Mitra Mojtahedi's research while affiliated with The University of Calgary and other places

Single-cell gene expression analysis: PCA from Fig 1 recolored for origin in the Sca1 population fraction. Colors of cells again (as in Fig 1) indicate treatment, but in addition, their provenience from the respective Sca-1 fraction in the progenitor population (d0). Rebellious cells are cells that shift toward the non-intended fate, i.e. EPO-treated cells moving towards the destination region of the differentiated myeloid cells in the top/right region and IL-3/GM-CSF-treated cells moving toward the prospective destination region of the differentiated erythroid cells in the bottom left region (see day 6 in top panel). To illustrate that the rebellious are recruited from the respectively primed states, the Sca1-fraction (LOW, MID, HIGH) in the progenitor population from which the cells originated was indicated by the color hue (see legend on left) for the day 3 stage: the darker, the higher was d0-Sca1 expression. Note that the perspective is slightly shifted from that of Fig 1 to show that there is no clean separation at d3 into two disjoint cluster. As previously noted (see MAIN TEXT), the Sca1HIGH cells are primed towards the myeloid cells, whereas Sca1LOW cells are primed towards the erythroid cells. Note that although at this time point at d3 the cloud spreads and begins to split, and there is no separation between IL-3/GM-CSF and EPO treated cells. However, rebellious cells tend to originate from the Sca1-fraction for which they are primed. For instance, Sca1LOW cells in the progenitor population which are known to be primed towards erythroid fate [4], are the source of the rebellious cells which despite treatment with IL-3/GM-CSF move towards the erythroid direction at d3 (light blue cells). Thus, priming determines fate more than instruction by the external signal. (JPG)

Manually curated model of gene regulatory network governing fate decision of CMP. Network of experimentally verified regulatory interactions of transcription factors involved in multipotency of the CMP state, fate decision and differentiation to the erythroid and myeloid lineages (S1 Table). The canonical GATA1-PU.1 circuit is highlighted in green. A few surface markers including c-kit (progenitor, grey box), EpoR (erythroid, red box) and CD11b (myeloid, blue box) were included in the network to control the cell differentiation behavior and used as markers for lineage commitment in experiments. The numbers point to the row in the S1 Table that contains the references. (JPG)

Evaluation of qPCR assays. Table lists all primer pairs and relevant information including IDs and amplicon length. All assays were inventoried. Identical PCR primers were used in the pre-amplification step and the subsequent singleplex qPCR step. In addition, the amplification efficiency and limit of detection (LOD) of the qPCR assays are given. To evaluate efficiency and LOD, a 1:2 serial dilution was prepared from 18 cycles pre-amplified product from 10 ng RNA purified from EML progenitor cell population. Amplification efficiency was calculated according to: [10^(1/-S)-1] × 100%. The slope was obtained by linear regression of the standards curve. Efficiency was determined as average of two biological replicates with 6 qPCR technical replicates each. The Cq value for the LOD is defined as the most diluted sample that results in positive amplification for 5 out of 6 replicates. (JPG)

Representation of an OpenArray plate used for single-cell qPCR. (A) Each OpenArray (Applied Biosystems) is the size of a microscope slide. It holds 48 groups (subarrays, red rectangular) of 64 holes of 33 nl volume in which one PCR reaction occurs. A hydrophilic layer is at the interior surface of each hole and a hydrophobic layer is at the exterior surface of the plate allowing for filling the hole by surface tension. In total, each array carries 3072 qPCR reactions. (B) Specific PCR primers are pre-immobilized in individual holes (by manufacturer, for customized assay patterns) and released by heat in the first cycle. (C) An example of the distribution of single-cell samples (SC) along with NTC (no template water control), IRC (inter-run calibrator) and 100-cell control (PC) samples on an OpenArray chip. (JPG)

Distinct trajectories of cell differentiation are observed upon stimulation of progenitor cells with cytokines in the PCA state space. Principal component projections in a total of ~1600 haematopoietic cells including progenitor (black), single-EPO treated (red-shades), single-IL3/GM-CSF treated (blue-shades) and combined-treated (purple-shades) in the first three components determined from expression of all 17 transcription factors and endogenous control genes. (B) Principal component loadings for PC 2 and 3 indicate the extent to which each gene contributes to the separation of cells along each component. (C) PCA weights of genes for the first three PCs reveals the importance of the individual genes to explain the difference between the different treatments and corresponding cell fate. (D) Cells in their attractor states still exhibit heterogeneous transcription profiles that can be traced back to individual genes. Cells treated with GM-CSF/IL-3 for 6 days are clearly located within the state space defined by the myeloid genes and cells treated by EPO exhibit 2 clusters where the lower one is governed by erythroid genes and the higher one by stemness genes. (E) Variance explained by principal components show that the first three components jointly explain more than 70% of variation in the data. (JPG)

