Transcriptome profiling and sequencing of differentiated human hematopoietic stem cells reveal lineage-specific expression and alternative splicing of genes

Abstract

Hematopoietic differentiation is strictly regulated by complex network of transcription factors that are controlled by ligands binding to cell surface receptors. Disruptions of the intricate sequences of transcriptional activation and suppression of multiple genes cause hematological diseases, such as leukemias, myelodysplastic syndromes, or myeloproliferative syndromes. From a clinical standpoint, deciphering the pattern of gene expression during hematopoiesis may help unravel disease-specific mechanisms in hematopoietic malignancies. Herein, we describe a human in vitro hematopoietic model system where lineage-specific differentiation of CD34⁺ cells was accomplished using specific cytokines. Microarray and RNAseq-based whole transcriptome and exome analysis was performed on the differentiated erythropoietic, granulopoietic, and megakaryopoietic cells to delineate changes in expression of whole transcripts and exons. Analysis on the Human 1.0 ST exon arrays indicated differential expression of 172 genes (P < 0.0000001) and significant alternate splicing of 86 genes during differentiation. Pathway analysis identified these genes to be involved in Rac/RhoA signaling, Wnt/B-catenin signaling and alanine/aspartate metabolism. Comparison of the microarray data to next generation RNAseq analysis during erythroid differentiation demonstrated a high degree of correlation in gene (R = 0.72) and exon (R = 0.62) expression. Our data provide a molecular portrait of events that regulate differentiation of hematopoietic cells. Knowledge of molecular processes by which the cells acquire their cell-specific fate would be beneficial in developing cell-based therapies for human diseases.

recent developments in stem cell biology have generated much excitement about the potential for regenerative medicine and cell-based therapies in a variety of clinical applications, such as treating Parkinson's disease, leukemia, and spinal cord injuries (23). Crucial to the success of these applications is the detailed understanding of how the cells remain stem cells and the cues that they require to differentiate and commit themselves to specific cell fates. Given that hematopoietic stem cells are a particularly interesting class of stem cells and a well-characterized cellular differentiation system, a number of studies have recently been undertaken to decipher their genetic program both in culture and in vivo (22).

Hematopoiesis is the process by which all the different cell lineages that form the blood and immune system are generated from a common pluripotent stem cell (28, 35, 36). A complex interplay between the intrinsic genetic processes of hematopoietic cells and their environment, including the effects of specific cytokines such as interleukins and granulocyte/monocyte stimulating factors, determines whether stem cells, lineage-specified progenitors, and mature blood cells self-renew, remain quiescent, proliferate, differentiate, or undergo apoptosis (1, 6–8, 11, 16–18, 24, 34, 37). Catastrophic consequences to aberrant hematopoiesis have been described in diseases such as leukemia, lymphoma, etc. (10, 12, 19, 21, 25, 26, 29). Hence, understanding the nature of the hematopoietic stem cells, as well as the molecular process by which these cells acquire their specific cell fate, is crucial for understanding disease pathogenesis and for the success of cell-based therapies.

As the phenotype of any given cell is ultimately the product of the genes, it is critical to identify gene expression patterns during lineage-specific differentiation. It is also believed that an important source of diversity in the transcriptome of differentiated cells is due to the splicing process in multiexon genes. Alternative splicing is thought to regulate differentiation through coordination of gene networks where each network coordinates a different cell function (4). Current studies of hematopoiesis have mostly examined gene level changes in expression and have not been extended to understand alternative splicing events.

The advancement of genomic technologies has now provided us a platform in the form of Genechip Human Exon 1.0 ST arrays and massively parallel sequencing to study the exome profile of cells. The current study exploits these advances in microarray and sequencing technologies for elucidating the global changes in the whole transcriptome and exome expression during ex vivo lineage-specific hematopoietic cell differentiation.

MATERIALS AND METHODS

Human Granulocyte Colony-stimulating Factor-mobilized CD34⁺ Peripheral Blood Cells

Human granulocyte colony-stimulating factor (G-CSF)-mobilized CD34⁺ peripheral blood cells (CD34⁺ PBCs) were collected by apheresis from healthy volunteers who were given 5 days of G-CSF (10 μg/kg per day). After CD34⁺ antigen-mediated selection with immunomagnetic beads (ISOLEX300i system; Baxter Healthcare, Deerfield, IL), purified CD34⁺ PBCs were collected and preserved in liquid nitrogen until use.

Suspension Cultures and Growth Factors

CD34⁺ PBCs were cultured in X-VIVO10 (BioWhittaker, Walkersville, MD) supplemented with 1% human serum albumin. At least 1 × 10⁶ CD34⁺ cells were assayed in six-well plates and incubated at 37°C and 5% CO₂ in a fully humidified atmosphere in air. Lineage-specific differentiation was induced on CD34⁺ cells using the method described in Komor et al. (15). In brief, lineage-specific differentiation was induced by the addition of growth factors (R&D Systems, Wiesbaden-Nordenstadt, Germany) stem cell factor (SCF, 50 ng/ml), Flt3-ligand (50 ng/ml), IL-3 (10 ng/ml), erythropoietin (10 U/ml) for erythropoietic differentiation; SCF (50 ng/ml), Flt3-ligand (50 ng/ml), IL-3 (10 ng/ml), G-CSF, and granulocyte/macrophage colony-stimulating factor (each, 10 ng/ml) for granulopoietic differentiation; SCF (50 ng/ml), Flt3-ligand (50 ng/ml), thrombopoietin (20 ng/ml) for megakaryopoietic differentiation. Differentiated cells were harvested on day 11 and purified by immunomagnetic beads using the MACS system using CD71⁺ microbeads for erythropoietic (E) group, CD15⁺ microbeads for granulopoietic (G) cells, and CD61⁺ microbeads for megakaryopoietic (M) cells (15).

Flow Cytometry

Uninduced CD34⁺ and cultured cells were characterized by dual-color immunofluorescence using a BD FACSCanto flow cytometer. E cells were characterized by staining with an anti-CD71 FITC antibody. Megakaryocytic cells were determined with an anti-CD61 FITC antibody, and G cells were analyzed with anti-CD15 FITC antibodies. Isotype-matched nonspecific antibodies were used as controls. Analysis gates were set to exclude dead cells and debris, with 10,000 viable cells analyzed per sample. Morphology of the flow-sorted differentiated cells was examined by Diff-Quik stain (Dade Behring, Newark, DE) following the manufacturer's protocols. Micrographs were taken with a Leica microscope at ×10 objective lens.

RNA Isolation

Total RNA was extracted using RNeasy mini kit (Qiagen, Valencia, CA) following the manufacturer's directions. Genomic DNA was removed by using the gDNA eliminator spin columns. The concentration of the isolated RNA was determined using the Nanodrop ND-100 spectrophotometer (Nanodrop Technologies, Wilmington, DE). Quality and integrity of the total RNA isolated were assessed on the Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA).

Target Preparation and Hybridization to Human Exon 1.0 ST Arrays

Labeling of samples for hybridization to the Exon array was performed according to Affymetrix GeneChip Whole Transcript Sense Target Labeling Assay. One microgram of total RNA was subjected to ribosomal RNA reduction following Invitrogen's ribominus reduction procedure. Double-stranded cDNA was synthesized using random hexamers incorporating a T7 promoter sequence to generate cRNA by in vitro transcription. Labeled single-stranded cDNA in the sense direction was generated from cRNA and used for array hybridization. Hybridization was performed at 45°C overnight, followed by washing and staining using FS450 Fluidics station. Scanning was carried out using the 7G GCS3000 scanner.

Microarray Data Collection and Annotation

Exon-level core robust multiarray average (RMA)-sketch intensity values for each of 12 chips were collected using Affymetrix Expression Console (EC) Software (Affymetrix, Santa Clara, CA). The 287,329 core probe-sets from EC were mapped to genomic address for human genome build hg18 using the Affymetrix probe-set selection region (downloaded from EC, March 2008). These probe-sets were then mapped to genes and exons using a reduced RefSeq gene model (Hg19, downloaded from http://genome.ucsc.edu). A gene model consists of all possible exons for a set of RefSeq transcripts for a single gene. The gene model is necessary to allow for testing of alternative splicing using a statistical approach, described below. Specifically, the reduced gene model consists of a set of nonoverlapping intervals that represent the possible exons within a gene. When more than one core probe-set maps to the same RefSeq exon interval, the average RMA intensity is calculated. This process yields 174,792 distinct exons grouped into 17,457 genes.

The data has been deposited in the Gene Expression Omnibus (GSE29989).

Principal Component Analysis

Data from 12 samples, after reduction to the RefSeq modeled genes and exons, were subjected to a principal component analysis (PCA) for detection of outliers. The four conditions (CD34⁺, E cells, granulocytes, megakaryocytes) were studied in triplicate. The bi-plot of PC1 vs. PC2, giving rise to 70% of the variability, revealed four distinct clusters, corresponding to the four conditions.

Analysis of Exon Arrays by ExonSVD

The Affymetrix Exon arrays offer the capability for alternative splicing detection, unlike older 3′ IVT arrays, because each exon of the gene is separately probed. The ExonANOVA, a three-factor, nested mixed-effect ANOVA, has been the widely used model for the analysis of exon arrays (http://www.partek.com, http://www.partek.com/html/products/pdf/Brochures/). The ExonANOVA model fits the following formula

$$y_{ijk}=\mu +A_i+A{C}_{ik}+C_k+{\beta }_{j\left(i\right)}+{\epsilon }_{ijk}$$

to the data.

The two fixed factors are the treatment effect (A) and exon effect (C). The random factor (β) is the sample within treatment effect. The fixed interaction between the treatment and exon effects (AC) determines if an alternative splicing event has occurred. This model makes a strong assumption of additivity, which may fail if the probe-set “sensitivity” is low and the data fall in the background range, or if the values become “saturated” at the high end of the data range. These two situations describe a dead or unresponsive probe-set, respectively, but there may be other causes of nonadditivity. Probe-sets exhibiting this behavior can strongly affect the alternative splicing signal and thus can falsely increase the number of detected alternative splicing events. Other assumptions required by the design of the ANOVA model are independence of cases and variance homogeneity. These last two assumptions are generally accounted for by proper experimental design and data transformations.

We introduce a new analytical method, termed the ExonSVD model (30), which aims to overcome the limitations inherent in the ExonANOVA. The ExonSVD model

$$y_{ijk}=\mu +A_i^{\prime }\ast D_k+E_{ik}+C_k+{\beta }_{j\left(i\right)}+{\epsilon }_{ijk}$$

has three new parameters compared with the ExonANOVA. The A′ and AC terms of the ExonANOVA model have been combined and then split into the three new parameters where A′ represents the treatment effect, D represents the probe sensitivity, and E represents the residual or deviation from the simple model. The E factor is tested for significance to determine if alternative splicing has occurred. This new model alleviates the need to detect and remove dead and unresponsive probe-sets. The P (P–E) values for the E term were generated by numerical simulation, fitting rational polynomials to the sum-of-squares curves, as estimates of degrees of freedom, so that one can generate a statistic with approximately an F distribution, and thereby obtain approximate P values.

