Progressive lengthening of 3′ untranslated regions of mRNAs by alternative polyadenylation during mouse embryonic development

Abstract

The 3′ untranslated regions (3′ UTRs) of mRNAs contain cis-acting elements for posttranscriptional regulation of gene expression. Here, we report that mouse genes tend to express mRNAs with longer 3′ UTRs as embryonic development progresses. This global regulation is controlled by alternative polyadenylation and coordinates with initiation of organogenesis and aspects of embryonic development, including morphogenesis, differentiation, and proliferation. Using myogenesis of C2C12 myoblast cells as a model, we recapitulated this process in vitro and found that 3′ UTR lengthening is likely caused by weakening of mRNA polyadenylation activity. Because alternative 3′ UTR sequences are typically longer and have higher AU content than constitutive ones, our results suggest that lengthening of 3′ UTR can significantly augment posttranscriptional control of gene expression during embryonic development, such as microRNA-mediated regulation.

The 3′ UTRs of mRNAs harbor various cis-acting (or cis) elements involved in posttranscriptional regulation of gene expression (1–3). Some cis elements are recognized by RNA-binding proteins that lead to control of mRNA localization, translation, or stability; some interact with microRNAs (miRNAs), which are ≈22-nucleotide (nt)-long RNAs that cause inhibition of translation and/or mRNA degradation (4, 5). Posttranscriptional gene regulation is believed to be widespread. For example, in humans, the AU-rich elements (AREs) involved in control of mRNA stability are found in 5–8% of all genes (6), and over 30% of genes are predicted to have miRNA target sites (7). Many studies have shown that posttranscriptional gene regulation is critical for embryonic development, with most data on miRNA-mediated regulations (8–10).

Over half of mammalian genes contain multiple polyadenylation sites, or poly(A) sites, that lead to transcript variants with different 3′ UTRs or coding regions (11). As such, transcript variants can contain different cis elements, such as miRNA target sites, resulting in different mRNA metabolism (12). Previous studies indicate that the alternative polyadenylation (APA) pattern of genes varies across human and mouse tissues (13–16), and can be affected by genomic imprinting (17). In addition, many APA events correlate with cell proliferation (14). Both regulation of polyadenylation factors and regulatory proteins have been found to modulate APA (18–20).

Results

We previously mapped poly(A) sites for a large number of human and mouse genes using cDNA/EST sequences (11). Most mouse genes with APA contain 2 poly(A) sites in the 3′-most exon, termed proximal and distal on the basis of their positions relative to the coding region (illustrated in Fig. 1A), which results in a constitutive region (named cUTR) and an alternative region (named aUTR). Overall, aUTRs are ≈90% longer than cUTRs and ≈60% longer than 3′ UTRs that are not affected by APA [named single UTR (sUTR)] (Fig. 1B). In addition, aUTRs typically have higher AU content than cUTRs and sUTRs (Fig. 1C).

Fig. 1.

Alternative polyadenylation (APA) leading to alternative 3′ UTRs. (A) Schematics of APA in the 3′-most exon and using SAGE tags and Affymetrix GeneChip probes to examine APA. Two transcript variants resulting from proximal and distal poly(A) sites are shown. The dotted vertical line separates 2 UTR regions, i.e., constitutive UTR (cUTR) and alternative UTR (aUTR). CDS, coding sequence; pA, poly(A) site; AAA, poly(A) tail. (B) Distribution of length for different UTR groups. The full 3′ UTRs for genes with APA are shown by cUTR + aUTR [4,139 genes, median = 1,288 nucleotide (nt)]. Constitutive and alternative regions are represented by cUTR (median = 358 nt) and aUTR (median = 685 nt), respectively. sUTRs are 3′ UTRs not affected by APA (7,242 genes, median = 439 nt). (C) AU content for different UTR sequences.

Interestingly, we noticed that many mouse EST sequences supporting proximal poly(A) sites came from cDNA libraries that were annotated with early development stages, particularly those before implantation. We calculated the usage of distal poly(A) sites relative to that of proximal sites for genes represented in all relevant EST libraries (listed in Table S1), and named the value relative usage of distal poly(A) sites supported by EST (RUD-EST). Intuitively, RUD-EST reflects the 3′ UTR length regulated by APA. As shown in Fig. 2A, genes expressed at preimplantation stages tend to have significantly shorter 3′ UTRs than those expressed at later stages. Overall, the median RUD values appear to increase along with developmental time, except for adult tissues, suggesting global lengthening of 3′ UTR during embryonic development.

