Conserved regulation of RNA processing in somatic cell reprogramming
Along with the reorganization of epigenetic and transcriptional networks, somatic cell reprogramming brings about numerous changes at the level of RNA processing. These include the expression of specific transcript isoforms and 3’ untranslated regions. A number of studies have uncovered RNA processing factors that modulate the efficiency of the reprogramming process. However, a comprehensive evaluation of the involvement of RNA processing factors in the reprogramming of somatic mammalian cells is lacking.
Here, we used data from a large number of studies carried out in three mammalian species, mouse, chimpanzee and human, to uncover consistent changes in gene expression upon reprogramming of somatic cells. We found that a core set of nine splicing factors have consistent changes across the majority of data sets in all three species. Most striking among these are ESRP1 and ESRP2, which accelerate and enhance the efficiency of somatic cell reprogramming by promoting isoform expression changes associated with mesenchymal-to-epithelial transition. We further identify genes and processes in which splicing changes are observed in both human and mouse.
Our results provide a general resource for gene expression and splicing changes that take place during somatic cell reprogramming. Furthermore, they support the concept that splicing factors with evolutionarily conserved, cell type-specific expression can modulate the efficiency of the process by reinforcing intermediate states resembling the cell types in which these factors are normally expressed.
KeywordsiPS cells Somatic cell reprogramming RNA processing Alternative splicing
(chemically) induced pluripotent stem cells
(delta) percent spliced in
Adipose progenitor cells
Human inducible fibroblast-like cells (constitutively express human Telomerase)
Hematopoietic progenitor cells
Mouse embryonic fibroblasts
Expression cassette harboring cDNAs for Myc, Klf4, Oct4 and Sox2
Spermatogonial stem cells
Transgenic MEFs with MKOS cassette
Transcripts per million
Integrated analyses of genomics, epigenomics, transcriptomics, and proteomics data are systematically unravelling the gene regulatory networks underlying the reprogramming of differentiated cells into induced pluripotent stem cells (iPSCs) [1, 2, 3, 4, 5, 6, 7], particularly the underlying ‘epigenetic landscape’ . These studies have improved our understanding of the dynamics of cell state transitions and of cell fate decisions, while the vast number of resulting data sets have enabled the development of computational models for predicting regulatory switches, as well as to facilitate cellular reprogramming and transdifferentiation [9, 10].
The ‘molecular roadmap’ of somatic cell reprogramming includes bursts of changes in mRNA, miRNA and histone modification levels during the first three days and at the end of reprogramming. Between these bursts is a less understood period of relatively little transcriptional change. The initial phase resembles a mesenchymal-epithelial transition (MET) , a process that takes place during the development of various organs as well as during cancer transformation. Many interactions that have been uncovered in the context of MET or its reversal, epithelial-mesenchymal transition (EMT), are also studied in the context of reprogramming . Interestingly, some of the important changes occur at the level of RNA processing through alternative splicing. For example, the inclusion of 10 variable exons in messages of the adhesion molecule Cd44, which is important for MET , is driven by ESRP1 and ESRP2 (epithelial splicing regulatory protein 1 and 2) . ESRP1 has recently been reported to also enhance somatic cell reprogramming [15, 16], partly through the alternative splicing of the Grhl1 (grainyhead like transcription factor 1) . Although evidence of deeply conserved regulators of pluripotency has started to emerge , a mammalian pluripotency network that includes splicing regulators has not been reconstructed so far.
Aiming to identify conserved regulators of somatic cell reprogramming we have analyzed mRNA sequencing data obtained in 14 reprogramming studies that were carried out in mouse, chimpanzee and human. We found that functional categories related to ‘RNA processing’ and ‘mRNA splicing’ were strongly over-represented among genes whose expression was higher in iPSCs compared to parental cells, in line with general changes in cell physiology that were reported previously [18, 19, 20]. Furthermore, we identified a set of 9 splicing-related genes, which exhibited a highly consistent pattern of expression in iPSCs and parental cells across all species and types of reprogramming methods. Among these, ESRP1 showed the strongest and most conserved increase in expression in iPSCs compared to parental cells. While the potentiating effect of ESRP1 on reprogramming efficiency had already been reported , here we demonstrate that ectopic expression of either Esrp paralog accelerates the reprogramming kinetics and increases reprogramming efficiency. Finally, we found that transcripts related to the cytoskeleton, cell adhesion and epigenetic regulation undergo splicing/isoform changes in both human and mouse systems. Our analysis supports the concept that splicing factors with an evolutionarily-conserved cell-type-specificity of expression enforce the identity of the corresponding cell types and can modulate reprogramming efficiency.
