Dynamic alternative polyadenylation during iPSC differentiation into cardiomyocytes

Graphical abstract

Abstract

Alternative polyadenylation (APA) is an important post-transcription regulatory mechanism widely occurring in eukaryotes and has been associated with special traits/diseases by several studies. However, the dynamic roles and patterns of APA in cell differentiation remain largely unknown. Here, we systematically characterized the APA profiles during the differentiation of induced pluripotent stem cells (iPSCs) to cardiomyocytes by the previously published RNA-seq data across 16 time points. We identified 950 differential APA events and found five dynamic APA patterns with fuzzy c-means clustering analysis. Among them, 3′UTR progressive lengthening is the main APA pattern over time, the genes of which are enriched in cell cycle and mRNA metabolic process pathways. By constructing the linear mixed-effects model, we also indicated that TIA1 plays an important role in regulating APA events with this pattern, including genes essential to cardiac function. Additionally, APA and polyA machinery activity with another pattern can immediately respond to developmental signal-mediated stimuli at the early differentiation stage and result in a sharp shortening of the 3′UTR. Finally, a miRNA-APA network is constructed and several hub miRNAs potentially regulating cardiomyocyte differentiation are detected. Our results show the complex APA mechanisms during the differentiation of iPSCs into cardiomyocytes and provide further insights for the understanding of APA regulation and cell differentiation.

1. Introduction

Alternative polyadenylation (APA) is an important post-transcriptional regulatory mechanism that widely occurs in eukaryotes [1], [2]. By recognizing the cleavage and polyadenylation signals from different positions, APA can generate transcript isoforms with different lengths of 3′ untranslated regions (3′UTRs), thereby affecting transcript diversity and stability, transportation and translation efficiency [2]. APA has been recognized as a key player in regulating differentiation [3]. Progressive lengthening of 3′UTR has been found during mouse embryonic development [4]. In addition, members of the core polyadenylation factors, such as PCF11 [5] and NUDT21 [6], have been demonstrated to control cell differentiation by regulating APA. Recent studies have revealed that the APA pattern may vary greatly among different tissues [7] and differentiation stages [8]. For example, patterns of polyA site usage have been identified to be time-specific during retinal development [8], suggesting the complexity of the APA mechanism during differentiation.

Induced pluripotent stem cells (iPSCs), as a kind of pluripotent stem cells, are produced by ectopic expression of reprogramming regulators in somatic cells. Due to the ability of their self-renewal and pluripotency, iPSCs can unlimitedly proliferate and directly differentiate into a specific type of cell, and therefore are widely used in disease modeling, drug screening and cell replacement therapy [9]. Cardiovascular diseases (CVDs) are the leading cause of death around the world. Cardiomyocytes derived from iPSCs could provide unlimited supplies to repair damaged hearts and improve cardiac function [10]. However, dynamic changes in the cellular phenotypes during cardiomyocyte differentiation may result in immature cardiomyocytes [11]. Besides, injecting iPSC-derived cardiomyocytes suffers risks of teratoma formation [12], [13] or de-differentiation into the initial cell state [13]. To solve these problems for further clinical application, it is necessary to gain an in-depth understanding of the molecular mechanisms controlling cell fate decisions during cardiomyocyte differentiation. APA has recently been shown to play an important role in regulating cell fate [3]. Furthermore, an increasing number of APA disorders have been found to take part in heart diseases such as heart failure [14], cardiac hypertrophy [15], [16] and congenital heart disease [17]. For example, the loss of the distal-specific NQO1 isoform was identified during cardiac hypertrophy. Expression of NQO1 from the distal isoform could reverse cellular and molecular events of hypertrophy in the cardiomyocytes [15]. However, research on APA mechanisms during cardiomyocyte differentiation is rarely reported.

To bridge this gap, we collected a high-resolution public RNA-Seq data panel derived from iPSC differentiation into cardiomyocytes at 16 time points [18] and systematically analyzed their APA profiles. In addition to the expected progressive lengthening of 3′UTR across differentiation, we also observed sharp shortening of the 3′UTR at the early differentiation stage. We also explored the APA regulators and their effects on miRNA regulation, and found several molecules which may be involved in APA regulation. In conclusion, our systematic analysis provided a high-resolution dynamic APA atlas of iPSC differentiation into cardiomyocytes and revealed potential mechanisms regulating APA. These results will provide new insights for the understanding of APA regulation and cardiomyocyte differentiation.