Fig. 2.

Progressive lengthening of 3′ UTR during mouse embryonic development. (A) Box plots of RUD-EST values for mouse developmental stages. The preimplantation group includes “cleavage,” “morula,” and “blastocyst.” Embryonic fetus corresponds to whole fetus, and embryonic tissues include different embryonic tissue types. Median values are connected by a red line. (B) Scatter plot of RUD-SAGE values vs. developmental time. Embryonic development is shown in days and postnatal development is shown in weeks. The 8th and 19th days are indicated by vertical gray lines. The data were fitted by a Lowess function (shown as a dotted red line). There are 32 tissue types in the graph. A total of 3,041 genes were used to calculate RUD-SAGE values. (C) Scatter plot of RUD-SAGE values vs. developmental time for brain tissues. All samples are shown as gray dots, and brain samples are boxed. Linear regression was used to fit the data. (D) As in C except that testis samples (boxed) were analyzed. Two linear regression lines were used, 1 for embryonic stages (green) and 1 for postnatal stages (blue). P0 was used for both stages. R² values for the regression lines are indicated in the graphs. (E) Box plots of RUD-array values for preimplantation cells. (F) As in E except that mixed tissues were used. (G) Normalized RUD-array values for 8 embryonic tissue types. The normalized RUD-array values are RUD values standardized within each sample set. For each set, all time points are relative to the middle point, which was set at 0. Linear regression was used to detect the overall trend [Pearson correlation coefficient (CC) is indicated in the graph]. NCC, neural crest cells. Data sets used for A are shown in Table S1, those used for B are shown in Table S2, and those used for E, F, and G are shown in Table S3.

To further explore regulation of 3′ UTR by APA in development, we used a serial analysis of gene expression (SAGE) data set that was generated for almost all stages of mouse development, from fertilized egg to postnatal development (21) (listed in Table S2). Because the SAGE tags were generated from the 3′-most recognition site of restriction enzyme NlaIII, which recognizes CATG, the usage of alternative poly(A) sites can be inferred (illustrated in Fig. 1A): SAGE tags falling into aUTRs and upstream regions can be considered to support the usage of distal sites and proximal sites, respectively. It is widely accepted that SAGE data are much more quantitative and sensitive than EST data. As such, a better understanding of regulation of 3′ UTR by APA can be expected. We calculated usage of distal sites relative to proximal sites (named RUD-SAGE) to gauge change of 3′ UTR length. We found that 3′ UTRs progressively lengthen during the entire course of embryonic development (Fig. 2B), as indicated by a trend line fitted by the Lowess function (22).

Because additional cis elements influencing mRNA stability can exist in aUTRs compared with cUTRs, the difference in the steady-state level of mRNA between APA variants may result from their difference in stability rather than synthesis (23). While it is not possible to dissect these 2 confounding factors for single genes, this issue can be addressed for global analysis of trend: We reasoned that randomly selected genes are not likely to share the same regulatory mechanism for mRNA stability across different conditions, and highly expressed genes or genes expressed at their high levels are not likely to be under the condition under which they are subject to degradation. The latter is particularly relevant to miRNA-mediated regulation, as mRNAs tend to be expressed at high levels under conditions that avoid their target miRNAs (24). We found that all gene selection methods resulted in similar trends (Fig. S1), indicating that the observed lengthening of 3′ UTR most likely results from regulation of APA. On the other hand, our result is not in disagreement with the possibility that certain mRNAs with lengthened 3′ UTRs are selectively degraded at some stages. In addition, the trend of 3′ UTR lengthening remains when we used only genes ubiquitously expressed across samples (Fig. S1), indicating that our finding is not the result of surveying different sets of genes at different stages.

Three phases of 3′ UTR lengthening can be discerned in development (Fig. 2B). The first phase is before the 8th embryonic day (E8.0), during which 3′ UTRs rapidly lengthen. The second phase is from E8.0 to birth (postnatal day 0, or P0), during which 3′ UTRs in general continue to lengthen but at a much slower rate than the first phase. The third phase corresponds to postnatal development, during which the 3′ UTR length overall does not appear to change over time. Because during mouse embryonic development, 3 germ layers, i.e., ectoderm, endoderm, and mesoderm, begin to develop into organs around E8.0, our result suggests that 3′ UTR lengthening in the first phase coordinates with initiation of organogenesis. Using the SAGE data we also observed that the mRNAs expressed in brain tissues tend to have progressively longer 3′ UTRs throughout embryonic and postnatal developmental stages (Fig. 2C), whereas 3′ UTRs of the mRNAs expressed in testes sharply shorten after birth (Fig. 2D). This result is not only consistent with previous reports which indicate that the mRNAs expressed in brain tissues tend to have longer 3′ UTRs and those in testes tend to have shorter ones than other tissues (13, 14), but also it provides a view of 3′ UTR dynamics in these 2 tissues from a developmental standpoint.

