Natural Cardiogenesis-Based Template Predicts Cardiogenic Potential of Induced Pluripotent Stem Cell Lines

Abstract

Background

Cardiac development is a complex process resulting in an integrated, multi-lineage tissue with developmental corruption in early embryogenesis leading to congenital heart disease. Interrogation of individual genes has provided the backbone for cardiac developmental biology, yet a comprehensive transcriptome derived from natural cardiogenesis is required to gauge innate developmental milestones.

Methods and Results

Stage-specific cardiac structures were dissected from eight distinctive mouse embryonic time points to produce genome-wide expressome analysis across cardiogenesis. In reference to this native cardiogenic expression roadmap, divergent iPSC-derived cardiac expression profiles were mapped from pro-cardiogenic 3-factor (SOX2, OCT4, KLF4) and less-cardiogenic 4-factor (plus c-MYC) reprogrammed cells. Expression of cardiac-related genes from 3F-iPSC differentiated in vitro at days 5 and 11 recapitulated expression profiles of natural embryos at days E7.5–E8.5 and E14.5–E18.5, respectively. In contrast, 4F-iPSC demonstrated incomplete cardiogenic gene expression profiles beginning at day 5 of differentiation. Differential gene expression within the pluripotent state revealed 23 distinguishing candidate genes among pluripotent cell lines with divergent cardiogenic potentials. A confirmed panel of 12 genes, differentially expressed between high and low cardiogenic lines, was transformed into a predictive score sufficient to discriminate individual iPSC lines according to relative cardiogenic potential.

Conclusions

Transcriptome analysis attuned to natural embryonic cardiogenesis provides a robust platform to probe coordinated cardiac specification and maturation from bioengineered stem cell-based model systems. A panel of developmental-related genes allowed differential prognosis of cardiogenic competency, thus prioritizing cell lines according to natural blueprint to streamline functional applications.

Introduction

Embryonic development of the heart encompasses a complex orchestration of events deployed in a predictable and sequential pattern.^1–4 Disruption of these organized events leads to developmental abnormalities underlying congenital heart disease^{5, 6} comprising a variety of manifestations for which molecular fingerprints are actively investigated.^7–9 With the intricacy of cardiac development, most studies focus on the dissection of discrete cardiogenic pathways as well as localized interactions between limited numbers of factors. However, integration of the multiple simultaneous processes and networks defining embryonic development of the heart is required to extend the molecular understanding of cardiac differentiation and congenital heart disease. Current knowledge based on comprehensive expression analysis of genes responsible for specific components of cardiac specification^{10, 11} would benefit from genome-wide approaches utilizing native tissue platforms to create a spatial-temporal expressome reflecting natural cardiac development.

Gene expression cascades, regulating the development of chamber-specific features of the heart, are incompletely recapitulated during in vitro cardiogenic differentiation of pluripotent stem cells¹² in part due to inconsistent cell culture conditions that expose the inevitable variability between stem cell lines. With the advent of induced pluripotent stem cell (iPSC) technology, sources for pluripotent stem cells have significantly expanded, providing a wide range of reagents to recreate patient-specific cardiac tissue.^{13, 14} However, nuclear reprogramming has been shown to impact downstream cardiogenicity of iPSC,^15–19 giving rise to lines with variable cardiogenic potential and threatening the reproducibility of disease-specific model systems. As the demand increases for “disease-in-a-dish” modeling efforts, bioengineered cardiac progeny are essential for the advancement of mechanistic and drug toxicity studies and ultimately safe and effective cell-based therapies for cardiovascular disease.^20–22 Therefore, a systematic understanding of natural cardiogenesis and the ability to target novel molecular switches that regulate the differentiation of the ventricular myocardium is required to identify conserved patterns of sequential gene expression. The molecular blueprint of ventricular myocardium could improve our understanding of deficiencies leading to congenital syndromes and augment the resources for regenerative strategies as well as disease modeling and drug testing. A cardiogenic roadmap calibrated according to natural embryonic cardiac tissue would establish developmental milestones and innate mechanisms to gauge normal and abnormal cardiogenic processes within bioengineered cardiac progeny.