2. Materials and methods

2.1. Data collection

We downloaded high-resolution public RNA-Seq data derived from iPSC differentiation into cardiomyocytes from GSE122380 [18]. The panel contains 297 samples across 16 time points. To confirm that TIA1 could regulate APA, we also downloaded BAM files of TIA1 shRNA (HepG2: ENCSR057GCF, K562: ENCSR694LKY) and control (HepG2: ENCSR603TCV, K562: ENCSR129RWD) from ENCODE [45]. Furthermore, to obtain miRNA expression data during cardiomyocyte differentiation, we downloaded three miRNA sequencing data of iPSC-derived cardiomyocytes from GSE60292 [56] and time-series miRNA microarray expression profiles across five time points (0, 3, 7, 10 and 14) from GSE35672 [19].

2.2. APA analysis

We used fastq-dump to convert the SRA format of all samples into FASTQ format. FastQC and Trimmomatic [20] were then used to perform quality control and trim on sequence files. We then aligned all sequence files to the hg19 using HISAT2 [21]. The obtained 297 BAM files were then used as input for DaPars v2.0 [22] to predict the de novo proximal polyA site and calculate the percentage of distal polyA site usage index (PDUI). The range of PDUI values is from 0 to 1. The larger PDUI represents the more transcripts using the distal polyA site, and vice versa. To ensure the accuracy of the prediction, we set the coverage threshold of the last exon ≥ 30 × as recommended by DaPars. In the PDUI matrix, the rows represent APA events, and the columns represent samples. If the number of PDUI missing values < 8 at a time point, we recognize the transcript with the true APA event at this time point. All APA events are the union of APA events across different time points. RNA-Seq read coverage was visualized using the R package Gviz [23].

We also applied DaPars v2.0 to obtain TIA1-dependent APA events by processing BAM files of TIA1 shRNA knockdown (TIA1_KD) and its control (TIA1_Control). TIA1-dependent APA events were defined as events with median-centered | PDUI TIA1_KD – TIA1_Control | ≥ 0.15.

2.3. Motif feature analysis

Sequences around proximal and distal polyA sites within ± 50 nt of each APA event were extracted by bedtools [24]. We then used DREME v4.11.2 [25] to separately enrich proximal and distal sequence features.

2.4. Principal component analysis

To identify different stages during cardiomyocyte differentiation, we utilized R function “prcomp” from the R package stats to perform a PCA analysis on median-centered PDUI values of the top 500 APA events ranked by standard deviation.

2.5. Characterization of APA patterns

We first identified differential APA events with the median-centered |ΔPDUI|≥0.15 across different time points. We next utilized the R package Mfuzz [26] to perform the fuzzy c-means clustering analysis on the median-centered PDUIs of differential APA events. The R function “fviz_nbclust” from R package factoextra, setting the option of the estimating method as “wss”, was used to determine the optimal number of clusters.

2.6. Gene set enrichment analysis

We generated an APA gene matrix based on all APA patterns. The columns of the matrix represent each APA pattern. Gene enrichment analysis was then performed by the “Express analysis” function in Metascape [27].

2.7. Differential gene expression analysis

We generated the matrix of raw counts of 297 samples using featureCounts [28] and converted it to the matrix of the transcripts per million (TPM). We then selected genes that were protein-coding, and had at least 10 samples TPM ≥ 0.1 and raw counts ≥ 6. In this way, 17,086 genes were finally reserved. We performed Wilcoxon signed-rank tests on TPM of day 0 samples and day 1 samples, and then obtained differentially expressed genes with FDR < 0.05.

2.8. Calculation of the polyadenylation machinery score

We used the method of Zingone, A. et al. [29] to calculate the polyadenylation machinery score. We first obtained the core polyadenylation regulatory gene list from biocarta_cpsf_pathway (M22041, https://data.broadinstitute.org/gsea-msigdb/msigdb/biocarta/human/h_cpsfPathway.gif). Next, we normalized the TPM expression matrix of each sample and extracted the median z-score of the biocarta_cpsf_pathway subset. Finally, the scores were calculated to represent PA machinery activity for the samples.

2.9. Strategy for screening factors specifically regulating APA in cluster 3

We proposed a new strategy to identify factors specifically regulating APA in cluster 3. We first performed differential expression analysis for longitudinal data on TREND factors in terms of the following criteria: (1) fold-change of expression across samples in different time points > 1.5; (2) coefficient of variation > 20 %; (3) median absolute deviation on the TPM values > 10. Next, we constructed the linear mixed-effects model using “lmer” from the R package lme4 to investigate the association between the longitudinal expression of differential TREND factors and PDUI of differential APA events. In each model, the PDUI value of each APA event was set as the dependent variable; gene expression and time as fixed effects. The cell line was set as the random effect to reduce noise from the genetic background. Therefore, each model was constructed as follows:

$$\mathrm{P}\mathrm{D}\mathrm{U}\mathrm{I}\mathrm{G}\mathrm{e}\mathrm{n}\mathrm{e}\mathrm{e}\mathrm{x}\mathrm{p}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{s}\mathrm{i}\mathrm{o}\mathrm{n}+\mathrm{T}\mathrm{i}\mathrm{m}\mathrm{e},\mathrm{r}\mathrm{a}\mathrm{n}\mathrm{d}\mathrm{o}\mathrm{m}=1|\mathrm{C}\mathrm{e}\mathrm{l}\mathrm{l}\mathrm{l}\mathrm{i}\mathrm{n}\mathrm{e}$$

APA events with FDR ≤ 0.05 were considered to be regulated by the corresponding TREND factors. Finally, we performed the Chi-Square test by setting other differential APA events as background to enrich cluster 3 specific regulators.