To corroborate the results obtained from EST and SAGE data, we analyzed a set of DNA microarray data related to mouse embryonic development (listed in Table S3). We have previously shown that many types of Affymetrix GeneChip arrays contain probes that target aUTR regions (25), making it possible to examine APA (illustrated in Fig. 1A). By comparing intensity values of the probes mapped to aUTRs with those mapped to upstream regions in the 3′-most exon across a given sample set, we derived a RUD-array value for each sample in a set. As shown in Fig. 2 E and F, lengthening of 3′ UTR by APA can be detected for preimplantation cells between E1.5 and E3.5, and mixed embryonic tissues between E10.5 and E14.5. Further, we analyzed 8 individual embryonic tissue types and found that, consistent with the SAGE analysis result, there is a global trend of 3′ UTR lengthening between E8.5 and P0 (Fig. 2G).

Notably, variations of 3′ UTR lengthening can be discerned among different tissues between E8.0 and P0, in both SAGE (Fig. 2B) and microarray (Fig. 2G) results, suggesting 3′ UTR lengthening is regulated differently in different tissue types in this phase of development. To examine how 3′ UTR lengthening coincides with regulation of other biological processes, we selected genes of which expression levels are significantly correlated with RUD values, either positively or negatively, and examined their associated gene ontology (GO) terms (Fig. S2). We considered the GO terms that are identified by both SAGE and microarray analyses to be the most significant and reliable. As shown in Table 1, we found that 3′ UTR lengthening coincides with upregulation of genes involved in morphogenesis and differentiation, such as “extracellular structure organization,” “regulation of cell differentiation,” “actin filament-based process,” and “cell morphogenesis,” and with downregulation of genes involved in proliferation, such as “DNA replication,” “cell cycle phase,” “DNA metabolic process,” and “cell division,” RNA processing, and translation.

Table 1.

GO entries associated with genes that have significant correlation with RUD-SAGE and RUD-array

Type	−log10(P)	GO ID, GO term
Associated with genes that positively correlate with RUD	3.51	GO:0006817, phosphate transport
	2.86	GO:0043062, extracellular structure organization
	2.60	GO:0007167, enzyme linked receptor protein signaling pathway
	2.19	GO:0007259, JAK-STAT cascade
	2.15	GO:0045595, regulation of cell differentiation
	2.02	GO:0030029, actin filament-based process
	1.97	GO:0030324, lung development
	1.79	GO:0040008, regulation of growth
	1.60	GO:0006182, cGMP biosynthetic process
	1.55	GO:0050982, detection of mechanical stimulus
	1.49	GO:0000902, cell morphogenesis
Associated with genes that negatively correlate with RUD	15.28	GO:0006260, DNA replication
	14.15	GO:0022403, cell cycle phase
	13.99	GO:0006259, DNA metabolic process
	12.73	GO:0051301, cell division
	11.95	GO:0006396, RNA processing
	10.72	GO:0006412, translation
	3.62	GO:0051716, cellular response to stimulus
	3.47	GO:0065002, intracellular protein transmembrane transport
	2.90	GO:0022607, cellular component assembly
	2.83	GO:0051726, regulation of cell cycle
	2.82	GO:0009262, deoxyribonucleotide metabolic process
	2.63	GO:0007017, microtubule-based process
	1.84	GO:0009117, nucleotide metabolic process
	1.62	GO:0006325, establishment or maintenance of chromatin architecture
	1.56	GO:0015931, nucleobase, nucleoside, nucleotide and nucleic acid transport

P-value in −log₁₀(P) is the geomean of P-values (Fisher's exact tests) from SAGE analysis and microarray analysis (Fig. S2). GO entries containing more than 500 genes were discarded, as they are too generic. To eliminate redundancy between GO entries, we required that each reported GO entry has at least 25% of associated genes that are not associated with GO entries with more significant P-values.

