreased contaminating signals from neighboring cells. It improved the recovery of marker signals across both isotopic (i.e., Multiplexed Ion Beam Imaging) and immunofluorescent (i.e., Cyclic Immunofluorescence) multiplexed images resulting in a marked improvement in cell-type classification.

Keywords: cell annotation; image correction; multiplexed tissue imaging; signal spillover; single-cell biology; spatial proteomics.

Figure 1

REDSEA Corrects Signal Spillover between Adjacent Cells. (A) A schematic representing the workflow and principles of REDSEA compensation. The spillover signal from neighboring cells is dynamically eliminated based on the fraction of the shared boundary between neighboring cells and the signal intensity. (B) Left: A representative 150 µm x 150 µm MIBI image of a rhesus macaque lymph node. Two mutually exclusive markers are shown (CD3, magenta; CD20, green), and the numerical counts of CD3 are indicated in each segmented cell before and after images were subjected to spillover subtraction using the following four methods: 1) spillover subtraction on the whole cell, 2) spillover subtraction on only the border region, 3) REDSEA compensation on the whole cell and 4) REDSEA only on the border regions. The CD3 and CD20 counts per cell are colored on the same scale for the segmented cells across compensation settings. Right: Zoomed images of the yellow boxed regions on the left. The yellow arrows indicate representative cells for which CD3 spillover was successfully corrected by all the four methods; red arrows indicate successful correction by all but the whole cell subtraction method, and blue arrows indicated successful correction only by REDSEA-based and not the other compensation methods. (C) A representative 1200 µm x 1200 µm MIBI image of a rhesus macaque lymph node subjected to spillover corrections as indicated above in (B). (D) Top: Biaxial plots of marker intensities of 68,739 single cells from segmented MIBI images of rhesus macaque lymph nodes. The percentage of single-positive (top left and bottom right quadrants), double-positive (top right quadrant), and double-negative (bottom left quadrant) cells are shown for each compensation method. Bottom: A log2 fold change plot (compensated over original non-compensated) for single-positive, double-positive, and double-negative gated populations.

Figure 2

REDSEA Reduces Non-Specific Spillover Signals. Left: Arcsine and square root transformed counts per cell for CD3, CD4, CD8a, CD20, and CD68 were plotted before and after REDSEA border compensation for each of the cell types identified. Right: Representative images of each cell type with marker counts before and after REDSEA compensation.

Figure 3

REDSEA Enriches for Cell-type-specific Signals and is Platform Agnostic. (A) Schematic of the workflow for calculation of enrichment and depletion of various cell types for each channel before and after REDSEA correction. Cells with no counts in the channel of interest after REDSEA correction were discarded, and the percentage composition of each cell type remaining was calculated. The relative change is the difference in percentage composition of each cell type before and after REDSEA correction. (B) Left: A representative 900 µm x 900 µm CyCIF image of a human tonsil. Three pairs of mutually exclusive markers are shown: CD3 (magenta) and CD20 (green); CD4 (magenta) and CD8a (green); and CD68 (magenta) and CD20 (green). The differences in percentage compositions between the REDSEA image and the original in counts of both markers per segmented cell are shown on a visual scale. Right: Biaxial plots of marker signals from each of the 6,295 single cells extracted from the segmented CyCIF images. The percentage composition of double-positive (top right quadrant) cells is shown for each compensation method.

Figure 4

REDSEA Improves Cell-type Annotation of MIBI images. (A) Single-cell marker values were determined from a single MIBI field of view of a rhesus macaque lymph node (400 µm x 400 µm) with no corrections (Original) or 1) spillover subtraction on the whole cell, 2) REDSEA compensation on the whole cell, 3) spillover subtraction on only the border region and 4) REDSEA only on the border regions. A single iteration of unsupervised cell type classification was performed with identical parameters on extracted single cell marker values under each of the conditions. The quantification of cell types identified in terms of fold change (left) and cumulative number (right) are represented here. Not determined (yellow) denotes cell types that could not be confidently assigned to clusters identified during classification. (B) Spatial positions of cell types identified under each condition are represented as phenotype maps. Phenotype maps for cell types identified without correction (Original), with whole cell-based spillover subtraction (Whole Cell Subtraction), with consensus-based manual annotation from three independent individuals (Manual Annotation) and with border-based REDSEA [REDSEA (Border)] are shown. (C) Pseudo-colored MIBI images containing various combinations of cell-type-specific markers from the same field of view.