2.10. miRNA-APA network construction

We normalized the miRNA expression raw count matrix of three iPSC-derived cardiomyocytes by reads per million mapped reads (RPM), and defined miRNAs with an average RPM ≥ 100 as highly expressed miRNAs in cardiomyocytes. We downloaded the miRNA binding site information from TargetScanHuman 7.2 [30], then mapped binding sites of highly expressed miRNAs in aUTR regions of differential APA events. To construct the miRNA-APA network, we next used STRING [31] to perform PPI network analysis on genes of each APA cluster. Gene pairs with interaction scores > 0.7 were reserved. Finally, we merged miRNA-APA with the PPI of each cluster. Cytoscape [32] was used to visualize the miRNA-APA network. We also constructed the miRNA-APA networks of different time processes during differentiation. We first obtained miRNA expression profiles from the microarray dataset on day 0, 3, 7, 10 and 14. Then, we identified differential APA events between day 0 and 3, day 3 and 7, day 7 and 10, day 10 and 14, respectively. Finally, we predicted target sites of expressed miRNAs on the aUTR regions of differential APA events and constructed the miRNA-APA network during differentiation.

2.11. Statistical analysis

Basic statistical analyses such as the Wilcoxon signed-rank test and Chi-Square test, were performed using R language. The linear mixed-effects model was constructed using the R package “lme4”. The p values for results of the linear mixed-effects model were calibrated using FDR, and the FDR threshold was 0.05. The correlation between PDUI of APA events with the expression level of TIA1 was calculated by the R package “ggpubr”.

3. Results

3.1. Global patterns of APA events during cardiomyocyte differentiation

We downloaded a public RNA-Seq panel (GSE122380) to study dynamic APA profiles in cardiomyocyte differentiation. These data were obtained from 19 well-characterized human Yoruba HapMap cell lines during differentiation from iPSCs to cardiomyocytes [18]. To capture the entire differentiation process, each cell line was sequenced by RNA-Seq every 24 h for 16 days and finally a total of 297 samples with high-quality RNA-Seq datasets were obtained [18] (Fig. 1). To obtain global APA profiles during cardiomyocyte differentiation, we used DaPars v2.0 [33], a well-known APA algorithm, to predict the proximal polyA site and percentage of distal polyA site usage index (PDUI) of each transcript. A total of 6255 APA events belonging to 4830 known genes were obtained and an average of 4097 APA events were found for each time point.

Fig. 1.

The workflow for the study of APA during cardiomyocyte differentiation.

Previous studies have recognized a series of regulatory cis-elements responsible for the recognition of poly(A) sites. Hence, we separately extracted sequences within ± 50 nt of each proximal and distal polyA site of APA events, and used DREME [25] to enrich motif features. cis-elements for cleavage and polyadenylation, such as AAUAAA, UGUA, G/U-rich, A/U-rich and U-rich sequences [2], were identified around both distal and proximal polyA sites. However, the motifs of these cis-elements have slightly different between distal and proximal polyA sites. For example, the main G/U-rich motif around distal polyA sites is GUGUGUGU, while CGUGU is the main G/U-rich motif around proximal polyA sites. In addition, percentages of these cis-elements were more present at distal 3′end, especially the polyadenylation signal (PAS) AAUAAA (54.9 % around distal 3′ends and 11.1 % around proximal 3′ends), which is consistent with previous studies [2] (Fig. 2A).

Fig. 2.

Global landscape of alternative polyadenylation (APA) during cardiomyocyte differentiation. (A) Enriched motifs around proximal and distal polyA sites of APA events. The barplot showed the percentage of each motif feature. (B) PCA analysis result. (C) Boxplot of 3′UTR differences between transcripts with and without APA events. (D) Median-centered |ΔPDUI| value between early (0–3 days) and late (12–15 days) states for each cell line to represent global APA variation. |ΔPDUI| values were classified into different bins of 3′UTR sizes. The error bar represents the standard error (SE) of Median-centered |ΔPDUI| values across 19 individuals. (E) Density plot of aUTR sizes classified by different bins of 3′UTR sizes.