We also found that 3′ UTRs progressively lengthen during differentiation of C2C12 myoblast cells to myotubes (Fig. 3A), a commonly used cell line model for myogenesis. To examine how this process is related to the one in embryonic development, we selected 2 gene sets, named positive correlation set (PCS, 74 genes) and negative correlation set (NCS, 59 genes), of which expression levels well correlated with both RUD-SAGE and RUD-array values of embryonic samples between E8.0 and P0 (Fig. S2 and Fig. S3; Table S4 and Table S5). Thus, they can be considered as biomarker genes that correlate with regulation of APA during embryonic development. As shown in Fig. 3B, we found that the expression of genes in PCS and NCS also correlated well with RUD values for C2C12 samples at different stages of differentiation, suggesting that regulation of 3′ UTR by APA during C2C12 differentiation can recapitulate the process in embryonic development. To further confirm this result, we used the Affymetrix mouse exon array to analyze APA during C2C12 differentiation, which allowed us to survey more genes than 3′ arrays (1,836 vs. 691). As shown in Fig. 3C, we found that 3′ UTRs significantly lengthened under the differentiation condition compared with the growth condition, and the lengthening coincided with regulation of genes in PCS and NCS (Fig. S4). The number of genes with lengthened 3′ UTRs is about 8- to 20-fold greater than those with shortened ones, depending upon the level of stringency for selection (Fig. 3 D and E).

Fig. 3.

Regulation of 3′ UTR during C2C12 differentiation. (A) Analysis of 3′ UTR lengthening during C2C12 differentiation using 3′ array (Affymetrix Mouse 430 v2.0) data. Lanes 1 and 2 correspond to 50% and 90% cell confluency, respectively, for which cells are under the growth condition; and lanes 3–8 correspond to 6 h, 1 day, 2 days, 3 days, 4 days, and 5 days after induction of differentiation, respectively. (B) Correlation between RUD-array values and expression of gene sets during C2C12 differentiation. The negative correlation set (NCS) contains 59 genes, and the positive correlation set (PCS) contains 74 genes. Selection of these sets is shown in Fig. S2. Expression values for genes in each set were standardized across samples and averaged. RUD-array values are those shown in A. Left, correlation between RUD-array values and gene expression of PCS. Right, correlation between RUD-array values and gene expression of NCS. (C) Analysis of 3′ UTR lengthening during C2C12 differentiation using exon array (Affymetrix Mouse Exon 1.0 ST). Growth and differentiation conditions approximately correspond to 50% confluency and between +1 and +2 days in A, respectively. (D) Numbers of genes with significant change of APA during C2C12 differentiation based on comparison of probes in aUTRs with those in cUTRs. Different FDRs were used to select significant genes. The ratio of number of genes with 3′ UTR lengthened to that with 3′ UTR shortened are indicated. (E) Genes with APA that can be examined by exon array are plotted. y axis, change of intensity (differentiation vs. growth) for probes in aUTRs; x axis, change of intensity for probes in cUTRs. Genes with significant change of APA (FDR = 10%) are colored. Red, 3′ UTR lengthening (284 genes); green, 3′ UTR shortening (41 genes). (F) Schematic of constructs used to examine APA in C2C12 cells. pCMV, CMV promoter; RFP, red fluorescent protein; IRES, internal ribosome entry site; EGFP, enhance green fluorescent protein; AAA, poly(A) tail. (G) Relative usage of poly(A) sites in C2C12 cells under growth and differentiation conditions. The qRT-PCR value for the EGFP region was normalized to that for the RFP region for each sample. The value for 77S.AE under growth condition was used as reference and set at 1. Error bars are standard deviations based on triplicates.

To corroborate our global analysis results, we used reporter constructs to examine regulation of APA in C2C12 cells. We used the pRiG vector, which can generate 2 transcript variants when a proximal poly(A) site is inserted in its cloning site (Fig. S5) (26). Two poly(A) sites, 77S.AE and 77S.AD, which were derived from the intronic poly(A) site of human CSTF3 were used (Fig. 3F). The sequence of 77S.AE contains UGUA, AUUAAA, U-rich, and UGUG elements, whereas that of 77S.AD lacks UGUG element, and, as such, 77S.AE is a stronger poly(A) site than 77S.AD (Fig. S5C). We transfected reporter constructs into C2C12 cells and maintained the cells either under growth or differentiation conditions (see Materials and Methods for detail). Using qRT-PCR with primers targeting either the upstream (RFP) or the downstream (EGFP) regions of proximal poly(A) site, the relative expression levels of 2 transcript variants resulting from APA were calculated. As expected, 77S.AD had lower polyadenylation activity than 77S.AE under both conditions (Fig. 3G). Significantly, polyadenylation took place more frequently at the proximal poly(A) site for both constructs under the growth condition than the differentiation condition (2.4 times for 77S.AE and 2.8 times for 77S.AD), suggesting that the polyadenylation activity at proximal sites is weaker under the differentiation condition. This observation was also confirmed by using another reporter construct pCβS-COX-2-proximal, in which the proximal poly(A) site of COX-2 gene was used (Fig. S6) (27). Taken together, these results directly support the notation that lengthening of 3′ UTRs during C2C12 myogensis is mediated by APA, most likely because of weakened activity of polyadenylation at proximal sites. Given the consistency between C2C12 myogenesis and embroynoic development in (i) lengthening of 3′ UTR over time and (ii) expression change of a set of biomarker genes (PCS and NCS), it is highly conceivable that the polyadenylation activity may also weaken over time during embryonic development, leading to lengthening of 3′ UTRs.