Herein, we leverage the genome-wide transcriptome of innate cardiac developmental biology in the mouse model and validate the utility of this spatial-temporal roadmap as a tool to predict in vitro cardiogenic potential of independent iPSC lines. By extracting a set of developmentally regulated genes that are differentially expressed in undifferentiated pluripotent lines stratified according to high and low functional cardiogenic potential, we have developed a predictive indicator of downstream cardiogenicity applicable for probing individual iPSC lines. The ability to diagnose and discriminate iPSC clones according to pathways conserved in native cardiogenicity provides an innovative approach to select individual cell lines for disease modeling, drug testing, and ultimately the efficient production of stage-specific cardiogenic progeny for personalized regenerative applications.

Methods

Tissue dissection and RNA isolation

Staged mouse embryos were harvested from timed mating pairs and confirmed using morphology and somite count as inclusion criteria for each developmental time point. Whole embryos (E7.5), heart tubes (E8.5), or left and right ventricles (E9.5, E12.5, E14.5, E18.5, newborn and adult mice) were micro-dissected and pooled into triplicate sample sets. Approximately 30–50 embryos/hearts were included in each replicate for E7.5, E8.5 and E9.5; 10–20 hearts were collected per E12.5 or E14.5 sample; 3–5 hearts were pooled for each E18.5 replicate and one was used for newborn and adult samples. R1 embryonic stem cells (ESC) were used as pluripotent reference. RNA was extracted using the Qiagen RNeasy kit. Due to high protein content and larger size of samples corresponding to newborns and adult animals, samples were fast-frozen and pulverized to optimize digestion in the first steps of the RNA extraction process.

Cell reprogramming

iPSC lines used for transcriptome analysis were derived from mouse embryonic fibroblasts (MEFs) exposed to viral cocktails containing either three (SOX2, OCT4 and KLF4) or four (plus c-MYC) reprogramming factors.^{16, 18, 23} Infective units were composed of pEX-QV (GAG/POL) expression vector and pMD-G packaging plasmid with individual expression constructs: psin-c-MYC, psin-KLF4, psin-OCT4 and psin-SOX2.²³ Five days after infection, cells were replated onto mitomycin-inactivated feeders and cultured in mouse ESC medium containing leukemia inhibitory factor (LIF) for 2–4 weeks. Compact and round colonies emerged within 2 weeks for 4 factor-reprogrammed cells or 3–4 weeks for 3 factor-reprogrammed cells. These colonies were individually picked and grown on inactive feeders in LIF-containing medium. Cell lines were split every 2–3 days and expanded in the pluripotent state prior to differentiation.

iPSC lines used for test and validation purposes were generated from multiple sources and different reprogramming strategies including MEFs, adult cardiac fibroblasts (CFs) or adult tail tip fibroblasts (TTFs) through lentiviral or doxycycline-inducible reprogramming. Specifically, line A was derived from CFs from Gt(ROSA)26Sor^{tm1(rtTA*M2)Jae}Col1a1^{tm3(tetO-Pou5f1,-Sox2,-Klf4,-Myc)Jae}/J mice carrying a doxycycline-inducible cassette for the aforementioned four reprogramming factors. Isolated fibroblasts were plated directly onto inactivated feeders and cultured in EmbryoMAX DMEM supplemented with 15% fetal bovine serum (FBS), LIF and 2 μg/mL doxycycline, with daily medium changes. Colonies were picked after 2–5 weeks and maintained in doxycycline-free medium after passage 4. Line B was generated from MEFs with an integrative approach using lentiviruses expressing Oct4, Sox2, and Klf4 (Open Biosystems). Line C was produced from TTFs and sample D was derived from CFs, both using the 3 factor-integrative approach. Colonies were picked between weeks 4 and 9 after infection. Three lines from each of the six combinatorial approaches were used for final validation. Undifferentiated cells were used for qRT-PCR analysis.