Expression analysis of somatic cell reprogramming systems across species
Publicly available RNA-Seq data sets analyzed in this study
Adipose progenitor cells, fibroblasts, hematopoietic progenitor cells, iPSCs
Fibroblasts, iPSCs (chemical)
Fibroblasts, iPSCs (retrovirus), iPSCs (Sendai virus)
Fibroblasts, iPSCs, lymphoblastoids (human only)
hiF, hiF-T, iPSCs
iPSCs, spermatogonial stem cells
Fibroblasts, iPSCs (chemical)
Evolutionarily conserved patterns of splicing factor expression in somatic cell reprogramming
We then sought to identify splicing factors with consistent expression changes across studies and species and found that 26 splicing factors exhibited robust and relatively large (median fold change ≥ 5 in studies of at least one species) expression changes (Fig. 2a), 19 having increased and 6 decreased expression in iPSCs compared to fibroblasts. Around a third of splicing-associated genes (9 out of 26) had a median fold change ≥ 5 in all three organisms: AHNAK (AHNAK nucleoprotein; also known as desmoyokin), CELF5 (CUGBP Elav-like family member 5), ESRP1, LGALS3 (galectin 3), MBNL1/2 (muscleblind like splicing regulator 1 and 2), RBM47 (RNA binding motif protein 47), SNRPN (small nuclear ribonucleoprotein polypeptide N), ZCCHC24 (zinc finger CCHC-type containing 24). Another six met the expression change cutoff in human and mouse: ELAVL2 (ELAV like RNA binding protein 2), RBFOX3 (RNA binding fox-1 homolog 3), ESRP2 (epithelial splicing regulatory protein 2), PSIP1 (PC4 and SFRS1 interacting protein 1), USP49 (ubiquitin specific peptidase 49), SRSF12 (serine and arginine rich splicing factor 12). This is considerably more than would be expected if expression changes were randomized over genes within individual species (Fig. 2b; P value for the overlap across all species <10-22; P values for pairwise comparisons ranging from ~10-10 to ~10-19). Only one gene, AFF2 (AF4/FMR2 family member 2, a gene associated with fragile X-linked mental retardation) increased in expression during the reprogramming of primate cells, but decreased during mouse cell reprogramming. The core set of 9 highly conserved splicing factors included genes coding for recently described regulators of reprogramming such as MBNL1/2 , ESRP1, RBM47 and ZCCHC24 . RBFOX2 (RNA binding fox-1 homolog 2), SON (SON DNA binding protein), SRSF2/3/11 (serine and arginine rich splicing factor 2, 3 and 11) and U2AF1 (U2 small nuclear RNA auxiliary factor 1), other splicing factors that have been experimentally linked to the efficiency of reprogramming [24, 25, 26, 27], showed considerably less pronounced changes in their mRNA levels in the systems that we analyzed (Fig. 2).
In all species, the gene with the largest and most consistent increase in expression in iPSCs compared to parental cells was ESRP1 (Fig. 2a). In contrast, its ESRP2 paralog underwent a similarly large expression change in primate iPSCs, but less pronounced in mouse iPSCs. Time series mRNA sequencing data from human and mouse reprogramming that are also available [15, 28] helped explain this discrepancy. Specifically, we found that while expression of both ESRP paralogs steadily and concurrently increased during the reprogramming of human cells (Additional file 6: Figure S3), in mouse cell reprogramming this pattern was shared by Esrp1, but not by Esrp2. Rather, the expression of Esrp2 increased only transiently during the reprogramming of mouse cells (Additional file 7: Figure S4). Of note, DDX25, ELAV2, and, to a lesser extent, CELF5 and USP49, exhibited an expression pattern similar to that of ESRP2, with their expression strongly increasing towards the final stage of reprogramming in human, but decreasing in mouse reprogramming systems (Additional file 6: Figure S3/ Additional file 7: Figure S4). RBFOX3, a neuronal marker, and RBM11 showed a similar increase in expression in human reprogramming systems, but their transcripts were almost entirely absent from mouse cells along the entire reprogramming timeline (sum of TPM <1; Additional file 6: Figure S3/ Additional file 7: Figure S4).
Splicing factors whose expression decreased upon reprogramming included MBNL1/2, whose siRNA-mediated knock-down was found to increase reprogramming efficiency . Down-regulated genes further encode the canonical splicing factor CELF2, the zinc finger-containing protein ZCCHC24, the neuroblast differentiation-associated scaffolding protein AHNAK that was recently found to bind RNAs , and the carbohydrate-binding protein LGALS3 that also functions in splicing . Interestingly, whereas these factors are expressed at high levels across a panel of human tissues , the splicing factors whose expression increases upon reprogramming often have a more restricted pattern of expression across human cell types (Fig. 2c). Furthermore, genes whose expression increased most upon reprogramming of primate somatic cells appear to also undergo large expression changes in a large number of cancers (Additional file 8: Figure S5). While RBFOX3, RBM11 (RNA binding motif protein 11) and DDX25 (DEAD-box helicase 25) are generally downregulated in cancers compared to normal cells, the dysregulation of CELF5, ELAVL2 and ESRP1 is dependant on the type of cancer. In particular, ESRP1 expression is heavily upregulated in (endo)cervical cancers (~192-fold), but strongly downregulated (~136-fold) in sarcomas.
