Page 52 - 2019_05-HaematologicaMondo-web

Z. Wu et al.
but the ratio of maximum expression of lncRNAs relative to housekeeping genes was about 4-fold higher in single cells than in bulk samples. By scRNA-seq, the maximum expression of lncRNAs was similar to that of both mRNAs and housekeeping genes (Figure 2C). Genes with high variance tended to be captured by the single cell analysis rather than by the bulk approach (Online Supplementary Figure S4). Thus, lncRNA expression appeared to be better detected among single cells due to an expression pattern of high cell-to-cell variation and cell-specificity.
We then sought to infer putative functions of defined lncRNAs in hematopoiesis by a comprehensive “guilt by association” approach (Online Supplementary Methods and Results), correlating expression of lncRNAs with protein- coding genes of known functions.4,15,39-41 Associated pro- tein-coding genes of defined lncRNAs across CD34+ cells were enriched in gene ontology (GO) terms related to myeloid cell differentiation, cell growth, and cellular func- tions including DNA repair, mRNA splicing, gene expres- sion, and epigenetic regulation (Figure 2D), implicating lncRNAs in the regulation of human hematopoiesis and associated cellular functions.
Stage- and lineage-specific expression of long noncoding RNAs in normal hematopoiesis
To obtain a profile of lncRNA expression in normal human hematopoiesis, we assessed lncRNA expression in 391 CD34+ cells from healthy donors. We first studied whether a lncRNA signature separated CD38- and CD38+ cell populations. lncRNAs detected with 20 reads in at least 20 cells were retained, and highly variable lncRNAs were used for stage-specific analysis (Online Supplementary Figure S5A). The method of t-distributed stochastic neighbor embedding (t-SNE) was adopted for non-linear dimension reduction based solely on batch-corrected (by Combat/SVA) lncRNA expression (Online Supplementary Figure S5B). In an unsupervised t-SNE plot, sorted CD38- cells formed a cluster distinct from CD38+ cells, while CD38+ cells were more dis- persed (Figure 3A). To determine stage specificity, we per- formed pair-wise comparison of lncRNA expression in CD38- cells relative to expression in CD38+ cells. lncRNA expression exhibited substantial differences in two stages (Online Supplementary Table S3); heatmaps of differentially expressed mRNAs and lncRNAs of CD38- and CD38+ pop- ulations are shown in Figure 3B.
We previously assigned single CD34+ cells to a cell type according to their protein-coding transcriptome profiles, based on gene expression data from flow cytometrically- sorted cell populations.42 The cell types to which the sin- gle cells were assigned included HSC, multilymphoid progenitor (MLP), megakaryocyte-erythroid progenitor (MEP), granulocyte-monocyte progenitor (GMP), pro-B cell (ProB), and earliest thymic progenitor (ETP).31 We applied weighted gene co-expression network analysis43 to assess the potential functions of lncRNAs in CD38+ and CD38- cells. When protein-coding and lncRNA-encoding genes were simultaneously analyzed, they clustered into seven unsupervised modules (Online Supplementary Table S4), and genes in individual modules were analyzed for GO term enrichment (Figure 3C). Genes in module 1 showed high enrichment of lymphocyte activation path- way genes, and their expression levels were higher in ProB and ETP than in other cell types. Genes in module 6 were enriched in the heme metabolic process, and they showed higher expression in MEP. These data suggest
roles of lncRNAs in hematopoiesis and lineage specificity of lncRNA expression.
By t-SNE, cells tended to cluster according to cell types (Figure 4A, right) and were coincident with the pattern of hematopoietic differentiation based on mRNA expression in pseudotime ordering (Figure 4A, middle).31 Thus lncRNAs appeared as powerful as their protein-coding counterparts in resolving subtypes of CD34+ cells. We then analyzed cell-type specificity of gene expression by cell- type variance (Figure 4B) and assessed a Jensen-Shannon score8 (JScore) (Figure 4C). lncRNA expression showed higher cell-type specificity than did mRNA expression (JScore, P=1x10-16). There was more cell-to-cell variation in lncRNA expression than in mRNA expression, even within the same cell type (Online Supplementary Figure S6). We investigated our dataset for lncRNA signatures in various lineages, using difference in expression in a lineage, relative to expression in all other subsets, by pairwise comparisons, at a threshold P<0.05 (Figure 4E and Online Supplementary Table S5). Heatmaps revealed that MLP had signatures of both mRNAs and lncRNAs similar to those of HSC, in con- trast to distinctive gene expression patterns in other lineag- es. These data were congruent with those of earlier stud- ies,31,42,44 and indicated that HSC and MLP defined by a tran- scriptome signature were enriched in a phenotypically characterized CD34+CD38- population, while the other lin- eages comprised the more heterogeneous CD34+CD38+ population. We examined overlap of lncRNA and mRNA expression among lineages: 94.8% of mRNAs were shared by at least five out of six lineages, but only 62.2% of lncRNAs were so widely expressed (Figure 4D, top panel); conversely, 81.4% of lineage-signature mRNAs were spe- cific to only one lineage, while 92.2% of lncRNAs were equivalently specific (Figure 4D, bottom panel). Again, lncRNA expression appeared more lineage-restricted than did the counterpart coding gene expression. In summary, we found lncRNA expression to be highly stage- and line- age-specific during early hematopoiesis.
To confirm our findings of potential novel lncRNAs and lineage-specific expression patterns of lncRNAs, we com- pared our results with a publicly available dataset.44 This scRNA-seq study was conducted with human HSPCs sort- ed based on cell surface antigens (GSE75478). Lineage-spe- cific lncRNAs (and mRNAs) defined in the current study were also detected and showed consistent lineage-specific expression in the two datasets (Online Supplementary Results and Online Supplementary Figures S7 and S8). We then assessed 39 lncRNAs and 14 mRNAs by quantitative RT-PCR of aliquots of whole transcriptome amplification from those 391 single CD34+ cells and another set of flow cytometry-sorted bulk samples (Online Supplementary Methods and Results; Online Supplementary Table S6). All 39 signature lncRNAs, including 20 novel lncRNAs, were detectable in single cells and bulk samples by quantitative RT-PCR, indicating expression in human CD34+ cells. We confirmed cell type assignment of single cells by expres- sion of well-recognized mRNAs (Online Supplementary Figure S9C) and confirmed lineage-specific expression for 35 out of 39 lineage signature lncRNAs in single cells. Moreover, their lineage-specific expression patterns in sin- gle cells were reproducible in independent sorted bulk samples (Online Supplementary Figure S9A,B). Expression of these lineage-specific lncRNAs in hematopoietic differen- tiation, by scRNA-seq and quantitative RT-PCR, is illus- trated in Figure 4F.
898
haematologica | 2019; 104(5)