Cell differentiation

Pluripotent cells were spontaneously differentiated following an embryoid body (EB) formation protocol in differentiation medium with 20% FBS in the absence of LIF.¹⁸ Drops containing ~600 cells in 25 μl differentiation medium were deposited on bacteriological plate lids and cultured for 48 h to force cell aggregation, followed by 48 h of suspension culture. At day 5 of differentiation, EBs were plated on gelatin and allowed to grow for 6 more days. Approximately 48 h after plating (day 7), EBs started displaying areas of rhythmical contraction corresponding to newly formed cardiac cells. Spontaneous contractile activity was quantified in biological triplicates, including 20–60 EBs per replicate (total of 90–140 EBs characterized per time point) for a total differentiation time of 11 days. Samples were collected at baseline (day 0), suspension EBs (day 5) and at the end of the differentiation process (day 11). Pluripotent lines used for validation were differentiated using aggrewells.²⁴ In brief, cells were trypsinized into a single cell suspension and centrifuged in microwell-containing plates at a density of 2.4 million cells to each aggrewell in differentiation medium. Two days later, aggregates were transferred to low adherence plates and cultured in suspension for 48 h. EBs were then plated onto gelatinized plates and beating activity was monitored for 20 days. Plates were individually monitored in a blinded fashion, such as cell line identity was only revealed at the end of the analysis. Statistical relevance of differences in beating activity was probed using general linear model (glm) in R programming.

qRT-PCR

cDNA was synthesized with the reverse transcriptase supermix kit from Invitrogen.¹⁸ All qRT-PCR was performed with an Illumina Eco-Real Time PCR system. For data processing, readings were normalized to Gapdh from the same plate. Expression values were made relative to the lowest expression value among all samples for one given gene. Error bars represent SEM. Primers included in the predictive panel can be found in Supplementary Table 1.

Statistical methods

Statistical microarray data analysis

Affymetrix Mouse Genome 430 2.0 arrays were used to screen for differential gene expression levels according to standard protocols in the Mayo Clinic Advanced Genomic Technology Center (AGTC) Gene Expression Core (GEC). Raw microarray image data were analyzed using several R and Bioconductor packages and custom R scripts for quality assessment/quality control (QA/QC) and differential analysis.^25–29 The QA/QC process included assessment of the raw microarray images, MAplot, normalized unscaled standard error, residual images from the RMA model, relative log expression, RNA degradation based on all the probes on the microarray using internal R scripts and R/Bioconductor packages. Standard Affymetrix quality metrics were also assessed, such as 3′/5′ ratios, background, Scaling factor, control probes, GAPDH and Percent Present calls. Differential analysis was performed using the Limma package in R.³⁰ Differential genes were selected based on p-value less than 0.01 after false discovery rate (FDR) control to correct for multiple comparisons and fold change (FC) greater than 4. All the hierarchical clustering analyses to generate the dendrograms were performed using R based on the algorithm of average Euclidean distance. Gene function enrichment analysis on differentially expressed genes was performed using MetaCore (GeneGo) software.^31–33 Raw microarray data is available in NCBI GEO database under accession number GSE43197.

A 72-gene panel was designed to extract the cardiac component (in terms of gene expression levels) from the mixed population resulting from spontaneous unguided differentiation, enabling comparison with the all-cardiac probes represented in the developmental roadmap. Genes were included based on cardiac-related functions as curated in GenMAPP pathway database and current knowledge of cardiac development. Representative pluripotency and gastrulation markers were included to visualize loss of the undifferentiation state and transition into maturing phenotypes.

Pearson’s correlation and hierarchical clustering heatmap

All correlation values in this manuscript are Pearson’s correlation which are calculated based on average gene expression across triplicate biological samples at each time point. Calculation and clustering were done using R program functions.