Selection of Lineage-specific and Differentially Expressed Genes

Using the ExonSVD, differentially expressed genes were detected using the P value for the A′ parameter (differential expression) and a fold-change cutoff. P values ≤10^−7 and fold changes greater than two for any of the three comparisons (E vs. CD34⁺, M vs. CD34⁺, G vs. CD34⁺) were required. Additionally, a gene was defined to be “specifically upregulated” in a particular cell type (E, M, or G) compared with CD34⁺ if it was significant at P < 10^−4, and the largest observed change was in that cell type, and the largest change was greater than twofold, and neither of the other cell types showed an upward change >1.4-fold.

Identification of Alternatively Spliced Genes

Alternatively spliced genes were defined as having a P-E (P value for alternative splicing) of ≤10^−7 and a largest deviation of twofold or more where the largest deviation is max_i,k|E_i,k − E_CD34,k|, where i ranges over E, G, and M, and k ranges over the exons in a gene.

Next Generation Sequencing RNA Transcript Analysis on SOLiD Sequencer

Library preparation.

Total RNA (2 μg) from CD34⁺ and E cells were depleted of rRNA and enzymatically fragmented using 1 unit of RNase III (Ambion) by incubation at 37°C for 10 min. The fragmented RNA was size selected using the flashPAGE fractionator (Ambion) to collect RNA fragments ranging in size from ∼50 to 150 nucleotides in length. The RNA fragments were then ligated to adaptors, converted into cDNA, and amplified by 15 cycles of PCR using the SOLiD RNA Expression Kit (Ambion). The PCR reactions were purified using the Qiagen Mini elute PCR purification kit and separated on a native Novex 6% TBE polyacrylamide gel (Invitrogen). PCR products ranging in size from ∼150 to 200 bp (corresponding to RNA fragment insert sizes of ∼60–110 nucleotides) were cut out of the gel, and the products eluted overnight and precipitated. The gel-purified material was quantitated by Nanodrop and prepared for emulsion PCR and sequencing on an Applied Biosystems SOLiD sequencer (version 3.0; Applied Biosystems, Carlsbad, CA).

Sequence read processing and alignment.

mRNA-Seq sequencing reads were analyzed using Applied Biosystems' whole transcriptome software tools (http://www.solidsoftwaretools.com/). Reads of length 50 bases originating from each sample were first aligned to the human genome (US National Center for Biotechnology Information Build 36.3) using Applied Biosystems' SOLiD System Analysis Pipeline Tool (Bioscope). The aligned reads were mapped to RefSeq exons downloaded from the UCSC Genome Browser (Human genome build hg18), and reads per kilobase per million reads (RPKM) values were obtained for each RefSeq exon. The RPKM calculations were adapted at the exon, gene, and transcript level and can be thought of as a normalized expression level (E) based on the read count across the region of interest.

The calculation for RPKM is as follows: RPKM = 10^9 * C/NL, where C = counts or number of reads falling in the exon, N = total mapped reads, and L = length of the transcript.

RPKM values for each of two CD34⁺ (CD34⁺_1, CD34⁺_2) and two erythropoietic (E_1, E_2) samples were obtained for further analysis with the microarray data.

RNAseq and Microarray Data Correlation Analysis

To compare the RNAseq data to the microarray data at the exon level, it was necessary to match each RefSeq exon RPKM value with the corresponding reduced model exon microarray RMA value (see materials and methods). A total of 149,765 RNAseqmicroarray value pairs were obtained. Correlations between the two methods on exon-level fold changes were computed. Gene-level fold changes were determined by averaging the log fold-change values across exons of each gene.

QPCR Analysis to Determine Gene Expression Values

First-strand cDNA was synthesized using 500 ng of RNA and random primers in a 20 μl reverse transcriptase reaction mixture using Invitrogen's Superscript cDNA synthesis kit (Invitrogen, Carlsbad, CA) following the manufacturer's directions. Quantitative real-time PCR assays were carried out with the use of gene-specific double fluorescently labeled probes in a 7900 Sequence Detector (PE Applied Biosystems, Norwalk, CT). In brief, PCR amplification was performed in a 384-well plate with a 20-μl reaction mixture containing 300 nm of each primer, 200 nm probe, 200 nm dNTP in 1× real-time PCR buffer and passive reference (ROX) fluorochrome. The thermal cycling conditions were 2 min at 50°C and 10 min at 95°C, followed by 40 cycles of 15 denaturation at 95°C and 1 min annealing and extension at 60°C. Samples were analyzed in duplicate and the C_T values obtained were normalized to the housekeeping gene β-actin. The comparative C_T (ΔΔC_T) method (33), which compares the differences in C_T values between groups, was used to achieve the relative fold change in gene expression between the four groups in the study.

RESULTS

In Vitro Differentiation of CD 34⁺ Cells

Lineage-specific cells were characterized by immunophenotyping as shown in Fig. 1A. The dot plots illustrate the expression of lineage-specific cell surface markers that define the undifferentiated CD34⁺ and differentiated E, G, and M cells, respectively. Figure 1A, top left, shows the unstained CD34⁺ cells on day 11 as a control for panels at top right and bottom with erythroid, granulocytic, and megakaryocytic cells. After 11 days in culture, out of the stained cells, 98% stained positive for CD71 representing erythropoietic differentiation, 98% stained positive for CD15 representing granulopoiesis, and 78% stained positive for CD61 representing megakaryopoietic differentiation. Flow-sorted differentiated cells were analyzed for cell morphology using the Diff-Quik stain set. Figure 1B shows the micrographs taken with a Leica microscope at ×10 objective lens, thereby confirming the purity of the cells selected for transcriptome analysis.

Fig. 1.

A: flow cytometry characterization of CD34⁺ and differentiated erythropoietic (E), granulopoietic (G), and megakaryopoietic (M) cells by dual-color immunofluorescence using a BD FACSCanto flow cytometer. Unstained CD34⁺ cells (top left); E cells stained with anti-CD71 FITC antibody (top right); G cells stained with anti-CD15 FITC antibody (bottom left); M cells stained with anti-CD61 FITC antibody (bottom right). B: morphology of differentiated CD34⁺ cells into E cells (a), G cells (b), and M cells (c).

QPCR Analysis to Determine the Expression Levels of Few Lineage-specific Genes

QPCR was carried out on undifferentiated CD34⁺ and purified differentiated cells (E, M, and G) to determine the expression levels of genes known to be specific for each of the cell type. Table 1 illustrates the fold change in expression levels of few genes ANK1, GYPA (specific for E cells); FCGRA, MIP2 (specific for G cells); and GPIBA and PF4 (specific for M cells). We observed a significant increase in the expression of genes that are specific for each of the cell type confirming the cytokine-induced differentiation of CD34⁺ cells to their specific lineages. The E group of cells showed a significant increase in the expression of ANK1 and GYPA compared with the CD34⁺. Genes known to be specific or abundant for M and G groups, such as FCGR2A and MIP2 for G group and GPIBA and PF4 for M group, showed significant increase in their expression in the respective cell types confirming the lineage-specific differentiation.

Table 1.

Gene expression and QPCR expression fold changes for some lineage-specific genes in hematopoietic differentiation

Symbol	Gene Title	FC ± SD
E-specific genes		E vs. CD34
ANK1	ankyrin 1	8,192.44 ± 204.81
GYPA	glycophorin A	78,612.44 ± 2,358.88
G-specific genes		G vs. CD34
FCGR2A	Fc fragment of IgG, low affinity receptor 2A	4.79 ± 0.09
MIP2	macrophage inflammatory protein 2	14.24 ± 0.23
M-specific genes		M vs. CD34
GPIBA	glycoprotein 1B-alpha	19.16 ± 0.38
PF4	platelet Factor 4	44.46 ± 0.88

Fold changes (FC) were calculated by comparing the ΔCT values of differentiated group to the ΔCT values of undifferentiated CD34. Values are given as means ± SD (n = 3) in each group.

Abbreviations: E, erythropoietic cells; G, granulopoietic cells; M, megakaryopoietic cells.

Microarray-based Confirmation of Differentiated Cells by Lineage-specific Gene Expression

PCA was first used to identify outliers within groups and to characterize the stem cells and differentiated cells based on their expression profile on microarrays. Principal component 1 (PC1) vs. principal component 2 (PC2) as plotted in Fig. 2A showed a clear segregation and clustering between the four groups of cells. We observed 60% variability in PC1, which accounted for the difference between the M cells and the other three cell types.

Fig. 2.

A: bi-plot of the first 2 principal components showing 82% of the variability within the data [principal component 1 (PC1, x-axis), principal component (PC2, y-axis)]. The bi-plot shows a clear separation of the 4 differentiated cell types. Most of the variability in the data represented by PC1 (59.8%) shows megakaryocytes separated from all other cell types. Principal component analysis calculated from averaged RefSeq exons robust multiarray average (RMA) values by RefSeq gene symbol. This ellipse is drawn around the grouping variable (i.e., cell type C-CD34⁺, E, G, and M) specified for each sample of the principal components. The ellipsoid is computed from the bivariate normal distribution fit to the X and Y variables in JMP statistical software (Cary, NC). The bivariate normal density is a function of the means and standard deviations of the X and Y variables, PC 1 and PC2, and the correlation between the 2 is calculated. B: confirmation of the in vitro differentiation of CD34⁺ cells by microarray analysis. During lineage-specific differentiation, the expression of well-known marker genes for each hematopoietic lineage was analyzed and the relative increase in expression compared with CD34⁺ was determined. x-Axis denotes the fold increase in expression of genes for the differentiated E, G, and M groups. y-Axis denotes the lineage-specific genes represented as gene symbol. The genes analyzed include erythropoietic-specific genes erythrocytic Ankyrin 1 (ANK1), Glycophorin A (GYPA), and Tropomodulin (TMOD1); granulopoietic-specific genes FCGR2A and FCGRA2B, Fc fragment of immunoglobulin G, low-affinity IIA and IIb, Cluster of Differentiation 300A (CD300A), and Macrophage Inflammatory Protein 2 (MIP2); megakaryopoietic-specific genes serine proteinase inhibitor, clade E (SERPINE1), Glycoprotein 1B alpha (GPIBA), Platelet factor 4 (PF4), and Prostaglandin-endoperoxide synthase 1 (PTGS1).