When considering genes annotated with the more general “RNA processing” gene ontology term, we identified a total of 53 genes with large and consistent expression changes (median fold change ≥5 across the comparisons of one or more species) between iPSCs and parental cells (Additional file 9: Figure S6A). Similarly to splicing factors, this is much more than expected by chance (Additional file 9: Figure S6B). Beyond the already discussed splicing factors, 9 additional ‘RNA processing’ genes emerged from this analysis, in all three species: ATXN1 (ataxin 1), EGFR (epidermal growth factor receptor), NR2F1 (nuclear receptor subfamily 2 group F member 1), and TGFB1 (transforming growth factor beta 1) had lower expression, while CHD7 (chromodomain helicase DNA binding protein 7), LIN28A/B (lin-28 homolog A and B), MDN1 (midasin AAA ATPase 1), and TRIM71 (tripartite motif containing 71) had higher expression in pluripotent compared to somatic cells. In particular, the LIN28A protein has been demonstrated to play important roles in reprogramming from very early on, being able to substitute for c-Myc in the “Yamanaka cocktail” . Recently, similar observations have been reported for the LIN28B paralog . Interesting, although these genes are annotated as genes involved in ‘RNA processing’, they are better for other functions, such as DNA or chromatin binding.
Overexpression of ESRPs enhances the reprogramming of mouse embryonic fibroblasts
We were intrigued by the discordant expression pattern of the ESRP2 paralog in human and mouse reprogramming systems and wondered whether in spite of its transient induction during the reprogramming of mouse cells, this protein can nevertheless increase reprogramming efficiency, as does its paralog. To answer this question we used a previously described reprogramming system of mouse embryonic fibroblasts (MEF) that can be induced to express the set of “MKOS” transcription factors (Myc, myelocytomatosis oncogene; Klf4, Kruppel-like factor 4 (gut); Oct4/Pou5f1, POU domain, class 5, transcription factor 1; Sox2, SRY (sex determining region Y)-box 2) from a genomically-integrated construct. Application of doxycycline to these transgenic MEFs (TNG-MKOS-MEFs) induced the expression of the MKOS factors as well as of mOrange. This allowed the monitoring of MKOS expression during the reprogramming process, while the endogenous Nanog-GFP  reporter enabled the detection and quantification of pluripotent stem cell colonies 15 days after MKOS induction.
ESRPs enhance MET and accelerate the reprogramming of TNG-MKOS-MEFs
As ESRPs are important for EMT  and as the converse MET is an essential stage in MEF reprogramming , we investigated the kinetics of the reprogramming process upon expression of either Esrp paralog, by monitoring levels of the CD44 and ICAM1 cell surface markers . During reprogramming, mouse fibroblasts convert from CD44+ / ICAM1- to CD44- / ICAM1- and then to CD44- / ICAM1+ cells, the latter population containing a substantial fraction of pluripotent cells. Flow cytometry-based analysis 5 days after doxycycline induction revealed that the proportion of CD44- / ICAM1- cells was 2-fold higher in Esrp1/2-transduced compared to Renilla luciferase-transduced TNG-MKOS-MEFs. We also observed a similar trend 2 days after doxycycline induction, but the data is somewhat noisy, likely due to variation in the onset of transgene expression following retroviral transduction. Taken together, these data indicate that both ESRPs accelerate the reprogramming of MEFs (Fig. 3c, d). To further determine whether an accelerated MET underlies the accelerated reprogramming, we measured the expression of epithelial marker Cdh1 (cadherin 1; E-cadherin)  by qRT-PCR. Indeed, we found that Esrp1/2-transduced cells expressed 2-fold higher levels of Cdh1 at day 2 and 3.5-7-fold higher levels at day 5, compared to non-transduced cells (Fig. 3e). The difference in the impact of Esrp2 and Esrp1 overexpression on Cdh1 levels at day 5 is consistent with a previous study demonstrating that the knockdown of Esrp paralogs has different effects on Cdh1 expression in the context of cancer cell motility . Conversely, the expression of the mesenchymal marker Cdh2 (cadherin 2; N-cadherin) was 1.5 fold lower at day 2 and ~2-fold lower at day 5 in ESRP1/2-transduced compared to non-transduced cells (Fig. 3f).
Cd44 isoform switching parallels ESRP-induced acceleration of reprogramming
A network of conserved splicing changes in human and mouse somatic cell reprogramming
Splicing patterns are known to change relatively rapidly during mammalian evolution . Nevertheless, given the strong conservation of the splicing regulators, we sought to identify genes and pathways in which reprogramming-related splicing events occur in both human and mouse. To evaluate the significance of alternative splicing changes between reprogramming starting and end points, we quantified the inclusion of annotated alternatively spliced exons in terms of ‘percent spliced in’ (PSI) values, a measure indicating the fraction of transcripts that are consistent with a specific splicing event among all transcripts generated from the corresponding gene, across all samples. Principal component analysis of PSI values of all quantified events again revealed the expected clustering of the samples by cell type (Additional file 11: Figure S8A/B), although the first two principal components explained only 15-18% of the variance. Gene ontology analysis of genes that underwent significant (P ≤ 0.05) splicing changes showed relatively little overall conservation across species (Additional file 11: Figure S8C).