Flow cytometry data for Fig 4 of main text. FCS-format data files for individual histograms of the two time courses (untreated and IL-3/GM-GM-CSF treated) as shown in Fig 4. Each folder represents a time point for the untreated and the IL-3/GM-CSF treated cells and contains two files for the sample histograms for Sca1 and the PI stained cells, respectively for the given time point. The day 0 (d0) folders also contain the data for the isotype control (3 files). The day 1 (d1) folders also contains the pre-sorting samples for both PI and Sca1 stained cells (4 files). Acronyms in folder/filenames:—‘EML’ = untreated EML cells—‘IL-3+GM-CSF’ or ‘MYL’ = cells treated to induce myeloid commitment—‘d3’ or ‘D3’ = flow cytometry performed at day 3 after treatment, as explained in the text—‘PI’ = stained with Propidium iodide (dead cells)—‘SCA1’ = stained for SCA1 Surface expression—‘LOW’ = cells sorted from the SCA1LOW fraction at d1 and recultured—‘UNSORTD’ = non-sorted cells as control. (ZIP)

Regulatory interactions in the curated GRN model of binary fate decision in CMP. Table of the regulatory interactions (either activating (A) or inhibiting (I)) between the genes. For each interaction, the literature is referenced (numbered list in the right panel). All interactions have been reported in for murine hematopoiesis. (JPG)

Technical noise associated with single-cell RT-qPCR is significantly smaller than biological cell-cell variability. (A) Quantification cycles (Cq) of 80 individual EML cells for GATA1 expression is reported. Values are means ± STD for up to 128 technical replicates. (B) Quantification cycles (Cq) of up to 110 technical replicates are presented for 3 selected single-cells. Single-cell Cqs of biological samples clearly show a broader distribution relative to that of technical replicates. (C) Box plots represent the variability in terms of CV for technical replicates averaged over 110 realizations of the real-time PCR-steps on the ds-cDNA and the distribution of CV across all 80 individual EML progenitor cells for the GATA1 expression. The biological variation was significantly larger than the technical noise (p-value 2.2e-28, Mann-Whitney U test). Similar results were obtained for PU.1 (not shown). (JPG)

Quantified dissimilarity between transcriptomes from micro-arrays between samples. Pair-wise dissimilarity between expression profiles (samples) was calculated based on the normalized gene expression levels for 6297 filtered genes (see Material and Methods) with 1–R where R is the Pearson’s correlation coefficient which ranges from 0 to 1, meaning that 0 correspond to highest similarity and 1 to most different expression. Bootstrapping was performed by randomly selecting 30% of the genes in any sample to calculate the pair-wise dissimilarity metric and repeating the procedure 10,000 times to generate the reported standard deviations. (JPG)

Gene expression in individual cells from the progenitor population and the α, β, and γ subpopulations. (A-D) Heatmap representation of gene expression profiles for the set of 17 genes of the curated network and 2 endogenous genes as control in total 216 single cells including 72 progenitor cells (panel A) and 48 single cells from each of the three subpopulations in the tri-modal Sca-1 population distribution on day 3 after GM-CSF/IL-3 treatment (Fig 3 in main text), α (B) β (C) and γ (D). Genes are ordered according to their reported biological role, as erythroid-associated (red box), stemness (green box), myloid-associated (blue box) and endogenous genes in all subplots. Based on the expressed genes, the β subpopulation seems to be committed to the myeloid lineage while the γ subpopulation is committed to the erythroid lineage. The α subpopulation exhibits an indeterminacy with a bias towards the myeloid lineage. (E) PCA of all attractor cells (d0 and d6) as shown in the S4 Fig combined with the cells from the α (yellow), β (green), and γ (pink) subpopulations support the above described similarity to the untreated EML, the GM-CSF/IL-3 stimulated and the EPO-stimulated cells, respectively. (F) Coefficient of variation CV of expression levels of distinct genes is used as a cell-specific quantity to expose population dispersion and has no direct physical meaning; it was calculated for each cell from the expression levels across all genes for each subpopulation. Histograms represent the number of cells at different level of the CV measure and show that cells in α subpopulation have higher spread of cellular CV values. (JPG)