The average PDUI values of APA events across time points are close to 0.5, indicating nearly half of transcripts with APA events use distal PAS (Supplemental Fig. S1A). We found the number of APA events across different time points changed during cardiomyocyte differentiation, including the instantaneous increase after day 0 and the decrease after day 7 (Supplemental Fig. S1B). Principal component analysis (PCA) on PDUI values of APA events classified samples into four different groups (Fig. 2B), which are day 0, days 1–5, days 6–8 and days 9–15. PC1 captured 48.9 % APA changes. According to the description of data-oriented experiments, on day 0, the WNT signaling was activated by CHIR99021 and differentiation was initiated. On day 7, spontaneous mechanical beatings of cells began to appear [18]. These findings suggest that APA change is consistent with cardiomyocyte differentiation progress and could reflect the transition of the differentiation status to a certain extent.

We further analyzed the association between APA changes and 3′UTR lengths. We first compared the 3′UTR length of transcripts having APA events with those of other transcripts and found the former transcripts having significantly longer 3′UTRs (Wilcoxon test P = 6.82 × 10⁻⁹²) (Fig. 2C). Furthermore, we calculated global APA variation during differentiation by the median-centered |ΔPDUI| of each APA event between the early differentiation state (0–3 days) and the late state (12–15 days). After dividing 3′UTR lengths into different bins, we found that |ΔPDUI| of APA genes with longer 3′UTRs tended to vary larger during differentiation, especially APA events with 3′UTR size ≥ 2000 nt (Fig. 2D). The alternative 3′ UTR (aUTR) length is defined as the distance between the distal polyA site and the proximal polyA site. Our results show most aUTR sizes of APA events are ∼ 300 nt. Meanwhile, as 3′UTRs lengthen, aUTRs also lengthen (R² = 0.91) (Fig. 2E and Supplemental Fig. S2). These findings indicate genes with longer 3′UTRs are more likely to be regulated by APA during cardiomyocyte differentiation.

3.2. Dynamics of APA events during cardiomyocyte differentiation

We next explored APA changes during cardiomyocyte differentiation. We defined APA events with median-centered |ΔPDUI| ≥ 0.15 across different time points as differential APA events, and identified 950 (15.2 %) differential APA events with 3′UTR changes (Fig. 3A). To identify the dynamic patterns of APA events during cardiomyocyte differentiation, we utilized Mfuzz [26] to perform the fuzzy c-means clustering analysis on the median-centered PDUI values of differential APA events. In total, five diverse APA patterns were discovered, some of which were hardly classified into monotonic trends. The main features of these APA patterns are as follows: (1) cluster 1: with lower PDUI at the middle time point (day 6–8); (2) cluster 2: PDUI progressive decrease during differentiation; (3) cluster 3: PDUI progressive increase during differentiation; (4) cluster 4: slight fluctuation during differentiation; (5) cluster 5: PDUI rapid decrease on day 0 and slightly increase after day 8. Among them, cluster 3 contained the largest number of APA events (n = 267) (Fig. 3B). The median 3′UTR lengths of genes in cluster 3 and cluster 5 were ∼ 2000 nt, which are relatively longer than others (Supplemental Fig. S3A).

Fig. 3.

Dynamics of APA changes during cardiomyocyte differentiation. (A) Heatmap of differential APA events. Each APA event was normalized by Z-score. (B) Tracks of different APA patterns during cardiomyocyte differentiation. (C) Heatmap of gene enrichment results of APA patterns. (D) and (E) Read coverage of PAFAH1B1 and TRIOBP visualized by R package Gviz. The y-axis represents time points during cardiomyocyte differentiation. The proximal polyA site was obtained from DaPars results. The one-way repeated measures ANOVA analyses were performed for PAFAH1B1 (P = 2.69 × 10⁻⁶) and TRIOBP (P = 2.93 × 10⁻⁴).

Functional enrichment analysis showed different APA patterns had specific pathways. Most cell cycle progression-related GO terms, such as cell cycle, mitotic cell cycle process and mitotic G2-G2/M phases, were significantly enriched in cluster 3. In addition, cluster 5 was enriched by differentiation-related pathways such as the WNT signaling pathway and regulation of cardiac muscle tissue growth. Other common APA pathways were identified in more than one pattern, such as the metabolism of RNA [1] and cellular responses to stress [34] (Fig. 3C).

We further focused on the genes in cluster 3, which had the most changed APA events and the most enriched pathways, and found several heart-related APA genes. For example, PAFAH1B1 and TRIOBP (Fig. 3D, E), were genes in the enriched cell cycle pathway of cluster 3. Both 3′UTRs of PAFAH1B1 and TRIOBP were reported to be shortened in dilated cardiomyopathy (DCM) heart compared to normal heart tissues [14], suggesting abnormal myocardial proliferation might be caused by APA dysregulation in cardiomyocytes. Furthermore, we found more heart function-essential genes in cluster 3, such as CRLS1 [35] and QKI [36] (Supplemental Fig. S4A, B). Notably, the APA event of CRLS1 has not been reported in other studies. These results suggested APA may play an important role in heart development, and some APA events are worthy of further study.