We have further conducted a reprogramming time course experiment in the system described in the previous sections, focusing on the early stages of the process (days 0-5 following doxycycline induction). A time point representing iPSCs (day 15) was also included. We have then selected those of the top 100 differentially spliced events identified for mouse (see above) that correspond to genes from included in the described network of conserved splicing changes (Fig. 5c) or predicted to be Esrp targets (Fig. 5e). Plotting the corresponding percent spliced in (PSI) values for each day and condition (overexpression of Esrp1, Esrp2 or Luciferase following retroviral transduction or no transduction) largely verified differential splicing of these events between TNG-MKOS-MEFs and iPSCs (Additional file 14: Figure S10A). However, although the introduction of Esrp1 or Esrp2 cDNAs demonstrably led to a considerable and specific increase in their respective gene expression levels in the early stages of reprogramming (up to day 4; Additional file 14: Figure S10B), we did not identify splicing patterns that were specific to their overexpression compared to control treatments (Additional file 14: Figure S10A). We have also specifically looked at the expression of Grhl1 isoforms, as Grhl1 splicing has previously been reported to be affected by Esrp1 during reprogramming (Cieply et al. 2016). These isoforms were not included in the SUPPA index of splicing events because the short Grhl1 isoform lacks multiple exons, and these complex splicing patterns are not analyzed by the SUPPA software. While both Grhl1 isoforms are largely absent in TNG-MKOS-MEFs (Additional file 14: Figure S10C; day 0) and only expressed at low levels in iPSCs (day 15), their expression steadily increases between days 1 and 5, reaching peak abundances of approximately TPM = 5 (Grhl1-201) and TPM = 8 (Grhl1-202). However, while the relative abundance of the isoforms (Additional file 14: Figure S10D; fraction of each isoform by total gene expression) appears to be changing between days 2, 5 and 15, a clear influence of elevated Esrp1 or Esrp2 levels could not be observed (Additional file 14: Figure S10D). One possible reason for the lack of a consistent effect of Esrp overexpression on the splicing of Grhl1 (Additional file 14: Figure S10D) and other genes (Additional file 14: Figure S10A) might be the technical limitations of accurate quantification of isoforms, particularly of those with low abundances, leading to inaccurate PSI estimations . However, it may also be possible that Esrp-induced changes in splicing patterns of the studied events are saturated at endogenous levels or affect splicing at later time points.
Stem cells have long held great promise for regenerative medicine . Thus, the demonstration that the gene regulatory program of fully differentiated mammalian cells can be overridden by temporarily expressing pluripotent stem cell-specific transcription factors to generate induced pluripotent stem cells (iPSCs) [45, 46] had a profound impact on basic and medical research . iPSCs are used as experimental models to study a wide range of diseases and to discover relevant drugs via chemical library screenings . They also serve as a basis for grafting organoids and tissues [49, 50, 51]. A first clinical trial involving patient-derived iPSCs is currently ongoing  and several more will be launched within the next decade . Despite this tremendous progress, our understanding of the process is still limited  and the efficiency of transcription factor-driven reprogramming remains in the percentage range . Approaches to increase it are in high demand.
Many different types of modulators have been already identified. Very potent is vitamin C, which substantially increases the efficiency of iPSC colony formation  by promoting epigenetic remodeling through histone and DNA demethylation [57, 58]. MicroRNAs of the embryonically-expressed miR-302/367 cluster also increase the efficiency of transcription factor-induced reprogramming, repressing targets that are involved in the cell cycle, chromatin remodeling, vesicular transport and epithelial-to-mesenchymal transition . The related miRNAs of the mouse-specific miR-290-295 cluster have been reported to contribute to the stabilization of the stem cell state by targeting epigenetic regulators [60, 61, 62, 63].
Following discoveries that core transcriptional drivers of reprogramming such as OCT4/POU5F1 and NANOG (Nanog homeobox) are regulated by alternative splicing [27, 64, 65], splicing factors have also been added to the repertoire of modulators of the reprogramming process. The general splicing factor SRSF2 has been linked to the pluripotency as it affects the balance of MDB2 (methyl-CpG binding domain protein 2) isoforms, which are part of a complex pluripotency circuit involving the OCT4/POU5F1 transcription factor and the miR-302/367 miRNAs . Conversely, SRSF11 acts as a roadblock for reprogramming, its RNAi-mediated knockdown leading to the rapid emergence of pluripotency-specific isoforms . MBNL1/2-dependent alternative splicing results in a change in the sequence-specificity of the FOXP1 transcription factor, with consequences for the efficiency of reprogramming [23, 67]. MBNL1 and MBNL2 are deeply conserved pluripotency regulators, in functional antagonism with CELF factors in the stem cells of planarian flatworms .
Many splicing factors that have so far been found to modulate reprogramming efficiency act as reprogramming roadblocks . This may be because many studies used small RNA-based screening to deplete individual factors in early stages of reprogramming. However, as we have found here, most splicing regulators increase in expression during the reprogramming time course, and one may expect them to rather promote reprogramming. Indeed, this was demonstrated for Esrp1 [15, 16], the splicing factor most strongly upregulated in iPSCs relative to parental cells (Fig. 2a).