In an effort to confirm established lineage-specific differentiation of CD34⁺ cells, we examined the microarray-based expression of differentiated cell-specific genes (Fig. 2B). As depicted in the figure, when CD34⁺ cells differentiate into erythrocytic cells, a 2- to 14-fold (log) increase in the expression of ankyrin, glycophorin A, and tropomodulin 1 (ANK1, GYPA, TMOD1) was observed, which code for erythrocyte membrane proteins. Similarly, the granulocyte-specific genes CD300A and MIP2 and low-affinity immunoglobulin gamma FC region receptor II-A and B proteins (CD300A, FCGR2 A and B) showed the highest fold change in G cells. Finally, megakaryocyte-specific genes prostaglandin-endoperoxide synthase 1, platelet factor 4, glycoprotein 1B alpha and serine (or cysteine) proteinase inhibitor, clade E (PF4, GPIBA, PTGS1, SERPINE1) had highest expression in megakaryocytic cells confirming the lineage specific differentiation of CD34⁺ cells.

Gene Level Analysis to Identify Differentially Expressed Transcripts During Hematopoietic Differentiation

Global gene-level analysis of transcripts comparing CD 34⁺ progenitor cells to the three differentiated lineages identified several genes with significant differences in expression between each of the differentiated cell groups. As mentioned in materials and methods, statistical filters were applied to find a total of 172 differentially expressed genes between any of the three cell types compared with CD34 progenitor cells. Selecting at a P-A ≤ 1e−7 and a twofold or more change, the following lists were generated: MvsCD34, 148 genes; GvsCD34, 52 genes; and EvsCD34, 70 genes. These lists show some overlap across the groups. Hierarchical clustering showed the differential expression pattern between these 172 transcripts across the lineage-specific stem cell types as depicted in Fig. 3. Comparison of these significantly altered transcripts by Venn diagram as shown in Fig. 4A shows that 26 transcripts are commonly differentially expressed in all three differentiated groups. Furthermore, 24 transcripts are found to be common to both the M and E groups, 14 are common between G and E, and four transcripts are found to be common between G and M. Six of the top ranking genes are a unique signature for erythrocyte differentiation, including SLC4A1, ALAS, EPPB9, HBB, SELENBP1, and GYPA. The unique granulocyte differentiation signature has only four genes including C20orf12, C19orf16, PRGC, and CEACAM6, while alterations in 90 transcripts appear to be unique for megakaryopoiesis. Table 2 lists all 172 transcripts that are common and unique to each cell lineage.

Fig. 3.

Hierarchical cluster analysis performed on the profiles of 172 differentially expressed genes. Genes were declared to be differentially expressed by comparing 3 lineage-specific cell types M, G, E to CD34⁺ stem cells and requiring >2-fold change and P value <10^7. The cluster shows E, G, and M relative to CD34⁺.

Fig. 4.

A: Venn analysis of the overlap between the 172 differentially selected genes based on P value and fold-change cutoff between M, G, and E groups relative to CD34⁺ (See Fig. 3). B: parallel plots of 331 specifically upregulated genes (See materials and methods). Specifically upregulated genes are those whose upregulation is >2-fold in 1 lineage but unchanged or upregulated <1.4-fold in the other 2 lineages. Top: upregulated transcripts specific for E cells relative to CD34⁺. Middle: upregulated transcripts specific for G cells relative to CD34⁺. Bottom: upregulated transcripts specific for M cells relative to CD34⁺.

Table 2.

List of differentially expressed genes during lineage-specific differentiation of hematopoietic stem cells