Fig. 4.

The effect of small molecules induced WNT signaling changes on APA and upstream polyA regulators. (A) Brief introduction of cardiomyocyte differentiation strategy. CHIR: WNT activater on day 0. WNT-C59: WNT inhibitor on day 3. (B) Boxplot of BRD7 PDUI values across different time points. (C) Read coverage of BRD7 visualized by R package Gviz. The y-axis represents day 0 and day 1. The one-way repeated measures ANOVA analysis was performed for BRD7 (P = 7.06 × 10⁻⁹). (D) Pie chart displaying expression changes of APA genes in cluster 5. (E) Heatmap displaying expression changes of core polyA regulators. FDR was obtained from the Wilcoxon signed-rank test. * q < 0.05; ** q < 0.01; *** q < 0.001; **** q < 0.0001. (F) Boxplot of polyA machinery activity scores.

3.3. Small molecule-induced WNT activation enhances polyA machinery activity and results in the transient shortening of 3′UTRs

The WNT signaling pathway has been proven as one of the key regulators of cardiogenesis [37]. The GSK3 inhibitor CHIR99021 is one of the most common molecules that activate the WNT signaling and is generally utilized to induce iPSC into the differentiation of cardiomyocytes [37], [38], [39]. According to the description of the data-oriented experiment, CHIR99021 was added on day 0–1 to activate WNT signaling and initiate differentiation [18] (Fig. 4A). Among clusters of APA patterns, we found the PDUI in cluster 5 rapidly decreased at the beginning of differentiation (Fig. 3B). And, GO/pathway analysis showed that APA events in cluster 5 were related to WNT activation (Fig. 3C). Other studies have demonstrated that BRD7 could function as an activator of the WNT signaling pathway by negatively regulating the GSK3B phosphotransferase activity [40]. We also visualized that the read coverages of the BRD7 distal 3′UTR on day 1 was significantly less than that on day 0 (Fig. 4B, C). 3′ends shortening could lead to the loss of functional elements, such as miRNA binding sites. In this way, some oncogenes could be activated by escaping the repression effect of miRNA [2]. Therefore, we further analyzed the gene expression between day 0 and day 1, and found 57.7 % of APA genes in cluster 5 up-regraded on day 1 (Fig. 4D).

The polyadenylation mechanism involves several trans-factors, such as CPSF, CstF, CFI and CFII [2]. Further analysis of gene expression between day 0 and day 1, showed that the expression of most core polyA regulators increased on day 1 (Fig. 4E). Since the mRNA polyadenylation machinery to an extent influences proliferation/differentiation and 3′UTR sizes [41], [42], we next quantified the polyadenylation machinery activity score for each sample, which was recently described by Zingone, A. et al. [29] (Fig. 4F). We found polyadenylation machinery activity was negatively correlated with PDUI of differential APA events (median of Spearman’s rank correlation coefficient (r) = -0.26). Meanwhile, the trend of polyadenylation machinery activity also well-matched APA changes in cluster 5 (Fig. 3B and Fig. 4F), indicating polyadenylation machinery activation may result in 3′UTR shortening. Interestingly, we found the change of polyadenylation machinery could influence APA of its own regulatory components, such as CPSF6 (Supplemental Fig. S4C). Together, we showed that WNT signaling changes regulated by small molecules could quickly affect APA and its upstream factors, thereby influencing cardiomyocyte differentiation.

