Neonatal cardiac dysfunction and transcriptome changes caused by the absence of Celf1

The RNA binding protein Celf1 regulates alternative splicing in the nucleus and mRNA stability and translation in the cytoplasm. Celf1 is strongly down-regulated during mouse postnatal heart development. Its re-induction in adults induced severe heart failure and reversion to fetal splicing and gene expression patterns. However, the impact of Celf1 depletion on cardiac transcriptional and posttranscriptional dynamics in neonates has not been addressed. We found that homozygous Celf1 knock-out neonates exhibited cardiac dysfunction not observed in older homozygous animals, although homozygous mice are smaller than wild type littermates throughout development. RNA-sequencing of mRNA from homozygous neonatal hearts identified a network of cell cycle genes significantly up-regulated and down-regulation of ion transport and circadian genes. Cell cycle genes are enriched for Celf1 binding sites supporting a regulatory role in mRNA stability of these transcripts. We also identified a cardiac splicing network coordinated by Celf1 depletion. Target events contain multiple Celf1 binding sites and enrichment in GU-rich motifs. Identification of direct Celf1 targets will advance our knowledge in the mechanisms behind developmental networks regulated by Celf1 and diseases where Celf1 is mis-regulated.

homozygous mice 22 . We detected altered electrophysiological and contractile functions at early postnatal stages that could explain reduced viability. Animals did not show abnormal cardiac functions at five weeks of age. Deep RNA-sequencing (RNA-seq) of Celf1 − /− heart samples identified extensive transcriptional changes at postnatal day 3 (PN3). We identified 45 alternative splicing events that are responsive to Celf1 depletion. Most of these events contain Celf1-CLIP tags and are enriched for GU rich motifs within the alternatively spliced regions and/or the flanking sequences suggesting that they are direct Celf1 splicing targets. Ion transport and circadian rhythm genes are significantly down-regulated in hearts from PN3 Celf1 − /− animals in comparison with wild type PN3 hearts. Furthermore, we identified a network of cell cycle genes that are significantly up-regulated in PN3 Celf1 − /− hearts. These genes are enriched for Celf1 binding sites based on CLIP-seq data supporting a regulatory role for Celf1 at neonatal stages in regulating the stability of mRNAs from cell cycle genes.

Results
Celf1 loss of function impacts viability. Celf1 protein is down-regulated more than ten times during postnatal heart development 15 . Therefore, we first evaluated Celf1 protein expression levels in hearts at neonatal (PN3) and later (PN38-42) stages from Celf1 − /− and Celf1+ /+ animals by Western blot assays. Celf1 protein was completely absent in hearts from Celf1 − /− neonates (Fig. 1a). Celf1 mRNA levels at PN3 were decreased 20-fold in Celf1 − /− hearts based on RNA-seq data. Celf1 protein expression in adult (PN38-42) Celf1 + /+ animals decreased approximately 40-fold and was not detected in Celf1 − /− mice ( Supplementary Fig S1). Only a slight up-regulation of the paralog Celf2 was observed and there was no change in level of Mbnl1, an RNA binding protein that co-regulates a subset of Celf1 targets (Fig. 1a).
As described previously 22 , Celf1 − /− animals were smaller than their wild type littermates throughout postnatal development (Fig. 1b) and the number of three-four week old Celf1 − /− animals from Celf1 − /+ matings was significantly lower than expected. At PN3, we observed that out of 75 mice (from ten Celf1 − /+ matings), 42 were heterozygous, 21 were wild type, and only 12 were homozygous. On the other hand, out of 107 PN21-28 mice from 15 Celf1 − /+ matings, 65 were heterozygous, 32 were wild type, and only 10 were homozygous (Supplementary Table S1). For litters evaluated at PN3, χ 2 did not reach significance, (Supplementary Table S1) although the data showed a trend toward a lower than expected number of Celf1 − /− animals. However, χ 2 was significant (p < 0.001) at PN21-28 demonstrating that fewer homozygous animals than expected survived (Supplementary Table S1). Taken together and consistent with a previous report 22 , we conclude that Celf1 loss of function leads to retarded growth and reduced viability.
We studied more in detail the neonatal animals measuring their body weight, heart weight and tibia length. We normalized heart weights and body weights to tibia length since we observed differences in body weight. Overall, we observed that both PN3-5 homozygous females and males exhibited significant smaller body and heart weights than wild type animals (Fig. 2c).
In conclusion we found that neonate cardiac functions and morphology are impacted by the absence of Celf1. Hearts of PN3-5 Celf1 − /− pups are smaller and show reduced function compared to those from wild type and heterozygous littermates, possibly contributing to reduced viability.