Discussion

By analyzing EST, SAGE, and microarray data sets, we have found that 3′ UTRs progressively lengthen during embryonic development. This global 3′ UTR lengthening is controlled by APA, and coordinates with initiation of organogenesis and aspects of embryonic development, including morphogenesis, differentiation, and proliferation. Notably, Sandberg et al. recently reported a negative correlation between expression of proliferation-related genes and 3′ UTR length during T cell activation and in a variety of tissues and cell lines (14). Consistent with this finding, we found that expression changes of genes associated with several proliferation-related GO terms have significant negative correlations with RUD values (Table 1). Thus, the coordination between 3′ UTR length and cell proliferation appears to be a widespread mechanism used in both embryonic development and growth/differentiation of adult cells across tissues. In addition, our study indicates that during embryonic development, 3′ UTR regulation by APA further coincides with control of genes involved in differentiation, morphogenesis, RNA processing, and translation (Table 1). In line with this notion, the PCS and NCS genes that we identified have very little overlap with the proliferation-related genes reported by Sandberg et al. (Fig. S7).

The molecular mechanism of APA regulation in embryonic development is yet to be elucidated. We found that expression of RNA processing genes as a group has negative correlation with 3′ UTR length. Consistent with this, our experimental assays using C2C12 differentiation as a model indicated that 3′ UTR lengthening is likely caused by weakening of mRNA polyadenylation activity. Previous studies of B cell differentiation identified CstF-64 as a critical regulator of APA (28). To explore this issue in C2C12 differentiation, we analyzed expression of genes whose protein products are part of the human pre-mRNA 3′ processing complex based on a recent proteomics study (29). We found that, in general, these genes are more downregulated than other genes during C2C12 differentiation (Fig. S8A). More than 20 genes are significantly downregulated, including several ones with known roles in polyadenylation (Fig. S8B). Interestingly, expression levels of genes encoding the 3 factors in the CstF complex were all downregulated and, intriguingly, a paralog of CstF-64, τCstF-64, was upregulated. Thus, the biochemical details of APA regulation in C2C12, and perhaps in embryonic development as well, can be rather complex, and may involve a network of events. More experimental assays are underway to unravel the regulatory mechanism controlling 3′ UTR length in development.

Many cis elements involved in posttranscriptional gene regulation are located in 3′ UTRs. Because aUTRs are ≈90% longer than cUTRs, APA can significantly regulate cis elements in 3′ UTRs. Consistent with previous reports (12, 14, 23), we found that, on average, aUTRs contain ≈30% more miRNA target sites that are conserved among human, mouse, rat, and dog, and much more if the conservation criterion is relaxed (Fig. S9A). Significantly, miRNA target sites located in aUTRs are surrounded by sequences that tend to be more AU rich and contain less stable RNA structures than those in cUTRs (Fig. S9 B and C), presumably because of the higher AU content in aUTRs than in cUTRs (Fig. 1C). Because these 2 parameters can influence the efficacy of miRNA targeting (30–32), our result suggests that miRNA target sites located in aUTRs can be better targets for miRNAs. In addition, lengthening of 3′ UTR can potentially change the distance of a target site to 3′ end, another parameter for the efficacy of miRNA targeting (30). It is also highly conceivable that 3′ UTR lengthening can impact on gene regulations by other types of cis elements located in 3′ UTRs, such as AREs (33), and coordinate their activities with development.