Gene Symbol	Gene Title	E vs. CD34	G vs. CD34	M vs. CD34	Groups
ARHGEF17	Rho guanine nucleotide exchange factor (GEF) 17	−2.38	−1.79	−2.41	E, G, M
CD177	CD177 molecule	−2.79	−2.71	−2.73	E, G, M
CDT1	chromatin licensing and DNA replication factor 1	1.50	1.46	−1.07	E, G, M
CMTM5	CKLF-like MARVEL transmembrane domain containing 5	3.30	1.64	5.79	E, G, M
DNTT	deoxynucleotidyltransferase, terminal	−3.94	−3.88	−4.12	E, G, M
FCGR3A	Fc fragment of IgG, low affinity IIIa, receptor (CD16a)	−2.71	−2.84	−2.77	E, G, M
GPSM1	G protein signaling modulator 1 (AGS3-like, C. elegans)	−1.40	−1.33	−2.52	E, G, M
HBG1	hemoglobin, gamma A	6.37	2.73	1.77	E, G, M
HBG2	hemoglobin, gamma G	7.06	3.84	2.62	E, G, M
ID1	inhibitor of DNA binding 1	−2.11	−2.10	−3.18	E, G, M
ITGB3	integrin, beta 3 (platelet glycoprotein IIIa, antigen CD61)	3.49	1.58	7.26	E, G, M
KCNA3	potassium voltage-gated channel, member 3	−2.03	−2.05	2.46	E, G, M
LAT	linker for activation of T cells	3.50	2.01	4.58	E, G, M
LSP1	lymphocyte-specific protein 1	−2.46	−1.72	−3.89	E, G, M
MDK	midkine (neurite growth-promoting factor 2)	−2.06	−1.98	−2.65	E, G, M
MGC29671		−1.12	1.49	−2.69	E, G, M
MGC35402		−3.03	−2.93	−2.81	E, G, M
MN1	meningioma (disrupted in balanced translocation) 1	−2.93	−2.78	−3.15	E, G, M
NPTX2	neuronal pentraxin II	−1.55	−1.94	−1.67	E, G, M
P2RY11	purinergic receptor P2Y, G-protein coupled, 11	−1.47	−1.47	−2.87	E, G, M
PCDH21	protocadherin 21	1.78	1.52	4.51	E, G, M
RNASE2	ribonuclease, RNase A family, 2	3.01	4.56	−1.54	E, G, M
S100A12	S100 calcium binding protein A12	−4.64	−4.53	−4.92	E, G, M
SH2D3C	SH2 domain containing 3C	−1.70	−1.19	−2.50	E, G, M
SLC11A1	solute carrier family 11	−2.45	−2.55	−2.75	E, G, M
UBE2C	ubiquitin-conjugating enzyme E2C	3.66	3.23	3.19	E, G, M
SLC4A1	solute carrier family 4, anion exchanger, member 1	3.42	−0.22	−0.56	E
ALAS2	aminolevulinate, delta-, synthase 2	3.26	0.41	−0.32	E
EPPB9		1.06	0.94	−0.31	E
HBB	hemoglobin, beta	3.66	0.76	−0.20	E
SELENBP1	selenium binding protein 1	2.13	0.36	−0.09	E
GYPA	glycophorin A (MNS blood group)	3.71	0.76	0.00	E
PRKG1	protein kinase, cGMP-dependent, type I	−1.84	−2.15	−0.82	E, G
CA1	carbonic anhydrase I	4.39	1.40	−0.48	E, G
H2AFX	H2A histone family, member X	1.59	1.18	−0.43	E, G
ERAF	erythroid associated factor	4.51	1.07	−0.08	E, G
ELA2		1.19	5.17	−0.05	E, G
PRTN3	proteinase 3	1.17	4.69	0.13	E, G
PRG1		1.14	2.65	0.49	E, G
IL9R	interleukin 9 receptor	3.00	2.24	0.69	E, G
NFKBIA	nuclear factor of kappa light polypeptide gene enhancer in B-cells inhibitor, alpha	−2.18	−1.52	0.75	E, G
MYL4	myosin, light chain 4, alkali; atrial, embryonic	3.30	1.12	0.80	E, G
CST7	cystatin F (leukocystatin)	1.46	4.08	0.83	E, G
EBP	emopamil binding protein (sterol isomerase)	2.08	2.44	0.94	E, G
H3/o		4.21	3.69	0.99	E, G
HIST2H3C	histone cluster 2, H3c	4.21	3.69	0.99	E, G
LITAF	lipopolysaccharide-induced TNF factor	−1.21	−0.11	−4.20	E, M
LAIR1	leukocyte-associated immunoglobulin-like receptor 1	−1.56	0.15	−3.97	E, M
TNFRSF1B	tumor necrosis factor receptor superfamily, member 1B	−1.67	−0.80	−3.65	E, M
CECR1	cat eye syndrome chromosome region, candidate 1	−1.95	0.51	−3.19	E, M
PRAM1	PML-RARA regulated adaptor molecule 1	−1.85	0.07	−2.69	E, M
KLF4	Kruppel-like factor 4 (gut)	−1.72	−0.86	−2.28	E, M
PTPRCAP	protein tyrosine phosphatase, receptor type, C-associated protein	−1.41	−0.67	−2.02	E, M
BEX2	brain expressed X-linked 2	−1.25	0.18	−1.98	E, M
HDC	histidine decarboxylase	1.45	0.93	−1.98	E, M
HBA2	hemoglobin, alpha 2	2.16	−0.78	−1.58	E, M
HBA1	hemoglobin, alpha 1	2.04	−0.70	−1.41	E, M
HIST1H2BL	histone cluster 1, H2bl	1.04	0.64	−1.09	E, M
FAM89A	family with sequence similarity 89, member A	1.41	0.59	−1.07	E, M
PPP1R3B	protein phosphatase 1, regulatory (inhibitor) subunit 3B	−1.35	−0.67	1.37	E, M
PPP1R14A	protein phosphatase 1, regulatory (inhibitor) subunit 14A	2.00	0.59	1.89	E, M
GATA1	GATA binding protein 1 (globin transcription factor 1)	2.31	0.62	2.30	E, M
CCND3	cyclin D3	1.04	0.35	2.33	E, M
THBS1	thrombospondin 1	1.44	−0.74	4.65	E, M
PF4	platelet factor 4	1.28	0.16	4.72	E, M
ITGA2B	integrin, alpha 2b	2.83	0.99	5.14	E, M
ARHGAP6	Rho GTPase activating protein 6	2.11	0.63	5.22	E, M
CTTN	cortactin	1.52	0.36	5.29	E, M
PPBP	proplatelet basic protein (chemokine (C-X-C motif) ligand 7)	1.24	0.07	5.67	E, M
RGS6	regulator of G protein signaling 6	1.61	0.38	6.27	E, M
C20orf112	chromosome 20 open reading frame 112	−0.75	−1.32	−0.70	G
C19orf10	chromosome 19 open reading frame 10	0.91	1.85	−0.44	G
PRG2	proteoglycan 2, bone marrow	0.40	2.15	−0.24	G
CEACAM6	carcinoembryonic antigen-related cell adhesion molecule 6	0.42	4.00	−0.11	G
FLJ11151		−0.89	−1.09	−2.82	G, M
NKG7	natural killer cell group 7 sequence	−0.81	2.01	−2.31	G, M
IGLL1	immunoglobulin lambda-like polypeptide 1	0.75	1.99	−2.13	G, M
CEBPA	CCAAT/enhancer binding protein (C/EBP), alpha	0.22	1.25	−1.59	G, M
TRIM58	tripartite motif-containing 58	0.05	−1.06	1.87	G, M
SEPT5	septin 5	−0.55	−1.83	3.14	G, M
ADCY6	adenylate cyclase 6	−0.24	−1.24	3.70	G, M
KIAA0513	KIAA0513	−0.01	−1.11	4.23	G, M
ZFP36L2	zinc finger protein 36, C3H type-like 2	−0.97	−0.87	−3.22	M
PRSSL1	protease, serine-like 1	−0.77	0.76	−3.18	M
SPI1	spleen focus forming virus (SFFV) proviral integration oncogene	−0.95	0.20	−2.68	M
TRIM14	tripartite motif-containing 14	−0.53	0.25	−2.61	M
UNQ501		−0.35	0.44	−2.34	M
GSTP1	glutathione S-transferase pi 1	−0.18	0.42	−2.19	M
RAB3D	RAB3D, member RAS oncogene family	−0.90	−0.32	−2.09	M
TIMM13	translocase of inner mitochondrial membrane 13	0.15	−0.04	−2.02	M
ITPA	inosine triphosphatase	0.24	0.14	−1.96	M
IFITM1	interferon induced transmembrane protein 1 (9–27)	0.35	0.14	−1.96	M
NHP2L1	NHP2 nonhistone chromosome protein 2-like 1	0.41	0.49	−1.93	M
ASB13	ankyrin repeat and SOCS box-containing 13	0.27	0.73	−1.89	M
SYNGR1	synaptogyrin 1	0.42	0.53	−1.83	M
ARL6IP4	ADP-ribosylation-like factor 6 interacting protein 4	0.26	0.25	−1.72	M
POFUT1	protein O-fucosyltransferase 1	0.55	0.68	−1.66	M
ZXDB	zinc finger, X-linked, duplicated B	−0.88	−0.79	−1.59	M
CEBPD	CCAAT/enhancer binding protein (C/EBP), delta	−0.59	0.89	−1.54	M
UBE2L6	ubiquitin-conjugating enzyme E2L 6	−0.16	−0.32	−1.49	M
SLIC1		−0.83	−0.41	−1.46	M
TSNARE1	t-SNARE domain containing 1	−0.49	−0.25	−1.42	M
ECHS1	enoyl Coenzyme A hydratase, short chain, 1, mitochondrial	0.54	0.59	−1.41	M
RPUSD1	RNA pseudouridylate synthase domain containing 1	0.08	0.12	−1.37	M
MRPS16	mitochondrial ribosomal protein S16	0.45	0.50	−1.35	M
CUEDC2	CUE domain containing 2	0.06	0.21	−1.28	M
PDXP	pyridoxal (pyridoxine, vitamin B6) phosphatase	0.54	0.24	−1.24	M
CKAP4	cytoskeleton-associated protein 4	0.78	0.83	−1.22	M
STK32C	serine/threonine kinase 32C	0.56	0.24	−1.21	M
ENDOG	endonuclease G	0.05	−0.09	−1.20	M
NDUFB7	NADH dehydrogenase (ubiquinone) 1 beta subcomplex, 7	0.36	0.38	−1.16	M
PALM	paralemmin	−0.14	0.26	−1.15	M
DCI	dodecenoyl-Coenzyme A delta isomerase	0.54	0.49	−1.13	M
ATP5D	ATP synthase, H+ transporting, delta subunit	0.47	0.44	−1.11	M
ABHD14A	abhydrolase domain containing 14A	0.36	0.73	−1.08	M
LOC92154		−0.16	−0.42	1.35	M
SNN	stannin	−0.70	−0.47	1.37	M
BCL2L2	BCL2-like 2	−0.32	−0.34	1.37	M
LIPC	lipase, hepatic	0.16	0.19	1.52	M
IRS2	insulin receptor substrate 2	−0.48	0.44	1.60	M
GLYATL1	glycine-N-acyltransferase-like 1	−0.14	−0.03	1.66	M
RGS10	regulator of G protein signaling 10	0.52	0.27	1.97	M
SRC	v-src sarcoma (Schmidt-Ruppin A-2) viral oncogene homolog (avian)	−0.74	−0.85	2.00	M
CDKN1A	cyclin-dependent kinase inhibitor 1A (p21, Cip1)	−0.63	−0.38	2.32	M
TULP2	tubby like protein 2	0.21	0.31	2.37	M
ADRA2A	adrenergic, alpha-2A, receptor	0.73	−0.47	2.37	M
MGC13057		0.94	−0.31	2.38	M
TAL1	T-cell acute lymphocytic leukemia 1	0.62	−0.78	2.38	M
C20orf175		0.45	−0.89	2.48	M
HTR2A	5-hydroxytryptamine (serotonin) receptor 2A	0.10	−0.08	2.52	M
NRGN	neurogranin (protein kinase C substrate, RC3)	0.57	0.18	2.53	M
BCL2L1	BCL2-like 1	0.69	−0.34	2.53	M
SEC14L5	SEC14-like 5 (S. cerevisiae)	0.15	0.16	2.57	M
TPM1	tropomyosin 1 (alpha)	0.78	−0.56	2.71	M
C20orf32		0.14	−0.21	2.75	M
C6orf188		−0.07	0.08	2.83	M
HEXIM1	hexamethylene bis-acetamide inducible 1	0.07	−0.12	2.95	M
ENDOD1	endonuclease domain containing 1	0.65	−0.57	3.00	M
GP5	glycoprotein V (platelet)	0.22	−0.01	3.05	M
PLA2G4C	phospholipase A2, group IVC (cytosolic, calcium-independent)	0.03	0.22	3.10	M
TUBA8	tubulin, alpha 8	0.97	0.13	3.10	M
IL21R	interleukin 21 receptor	0.90	0.08	3.16	M
PANX1	pannexin 1	0.57	0.06	3.22	M
C5orf4	chromosome 5 open reading frame 4	0.58	−0.45	3.29	M
PF4V1	platelet factor 4 variant 1	0.91	0.19	3.36	M
SERPINE1	serpin peptidase inhibitor, clade E	0.46	0.02	3.38	M
CXCL3	chemokine (C-X-C motif) ligand 3	0.16	0.16	3.52	M
SLC22A17	solute carrier family 22, member 17	0.69	−0.54	3.59	M
PDGFB	platelet-derived growth factor beta polypeptide	0.05	0.29	3.63	M
F2RL2	coagulation factor II (thrombin) receptor-like 2	0.41	−0.28	3.65	M
MFAP3L	microfibrillar-associated protein 3-like	0.16	−0.08	3.66	M
SDPR	serum deprivation response	0.72	−0.40	3.66	M
EGLN3	egl nine homolog 3 (C. elegans)	0.88	0.14	3.86	M
CD9	CD9 molecule	−0.09	−0.73	3.87	M
GABRE	gamma-aminobutyric acid (GABA) A receptor, epsilon	0.24	0.36	3.96	M
ESAM	endothelial cell adhesion molecule	0.81	−0.44	4.22	M
CCR4	chemokine (C-C motif) receptor 4	0.85	0.26	4.32	M
GNAZ	guanine nucleotide binding protein (G protein), alpha z polypeptide	0.37	−0.01	4.52	M
GP9	glycoprotein IX (platelet)	0.81	−0.06	4.55	M
LRRC32	leucine rich repeat containing 32	0.06	−0.49	4.56	M
C6orf21		0.82	0.10	4.82	M
TRPC6	transient receptor potential cation channel, subfamily C, member 6	0.88	0.12	4.84	M
TSPAN9	tetraspanin 9	0.50	0.26	4.88	M
PCSK6	proprotein convertase subtilisin/kexin type 6	0.30	0.19	4.97	M
PDE5A	phosphodiesterase 5A, cGMP-specific	0.10	−0.06	5.10	M
ABCC3	ATP-binding cassette, subfamily C (CFTR/MRP), member 3	−0.08	−0.30	5.25	M
ITGB5	integrin, beta 5	0.47	0.06	5.40	M
HPSE	heparanase	0.62	−0.32	5.69	M
CD226	CD226 molecule	0.10	−0.24	5.75	M
VWF	von Willebrand factor	0.44	−0.47	5.98	M
PDE3A	phosphodiesterase 3A, cGMP-inhibited	0.63	−0.11	6.18	M
CLEC1B	C-type lectin domain family 1, member B	0.66	0.67	6.62	M

FC values are given as log 2 means from an n of 3 per group. P < 0.0000001.

Identification of Differentially Regulated Genes During Lineage-specific Differentiation to E, G, and M Cells

In an effort to identify transcripts that are altered during cell differentiation, we specifically examined “upregulated” genes (see materials and methods) compared with CD34⁺ cells. We identified three patterns (Fig. 4B) of upregulation. Erythropoietic differentiation resulted in upregulation of 32 transcripts. When CD 34 cells differentiated into G cells, 30 genes were found to be significantly upregulated, and megakaryopoietic differentiation resulted in a much larger modulation of genes as identified by the upregulation of 269 transcripts. A partial list of these genes are tabulated in Tables 3, 4, and 5. (A complete list of these altered genes is shown as Supplemental Table S1).¹

Table 3.

Expression of genes highly specific for E cells

Gene Symbol	Gene Title	FC - E vs. CD34
TMEM56	transmembrane protein 56	3.60
SLC4A1	solute carrier family 4, anion exchanger, member 1	3.42
ALAS2	aminolevulinate, delta-, synthase 2	3.26
IRF6	interferon regulatory factor 6	2.51
DNAJA4	DnaJ (Hsp40) homolog, subfamily A, member 4	2.49
ADD2	adducin 2 (beta)	2.47
GYPB	glycophorin B (MNS blood group)	2.36
GALNT5	UDP-N-acetyl-alpha-d-galactosamine:polypeptide5	2.21
HBA2	hemoglobin, alpha 2	2.16
SELENBP1	selenium binding protein 1	2.13
HBA1	hemoglobin, alpha 1	2.04
ATP7B	ATPase, Cu2+ transporting, beta polypeptide	1.95
SLC14A1	solute carrier family 14 (urea transporter), member 1	1.85
LBH	limb bud and heart development homolog	1.83
GYPE	glycophorin E	1.81
SEC14L4	SEC14-like 4	1.73
ACSBG1	acyl-CoA synthetase bubblegum family member 1	1.67
PCSK9	proprotein convertase subtilisin/kexin type 9	1.60
TGM2	transglutaminase 2	1.55
SLC25A21	solute carrier family 25	1.42
ITGB4	integrin, beta 4	1.30
LEF1	lymphoid enhancer-binding factor 1	1.29
PISD	phosphatidylserine decarboxylase	1.18
NMNAT3	nicotinamide nucleotide adenylyltransferase 3	1.14
ICAM4	intercellular adhesion molecule 4	1.08
TRIB2	tribbles homolog 2	1.05
S100A4	S100 calcium binding protein A4	1.04
TRIM29	tripartite motif-containing 29	1.03
ZYX	zyxin	−1.07
BTG1	B-cell translocation gene 1, anti-proliferative	−1.08
ZBTB34	zinc finger and BTB domain containing 34	−1.10
TCEA2	transcription elongation factor A (SII), 2	−1.18
CXCR4	chemokine (C-X-C motif) receptor 4	−1.21
TSC22D3	TSC22 domain family, member 3	−1.37

List of genes highly specific for E cells. FC values are given as log 2 means for an n of 3 per group.