Celf1 regulates alternative splicing in neonatal hearts. To identify the transcriptional and posttranscriptional networks regulated by Celf1 in neonatal hearts, we performed RNA-seq using polyadenylated mRNA isolated from ventricles of Celf1 − /− and Celf1 + /+ animals at PN3 and from Celf1 + /+ animals at PN38 at which point Celf1 levels have decreased ( Supplementary Fig. S1). The high quality of the RNA-seq data was reflected by efficient mapping to the genome for all samples (85-86%) and sufficient depth for alternative splicing analysis (> 179,000,000 paired end reads) (Supplementary Table S6). Alternative splicing data (percent spliced in, PSI) from two replicates of homozygous or two wild type samples showed high levels of correlation (Pearson = 0.98 for Celf1 + /+ animals, Pearson = 0.95 for Celf1 − /− animals) (Fig. 3a).
From the 45 alternative splicing events responsive to Celf1 depletion at PN3 only nine events were also regulated during development between PN3 and PN38 when Celf1 levels decrease (five of them in the same direction as development and four in the opposite direction) (Fig. 3f). Gene ontology analysis of the genes with alternative splicing changes in PN3 Celf1− /− animals revealed that the most significantly enriched categories were related to chromatin organization (purple), cytoskeleton functions and cell-cell contact (green), and lipid and glucose metabolism (yellow) (Fig. 3g).
Alternative splicing events sensitive to Celf1 depletion contain Celf1 CLIP tags and are enriched in GU rich motifs. We next validated the alternative splicing data by reverse transcription (RT) PCR experiments. We designed primers annealing in the constitutive flanking regions of 16 alternative regions regulated by Celf1 depletion at PN3 (Supplementary Table S8). We performed RT-PCR using heart RNA from Celf1 + /+ and Celf1−/− animals at PN3 and PN38-42 ( Fig. 4a,b, and Supplementary Fig. S3). The correlation between Δ psi (psi Celf1− /− -psi Celf1+ /+ ) at PN3 obtained by RT-PCR and by RNA-seq was high (Pearson coefficient = 0.89) confirming the validity of quantitative splicing analysis from the RNA-seq data ( Fig. 4c and Supplementary Table S8).
Celf1 binds GU rich motifs in regions in close proximity to the regulated splicing event 25 . We thus analyzed the motifs enriched in 22 alternative regions and 500 bp up-and downstream using MEME software 26 . A GU rich motif was most significantly enriched being present in 16 out of the 22 splicing events analyzed ( Supplementary  Fig. S4a,b). Interestingly a GA rich motif was also enriched ( Supplementary Fig. S4a). Motif enrichment analysis was used as a first approach to investigate the potential splicing targets of Celf1. However, to better evaluate direct Celf1 effects on alternative exons we looked for already available experimental evidence of Celf1 binding to pre-mRNAs. High-throughput sequencing of RNA isolated by crosslinking immunoprecipitation (HITS-CLIP) 27 allows global identification of targets for specific RNA binding proteins in cells. We used a Celf1 HITS-CLIP data set from the mouse C2C12 cell line 28 to analyze Celf1-CLIP tags within or in close proximity to the alternative splicing events sensitive to the absence of Celf1 in neonatal hearts. Out of the 45 events we found that 31 (69%) contained at least one Celf1-CLIP tag within the alternative region + /− 500 bp up-or downstream ( Supplementary Fig. S4c, left). The presence of Celf1-CLIP tags was significantly higher (p = 0.001) within + /− 500 bp around alternative spliced regions affected by Celf1 − /− than around 50 alternative splicing events responsive only to Mbnl1 − /− ( Supplementary Fig. S4c, right). Four events (9%) contained Celf1-CLIP tags in the flaking 500-800 bp, and the remaining 10 events (22%) did not contain any Celf1-CLIP tag within the alternative region or the flanking sequences (+ /− 800 bp) ( Supplementary Fig. S4c).