3.4. The APA pattern facilitates the identification of RBP regulators

Cluster 3 had the most differential APA events. We next asked whether there were other unrevealed factors regulating APA in cluster 3. To solve this, we designed a new computational strategy based on a list of 174 factors potentially regulating transcriptome 3′ end diversity (defined as TREND factors by Ogorodnikov et al. [5] to predict cluster 3 special regulators (See methods). We found that TREND factors preferentially regulated APA genes in cluster 3 and cluster 5 (Fig. 5A). Among 90 differentially expressed TREND factors during differentiation, 14 factors were identified specifically regulating cluster 3 (Chi-square test, odds ratio ≥ 1 and FDR ≤ 0.05), including core polyA genes, such as PABPC1 and PPP1CB (Fig. 5B). We noticed that TIA1, CIRBP and RBM5 regulated the most number of cluster 3 APA events. Inconsistent with the down-regulation of many APA core factors such as most CPSF and CstF, TIA1, CIRBP and RBM5 gradually up-regulated APA events during cardiomyocyte differentiation (Supplemental Fig. S5), suggesting different kinds of factors might exist to affect APA. Among them, CIRBP [43], [44] was recently recognized as the auxiliary APA factor [34]. Up-regulation of CIRBP promotes longer transcript formation under low temperature, which is consistent with our results [43]. As TIA1 has correlations with most cluster 3 APA events (Fig. 5C), we further wondered whether TIA1 has a similar function that directly leads to 3′UTR lengthening. Thus, we analyzed APA events in TIA1 shRNA RNA-Seq data of K562 and HepG2 cell lines from ENCODE [45]. We found that TIA1 knockdown resulted in global 3′UTR shortening (Fig. 5D, E), suggesting TIA1 prefers to select distal PASs. As TIA1 is critical for eukaryotic stress response and stress granule formation [46], we also found the PDUI of several genes involved in the stress response, such as TIMP2 and TMEM248, correlated well with TIA1 expression (Supplemental Fig. S6A, B). Furthermore, several genes essential to heart function and differentiation were identified to be regulated by TIA1, such as WDR1 [47], VAPB [48], RBM8A [49] and EIF2S1 [50] (Fig. 5C; Supplemental Fig. S4D-G), suggesting TIA1 may play an important role in the cardiomyocyte differentiation progression. Interestingly, both TIA1 and CIRBP could affect cell proliferation [51], [52] and result in the alteration of the cell cycle [53], [54], which may further help to optimize the strategy of iPSC differentiation into cardiomyocytes.

Fig. 5.

New pipeline to predict cluster 3 special regulators from TREND factors and TIA1-regulated APA results. (A) Percentages of differential APA events regulated by 90 TREND factors. Each row represents the specific factor that significantly regulates the percentage of APA events in different clusters. (B) Enrich regulators in cluster 3 by Chi-Square test. (C) TIA1-regulated APA events in cluster 3. The x-axis represents the Spearman correlation coefficient (r) between the expression of TIA1 and PDUI of the APA event. (D) APA global change after TIA1-knockdown in K562 cell line. (E) APA global change after TIA1-knockdown in HepG2 cell line.

Fig. 6.

Statistics of miRNAs in aUTRs and the miRNA-APA network. (A) Distribution of miRNA binding sites in aUTRs of differential APA events. (B) Distribution of miRNA binding sites in aUTRs of each APA pattern. The bar represents the total number of miRNA binding sites of genes in each cluster. The dotted line represents the average number of miRNA binding sites of genes in each cluster. (C) miRNA-APA network. The lines represent the APA gene-affected miRNA and gene-gene with STRING PPI > 0.7.

3.5. miRNA-APA network during cardiomyocyte differentiation

APA events could cause the gain/loss of functions of miRNAs and previous studies have shown miRNAs are essential to embryonic cardiac morphogenesis [55]. Thus, we further analyzed the change of miRNA binding sites which would be caused by APA events during cardiomyocyte differentiation. We first collected a set of iPSC-derived cardiomyocyte miRNA expression data (GSE60292) [56], and used the RPM value to normalize the miRNA expression. We selected highly expressed miRNAs with RPM ≥ 100 in cardiomyocytes, and then mapped these miRNA binding sites within the aUTRs of differential APA events. In this way, we totally found 143 cardiomyocyte important miRNA sites potentially gain or loss caused by 830 differential APA events. Among these APA events, 54.4 % owned>10 miRNA sites on their aUTRs. All changed APA genes averagely contained 32 highly expressed miRNA sites (Fig. 6A). As with the longer aUTRs than other APA patterns (Supplemental Fig. S3B), cluster 3 and cluster 5 also owned larger numbers of total and average miRNA binding sites (Fig. 6B). We next used STRING [31] to identify protein–protein interactions (PPI) for each APA pattern. The results were then used to construct the miRNA-APA network (Fig. 6C). The top three miRNAs regulating the network were miR-3613-3p, miR-590-3p and miR-551b-5p. Of them, miR-590-3p has been shown to regulate the differentiation of several cardiac lineages including cardiomyocytes. [57], [58], [59], while the associations between miR-3613-3p/miR-551b-5p and cardiomyocyte differentiation have not been reported. However, miR-3613-3p was reported to play an important role in cardiac fibrillation [60] and could affect cell proliferation and cell cycle [61]. And miR-551b-5p was reported to take part in cardiomyocyte autophagy in diabetic cardiomyopathy [62]. Since these miRNAs are related to the biological processes of cardiomyocytes, they may also have the potential to regulate cardiomyocyte differentiation. On the other hand, we found several miRNAs, such as miR-141-3p, miR-204-5p and miR-101-3p, were specifically bound to cluster 5 (Supplemental Fig. S7). Cluster 5 owned most WNT genes, and these miRNAs were also reported to inhibit WNT signaling and affect differentiation [63], [64], [65].