Table 4.

Expression of genes highly specific for G cells

Gene Symbol	Gene Title	FC - G vs. CD34
BPI	bactericidal/permeability-increasing protein	5.23
CEACAM6	carcinoembryonic antigen-related cell adhesion molecule 6	4.00
CD24	CD24 molecule	3.67
CLEC5A	C-type lectin domain family 5, member A	2.88
IL2RA	interleukin 2 receptor, alpha	2.58
BEX1	brain expressed, X-linked 1	2.42
FAM107B	family with sequence similarity 107, member B	2.17
PRG2	plasticity-related gene 2	2.15
PRG2	proteoglycan 2, bone marrow	2.15
ELOVL3	elongation of very long chain fatty acids	2.10
SLPI	secretory leukocyte peptidase inhibitor	2.04
NKG7	natural killer cell group 7 sequence	2.01
GLT25D2	glycosyltransferase 25 domain containing 2	1.84
RHOU	ras homolog gene family, member U	1.75
P2RY2	purinergic receptor P2Y, G protein-coupled, 2	1.75
GPR97	G protein-coupled receptor 97	1.72
JDP2	Jun dimerization protein 2	1.68
RBP4	retinol binding protein 4, plasma	1.68
SERPINB2	serpin peptidase inhibitor, clade B, member 2	1.63
CD48	CD48 molecule	1.56
CDA	cytidine deaminase	1.47
HIP1	huntingtin interacting protein 1	1.47
LPO	lactoperoxidase	1.45
ANPEP	alanyl (membrane) aminopeptidase	1.37
CILP2	cartilage intermediate layer protein 2	1.36
LYZ	lysozyme (renal amyloidosis)	1.31
NT5DC3	5′-nucleotidase domain containing 3	1.27
CLEC11A	C-type lectin domain family 11, member A	1.23
SERPINB8	serpin peptidase inhibitor, clade B, member 8	1.21
FGFR1	fibroblast growth factor receptor 1	1.20
PADI2	peptidyl arginine deiminase, type II	1.16
DERL3	Der1-like domain family, member 3	1.16
MFAP4	microfibrillar-associated protein 4	1.12
FUT4	fucosyltransferase 4 alpha (1,3) fucosyltransferase	1.05
GRAMD1B	GRAM domain containing 1B	1.01
CBFA2T3	core-binding factor, runt domain, alpha subunit 2;	−1.03
ARHGEF12	Rho guanine nucleotide exchange factor (GEF) 12	−1.10

List of genes highly specific for G cells. FC values are given as log 2 means for an n of 3 per group.

Table 5.

Expression of genes highly specific for M cells

Gene Symbol	Gene Title	FC - M vs. CD34
VWF	von Willebrand factor	5.98
CD226	CD226 molecule	5.75
PKHD1L1	polycystic kidney and hepatic disease 1	5.49
ITGB5	integrin, beta 5	5.40
ABCC3	ATP-binding cassette, subfamily C	5.25
PDE5A	phosphodiesterase 5A, cGMP-specific	5.10
PCSK6	proprotein convertase subtilisin/kexin type 6	4.97
TUBB1	tubulin, beta 1	4.93
TSPAN9	tetraspanin 9	4.88
LRRC32	leucine rich repeat containing 32	4.56
GNAZ	guanine nucleotide binding protein (G protein)	4.52
PTPRJ	protein tyrosine phosphatase, receptor type, J	4.47
MMD	monocyte to macrophage differentiation-associated	4.45
EGF	epidermal growth factor (beta-urogastrone)	4.43
VEGFC	vascular endothelial growth factor C	4.42
MYOM1	myomesin 1, 185 kDa	4.37
KIAA0513	KIAA0513	4.23
TMEM40	transmembrane protein 40	4.13
GP1BA	glycoprotein Ib (platelet), alpha polypeptide	3.99
GABRE	gamma-aminobutyric acid (GABA)	3.96
ARHGAP21	Rho GTPase activating protein 21	3.96
TSPAN18	tetraspanin 18	3.93
DAB2	disabled homolog 2,	3.90
SLC9A9	solute carrier family 9	3.90
CD9	CD9 molecule	3.87
LIPH	lipase, member H	3.80
CCL5	chemokine (C-C motif) ligand 5	3.77
LGMN	legumain	3.77
MFAP3L	microfibrillar-associated protein 3-like	3.66
F2RL2	coagulation factor II (thrombin) receptor-like 2	3.65
PDGFB	platelet-derived growth factor beta polypeptide	3.63
DENND2C	DENN/MADD domain containing 2C	3.57
CXCL3	chemokine (C-X-C motif) ligand 3	3.52
C1orf71	chromosome 1 open reading frame 71	3.47
GNG11	guanine nucleotide binding protein gamma 11	3.42
SERPINE1	serpin peptidase inhibitor, clade E	3.38
GRIK4	glutamate receptor, ionotropic, kainate 4	3.37
OR2G3	olfactory receptor 2, subfamily G, member 3	3.36
SLC6A4	solute carrier family 6	3.36
DGKD	diacylglycerol kinase, delta 130 kDa	3.35
AQP10	aquaporin 10	3.34
IRAK2	interleukin-1 receptor-associated kinase 2	3.31
ABLIM3	actin binding LIM protein family, member 3	3.26
MRVI1	murine retrovirus integration site 1 homolog	3.25
CD40LG	CD40 ligand	3.24
NEXN	nexilin (F actin binding protein)	3.22
SEPT5	septin 5	3.14
PAPSS2	3′-phosphoadenosine 5′-phosphosulfate synthase 2	3.14
PLA2G4C	phospholipase A2, group IVC	3.10
ITPK1	inositol 1,3,4-triphosphate 5/6 kinase	−2.00
UCK2	uridine-cytidine kinase 2	−2.01
ALDH1B1	aldehyde dehydrogenase 1 family, member B1	−2.02
TIMM13	translocase of inner mitochondrial membrane 13	−2.02
NSMCE1	nonSMC element 1 homolog (S. cerevisiae)	−2.03
ANKH	ankylosis, progressive homolog (mouse)	−2.04
MRPL48	mitochondrial ribosomal protein L48	−2.05
HIST1H1D	histone cluster 1, H1d	−2.07
TIMM50	translocase of inner mitochondrial membrane 50	−2.07
DIP	interstitial pneumonitis, desquamative, familial	−2.07
NIPSNAP3A	nipsnap homolog 3A (C. elegans)	−2.07
UBL7	ubiquitin-like 7 (bone marrow stromal cell-derived)	−2.08
EBPL	emopamil binding protein-like	−2.09
DENND1A	DENN/MADD domain containing 1A	−2.09
NT5C3L	5′-nucleotidase, cytosolic III-like	−2.09
COMMD9	COMM domain containing 9	−2.10
NDUFA12	NADH dehydrogenase 1 alpha subcomplex, 12	−2.11
POLE4	polymerase (DNA-directed), epsilon 4	−2.12
LIG3	ligase III, DNA, ATP-dependent	−2.12
KIAA0664	KIAA0664	−2.13
IGLL1	immunoglobulin lambda-like polypeptide 1	−2.13
MRPL16	mitochondrial ribosomal protein L16	−2.14
DFFA	DNA fragmentation factor, 45 kDa, alpha	−2.16
RPUSD4	RNA pseudouridylate synthase domain 4	−2.16
TFAP4	transcription factor AP-4	−2.19
GSTP1	glutathione S-transferase pi 1	−2.19
GLT25D1	glycosyltransferase 25 domain containing 1	−2.19
C9orf123	chromosome 9 open reading frame 123	−2.20
MRPL38	mitochondrial ribosomal protein L38	−2.21
MFHAS1	malignant fibrous histiocytoma sequence 1	−2.22
MRPL21	mitochondrial ribosomal protein L21	−2.22
MYBBP1A	MYB binding protein (P160) 1a	−2.23
MPO	myeloperoxidase	−2.23
NANS	N-acetylneuraminic acid synthase	−2.25
ALDH3A2	aldehyde dehydrogenase 3 family, member A2	−2.27
SLC35F2	solute carrier family 35, member F2	−2.29
ZNF259	zinc finger protein 259	−2.29
ALOX5AP	arachidonate 5-lipoxygenase-activating protein	−2.33
TGFBR1	transforming growth factor, beta receptor 1	−2.34
SPN	sialophorin	−2.35
FBL	fibrillarin	−2.37
POLR1E	polymerase (RNA) I polypeptide E, 53 kDa	−2.37
SFXN4	sideroflexin 4	−2.38
GYPC	glycophorin C	−2.40
CAT	catalase	−2.40
COMMD2	COMM domain containing 2	−2.61
IGFBP7	insulin-like growth factor binding protein 7	−2.78
RCN1	reticulocalbin 1, EF-hand calcium binding domain	−2.81
SPG21	spastic paraplegia 21	−2.86
B3GNT5	UDP beta-1,3-N-acetylglucosaminyltransferase 5	−3.21
IL2RG	interleukin 2 receptor, gamma	−3.29
OSM	oncostatin M	−3.34
HINT1	histidine triad nucleotide binding protein 1	−3.42
ANXA2	annexin A2	−3.58

List of genes highly specific for M cells. FC values are given as log 2 means for an n of 3 per group.

Gene Ontology Analysis

Gene lists specific for each differentiated groups were subjected to gene ontology analysis to determine their molecular functions. This analysis yielded seven major functional groups including binding, catalytic, signal transducer, transcription regulator, structural molecule, enzyme regulator, and transporter activity as represented in Fig. 5. Genes with binding function were comparable between the three hematopoietic lineages. Genes with catalytic activity were found to be expressed at a higher percentage in E group; percentage of genes with signal transduction function was the highest in G group; Genes with functions such as enzyme activity and transporter activity were found to be expressed at a higher percentage in M groups.

Fig. 5.

Functional classification of differentially expressed genes during lineage-specific differentiation of CD34⁺ hematopoietic cells to E, G, and M cells by Ingenuity Pathway Analysis. The functional classification of a specific gene may be redundant, resulting from the assignment of 1 gene to more than 1 category.

Validation of Microarray Data by QPCR

To confirm the expression data obtained from the exon microarray studies, we analyzed the expression of a selection of few significantly up regulated genes in each group by real-time PCR. Table 6 illustrates the fold changes in the expression of transcripts between E, G, and M vs. CD34⁺ from microarray and QPCR analyses. A high degree of correlation was observed between the microarray data and the QPCR data.

Table 6.

