Bmi1 suppresses protein synthesis and promotes proteostasis in hematopoietic stem cells

In this study, Burgess et al. conditionally deleted Bmi1 from adult hematopoietic cells and found that this slowly depleted hematopoietic stem cells (HSCs) and that, rather than inducing senescence, Bmi1 deficiency increased HSC division. Overall, they found that Bmi1 promotes HSC quiescence by negatively regulating ARX expression and promoting proteostasis by suppressing protein synthesis.


Analysis of single cell RNA sequencing data
Raw sequencing reads from individual samples were demultiplexed, counted, and aggregated using 10x Genomics Cell Ranger 4.0.0 with the mouse mm10 reference genome.
Stressed or dying cells with >5% of their transcripts from mitochondrial genes were excluded from analysis using R 4.0.2 with Seurat 3.2 on Cell Ranger's filtered feature barcode matrix of Unique Molecular Identifier (UMI) counts. Filtered samples were then normalized and integrated using Seurat's variance stabilizing transformations for single cell UMI data (SCTransform). HSC clustering was performed using Seurat 3.2 (Stuart et al. 2019) with and without cell cycle scoring and regression (Tirosh et al. 2016). LT-HSC and ST-HSC signature scores were calculated using published gene lists (Pei et al. 2020) and scoring differences between cells were tested by a linear mixed-effects model followed by the False Discovery Rate method for multiple comparisons adjustment using R with the Ime4 package. Differential gene expression analysis was performed by Wilcoxon rank sum tests using Seurat, and gene set enrichment analyses were performed using GSEA 4.1.0 (Mootha et al. 2003;Subramanian et al. 2005) on the pre-ranked, log-transformed gene fold changes. GO Term enrichment of cell marker genes was performed using clusterProfiler 3.18.1 (Yu et al. 2012).

Analysis of bulk RNA sequencing data
RNAseq data were analyzed based on the Tuxedo protocol (Trapnell et al. 2012;Pertea et al. 2016), and the quality of raw reads were assessed by FASTQC 0.11 (Andrews 2010). Raw reads were trimmed using TrimGalore 0.6 and mapped to the Ensembl GRCm38 mouse reference genome using TopHat2 (Kim et al. 2013) with Bowtie2. Mapped reads were quality-filtered using SAMtools 1.9 (Li et al. 2009) and quantified using Cufflinks 2 (Trapnell et al. 2010;Trapnell et al. 2012). Quantified mapped reads were normalized to fragments per 10000 exonic bases per million mapped reads (FPKMs) and gene expression levels measured using Cuffnorm. Differential gene expression was assessed using Cuffdiff. Expressed ribosome genes were determined as protein coding genes with FPKM>1. The correlation among genes that were differentially expressed between control and Bmi1 deficient HSCs in the single cell RNA sequencing and bulk RNA sequencing was tested using only genes with detectable expression in both datasets. The statistical significance of the correlation was assessed using a Pearson correlation test.

Supplemental statistical methods
To analyze the statistical significance of differences among groups, we first tested if data were normally distributed and if variance was similar among groups. To test for normality, we performed the D'Agostino Omnibus test when n≥20 or the Shapiro-Wilk test for smaller sample numbers. To test for significant differences in variability among groups, we performed an F-test (for experiments with two groups) or Levene's median test (more than two groups). When the data significantly deviated from normality or variability significantly differed among groups, we log2transformed the data and tested again for normality and variability. If the transformed data did not significantly deviate from normality and equal variability, we performed parametric tests on the transformed data. If log2-transformation was not possible or the transformed data still significantly deviated from normality or equal variability, we performed non-parametric tests on the nontransformed data. The dagoTest and shapiroTest functions of the fBasics package were used to perform the normality tests, and the leveneTest function of the Companion to Applied Regression package was used to perform the Levene's median test for variance.
When data or log2-transformed data were normal and equally variable, statistical analyses were performed using Student's t-tests (when there were two groups), one-way ANOVAs (when there were more than two groups), or two-way ANOVAs (when there were two or more groups with multiple tissues or cell populations or time points). When the data or log2transformed data were normal but unequally variable, statistical analyses were performed using Welch's t-tests (when there were two groups) or Welch's one-way ANOVAs (when there were more