In summary, we identified alternative splicing events that are likely to be direct Celf1 targets in neonatal mouse heart. This conclusion is mainly based on the fact that they i) respond to Celf1 depletion in neonatal hearts, and ii) contain Celf1-CLIP tags within the alternative exons and/or the flanking regions.   Table S9). Correlation analysis of the gene expression levels expressed as FPKM (fragments per kilobase of transcript per million mapped reads) for Celf1 − /− and Celf1 + /+ biological replicates at PN3 showed high Pearson coefficients (1.00 and 0.91, respectively) demonstrating reproducibility (Fig. 5a). We next analyzed the number of transcripts differentially expressed (≥ 1.5 fold, with FDR ≤ 0.05) in Celf1 − /− compared to Celf1 + /+ hearts at PN3 and those differentially expressed during heart development between PN3 and PN38 in Celf1 + /+ mice (≥ 2.0 fold, with FDR ≤ 0.05; threshold differences are due to the extensive number of genes developmentally regulated). There were 483 genes differentially expressed in Celf1 − /− compared to Celf1 + /+ in PN3 hearts (287 down-regulated and 198 up-regulated) (Fig. 5b, left). During Celf1 + /+ mouse postnatal heart development 3,023 genes were down-regulated and 976 genes were up-regulated (Fig. 5b, right).
Global gene ontology analysis of the genes differentially expressed in Celf1 − /− hearts at PN3 revealed a significant enrichment in functions related to cell cycle and proliferation (up-regulated) and ion transport, immune response, and circadian rhythm (down-regulated) (Fig. 5c and Supplementary Tables S10 and S11). Celf1 has been shown to bind 3′ UTRs of transcripts and regulate mRNA stability [28][29][30] . To test the hypothesis that the absence of Celf1 in PN3 hearts affected mRNA levels of cell cycle genes we evaluated the presence of Celf1-CLIP tags within all of the 31 genes listed in the cell cycle category that was one of the most significant and the category with the highest number of genes (cell cycle, p = 2E-31) (Supplementary Table S10). We analyzed two groups of genes side by side: i) the 31 cell cycle genes (group a), and ii) a set of randomly selected 31 genes that did not show mRNA expression changes in Celf1− /− mice (group b). Visual inspection of all 62 genes using the UCSC genome browser revealed that cell cycle genes (group a) contained more Celf1-CLIP tags per gene in their 3′ UTRs in comparison with genes that were not affected by Celf1 loss (group b) (Fig. 6 show two examples of each group).
This observation led us to systematically compute the Celf1-CLIP tags present within the 3′ UTRs and those located within intronic and exonic regions for each of the cell cycle genes of group a (left side of Supplementary Table S12). We performed similar analysis in the 31 genes randomly selected that were unaffected by Celf1 deletion (group b) (right side of Supplementary Table S12). The analysis revealed that Celf1-CLIP tags are significantly more prevalent in 3′ UTRs of the regulated cell cycle genes (group a) compared to the set of control genes (group b) that are not affected by Celf1 loss of function (Fig. 6b). There were no significant differences in the presence of Celf1-CLIP tags within intronic or exonic regions of these two sets of genes (group a versus group b) however the total number of CLIP tags (3′ UTR + introns and exons) was significantly different reflecting the differences in the 3′ UTRs (Fig. 6b). In addition the correlation between Celf1-CLIP tags and the level of up-regulation (fold change) was higher for the 3′ UTRs (Pearson = 0.5) than for the intronic or exonic regions (Pearson = − 0.1) and total (Pearson = 0.2) (Fig. 6c) consistent with a correlation between Celf1 binding and mRNA levels.
Although our experiments and data cannot rule out other mechanistic scenarios, one possible explanation is that Celf1 may regulate the mRNA stability of a subset of cell cycle genes by binding to the 3′ UTRs. Assuming the hypothesis that Celf1 binding promotes mRNA decay 31 , mRNAs would be stabilized in the absence of Celf1 consistent with the observed mRNA up-regulation in PN3 Celf1− /− hearts. Further molecular experiments will be necessary to confirm this hypothesis and/or identify more complex mechanistic explanations for our global and high-throughput findings.

Discussion
We have characterized the Celf1 − /− mice complementing previous work from Luc Paillard and colleagues 22,32 . While past studies focused on the physiological impact of Celf1 depletion in growth, spermatogenesis, and skeletal muscle, we provide characterization of the cardiac features in Celf1− /− neonatal heart when Celf1 protein is normally high. Celf1 is a RNA binding protein that orchestrates multiple transcriptional and posttranscriptional programs important in normal development and disease. The mis-regulated expression of CELF1 seen in several diseases highlights the potential physiological importance of these coordinated networks. While the transcriptional and posttranscriptional effects of CELF1 over-expression have been studied in vivo [15][16][17] , the effects of Celf1 loss of function have not. This is particularly important in cardiac biology because Celf1 is down-regulated during postnatal development and its up-regulation in adult cardiomyocytes leads to heart failure 15,16,18 .