While conventional human iPSCs and mouse iPSCs represent ‘primed’ and ‘naive’ pluripotent states, respectively, requiring distinct signaling pathways for their self-renewal , the differences between these states and the species-specific aspects of reprogramming are not fully understood. Our analysis revealed that among the core splicing regulators, ESRP2 has a different expression dynamics in mouse compared to human, mirroring the monotonic increase in expression of its ESRP1 paralog in human reprogramming systems, but undergoing only transient induction in mouse (Additional file 7: Figure S4). Nevertheless, we found that just as ESRP1, ESRP2 also accelerated the kinetics and increased the efficiency of mouse cell reprogramming (Fig. 3a, b, c and d).
Although splicing networks appear to evolve relatively fast and splicing events are not strongly conserved in evolution , we identified a core set of conserved splicing targets that underwent significant changes across a large number of reprogramming systems. This set is likely incomplete, as the human and mouse transcriptomes that we used for our analysis of isoform usage are still not fully annotated. Furthermore, methods to quantify isoform usage are still in development, and the quantifications are not very accurate, particularly for low abundance transcripts . Nevertheless, we identified conserved splicing targets in excess of what is expected by chance. They encode proteins that promote remodelling at the cellular and chromatin level, and are themselves subject to many regulatory influences. Cd44 is one example: although the gene is strongly down-regulated at the level of transcription in the early stages of reprogramming, splicing of the produced transcripts also changes, leading to isoforms that contain epithelium-specific variable exons (CD44v) and are expressed at low levels. Also heavily regulated are factors that deposit or read DNA methylation marks, such as DNTM3B (DNA methyltransferase 3 beta): apart from undergoing alternative splicing, its expression is further modulated by embryonically-expressed miR-290-295 miRNAs as well as MECP2 (methyl CpG binding protein 2), which is in turn regulated by the related family of miR-302/367 miRNAs . Isoform switching and miRNA-dependent regulation has been reported for other methyl-CpG binding domain proteins as well . Multiple layers of regulation probably ensure that the activity of these epigenetic regulators is precisely controlled.
Through a comprehensive analysis of RNA sequencing data sets obtained in studies of mouse, chimpanzee and human somatic cell reprogramming, we uncovered an extensive set of splicing factors that undergo conserved changes in expression. In contrast to ESRP1, which undergoes strong upregulation across all reprogramming studies, the pattern of expression of its paralog, ESRP2, differs between species. Nevertheless, both of these proteins accelerate the kinetics and increase the reprogramming efficiency of MEFs. The RNA sequencing data further allowed us to reconstruct networks of splicing targets that are conserved between species. They correspond to proteins involved in subcellular structure and traffic as well as in DNA modification. Although transcription factors have been instrumental in changing cell fates, the efficiency of somatic cell reprogramming remains limited. A variety of molecules ranging from small metabolites and miRNAs to splicing regulatory proteins have been found to modulate the process. An improved understanding of these factors’ functions will enable a more controlled and efficient engineering of cell identity.
Gene and protein nomenclature
Gene symbols in this manuscript are italicized when referring to the gene itself, as well as derived transcripts and cDNAs (e.g. the ESRP1 isoforms), but not when the corresponding proteins are referenced (e.g. the ESRP1 protein). Only when unambiguously referring to mouse genes, cDNAs and mRNAs, symbols are specified in title case (e.g. Esrp1). In all other cases, including in cases where general statements are made about genes that are conserved across primates and rodents, symbols are denoted in all capitals (e.g. ESRP1 for the gene and ESRP1 for the protein). When a gene is first referenced, the corresponding human name/description is specified in parentheses right after the symbol. In some cases, popular symbols or names are indicated in addition to the (latest) official symbols/names.
RNA-Seq study search and selection
To obtain previously released data on iPSC reprogramming and differentiation, the Gene Expression Omnibus  was queried for relevant keywords ('iPSC' OR 'iPSCs' OR 'iPS cells'; 'somatic' AND 'reprogramming'; 'induced' AND 'reprogramming') on September 14th, 2016. Moreover, to be able to carry out the downstream analyses, we required that the data were derived from organisms whose genome sequence and corresponding gene annotations are available at Ensembl [22, 70]. We inspected the corresponding studies that had Illumina RNA-Seq-based sequencing data deposited at the NCBI Sequence Read Archive  and retained only those studies that include at least two replicates for each of iPSCs and the corresponding tissues or cell lines of origin. The Illumina RNA-Seq data of all included studies were downloaded from the NCBI Sequence Read Archive and converted to FASTQ format using the SRA Toolkit .