Cardiogenic score calculation

The cardiogenic score was calculated based on qRT-PCR values for 12 genes. First, in order to normalize the scale for individual genes, expression values were re-calculated as percentages, using the highest value within each gene as 100%. Genes naturally upregulated in pluripotent stem cells and initial stages of development were designated as “positive predictors”, while genes showing low values in early cardiogenesis were designated as “negative predictors”. The average contribution for all the negative predictors was subtracted from the average contribution of all the positive predictors, resulting in a composite cardiogenic score.

Results

Natural cardiogenesis roadmap revealed through high content microarray analysis of stage-specific embryonic heart samples

To establish a spatial-temporal expressome of natural cardiogenesis, stage-specific embryonic cardiac structures were dissected, collected, and profiled. Samples ranging from pluripotent cells to adult left and right ventricular tissue, including whole embryos at day E7.5, heart tubes from E8.5 embryos, as well as left and right ventricles were surgically obtained from developing mice at E9.5, E12.5, E14.5, E18.5, newborns and adults (Figure 1A) and subjected to microarray analysis. QA/QC analyses of isolated RNA and microarray raw data ensured that the surgical procedure required for harvesting and the microarray hybridization process did not compromise bioinformatic analysis and data interpretation, as illustrated by parallel slopes in the RNA degradation plots showing comparable good quality for all studied samples (Figure 1B). Unsupervised hierarchical clustering based on all the probesets in the microarrays, representing relative distance among samples, defined developmental units grouping pre-cardiac samples (undifferentiated cells and E7.5 whole-embryos), heart tube (E8.5 and E9.5), mid-gestation (E12.5 and E14.5), perinatal (E18.5 and newborn), and adult-derived samples (Figure 1C). In this way, a temporally regulated dataset fulfilling stringent quality standards established a cardiogenic roadmap calibrated at discrete developmental stages within natural embryogenesis.

Figure 1.

Establishment of a cardiac development roadmap by high content screening of embryonic cardiac tissue. A. Stage-specific samples subjected to microarray analysis included mouse embryonic stem cells, mouse embryonic stage E7.5 whole embryos, heart tubes from E8.5 and independent samples for left and right ventricles (LV and RV) of E9.5, E12.5, E14.5, E18.5, newborn and adult mouse hearts. B. RNA degradation plot for QA/QC. Each line represents one sample with probes ordered (on x-axis) according to relative locations on transcripts. Probe intensities were calculated across whole chip for all probes at that position. Lines were shifted for visualization. C. Unsupervised hierarchical clustering of cardiac development samples based on all the probesets in the microarray, with different tones of the same color representing RV and LV for the same stage. Bar height correlates with differences among samples.

Stage-specific expressome gauges cardiogenic potential among pluripotent cell lines