Validation of microarray data by QPCR

Gene Symbol		E vs. CD34	G vs. CD34	M vs. CD34	P Value
	Solute carrier family 4, anion exchanger, member 1
SLC4A1	exon array	10.68	0.86	0.68
	QPCR	117.26	1.14	0.03	< 0.001
	Proteoglycan 2, bone marrow
PRG2	exon array	1.32	4.45	0.85
	QPCR	1.28	49.70	1.23	< 0.05
	C-type lectin domain family 1, member B
CLEC1B	exon array	1.58	1.59	98.22
	QPCR	19.08	5.21	222.86	< 0.05
	Phosphodiesterase 3A, cGMP-inhibited
PDE3A	exon array	1.55	0.93	72.41
	QPCR	1.49	0.52	6.96	< 0.05
	Spleen focus forming virus proviral integration oncogene
SPI1	exon array	0.52	1.14	0.16
	QPCR	0.38	1.21	0.42	<0.001
	Translocase of inner mitochondrial membrane 13
TIMM13	exon array	1.11	0.97	0.25
	QPCR	1.14	0.51	0.13	<0.005
	von Willebrand factor
VWF	exon array	1.35	0.72	63.13
	QPCR	0.26	0.04	29.16	<0.05
	Elastase 2
ELA2	exon array	2.28	35.91	0.97
	QPCR	47.50	192.67	104.69	<0.001
	Erythroid-associated factor
ERAF	exon array	22.84	2.09	0.95
	QPCR	315.73	109.89	0.13	<0.001
	Proteinase 3
PRTN3	exon array	2.25	25.79	1.09
	QPCR	33.59	84.44	57.28	<0.001
	Rho guanine nucleotide exchange factor (GEF) 17
ARHGEF17	exon array	0.19	0.29	0.19
	QPCR	0.03	0.04	0.05	<0.001
	Fc fragment of IgG, low affinity IIIa, receptor
FCGR3A	exon array	0.15	0.14	0.15
	QPCR	0.01	0.06	0.01	<0.001
	Integrin, beta 3 (platelet glycoprotein IIIa, antigen CD61)
ITGB3	exon array	11.22	2.98	152.96
	QPCR	3.65	1.01	42.22	<0.001
	Solute carrier family 11 member 1
SLC11A1	exon array	0.18	0.17	0.15
	QPCR	0.01	0.06	0.01	<0.05
	Linker for activation of T lymphocytes
LAT	exon array	11.16	4.01	23.92
	QPCR	11.47	6.77	15.24	<0.001

Validation of microarray data by QPCR. FC in the expression for some differentially expressed genes are given as means for 3 samples in each group from exon array analysis and QPCR.

Alternative Splicing Events During Lineage-specific Hematopoietic Differentiation

We identified 86 alternatively spliced genes among the differentiated cells as shown in Tables 7, 8, and 9. An alternatively spliced gene based on this analysis is a gene with at least one exon whose behavior deviates by a certain magnitude relative to the other exons within the gene. In comparing across the three lineages, we observed 31 alternatively spliced genes in common including CAST, EPHX1, FANCA, CLCN7, PDE4D, PDE4DIP. E and G groups showed an overlap of 14 alternatively spliced genes, while 11 genes were common to E and M groups. G and M groups showed the lowest overlap of five spliced genes. Splicing was unique to 13 genes (ALDOA, ASS1, ATP5H, CLEC12A, ELMO1, NUMA1, PLK4CA, PTPN6, PTPRA, SLC39A4, SMG1, UBE2D3, XLT1) in the M group, two genes (CD97, RGR4275) in G group, and 10 genes (CMTM5, COL6A2, CYBR53, DNMT33, LPHN1, PHC2, SLC2A14, SORL1, STAB1, TPM1) in the E group. These gene lists also include previously identified alternatively spliced genes CAST, CLMN, EPHX1, GAB1, and vWF.

Table 7.

Alternatively spliced genes in E group

Gene Symbol	Gene Title	E vs. CD34
GAB1	GRB2-associated binding protein 1	2.73
AKR1C2	aldo-keto reductase family 1, member C2	2.62
COL24A1	collagen, type XXIV, alpha 1	2.62
SLC2A14	solute carrier family 2 member 14	2.46
EPHX1	epoxide hydrolase 1, microsomal	2.43
PARL	presenilin associated, rhomboid-like	2.34
GRAP2	GRB2-related adaptor protein 2	2.20
ORC4L	origin recognition complex, subunit 4-like	2.18
RBBP6	retinoblastoma binding protein 6	2.01
TYRO3	TYRO3 protein tyrosine kinase	1.82
CTNND1	catenin (cadherin-associated protein), delta 1	1.82
STAB1	stabilin 1	1.66
VWF	von Willebrand factor	1.57
CLCN7	chloride channel 7	1.55
COL6A2	collagen, type VI, alpha 2	1.55
NID2	nidogen 2 (osteonidogen)	1.44
RPL21	ribosomal protein L21	1.36
ATP5O	ATP synthase, H+ transporting, O subunit	1.26
RARA	retinoic acid receptor, alpha	1.24
RABGAP1L	RAB GTPase activating protein 1-like	1.22
TSC22D3	TSC22 domain family, member 3	1.22
SORL1	sortilin-related receptor, L	1.17
LDB1	LIM domain binding 1	1.13
HMHA1	histocompatibility (minor) HA-1	1.07
BOLA2	bolA homolog 2 (E. coli)	1.06
C13orf3	chromosome 13 open reading frame 3	1.06
COL18A1	collagen, type XVIII, alpha 1	0.89*
ATP5H	ATP synthase, H+ transporting, subunit d	0.89*
ALDOA	aldolase A, fructose-bisphosphate	0.81*
SMG1	SMG1 homolog, phosphatidylinositol 3-kinase-related kinase	0.78*
AKAP13	A kinase (PRKA) anchor protein 13	0.75*
UBE2D3	ubiquitin-conjugating enzyme E2D 3	0.63*
CLEC12A	C-type lectin domain family 12, member A	0.58*
PTPRA	protein tyrosine phosphatase, receptor type, A	0.56*
ELMO1	engulfment and cell motility 1	0.50*
HADHB	hydroxyacyl-Coenzyme A dehydrogenase beta subunit	0.45*
PTPN6	protein tyrosine phosphatase, nonreceptor type 6	0.26*
XYLT1	xylosyltransferase I	−0.31*
MPO	myeloperoxidase	−0.31*
NUMA1	nuclear mitotic apparatus protein 1	−0.44*
CD97	CD97 molecule	−0.45*
IL16	interleukin 16 (lymphocyte chemoattractant factor)	−0.58*
ASS1	argininosuccinate synthetase 1	−0.75*
SLC39A4	solute carrier family 39 (zinc transporter), member 4	−0.78*
CMTM5	CKLF-like MARVEL transmembrane domain containing 5	−1.04
ADAMTS13	ADAM metallopeptidase with thrombospondin,motif 13	−1.12
BPI	bactericidal/permeability-increasing protein	−1.16
PDE4D	phosphodiesterase 4D, cAMP-specific	−1.21
AARSD1	alanyl-tRNA synthetase domain containing 1	−1.23
CYB5R3	cytochrome b5 reductase 3	−1.31
RTN4	reticulon 4	−1.34
PHC2	polyhomeotic homolog 2 (Drosophila)	−1.35
PDE4DIP	phosphodiesterase 4D interacting protein	−1.39
RGS3	regulator of G-protein signaling 3	−1.40
LPHN1	latrophilin 1	−1.42
UBAP2	ubiquitin associated protein 2	−1.47
TPM1	tropomyosin 1 (alpha)	−1.53
WIPF1	WAS/WASL interacting protein family, member 1	−1.57
FANCA	Fanconi anemia, complementation group A	−1.58
ATP2A3	ATPase, Ca2+ transporting, ubiquitous	−1.60
TRIM16	tripartite motif-containing 16	−1.63
EPB49	erythrocyte membrane protein band 4.9	−1.69
CR1	complement component (3b/4b) receptor 1	−1.73
SIGLEC12	sialic acid binding Ig-like lectin 12	−1.77
DNMT3B	DNA (cytosine-5-)-methyltransferase 3 beta	−1.79
MYLK	myosin light chain kinase	−1.79
TMEM49	transmembrane protein 49	−1.84
C1orf113	chromosome 1 open reading frame 113	−1.85
TLE1	transducin-like enhancer of split 1	−1.86
PTK2	PTK2 protein tyrosine kinase 2	−1.88
CD44	CD44 molecule (Indian blood group)	−1.97
CLMN	calmin (calponin-like, transmembrane)	−2.04
TPCN2	two pore segment channel 2	−2.11
SLC25A3	solute carrier family 25 member 3	−2.30
CAST	calpastatin	−2.60
SMARCAD1	SWI/SNF matrix-associated regulator of chromatin	−2.85
TPD52L2	tumor protein D52-like 2	−2.90

List of alternatively spliced genes. The largest deviation of 2-fold over the exons in a gene for the E group is shown (log base 2 scale), The boldfaced genes are unique to the E group comparison.

^*Genes alternatively spliced in one of the other 2 groups.

Table 8.