The alternative splicing events that we found to be responsive to the absence of Celf1 in PN3 heart were also enriched for Celf1-CLIP tags within the variable region or in the flanking sequences together with GU rich preferred binding motifs for Celf1. These data suggest that the identified splicing events are likely to be direct Celf1 targets. Many of these splicing events are regulated by Celf1 depletion in C2C12 cells (unpublished data), reducing concern of secondary effects in vivo.
In terms of gene expression, it is interesting to note that it was previously shown that a dominant negative CELF protein expressed in heart produced a mild phenotype in young animals that resolved with aging 33 . Similarly to that study, our results also described the spontaneous recovery of a heart phenotype. Overlapping analyses indicate that only three of the genes affected by the mild or severe CELF1 dominant negative were also affected in the Celf1− /− PN3 hearts using the 2.0 fold change in gene expression. For all three, the gene expression was affected in the same direction for the dominant negative and Celf1− /− hearts consistent with a loss of function. However the relatively small number of genes precludes making firm conclusions.
We identified a set of cell cycle genes that are up-regulated in the absence of Celf1 in PN3 hearts. We also provided evidence that the 3′ UTRs of the cell cycle gene mRNAs are enriched for Celf1 binding sites. This binding is expected to destabilize the mRNA and therefore the absence of Celf1 at early neonatal stages is expected to stabilize cell cycle transcripts consistent with the up-regulation we observed. In heart and other tissues such as liver and brain, cell cycle and cell proliferation genes are strongly down-regulated after birth to allow cell differentiation and tissue maturation for adult functions 16,34,35 . The fact that cell cycle genes are up-regulated when Celf1 is absent at PN3 together with the Celf1-CLIP tag presence in their 3′ UTRs is compatible with the idea that Scientific RepoRts | 6:35550 | DOI: 10.1038/srep35550 Celf1 is exerting a destabilizing role on those transcripts. However, we cannot rule out other molecular mechanisms involved in the up-regulation of cell cycle genes observed in neonatal hearts when Celf1 is absent. Among these other mechanisms we should mention regulators acting in concert with Celf1 and transcriptional modifiers (transcription factors, epigenetics modifiers, etc.) regulated by Celf1 depletion that are the direct drivers of the changes in cell cycle genes. In addition, during normal postnatal development Celf1 is down-regulated 15 as are the cell cycle genes 16 leading to an apparent discrepancy However, postnatal heart development is highly dynamic with regard to gene expression changes 16 (Fig. 5). This apparent discrepancy can be likely due to a scenario in which Celf1 contributes to maintaining a level of mRNAs by acting in combination with additional regulatory factors that become dominant repressors of mRNA, chromatin or transcription in the postnatal to adult  Table S12 left side shows group a and right side shows group b). Total Celf1-CLIP refers to the sum of those in the 3′ UTRs and those in the intron and exon regions. Results are shown as mean ± s.e.m. *p ≤ 0.05 Student t-test (2 tails). n.s: not significant. (c) Correlation graphs between the Celf1-CLIP tags and up-regulation levels (fold change) (3′ UTRs, the exonic and intronic regions, and total). Scientific RepoRts | 6:35550 | DOI: 10.1038/srep35550 transition. Although our work open important new molecular questions to address experimentally, this work provides a new high-throughput view of the transcriptional dynamics in the absence of Celf1 in neonatal hearts.

Methods
Animals. All animals were handled following the NIH Guidelines for Use and Care of Laboratory Animals that were approved by the Institutional Animal Care and Use Committee (IACUC) at Baylor College of Medicine. Dr. Luc Paillard from the Centre National de la Recherche Scientifique (CNRS, France) kindly provided us the Celf1 − /+ heterozygous mice (mixed 129Sv/BL6 strain) 22 . Animals were anesthetized before cervical dislocation (older than PN21) or decapitation (neonatal) and the hearts were removed. Blood and atria were carefully removed and the ventricles were flash frozen in liquid nitrogen and kept at − 80 °C until use. Littermates were used for comparisons between genotypes for physiological and molecular analyses to standardize differences between animals and reduce variability as much as possible due to the mixed strain background. While we did not perform a formal survival analysis, an analysis of animals not selected for euthanasia provides a sense of reduced Celf1− /− survival detected by PN3 suggesting either reduced fetal or neonatal survival and a fraction of animals survive beyond two months of age ( Supplementary Fig. S5).