Genomes and other public resources
Genome sequences, gene annotations and sequences of mature mRNAs for human (GRCh38), mouse (GRCm38) and chimpanzee (CHIMP2.1.4) were obtained from Ensembl , release 84. On the genome level, unassembled regions, haplotype and patch regions were disregarded. Genes annotated on mitochondrial DNA and regions not assembled into chromosomes were dropped from the gene annotations and only genes of the following biotypes were kept: 'protein_coding', 'lincRNA', 'processed_transcript', 'antisense'. Human and mouse transcripts were further filtered according to their transcript support level (unavailable for chimpanzee transcript annotations), with only transcripts of support levels 1 through 3 being retained. This amounts to the minimal requirement that the transcript is supported by a single non-suspect EST. For analyses of tissue-dependent gene expression, we used the data (in transcripts per million; TPM, in a each sample, log2-transformed) from The Human Protein Atlas (release 16) . Log-transformed expression data for select genes in cancer and corresponding normal tissue were obtained from The Cancer Genome Atlas  through the FireBrowse ‘mRNASeq samples API’ . Log2 fold changes for available tumor versus normal tissue comparisons were computed based on the obtained median expression values.
Genome sequences, transcript sequences and gene annotations were indexed using STAR v2.4.1c , kallisto v0.42.3  and SUPPA v2.1  for read-to-genome mapping, estimation of transcript abundance and quantification of exon inclusion, respectively. Gene annotations were provided during STAR index generation. Cutadapt v1.8.3  was used to remove poly(A) tail fragments from reads of all sequencing libraries. Sequenced reads, in FASTQ format, were aligned to the genome with the STAR aligner (Figure S1C/D for mapping rates). Transcript abundances were estimated with kallisto, and based on these, the relative usage of transcript isoforms and alternative splicing events was quantified with SUPPA. The means and standard deviations of the fragment length distributions required by kallisto for estimating transcript abundances from single-end sequencing libraries were set to 300±100 for single-end libraries obtained from the NCBI Sequence Read Archive. The indexing of genome resources and processing of RNA-Seq samples was performed with the help of the Anduril workflow framework v1.2.23 . Estimates of gene expression were obtained by summing transcript abundances and raw read counts (from kallisto; see data sets 1-6 in ) for transcripts and reads corresponding to individual genes. Before calculating principal components, the gene-by-sample gene expression matrix (in transcripts per million, TPM) for a given organism was first log2-transformed and then zero-centered by columns and rows.
Differential expression and splicing analyses
To compare gene expression between sample types of interest (e.g. reprogrammed versus differentiated cells from a given study), gene-level read count estimates  for the corresponding samples (Additional file 2) were used as input to the R/Bioconductor package edgeR v3.12.0  for differential gene expression analysis (see data sets 7-12 in  for comprehensive summaries of log fold changes and false discovery rates). Unless mentioned otherwise, genes with a false discovery rate (Benjamini-Hochberg method) of less than 0.05 were considered differentially expressed. Gene set enrichment analyses were performed with Ontologizer 2.1 (Build: 20160628-1269)  and calculation method “Term-For-Term”. The ontology file (downloaded on 2016-12-24; in OBO 1.2 format) was obtained from The Gene Ontology Consortium . Genes of a given organism associated with each GO term were obtained from Ensembl BioMart  Archive March 2016 version (corresponding to Ensembl release 84) and converted to the GAF 2.0 format by a custom script. ‘Percent spliced in’ (PSI) measures calculated with SUPPA for each annotated alternative splicing event (see data sets 13 and 14 in  for PSI values in human and mouse, respectively) were further supplied to SUPPA's diffSplice function  to identify differentially spliced events in comparisons of interest (see data sets 15-18 in  for comprehensive summaries of the resulting ΔPSI and associated P values).
Orthologous genes & conservation analyses
Orthologous genes were obtained from Ensembl BioMart  (Archive March 2016 version, corresponding to Ensembl release 84) and filtered for one-to-one relationships. Using the resulting table of corresponding Ensembl identifiers for human, mouse and chimpanzee (differential gene expression analysis only) orthologs, as well as the Ensembl gene annotations (release 84, see above) for individual species, Ensembl identifiers were converted to human gene symbols. Only those genes were kept whose Ensembl identifier could be unambiguously matched to a unique gene symbol.
Reproducing the computational analyses
Generally, the indicated tools were ran according to their primary use cases, thus requiring no or only very little modification of default values. In other words: if at all, default values Exhaustive information on how computational analyses were performed and instructions for replicating the analyses are available on GitHub .
Murine transgenic Nanog-GFP embryonic stem cells were generated from the E14Tg2a mouse ES cell line in the Smith lab as described previously (TNG-ESCs)  and donated to the Kaji lab. TNG-MKOS-ESCs were derived from TNG-ESCs by gene targeting the Sp3 locus with a vector containing a 2A peptide-linked "Yamanaka factors" (MKOS) gene cassette followed by ires-mOrange under a tetracycline inducible promoter and a reverse tetracycline transactivator (rtTA) under a constitutive CAG promoter, as described previously . Chimeric mouse embryos were generated with TNG-MKOS-ESCs via morula aggregation. TNG-MKOS-MEFs and wild-type MEFs were isolated at E12.5 from chimeric and wild-type embryos, respectively, as described previously . NMuMG mouse mammary gland epithelial cells  and Phoenix-ECO cells were obtained from the American Type Culture Collection (ATCC; numbers CRL-1636 and CRL-3214, respectively).