Taken together, our results suggest that 3′ UTR lengthening is part of the gene regulatory program in embryonic development, which can significantly augment posttranscriptional gene regulation via cis elements in 3′ UTRs. On the other hand, it can be conjectured that the global 3′ UTR length in a given cell reflects its developmental state.

Materials and Methods

Analysis of EST Data.

We used cDNA library information obtained from the UniGene database to annotate usage of EST-supported poly(A) sites at different developmental stages and in different tissue types, as previously described (13). Libraries with fewer than 100 genes with APA were discarded. The RUD-EST value for a cDNA library is the averaged usage of distal poly(A) sites minus the averaged usage of proximal sites across all genes.

Analysis of SAGE Data.

SAGE tags were downloaded from the SAGE Genie database (34), and were mapped to aUTR sequences in the mouse genome. We used the “Best Gene” file of SAGE Genie to map SAGE tags to genes. The RUD-SAGE value for a given SAGE library is the averaged percentage of tags mapped to aUTRs minus the averaged percentage of tags mapped to upstream regions across all genes. For gene expression analysis, tags mapped to all regions of a gene were combined. Tag counts in each library were normalized to tag per million (TPM). For each gene, the TPM values across all libraries were standardized, i.e., minus mean value and divided by standard deviation.

Analysis of DNA Microarray Data.

We mapped Affymetrix GeneChip Mouse 430 (v2.0) probes to 3′-most exons in the mouse genome. For each gene in a sample of a set, the ratio of the average intensity value of probes targeting aUTR to that of other probes was normalized across all samples in the set. The RUD-array value for a sample is the averaged normalized ratios across all genes. Raw probe intensities were normalized by the robust multiarray average (RMA) method using the Affymetrix Power Tools (APT) program. Probes with raw intensity value less than 5 or belonging to probe sets that were annotated with “Absent” by Affymetrix MAS 5.0 were discarded. Data sets were downloaded from the NCBI GEO database and are listed in Table S3.

For exon array data, probe sets mapped to 3′-most exons were analyzed. We used the GC-bin method for background correction and applied quantile normalization. Probe set intensities and gene expression levels were derived from the probe logarithmic intensity error model (PLIER). Genes with absolute intensity values less than 50 were considered not expressed and discarded, and probe sets with detection above background (DABG) P value <0.05 for at least 1 biological condition were used. For genes with APA, we also compared the probe sets targeting cUTRs with those targeting aUTRs with respect to intensity differences between differentiation and growth. SAM analysis was carried out to derive false discovery rate (FDR) for selecting significant probe sets at different stringency levels (35).

Experimental Assays.

The pRiG-77S.AD and pRiG-77S.AE vectors were constructed by cloning fragments containing the poly(A) site region in pRiG-CstF77.S(+) into the pRiG vector (26), using PCR (5′-CGATCTCGAGGTGTCTTACCTTTATTTTGTA and 5′-GGCCGGATCCTTCTAAATAAATGACACA for 77S.AE, and 5′-CGATCTCGAGGTGTCTTACCTTTATTTTGTA and 5′-GGCCGAATTCACAAGTAAATAAAAGGCT for 77S.AD), and restriction enzymes (XhoI and BamH I for 77S.AE and XhoI and EcoR I for 77S.AD). C2C12 cells were maintained at 20–70% confluency in Dulbecco's Modified Eagles Medium (DMEM) supplemented with 10% fetal bovine serum. Transfection was carried out using jetPEI (Polyplus-transfection) when the confluency of cells was ≈50%. Cells were split 12 h after transfection. To induce differentiation, cells were switched to DMEM + 2% horse serum (Sigma) after 10 h (≈100% confluency). After 24 h, cells were harvested, and total cellular RNAs were extracted using TRIzol (Invitrogen) and treated with DNase I (Roche). mRNAs were reverse transcribed using oligo(dT) primers, and quantitative real-time PCR (qRT-PCR) was carried out using the Power SYBR Green PCR Master Mix (Applied Biosystems) with primers targeting RFP and EGFP regions (5′-GCCCCGTAATGCAGAAGAAG and 5′-CTTCAGGGCCTTGTGGATCT for RFP, and 5′-GGGCACAAGCTGGAGTACAACT and 5′-ATGTTGTGGCGGATCTTGAAG for EGFP). For the experiment using Affymetrix GeneChip Mouse Exon 1.0 ST, total RNAs were extracted by RNeasy Mini Kit (Qiagen) when cells were at 60–70% confluency (growth condition) or 24 h after induction of differentiation (differentiation condition). Sample processing was carried out according to manufacturer's protocol.