Fig 2: Identification of a panel of surface markers that can further purify MICB+PVR+ MKPs.(A) Twenty index sorts of A1ATD1 MKs. Rank indicates the first occurrence of a marker in a depth-first search of the decision trees making up the bagging model; genes with consistently low rank are discriminative for a given replicate. (B) Determining cutoffs for MKP markers based on all index sorts of A1ATD1 MKs. Outliers were excluded, and the 20 datasets were normalized and combined. The resulting dataset was then divided into 60 bins based on PVR expression and the colonies and noncolonies counted in each bin. This allowed the calculation of (1 - proportion colony formation per bin) (y axis). Cells were then ordered with decreasing PVR expression, and the proportion of cells with higher than this value of PVR is plotted on the x axis. The dataset was then subsetted on this expression level of PVR (top 20%), and the analysis was repeated for MICB, CD51, CD168, and VEGFR1. Areas with maximum colony formation are shown with shaded boxes. (C) Final sort strategy for MKPs. Cells are first purified as MICBhiPVRhi, then as CD51hiCD168hi, and lastly as VEGFR1lo.

Fig 3: Single-cell sequencing of MKPs allows the selection of further candidate markers.(A) UMAP of 10x Genomics scRNA-seq of D27 A1ATD1 MKs sorted for the top 50% MICB, top 26% PVR, and bottom 33% CD42 expression. (B and C) UMAP of single-cell transform–normalized MICB and PVR expression. (D to F) UMAP of un-normalized GATA1, TAL1, and FLI1 expression from lentiviral transgenes plotted on the gene expression UMAP. (G to J) UMAP of single-cell transform–normalized ANXA1, HMMR, ITGAV, and FLT1 expression.