Alternative splicing validation by RT-PCR. RT reactions were performed using the High Capacity cDNA RT Kit (Applied Biosystem #4368814) and PCRs using GoTaq DNA Polymerase (Promega #M7123). In both cases, manufacturer protocols were followed. PCR program contained the following steps: (i) 95 °C for 1 min 45 s, (ii) 28 cycles of 95 °C for 45 s, 57 °C for 45 s and 72 °C for 1 min, (iii) 72 °C for 10 min, and (iv) 25 °C for 5 min. Primer sequences (Sigma) for the alternative splicing events evaluated are described in Table S8. PCR products were analyzed by 6% PAGE. We quantified the percentage spliced in (psi) 37 of the alternative regions by densitometry using ImageJ plugin for gel analysis and following equation 1 (Eqn 1).
inclusion band inclusion band skipping band (1) Genotyping. DNA was extracted from tail clips using Direct PCR lysis reagent (Viagen #102-T) following manufacturer protocols. PCR reactions were then performed using GoTaq DNA Polymerase (Promega #M7123) and the following primers (Sigma): i) for the Celf1 mutant allele (expected size 690 bp), Celf1-ko-mut-F 5′ -GAA TTA TGG CCC ACA CCA GT-3′ and Celf1-ko-mut-R 5′ -GAG GGT TTT GGC TCC TAT CC-3′ and ii) for the wild type Celf1 allele (expected size 490 bp), Celf1-ko-wt-F 5′ -GGA CCA CCA GAG CTA CAG ACA-3′ and Celf1-ko-wt-R 5′ -ACC ACC CAG ACC AAC CAG AT-3′ . In both cases we multiplexed these PCRs using beta casein as an internal loading control with the following primers: Csn2-F 5′ -GAT GTG CTC CAG GCT AAA GTT-3′ and Csn2-R 5′ -AGA AAC GGA ATG TTG TGG AGT-3′ (expected band size 525 bp Computational processing of RNA-seq data. Paired-end reads were aligned to the mouse genome (mm9) using TopHat 2.0.5 38 . RSEM 39 was used for differentially expressed gene analysis and FPKM calculation. In this manner we computed the number of fragments that mapped into Ensembl gene models, and followed by edgeR 40 we obtained the differentially expressed genes from those showing a false discovery rate FDR < 0.05. Gene expression was quantified by FPKM 41 . Differential alternative splicing events were identified utilizing the SpliceTrap 42 that is based on Ensemble 65 gene model. Alternative splicing was quantified by the percentage of mRNAs containing an alternative region known as psi value 37 . The events with psi changes between two conditions |Δ psi| ≥ 15% were considered differential splicing events.
Gene ontology analysis. The Database for Annotation, Visualization and Integrated Discovery (DAVID) v6.7 43,44 was used for gene ontology analysis. We considered significant p values ≤ 0.05.
Motif analysis. MEME version 4.9.0 26 (motif> 6-15 bp) software was used for motif analysis (maximum number of motifs was set at ten, and any number of repetitions was allowed). The E-value is the enrichment of a motif based on the following parameters: background frequencies, log likelihood ratio, width, sites, and training set size.
Celf1 HITS-Clip data analysis. Celf1 HITS-CLIP data are available from the murine C2C12 cell line 28 .
These Celf1 HITS-CLIP data were downloaded from the European Nucleotide Archive (ENA, accession number ERP00078912). The 4-bp tags were trimmed and we removed the sequences composed primarily of Illumina adapter. These pre-processed reads were then aligned to the mouse genome (mm9) using Bowtie13 software 45 (allowing two mismatches). The reads with identical 5′ starts were collapsed into a single read to avoid potential PCR duplicates. Therefore, only unique mapped reads were considered Celf1 binding CLIP tags.

Electrocardiograms and echocardiograms. ECGs and echocardiograms were performed in the
Mouse Phenotyping Core at Baylor College of Medicine. Echocardiograms were performed using a Vevo 770 Visualsonics high-resolution ultrasound system with a 707B probe for the cardiac analysis. The animals were anesthetized with 2.5% isoflurane mixed with 100% oxygen and maintained with 1.0-1.5% isoflurane mixed with 100% oxygen during imaging. Images were obtained by an experienced imager in the short axis confirmation. All data was quantified using the Visualsonics Vevo analysis software package. Electrocardiograms were measured using the ECGenie (MouseSpecifics) equipment. Mice were placed atop the ECGenie pad and were allowed to acclimate for 5-7 min before data collection. Two segments of ECG data were then obtained for each mouse. All data was analyzed offline using the ECGenie software package.

Statistics.
Results are expressed as the mean ± s.e.m, Except for the χ 2 test shown in Supplementary Table S1, p-values were estimated by the Student´s T-test (two tails). p ≤ 0.05 was considered significant.