Extracting a cellular hierarchy from high-dimensional cytometry data with SPADE

Abstract

The ability to analyze multiple single-cell parameters is critical for understanding cellular heterogeneity. Despite recent advances in measurement technology, methods for analyzing high-dimensional single-cell data are often subjective, labor intensive and require prior knowledge of the biological system. To objectively uncover cellular heterogeneity from single-cell measurements, we present a versatile computational approach, spanning-tree progression analysis of density-normalized events (SPADE). We applied SPADE to flow cytometry data of mouse bone marrow and to mass cytometry data of human bone marrow. In both cases, SPADE organized cells in a hierarchy of related phenotypes that partially recapitulated well-described patterns of hematopoiesis. We demonstrate that SPADE is robust to measurement noise and to the choice of cellular markers. SPADE facilitates the analysis of cellular heterogeneity, the identification of cell types and comparison of functional markers in response to perturbations.

Figure 1

Flowchart of SPADE, and SPADE analysis of a simulated dataset. (i) A simulated 2-parameter flow cytometry dataset, with one rare population and three abundant populations. (ii) Result of density-density downsampling of the original data. (iii) Agglomerative clustering result of the downsampled cells. (iv) Minimum spanning tree that connects the cell clusters. (v) Colored SPADE trees. Nodes are colored by the median intensities of protein markers, allowing visualization of the behaviors of the two markers across the entire heterogeneous cell population.

Figure 2

SPADE applied to mouse bone marrow flow cytometry data. (a) Known hematopoietic hierarchy in mouse bone marrow, and the colored SPADE tree derived from the mouse bone marrow data. Each tree was colored by the median intensity of one individual marker. (b) Traditional gating analysis on the mouse bone marrow data. For each gated population, one SPADE tree was drawn, where each node was colored by the percentage of gated cells in that node. Thus, the color of each tree represents which part of the tree is populated by the cells in the corresponding gate. This comparison shows the concordance between SPADE and gating results.

Figure 3

SPADE applied to human bone marrow dataset of 30 experiments with 2 overlapping staining panels and multiple experimental conditions. (a) Experiment and staining panel design of this human bone marrow dataset. (b) SPADE tree derived from this dataset. The SPADE tree was annotated according to the intensities of the 13 core surface markers.

Figure 4

SPADE tree colored by two NK specific markers CD7 and CD16, which were not used to derive the SPADE tree. The color patterns indicate that the nodes in the gray circle are NK cells. This result shows that SPADE clustered NK cells without using any NK specific markers.

Figure 5

SPADE tree that describes the cell-type-dependent behavior of one functional marker in response to one stimulus. (a) after stimulation with TNF, phosphorylated MAPKAPK2 was observed in myeloid and NK cell types, but not in other cell types. (b) after stimulation with LPS, degradation of total IkB was restricted to the monocytoid lineage. (c) TPO-induced phosphorylated STAT5 was observed in HSCs and CD123++, but not other cell types. (d) GM-CSF-induced phosphorylation of pSyk was observed only in myelocytes.