In addition, we further analyzed the mainly related miRNAs during differentiation. We first identified expressed miRNAs at day 0, 3, 7, 10, and 14. Then, we predicted the miRNA binding sites which may be affected by the change of aUTR during the time process. We found 914, 511, 454 and 288 miRNA sites potentially gain or loss during day 0–3, 3–7,7–10, 10–14 differentiation, respectively. By constructing miRNA-APA networks for each time process, we found that miRNA target sites affected by APA in differentiation are time-specific (Supplemental Fig. S8). Of related miRNAs, miR-590-3p was identified in day 0–3, 7–10, 10–14 differentiation, indicating that miR-590-3p might be an important mediator of APA events during the differentiation.

Taken together, these results illustrate that the miRNA-APA network we constructed contains a series of miRNAs that regulate the biological functions of cardiomyocytes, among which miR-590-3p has been shown to affect cardiomyocyte differentiation. Other miRNAs such as miR-3613-3p and miR-551b-5p can be used as novel regulators of cardiomyocyte differentiation for further studies.

4. Discussion

As a common post-transcriptional mechanism regulating the 3′ends of transcripts, APA has been proven essential to cell fate decision [3]. In previous studies, APA could reflect differences in cancer/normal cells [66] and switching of proliferative/arrested cell state [42]. Furthermore, APA was considered the superior indicator for classifying cell types compared to the expression [67]. In many APA-related analyses for differentiation, due to the insufficient temporal resolution of differentiation data, most APA trends during differentiation have been reported to be monotonous [2], [4]. To fully characterize the relationship between APA and cell differentiation, it is necessary to reveal the variation of APA in a more delicate way. In this study, by collecting high-resolution expression data and using c-means clustering to systemically construct APA profiles of cardiomyocyte differentiation, we accurately captured multiple APA patterns with diverse trends, indicating the complexity of APA during cardiomyocyte differentiation. We found that APA was sensitive to the switch of differentiation stages. Meanwhile, different APA patterns corresponded to different differentiation processes. These findings further suggested that APA could well characterize the cell differentiation state.

In our study, we observed a sudden shift in APA within 24 h. The massive changes during this stage are most likely caused by the massive alliterations in signaling cascades since true differentiation might take more time. According to the description of data-oriented experiments, WNT signaling is induced by the small molecule CHIR99021. Thus, we indicated that APA could rapidly respond to CHIR99021-mediated WNT signaling activation at the early differentiation stage, thereby leading to 3′UTR shortening and upregulation of some APA genes, including the WNT coactivator BRD7. Similarly, it has been reported that mTOR signal activation leads to 3′UTR shortening and enhances the translation of specific mRNAs [70]. In addition, the core APA factors PCF11 [5] and NUDT21 [73] were shown to regulate key genes of the WNT and NF-KB pathways, respectively. These results show APA plays an important role in response to multiple pathways and signal transmission. However, the APA quickly responded to the small molecules still needs further experimental validation.

CHIR99021 might affect the expression of core APA factors, and further mediate APA changes during 24 h. Also, more and more APA factors are reported to play an important role in differentiation [5], [6]. Molecular and drug design based on these factors opened a new way to increase efficiencies of differentiation in vitro [3]. For example, previous studies showed that small molecules T4 or T5 could suppress protein levels of PABPN1, thereby resulting in 3′UTR shortening [74]. However, there are few APA studies involving compounds or drugs. Therefore, the design of novel compounds that regulate APA core factors might help promote the development of APA in the field of regenerative medicine.

Most studies focused on core APA factor-mediated changes in 3′UTR length [2]. However, several non-core factors have also been found to regulate APA events [34]. In our study, we found that TIA1 and CIRBP specifically regulate the progressive 3′UTR lengthening during cardiomyocyte differentiation. CIRBP has been reported as the APA cofactor [34]. However, the association between TIA1 and APA has not been well studied. Using TIA1-knockdown data, we confirmed the role of TIA1 in choosing distal PASs. The previous study showed that TIA1 could function as the stress response factor and promote the formation of stress granule, which is used to store the translationally stalled mRNAs [68]. As APA is a good responder to stress and differentiation, it indicates that TIA1 might be a link between stress and differentiation. For example, Zheng et al. identified that TIA1 preferentially binds long isoforms, which is consistent with our results in cluster 3. In addition, APA under stress and differentiation has the same genes with long aUTRs [68]. In our research, some TIA1-regulated target APA events were also reported under cellular stress, such as TIMP2 and TMEM248. Interestingly, both CIRBP and TIA1 have the ability to control cell fate [51], [52], indicating these factors with important APA regulatory functions can be used as potential markers for the optimization of differentiation strategies in vitro.

