Transient expansion and myofibroblast conversion of adipogenic lineage precursors mediate bone marrow repair after radiation

Radiation causes a collapse of bone marrow cells and elimination of microvasculature. To understand how bone marrow recovers after radiation, we focused on mesenchymal lineage cells that provide a supportive microenvironment for hematopoiesis and angiogenesis in bone. We recently discovered a nonproliferative subpopulation of marrow adipogenic lineage precursors (MALPs) that express adipogenic markers with no lipid accumulation. Single-cell transcriptomic analysis revealed that MALPs acquire proliferation and myofibroblast features shortly after radiation. Using an adipocyte-specific Adipoq-Cre, we validated that MALPs rapidly and transiently expanded at day 3 after radiation, coinciding with marrow vessel dilation and diminished marrow cellularity. Concurrently, MALPs lost most of their cell processes, became more elongated, and highly expressed myofibroblast-related genes. Radiation activated mTOR signaling in MALPs that is essential for their myofibroblast conversion and subsequent bone marrow recovery at day 14. Ablation of MALPs blocked the recovery of bone marrow vasculature and cellularity, including hematopoietic stem and progenitors. Moreover, VEGFa deficiency in MALPs delayed bone marrow recovery after radiation. Taken together, our research demonstrates a critical role of MALPs in mediating bone marrow repair after radiation injury and sheds light on a cellular target for treating marrow suppression after radiotherapy.

Computational cell cycle analysis was conducted as described previously (21), we used a core set of 43 S and 54 G2/M genes defined previously (62). First, the genes that are expressed in less than 5% of total cells were removed, resulting in 34 S and 45 G2/M genes for the following cell cycle analysis. Second, we define proliferative cells if the cell express S gene set or G2/M gene set, that is, there is a significant difference between the expression value of S genes and G2/M genes. Since the null distribution for the S gene set and G2/M gene set is unknown, we designed a permutation test which does not assume a null-distribution. To be more specific, for each cell, we resampled 34 genes from the 79 genes (34+45) as "S genes" and the rest 45 genes as "G2/M genes". This resampling was repeated 5000 times and each time, we calculate the difference between the mean expression value of "S genes" and the mean expression values of "G2/M genes". The original difference between the mean expression value of S genes and the mean expression value of G2M was compared with the difference between resampling results to get a significant score (p-values). To correct for multi-test, FDR corrections were applied to each cluster. At last, FDR p-value < 0.05 was used as a cutoff for distinguishing proliferative or nonproliferative cells.
To computationally delineate the developmental progression of bone marrow mesenchymal cells and order them in pseudotime, we used the algorithms implemented in the Monocle 2 package (63). We include mesenchymal lineage cells with no chondrocytes from separated or integrated dataset of different group (NR or R) mice for the analysis. We decided the genes that define the cell differentiation trajectory by selecting genes with high dispersion across cells, using a parameter of "mean_expression >= 0.05 & dispersion_empirical >= 2 * dispersion_fit".
The gene list was further used for dimensional reduction to generate the trajectory reconstruction using the nonlinear reconstruction algorithm DDRTree. dimensional reduction after the PCA were calculated for individual or integrated datasets. Then Seurat objects were transformed into SingleCellExperiment objects. Slingshot trajectory analysis were conducted using the Seurat clustering information and with dimensionality reduction produced by UMAP.
The SCENIC algorithm was used to assess the regulatory network analysis regard to transcription factors (TFs) and discover regulons (TFs and their target genes) in individual cells.
Following the standard pipeline, the gene expression matrix with gene names in rows and cells in columns was input to SCENIC (version 0.9.1) (65). The genes were filtered with default parameter, and co-expressed genes for each TF were constructed with GENIE3 software, followed by calculating of Spearman's correlation between TFs and their potential targets, and then the "runSCENIC" procedure assisted to generate the GRNs (also termed regulons). Finally,                   Immunofluorescent staining of bone marrow from Adipoq/Td mice shows a CD150 + HSCP (pointed by a white arrowhead) resides next to a Td + cell (pointed by a yellow arrow). ColIV is a marker for both endothelial cells and MALPs. A white arrow points to a pericytic Td + cell.  Violin plots show the expression patterns of Alpl in normal (NR) and irradiated (R) scRNA-seq datasets.

Supplementary Table 1
Mouse real time RT-PCR primer sequences used in this study.

Gene
Forward primer Reverse primer