Alternatively spliced genes in G group

Gene Symbol	Gene Title	G vs. CD34
PARL	presenilin associated, rhomboid-like	2.23
RBBP6	retinoblastoma binding protein 6	2.11
COL24A1	collagen, type XXIV, alpha 1	1.81
RGR4275		1.75
GAB1	GRB2-associated binding protein 1	1.68
TYRO3	TYRO3 protein tyrosine kinase	1.61
HADHB	hydroxyacyl-Coenzyme A dehydrogenase/beta	1.60
GRAP2	GRB2-related adaptor protein 2	1.56
ATP5O	ATP synthase, H+ transporting,	1.50
PDE4D	phosphodiesterase 4D, cAMP-specific	1.50
ORC4L	origin recognition complex, subunit 4-like	1.39
VWF	von Willebrand factor	1.34
MPO	myeloperoxidase	1.32
RPL21	ribosomal protein L21	1.27
C13orf3	chromosome 13 open reading frame 3	1.27
EPHX1	epoxide hydrolase 1, microsomal	1.17
RARA	retinoic acid receptor, alpha	1.15
AKAP13	A kinase (PRKA) anchor protein 13	1.14
CTNND1	catenin (cadherin-associated protein), delta 1	1.08
AKR1C2	aldo-keto reductase family 1, member C2	0.96*
STAB1	stabilin 1	0.94*
ADAMTS13	ADAM metallopeptidase with thrombospondin 13	0.88*
BOLA2	bolA homolog 2 (E. coli)	0.87*
ELMO1	engulfment and cell motility 1	0.84*
ATP5H	ATP synthase, H+ transporting, mitochondrial	0.78*
SLC2A14	solute carrier family 2 member 14	0.76*
SMG1	SMG1 homolog, phosphatidylinositol 3-kinase	0.76*
ALDOA	aldolase A, fructose-bisphosphate	0.71*
DNMT3B	DNA (cytosine-5-)-methyltransferase 3 beta	0.68*
LDB1	LIM domain binding 1	0.68*
SORL1	sortilin-related receptor, L (DLR class)	0.63*
PTPRA	protein tyrosine phosphatase, receptor type, A	0.62*
TLE1	transducin-like enhancer of split 1	0.57*
SLC39A4	solute carrier family 39 member 4	0.50*
UBE2D3	ubiquitin-conjugating enzyme E2D 3	0.48*
TSC22D3	TSC22 domain family, member 3	0.44*
NUMA1	nuclear mitotic apparatus protein 1	0.43*
CLEC12A	C-type lectin domain family 12, member A	0.39*
HMHA1	histocompatibility (minor) HA-1	0.25*
PTPN6	protein tyrosine phosphatase, 6	−0.22*
XYLT1	xylosyltransferase I	−0.34*
WIPF1	WAS/WASL interacting protein family, member 1	−0.41*
LPHN1	latrophilin 1	−0.50*
PHC2	polyhomeotic homolog 2 (Drosophila)	−0.61*
TPM1	tropomyosin 1 (alpha)	−0.66*
CYB5R3	cytochrome b5 reductase 3	−0.74*
COL6A2	collagen, type VI, alpha 2	−0.75*
CMTM5	CKLF-like MARVEL transmembrane domain 5	−0.82*
ASS1	argininosuccinate synthetase 1	−0.88*
CLMN	calmin (calponin-like, transmembrane)	−0.91*
BPI	bactericidal/permeability-increasing protein	−0.96*
SIGLEC12	sialic acid binding Ig-like lectin 12	−1.04
CD97	CD97 molecule	−1.20
FANCA	Fanconi anemia, complementation group A	−1.22
TRIM16	tripartite motif-containing 16	−1.30
CR1	complement component (3b/4b) receptor 1	−1.31
IL16	interleukin 16 (lymphocyte chemoattractant factor)	−1.32
COL18A1	collagen, type XVIII, alpha 1	−1.39
TMEM49	transmembrane protein 49	−1.41
CAST	calpastatin	−1.42
UBAP2	ubiquitin associated protein 2	−1.44
RTN4	reticulon 4	−1.44
MYLK	myosin light chain kinase	−1.45
RGS3	regulator of G protein signaling 3	−1.48
PTK2	PTK2 protein tyrosine kinase 2	−1.56
EPB49	erythrocyte membrane protein band 4.9	−1.57
RABGAP1L	RAB GTPase activating protein 1-like	−1.58
PDE4DIP	phosphodiesterase 4D interacting protein	−1.63
NID2	nidogen 2 (osteonidogen)	−1.65
CD44	CD44 molecule (Indian blood group)	−1.68
TPCN2	two pore segment channel 2	−1.71
ATP2A3	ATPase, Ca2+ transporting, ubiquitous	−1.84
AARSD1	alanyl-tRNA synthetase domain containing 1	−1.90
TPD52L2	tumor protein D52-like 2	−2.16
C1orf113	chromosome 1 open reading frame 113	−2.22
SLC25A3	solute carrier family 25 member 3	−2.79
SMARCAD1	SWI/SNFmatrix-associated regulator of chromatin	−3.44

List of alternatively spliced genes. The largest deviation of 2-fold over the exons in a gene for the G qroup is shown (log base 2 scale). The boldfaced genes are unique to the G group comparison.

^*Genes alternatively spliced in one of the other 2 groups.

Table 9.

Alternatively spliced genes in M group

Gene Symbol	Gene Title	M vs. CD34
PARL	presenilin associated, rhomboid-like	3.26
ATP5O	ATP synthase, H+ transporting, O subunit	2.69
TYRO3	TYRO3 protein tyrosine kinase	2.55
COL24A1	collagen, type XXIV, alpha 1	2.27
CR1	complement component (3b/4b) receptor 1	2.26
XYLT1	xylosyltransferase I	2.18
PDE4DIP	phosphodiesterase 4D interacting protein	2.12
CLEC12A	C-type lectin domain family 12, member A	2.12
C13orf3	chromosome 13 open reading frame 3	2.10
CTNND1	catenin (cadherin-associated protein), delta 1	1.99
AARSD1	alanyl-tRNA synthetase domain containing 1	1.94
ASS1	argininosuccinate synthetase 1	1.93
AKR1C2	aldo-keto reductase family 1, member C2	1.88
NUMA1	nuclear mitotic apparatus protein 1	1.79
EPHX1	epoxide hydrolase 1, microsomal (xenobiotic)	1.77
RGS3	regulator of G protein signaling 3	1.71
BPI	bactericidal/permeability-increasing protein	1.60
HMHA1	histocompatibility (minor) HA-1	1.52
HADHB	hydroxyacyl-Coenzyme A dehydrogenase, beta	1.50
COL18A1	collagen, type XVIII, alpha 1	1.46
IL16	interleukin 16	1.44
PTPN6	protein tyrosine phosphatase,nonreceptor 6	1.40
MPO	myeloperoxidase	1.37
RBBP6	retinoblastoma binding protein 6	1.35
SMG1	SMG1 homolog, phosphatidylinositol 3-kinase	1.31
ORC4L	origin recognition complex, subunit 4-like	1.29
ATP5H	ATP synthase, H+ transporting, subunit d	1.28
TSC22D3	TSC22 domain family, member 3	1.21
CLCN7	chloride channel 7	1.15
BOLA2	bolA homolog 2 (E. coli)	1.10
UBE2D3	ubiquitin-conjugating enzyme E2D 3	1.10
SIGLEC12	sialic acid binding Ig-like lectin 12	1.10
SLC39A4	solute carrier family 39 zinc transporter 4	1.10
RARA	retinoic acid receptor, alpha	1.03
ELMO1	engulfment and cell motility 1	1.01
LDB1	LIM domain binding 1	1.01
VWF	von Willebrand factor	0.95*
RPL21	ribosomal protein L21	0.87*
DNMT3B	DNA (cytosine-5-)-methyltransferase 3 beta	0.87*
LPHN1	latrophilin 1	0.81*
COL6A2	collagen, type VI, alpha 2	0.66*
STAB1	stabilin 1	0.66*
GRAP2	GRB2-related adaptor protein 2	0.62*
GAB1	GRB2-associated binding protein 1	0.57*
TPM1	tropomyosin 1 (alpha)	0.44*
SLC2A14	solute carrier family 2 member 14	0.41*
SORL1	sortilin-related receptor, L(DLR class)	0.30*
EPB49	erythrocyte membrane protein band 4.9	0.23*
PHC2	polyhomeotic homolog 2	−0.09*
CMTM5	CKLF-like MARVEL transmembrane domain 5	−0.10*
TMEM49	transmembrane protein 49	−0.38*
CD44	CD44 molecule (Indian blood group)	−0.41*
CD97	CD97 molecule	−0.43*
SLC25A3	solute carrier family 25, member 3	−0.76*
CYB5R3	cytochrome b5 reductase 3	−0.78*
RTN4	reticulon 4	−0.82*
NID2	nidogen 2 (osteonidogen)	−0.85*
MYLK	myosin light chain kinase	−0.91*
PTK2	PTK2 protein tyrosine kinase 2	−0.93*
UBAP2	ubiquitin associated protein 2	−1.03
ALDOA	aldolase A, fructose-bisphosphate	−1.08
ADAMTS13	ADAM metallopeptidase with thrombospondin	−1.10
AKAP13	A kinase (PRKA) anchor protein 13	−1.25
PTPRA	protein tyrosine phosphatase, receptor typeA	−1.39
CLMN	calmin (calponin-like, transmembrane)	−1.47
TPCN2	two pore segment channel 2	−1.51
FANCA	Fanconi anemia, complementation group A	−1.70
TPD52L2	tumor protein D52-like 2	−1.72
RABGAP1L	RAB GTPase activating protein 1-like	−1.76
C1orf113	chromosome 1 open reading frame 113	−1.78
ATP2A3	ATPase, Ca2+ transporting, ubiquitous	−1.83
CAST	calpastatin	−1.89
WIPF1	WAS/WASL interacting protein family, 1	−1.97
PDE4D	phosphodiesterase 4D, cAMP-specific	−2.06
TRIM16	tripartite motif-containing 16	−2.27
SMARCAD1	chromatin DEAD/H box 1	−2.89
TLE1	transducin-like enhancer of split 1	−3.24

List of alternatively spliced genes. The larqest deviation of 2-fold over the exons in a gene for the M group is shown (log base 2 scale). The boldfaced genes are unique to the M group comparison.

^*Genes alternatively spliced in one of the other 2 groups.

Important Pathways and Networks Affected by Alternative Splicing

We analyzed these 86 lineage-specific alternatively spliced genes using Ingenuity Pathway Analysis software (http://www.ingenuity.com) for further insights into their potential functional roles during hematopoietic differentiation. Not surprisingly, these analyses of showed preferential enrichment of biological processes related to hematological system development and function. The Ingenuity Pathway Analysis implicated alternative splicing in aspects of cellular development, molecular transport, cellular function and maintenance, cell-to-cell signaling and interaction, hematological system development and function, immune cell trafficking, cell-mediated immune response, antigen presentation and protein trafficking.

The top five canonical pathways were found to be Rac and RhoA signaling, leukocyte extravascular signaling, and wnt/β-catenin signaling, alanine and aspartate metabolism, and regulation of actin based motility by Rho.

Alternative Splicing During Erythroid Differentiation by RNA Sequence Analysis

Along with microarray analysis, we conducted a massively parallel sequencing study of the transcriptome using the SOLiD next generation sequencing platform on the CD34⁺ and differentiated E cells. Data generated from this study allowed for direct comparison and validation of the microarray data at both the gene and exon levels and offered the opportunity to assess the reliability of this emerging sequencing technology for transcriptome analysis. Using the RPKM measurements for the level of expression of an exon, or transcript, we observed a strong correlation between microarrays and RNA sequencing. At the gene level we observed a correlation of R = 0.72 (Fig. 6A) and at the exon level we observed an R value of 0.62 (Fig. 6B). We further interrogated several genes that had high correlations (R values range: 0.83–0.91) between the two platforms including TPD5L2, GAB1, SLC25A3, AND CAST. The fold-change measurements for the exons of these genes using both technologies are illustrated in Fig. 7, A–D. Shown in the figure panels, each gene tends to have at least one exon deviating by a large degree away from the behavior of the other exons, suggesting an alternative splicing event. Strikingly, each platform shows the same exon deviating by a large magnitude of change, indicating that these are very likely due to alternative splicing. This is further suggested by the RefSeq intron/exon isoforms plotted directly below the relative abundance for each transcript which show the inclusion (Fig. 7, B and D) or exclusion (Fig. 7, A, C, and D) of a specific exon corresponding to an alternatively spliced RefSeq isoform. In the gene TPD52L2 as illustrated in Fig. 7A, Exon 3 shows a deviation in microarray data only, but its magnitude is modest (less than twofold), and it does not correspond to a known splice variant. Thus it is likely due to experimental noise of the microarray. Similarly for the gene GAB1 as shown in Fig. 7B, Exon 2 shows a deviation for RNAseq data only with somewhat smaller magnitude than for Exon 8. This may be evidence for a novel transcript (exclusion in E), but as it is not confirmed by another method, it remains speculative. Thus RNAseq data confirms our microarray findings for statistically significant gene expression changes and alternatively spliced genes observed during erythropoietic differentiation.