In this study, we identified APA profiles and associated factors across different time points, however, several limitations should be noted. First, the identification of APA events is based on RNA-sequencing data and bioinformatics tools, which may generate false positive events to some extent. Further, the associations between APA and other factors were also calculated based on correlation analyses and bioinformatics prediction. In addition, during the miRNA-APA network analysis, we noticed that unavoidable impacts, such as poor matching of time points and batch effects may exist when merging data from different differentiation experiments. Thus, the identification of APA events and the relationship between APA and other miRNAs still need additional experimental experiments.

5. Conclusion

In this study, we established high-resolution dynamic APA profiles with various patterns during cardiomyocyte differentiation. We observed distinct APA patterns associated with specific differentiation stages, including 3′UTR rapidly shortening at the early differentiation stage, and 3′UTR progressive lengthening during the whole differentiation stage. We found APA and most cleavage and polyadenylation factors can rapidly respond to WNT signaling stimuli mediated by CHIR99021 during the early differentiation phase of cardiomyocytes. This result not only indicates that developmental signaling could contribute to 3′UTR rapidly shortening by leading to the upregulation of known core APA factors, but also demonstrates the feasibility of small molecules to regulate APA. On the other hand, we identified TIA1 might play an important role in regulating 3′UTR progressive lengthening during cardiomyocyte differentiation. We discovered TIA1 could regulate some heart-essential genes, such as WDR1 during differentiation. Meanwhile, the expression trend of TIA1 during differentiation differs from that of most known cleavage and polyadenylation factors, suggesting that there are multiple mechanisms regulating APA during cardiomyocyte differentiation. Finally, we constructed a miRNA-APA network specific to cardiomyocytes to explore the effect of 3′UTR changes on important miRNAs during cardiomyocyte differentiation. Taken together, we in-depth investigated the dynamic patterns and potential mechanism of APA during cardiomyocyte differentiation. To our best knowledge, this is the first research systemically evaluating the role of APA in cardiomyocyte differentiation. Our findings could provide new insights for the design of differentiation strategies and cardiac disease models.

CRediT authorship contribution statement

Yanbo Yang: Conceptualization, Methodology, Visualization, Writing - original draft, Writing - review & editing. Xiaohong Wu: Methodology, Writing - review & editing. Wenqian Yang: Conceptualization, Methodology. Weiwei Jin: Conceptualization, Methodology. Dongyang Wang: Methodology. Jianye Yang: Methodology. Guanghui Jiang: Methodology. Wen Zhang: Conceptualization. Xiaohui Niu: Conceptualization, Methodology, Writing - review & editing. Jing Gong: Conceptualization, Methodology, Supervision, Writing - review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article:

Supplementary Fig. S1

Statistics of PDUI values and numbers for APA events. (A) Median-centered PDUI value across different time points during cardiomyocyte differentiation. (B) The numbers of APA events across different times during cardiomyocyte differentiation.

Supplementary Fig. S2

3’UTR size (x-axis) and aUTR size (y-axis) of each APA event. The correlation coefficient (r) and P value were calculated by Pearson analysis.

Supplementary Fig. S3

3’UTR and aUTR lengths of APA clusters. (A) Boxplot of 3’UTR lengths. (B) Boxplot of aUTR lengths.

Supplementary Fig. S4

APA event visualization of seven important genes by R package Gviz. (A-B) APA events of CRLS1 and QKI genes which are important to heart-related functions during cardiomyocyte differentiation. (C) APA events of CPSF6 gene, an APA factor which might result in APA change in its own gene. (D-G) APA events of WDR1, VAPB, RBM8A and EIF2S1 which might be regulated by TIA1 expression and are essential to cardiac functions. The y-axis represents time points during cardiomyocyte differentiation. The proximal polyA site was obtained from DaPars results. The one-way repeated measures ANOVA analyses were performed for these APA events. The p values are 1.41×10⁻¹⁴ for CRLS1, 9.26×10⁻¹² for QKI, 2.04×10⁻⁴ for CPSF6, 3.36×10⁻⁴ for WDR1, 7.72×10⁻⁵ for VAPB, 1.39×10⁻¹⁷ for RBM8A and 4.65×10⁻¹⁰ for EIF2S1, respectively.

Supplementary Fig, S5

Heatmap of the expression of core APA factors and TIA1, CIRBP and RBM5.

Supplementary Fig. S6

(A) Positive correlation between PDUI values of TIMP2 and the expression of TIA1. (B) Positive correlation between PDUI values of TMEM248 and the expression of TIA1.

Supplementary Fig. S7

miRNAs in the miRNA-APA network across different APA clusters.

Supplementary Fig. S8

miRNA-APA networks during day 0-3 (A), 3-7 (B),7-10 (C), 10-14 (D) differentiation.