Phoenix-ECO cells and TNG-MKOS-ESCs were cultured in Glasgow Minimum Essential Medium (GMEM; Sigma Aldrich, G5154), supplemented with 10% fetal bovine serum (Invitrogen, 10270-106, Lot 41A1520K), 1X non-essential amino acids (100X, Invitrogen, 11140-036), 1X Pen/Strep antibiotics, 1 mM sodium pyruvate (Invitrogen, 15140-122), 0.1 mM 2-mercaptoethanol (Life Technologies, 31350010), 2mM L-glutamine (Invitrogen, 25030-024) and 100,000U/ml of leukemia inhibitory factor (LIF) (from Kaji lab, SCRM, University of Edinburgh). This medium hereafter is referred to as normal medium. Normal medium was supplemented with FGF-2 (5 ng/ml, Preprotec, 100-18B) and heparin (1 μg/ml, Sigma Aldrich, H3149) to culture wild-type and transgenic MEFs (TNG-MKOS-MEFs). Reprogramming medium consisted of normal medium with 1 μg/ml doxycycline (Sigma Aldrich, D9891-1G), 10 μg/ml vitamin C (Sigma Aldrich, 1000731348) and 500 nM Alk5 inhibitor (A83-01, Tocris Bioscience, 2939). NMuMG cells were cultured in Dulbecco's modified Eagle's medium (DMEM; Sigma Aldrich, D5671) with high glucose and L-glutamine, supplemented with 10% fetal bovine serum (Sigma-Aldrich, F7524).
Retroviral vectors and transductions
Retroviral vectors were constructed via an attR1/R2 Gateway cloning cassette (Invitrogen). Esrp1 and Esrp2 cDNAs were PCR-amplified from NmuMG cells; Renilla luciferase cDNA was PCR amplified from psiCHECK-2 (Promega AG, C802A). See Additional file 5 for a list of the used primers. cDNAs were cloned into the pENTR2B Gateway entry vector (Invitrogen, A10463) using EcoRI sites. Mutation-free inserts were gateway-cloned into a pMXs retro-vector  using LR clonase (Invitrogen, 11791-020). Pseudo-retroviral particles were produced by transfecting 10 μg of pMXs retro-vectors expressing transgenes, separately for 48 hrs in a 100 mm plate seeded with 2.2x106 Phoenix-ECO cells (ATCC, CRL3214). Medium collected after 48 hrs of transfection was filtered with 0.45 micron filter and mixed with polybrene (TR-1003-G, Millipore) for transductions.
Reprogramming of transgenic mouse embryonic fibroblasts
TNG-MKOS-MEFs were isolated from 12.5 dpc chimeric embryos and contribution levels of transgenic cells were measured by mOrange expression upon doxycycline administration. For all reprogramming experiments conducted in this study, 3% TNG-MKOS-MEFs were seeded with 97% of wild-type MEFs in a gelatinized 6 well culture plate. A total of 1x105 cells (3000 TNG-MKOS-MEFs and 97000 WT-MEFs) was seeded in each well of a 6 well plate. Each well was transduced with 2 ml of retroviral particles expressing Esrp1, Esrp2 and Renilla luciferase for 4 hours with polybrene (1ug/ml, Millipore, TR-10003-G). After 4 hours medium was replaced with reprogramming medium to induce reprogramming of TNG-MKOS-MEFs. Expression of the transgenes was measured by qRT-PCR after 4 days of transduction and induction of the reprogramming experiment (Figure S7A). Colony counting was done after 15 days of reprogramming induction. The number of colonies was determined by manual counting and validated with the Celigo software.
Quantitative real-time PCR and RT-PCR
For qRT-PCRs assays, total RNA was extracted using TRI reagent (Sigma Aldrich, T9424) and subjected to DNase digestion with RQ1 DNase (Promega, M6101) followed by phenol-chloroform (Sigma Aldrich, P3803) purification. 1μg of total RNA was used for cDNA synthesis using SuperScript III (Invitrogen, 18080-044) reverse transcriptase according to the manufacturer’s protocol. 10 ng of cDNA per sample was added to Power SYBR Green PCR Master Mix (Applied Biosystems). Assays were performed in triplicates. See Additional file 15 for a list of the used assays and primers. Unless specified otherwise, GAPDH levels were used as internal control, and the samples from non-transduced controls or MKOS-ESCs  were used as external controls. Relative fold changes were calculated according to the ΔΔCT method . RT-PCR products were stained with RedSafe (iNtRON Biotechnology) and run on a 1.5% agarose gel. DNA was visualized at a wavelength of 290 nm. For the Cd44 RT-PCR assays, a Cd44 isoform (ENSMUST00000005218) was PCR amplified (see Additional file 15 for the primer sequences) and purified by the QIAEX II gel extraction kit (Qiagen, 20021). The DNA concentration was quantified by spectrophotometry (NanoDrop). A series of 10-fold dilutions of this DNA was used as a template for the generation of a standard curve in which CT values were plotted against the concentration of the template. The amounts of cDNA to use in order to obtain equal amounts of Cd44 in each RT-PCR assay were calculated by linear regression from the standard curve. Pixel intensities of each band of the resulting gel photos were quantified by the AzureSpot analysis software (Azure Biosystems).