Fig. 6.

Comparison of RNAseq to microarray fold change for E vs. CD34⁺. Correlation analysis between gene level and exon level microarray (x-axis) and SOLiD (y-axis) data. Red line shows the line of identity. A: gene level correlation analysis. Log10 fold-change RNAseq vs. Log10 fold-change microarray of gene level data (n = 14,105, R = 0.72). Each point represents a gene. B: exon level correlation analysis. Log10 fold-change RNAseq vs. Log10 fold-change microarray of microarray data of gene level data (n = 149,765, R = 0.62). Each point represents an exon.

Fig. 7.

Exon-specific fold change (E over CD34⁺ expression) comparing RNAseq (blue) vs. microarray (red). A: gene TPD52L2. Correlation, R = 0.86. Exon 7 (red) signals AS event on both platforms. Six known RefSeq transcripts are plotted under the overlay plot. Exon 7 is excluded in 2 known RefSeq isoforms (NM_199359 and NM_003288), agreeing with both the microarray and RNAseq data. B: gene GAB1. Correlation, R = 0.83. Exon 8 (green) signals an alternative splicing (AS) event on both platforms. Two known RefSeq transcripts are plotted under the overlay plot. Exon 8 is included in isoform NM_207123, agreeing with both the microarray and RNAseq data. C: gene SLC25A3. Correlation, R = 0.91. Exon 2 (red) signals an AS event on both platforms. Three known RefSeq transcripts are plotted under the overlay plot. Exon 2 is found to be alternatively spliced in transcript NM_005888. D: gene CAST. Correlation R = 0.83. Exons 2 and 3 (red), and 4 (green) signal an AS event on both platforms. Ten known RefSeq transcripts are plotted under the overlay plot. Exons 2 and 3 show downregulation and are unique to 5 of the RefSeq transcripts. Exon 4 shows upregulation on both platforms and appears to be the starting exon on 5 RefSeq transcripts.

DISCUSSION

Blood cells share numerous functional properties (cell motility, immune functions) that distinguish them from differentiated cells of solid tissues. These specific functions are acquired during hematopoietic cell differentiation, and the differentiated cells become fully operative the moment they leave bone marrow or other organs of the immune system toward the peripheral circulation.

Previous reports suggest that the self-renewal and differentiation of hematopoietic cells is not likely be governed by a single or few factors but rather by the integration of many integrating signal inputs affecting gene transcription including chromatin regulation, transcription factors, alternative splicing, and posttranslational modification. In particular, alternative splicing of exons is believed to contribute extensively to transcript and protein complexity in differentiated stem cells. While it is acknowledged that alternative splicing is a major determinant affecting global transcript protein complexity, this has not been examined in functional studies of hematopoietic cell differentiation. Hence, we undertook this study to gain insight into transcriptome and exome of hematopoietic process by using an in vitro human hematopoietic model system that permits analysis of CD34⁺ differentiation into major blood cell lineages.

Exploiting the QPCR and microarray technology, we analyzed the cytokine induced differentiation of CD34⁺ progenitors to identify gene signatures and to determine the degree to which alternative splicing might regulate this process. In an effort to confirm and validate this in vitro model we analyzed the parent and differentiated cells by flow cytometry for specific cell surface markers such as CD71, CD15 and CD66 specific for E, G, and M, respectively, and lineage-specific expression of transcripts such as PTGS1, SERPINE1, GPIBA, PF4 (specific for M), FCGRA, MIP2, FCGR2B, and CD300A, (specific for G), and GYPA, TMOD1, ANK1 (specific for E). These analyses showed increased expression of lineage-specific transcripts and proteins in the differentiated groups compared with CD34⁺ cells, indicating that the day 11 cultures are indeed differentiated cells comparable to those found in the peripheral blood. Our observed data also correlate with other published reports on the lineage-specific gene expression confirming the identity of the cells being studied here.

Initial microarray analysis detected 172 genes that are significantly modulated during differentiation. These transcripts by ingenuity pathway analysis showed them to be involved in cell motility, immune system development, and cell signaling as would be expected for developing hematopoietic cells. CD34⁺ cells on the other hand showed an upregulated expression of genes such as ARHGEF17, CD177, GPSM1, ID1, S100A12, and SLC11A1, and these genes appear to participate in cell signaling processes. In E cells, genes such as SLC4A1, ALAS2, HBB, GYPA, and SELENBP1 are overexpressed. While this red cell signature is expected for SLC4A1 (band 3 red cell membrane protein), HBB (Hemoglobin subunit), and GYPA (glycophorin a membrane protein), SELENBP1 has only been implicated in the pathogenesis of cancer and neuronal disorders. The G cells showed very few overexpressed genes upon differentiation and these genes included PRG2, CEACAM6. Interestingly, CD34⁺ cells differentiation into megakaryocytes demonstrated significant modulations for a diversity of genes, 90 in total, which included 32 downregulated genes and 58 upregulated transcripts. These genes as shown in Table 2 are associated with regulation of cell proliferation, cell cycle signaling, and immune system development. Whether or not individual genes within these signatures play a functional role in hematopoietic differentiation will require additional studies.

We further analyzed the dataset to generate highly selective gene lists that would predict the nature of the differentiated cell types and the parent stem cell. The criteria that were used to determine selectivity, in order for a gene to be selectively up for either of the three comparisons (E vs. CD34⁺, M vs. CD34⁺, G vs. CD34⁺) were to have a P < 0.0001 and at least twofold up for the comparison of interest and unchanged or upregulated <1.4-fold in the other two comparisons. Applying these criteria, we were able to generate three different clusters that showed an upregulation of 30 transcripts highly selective for the E group, 32 for the G group, and 269 for the M group. Gene ontology analysis of these highly selective genes classified them to be involved in seven major functions such as binding, catalytic, signal transducer, transcription regulator, structural molecule, enzyme regulator, and transporter activity. However, it should be noted that the functional classification of a gene may be redundant and few genes could be classified into many different functional categories.

Of particular interest, we observed that during erythropoietic differentiation, the expression of GTPase activator proteins are upregulated. These GTPase activator proteins are known to be involved in actin cytoskeleton organization, membrane trafficking, gene expression, and cell proliferation. We also observed and validated increased expression of several genes associated with homeostasis and platelets during megakaryopoietic differentiation including CD44, TPM1, vWF, GP5, PDGFB, F2RL2, and ELMO1.

Having defined the transcriptome of the differentiated hematopoietic cells, we then examined the differences in exon expression that would represent putative alternative splicing. Using the ExonSVD model, we identified 86 known genes to be alternately spliced among the differentiated cells compared with progenitors. These genes include 31 transcripts that are common to the E, G, and M groups. E and G groups showed an overlap of 14 alternatively spliced genes, while G and M groups showed the lowest overlap of five spliced genes. Eleven spliced genes were found to be common to E and M groups, which are of potential functional importance as erythrocytes and megakaryocytes share a common progenitor. Lineage-specific splicing was observed in 13 genes in the megakaryocytes, while only two genes were specific for the granulocytic group. Red cell differentiation identified 10 genes with differentially expressed exon transcripts. This is the first report showing alternative splicing events during lineage-specific in vitro hematopoietic differentiation. Our gene list also includes alternatively spliced genes (CAST, CLMN, EPHX1, GAB1, vWF) that have previously been identified by others using PCR, SAGE, and sequencing technologies. Importantly, we have identified additional novel spliced genes in the current study as shown in Tables 7, 8, and 9. Finally, we were able to validate our microarray-based detection of alternative splicing with the whole transcriptome sequencing on a next generation massively parallel sequencing platform with a significant degree of correlation. It is likely that this sequence-based expression profiling would become a widely used platform future transcriptome studies.

Functional pathway analysis of these 86 alternatively spliced genes showed preferential enrichment of biological processes related to hematological system development and function, molecular transport, cellular function and maintenance, cell-to-cell signaling and interaction, immune cell trafficking, cell-mediated immune response, antigen presentation, and protein trafficking. Most importantly, the top five canonical pathways were found to be Rac and RhoA signaling, leukocyte extravascular signaling and wnt/B-catenin signaling, alanine and aspartate metabolism, and regulation of actin-based motility by Rho.

The Rac signaling pathway has a role in many cellular functions including cell motility and adhesion, cell growth and proliferation, and cell survival and apoptosis (5, 13, 20, 27). Rac proteins constitute a subgroup of the Rho family of small GTPases and include Rac1, Rac2, Rac3, and the splice variant of Rac1, Rac1b. By acting as molecular switches, they control a variety of signal pathways that are essential for cell functions (5, 13, 20, 27). Rac GTPases are key regulators of the actin cytoskeleton, cell-cycle progression and gene transcription, cell survival and apoptosis, and the NADPH oxidase for producing reactive oxygen species. Aberrant Rac signaling is found in some human cancers as a result of changes in the GTPase itself or in its regulation loops (38).

Rho signaling pathway, another pathway that was found to be modulated during hematopoietic differentiation, orchestrates cellular processes as diverse as cell migration, cell-cycle progression and cytokinesis, microbial killing (through phagocytosis and NADPH oxidase activity), and agonist-regulated gene transcription. In particular, Rac- and Rho-induced effects, which correlate with membrane protrusion and contractility, respectively, antagonize each other in a variety of cell types.

Wnt proteins are a family of highly conserved signaling factors controlling cell fate and differentiation during development including regulation of signals regulating the self-renewal and differentiation interface in hematopoietic stem cells (31). Alterations of the Wnt/β-catenin signaling pathway are known to be associated with the tumorigenesis of tissues with a high renewal potential such as that of bone marrow-hematopoietic tissue (2, 3, 9, 14, 32). Moreover, different human wnt genes show a complex organization and pattern of expression with alternative promoters and RNA splicing responsible for the expression of isoforms. Therefore, though less attention has thus far been paid to the regulation of Wnt expression, such an analysis appears to be required to understand and define the respective role of individual Wnt proteins, if not individual Wnt isoforms, in the control of cell fate, differentiation, and tissue regeneration. Further studies are still needed to determine if alternatively spliced isoforms of Wnt pathway genes play a functional role leading to hematopoietic cell differentiation.

In summary, we used an in vitro model to analyze differentiating hematopoietic stem cells and applied microarray and sequencing technologies to generate detailed expression and exon profiles during lineage-specific differentiation of cells. Findings for erythroid differentiation have been validated and extended with next generation sequencing technology. Knowledge about the specific transcriptional programs during normal hematopoiesis may contribute to further understanding of the complex process of hematopoietic stem cell development to define new pathophysiological pathways that can possibly be used for the strategy of target-specific treatment in the near future.

GRANTS

Funding by Intramural Research, NHLBI and CIT, NIH.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).