Flow cytometry, cell sorting and time course mRNA-sequencing
Double staining of CD44 and ICAM1 was performed as previously described [36, 88]. Briefly, TNG-MKOS-MEFs infected with retroviral particles and treated with doxycycline to induce reprogramming as described above, were trypsinized, filtered through a cell strainer and stained with fluorophore-conjugated antibodies of ICAM1-Biotin (eBioscience, 13-0541-81, dilution-1:100), CD44-APC (eBioscience, 17-0441-81, dilution-1:100) and Streptavidin-PE-Cyanine7 (eBioscience, 25-4317-82, dilution-1:500) for 30 minutes, followed by two washing steps with FACS buffer (2% FCS in PBS). TNG-MKOS-MEFs were re-suspended in FACS buffer and analyzed on an LSRFortessa Cell Analyzer (BD Biosciences). Single-stained cells were used as controls. The data was analyzed using FlowJo software (FlowJo LLC).
For the time course mRNA-sequencing libraries, 3 to 15% TNG-MEFs were seeded and transduced with retroviral particles expressing transgenes. After inducing reprogramming, around 3x105 mOrange positive cells were sorted on FACSAria IIIu cell sorter (BD Biosciences) and used the same for library preparation as described in .
We would like to thank Christina J. Herrmann for helping with the initial reprogramming meta-analysis and with establishing the Anduril workflow engine in our group. We would further like to thank Andreas J. Gruber and Maciej Bak for ideas and fruitful discussions concerning the splicing analysis. We thank Sergio Menendez, Tyson Ruetz and Sarah Brightwell from the Kaji lab for their assistance with reprogramming experiments. We also thank Verena Jäggin and Telma Lopez from the Department of Biosystems Science and Engineering, Single cell facility, ETH Zurich, for helping us with flow cytometry experiments, as well as the Imaging Core Facility, Biozentrum, University of Basel, and Liliane Todesco for her technical assistance with iPSC differentiation experiments. The computations used the sciCORE scientific computing core facility at the University of Basel . The results published here are in part based upon data generated by the TCGA Research Network .
The work was supported by the European Research Council Starting grant 310510-WHYMIR to MZ. KK is an MRC senior non-clinical fellow (MR/N008715/1). The funding bodies had no role in the design of the study, the collection, analysis, and interpretation of data, or in the preparation of this manuscript.
Availability of data and materials
All data generated or analysed during this study are included in this published article and its supplementary information files. Raw data for plots in Fig. 3a, c, d, e, f, Fig. 4d, S7A/C/D are available in Additional file 16. Raw RNA-Seq data generated and/or analysed during the current study are available in the NCBI Sequence Read Archive  under accessions SRP011318, SRP016568, SRP026281, SRP033561, SRP033569, SRP033700, SRP045688, SRP045999, SRP049340, SRP052014, SRP056571, SRP059670, SRP063867, SRP064357, and SRP150302. Analyzed RNA-Seq data are available at Zenodo (https://zenodo.org) under accession 1303234 .
MZ, AK, and APS conceived and designed the study. AK analyzed RNA-Seq and related data with input from MZ and APS. APS performed the experiments and analyzed the corresponding data with input from MZ, AK and KK. AK, APS, KK and MZ interpreted results and wrote the manuscript. All authors read and approved the final manuscript.
Ethics approval and consent to participate
All animal experiments for the iPSC generation were approved by the University of Edinburgh Animal Welfare and Ethical Review Body, performed at the University of Edinburgh, and carried out according to regulations specified by the Home Office and Project License.
Consent for publication
The authors declare that they have no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
- 17.Solana J, Irimia M, Ayoub S, Orejuela MR, Zywitza V, Jens M, et al. Conserved functional antagonism of CELF and MBNL proteins controls stem cell-specific alternative splicing in planarians. elife. 2016;5. https://doi.org/10.7554/eLife.16797.
- 39.Bebee TW, Park JW, Sheridan KI, Warzecha CC, Cieply BW, Rohacek AM, et al. The splicing regulators Esrp1 and Esrp2 direct an epithelial splicing program essential for mammalian development. elife. 2015;4. https://doi.org/10.7554/eLife.08954.
- 71.SRA Tools. Github. https://github.com/ncbi/sra-tools. Accessed 12 June 2017.
- 79.Kanitz A, Syed AP, Kaji K, Zavolan M. Conserved regulation of RNA processing in somatic cell reprogramming – supplementary data; 2018. https://doi.org/10.5281/zenodo.1193699.
- 86.Kanitz A, Syed AP, Kaji K, Zavolan M. Conserved regulation of RNA processing in somatic cell reprogramming – code repository. GitHub. https://github.com/zavolanlab/SplicingReprogramming. Accessed 7 Feb 2018.
- 91.sciCORE - Center for Scientific Computing. University of Basel. https://scicore.unibas.ch/. Accessed 7 Feb 2018.
- 92.The Cancer Genome Atlas. National Cancer Institute. https://cancergenome.nih.gov/. Accessed 7 Feb 2018.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.