Nuclear reprogramming allows derivation of iPSC from somatic starting sources.³⁴ However, varying degrees of cardiogenicity have been reported depending on the reprogramming factors used to reinstate pluripotency.^{16, 17} Herein, somatic cells were reprogrammed using the pluripotency factors (SOX2, OCT4, KLF4 and c-MYC; 4F-iPSC) or three factors (SOX2, OCT4 and KLF4; 3F-iPSC) to achieve the pluripotent state according to stringent pluripotency criteria.¹⁶ Utilizing a pseudo-3-D culturing system based on embryoid body formation, pluripotent stem cells were differentiated towards cardiac lineages and monitored for beating activity as a functional surrogate for proper cardiogenesis (Figure 2A). Beating activity has been shown to correlate with levels of expression of cardiac transcription factors and protein expression³⁵ and provides a readout of electromechanical integration and functionality of derived cardiac cytotypes. Established R1 ESCs, extensively characterized for pluripotency and cardiac differentiation,^36–38 were used as a reference pluripotent stem cell line. In agreement with previous in vitro work,¹⁶ contracting cardiac tissue appeared in differentiating 3F-iPSC following the time-course consistent with R1 ESCs (Figure 2B), while 4F-iPSC did not spontaneously produce a beating cardiac phenotype according to the standardized culture conditions used herein (Figure 2B). Despite previously demonstrated cardiogenic potential of 4F-iPSC in the contexts of diploid aggregation¹⁶ and myocardial regeneration,³⁹ the in vitro environment was capable of distinguishing the innate difference in the cardiogenic potentials between 3F-iPSC and 4F-iPSC lines according to production of beating cardiac tissue under equivalent conditions. Based on this phenotypic observation of functional beating activity, we compared the in vitro expression profiles of pluripotent stem cell lines and the established natural cardiac development roadmap to assess whether the assayed cell lines followed canonical differentiation pathways as defined in the natural embryo. To validate the utility of the in vitro model system of tissue-specific differentiation, a panel of 72 markers including pluripotency, gastrulation, early cardiac and late cardiac markers was selected from the roadmap of natural cardiogenesis in the mouse model (Supplementary Table 2 and Figure 2C). The obtained dataset showed age-dependent changes in gene expression as heart development progressed, with pluripotency patterns being replaced by more mature expression patterns. (Figure 2C). Gene expression heatmap for the same group of cardiac-related markers in R1-ESCs at day 0, 5 and 11 of spontaneous differentiation (Figure 2D) highlighted expression patterns corresponding to developmental stages R1, E7.5 and E12.5-newborn respectively (Figure 2C and Supplementary Figure 1). These results indicate that in vitro differentiation of ESCs mimics stage-specific cardiogenesis, validating the comparative approach to monitor in vitro cardiac differentiation from bioengineered iPSC according to an embryonic expressome roadmap.

Figure 2.

Cardiac development roadmap categorizes cardiac potential of pluripotent lines. A. Embryoid body (EB)-based spontaneous differentiation protocol applied to R1 embryonic stem cells (R1), 4 factor-reprogrammed induced pluripotent stem (4F-iPSC) cells and 3 factor-iPS cells (3F-iPSC). B. Spontaneous beating activity in differentiating populations in vitro (n=4 for 3F and 4F-iPSC). C. Gene expression matrix for a panel of 72 pluripotency and cardiac-related genes extracted from the cardiac development roadmap including heart samples from embryonic stages (abbreviated as E) ranging from E7.5 to adult. D. Expression levels for the 72 gene-panel in differentiating R1 cells. E. Time-course of expression for the cardiogenic panel in R1 embryonic stem cells (R1), 3F-iPSC (3F) and 4F-iPSC (4F). F. Unsupervised hierarchical clustering of pluripotent cells and their progeny based on the 72 gene-panel. Day of differentiation is abbreviated as “d” for all panels.

To determine if the cardiogenic roadmap is capable of distinguishing between high- and low-cardiac potential of iPSC cell clones, expressome profiles for 3F-iPSC and 4F-iPSC differentiated in vitro were extracted according to the panel of cardiac-related genes (Figure 2E) and compared to the reference ESC-derived dataset. The overlay between expressomes derived from pluripotent stem cells uncovered equivalent global gene expression profiles at baseline across the three independent pluripotent lines (Figure 2E, left). However, as in vitro differentiation progressed, divergent expression profiles became apparent starting at day 5 for 4F-iPSC (Supplementary Figure 1), which failed to upregulate key gastrulation (Lhx1), endoderm (Sox17), and early mesendoderm/cardiac precursor markers (Cxcr4, Kdr, Tbx20, Mesp1, Isl1; Figure 2E, middle), while maintaining expression patterns that preserved the pluripotent fingerprint. Based on correlation values, 4F-iPSC showed higher similarity to R1 (undifferentiated cells in roadmap) after 5 days of in vitro differentiation, while 3F and reference ESC (day 5) advanced toward a gastrulation/pre-cardiac profile similar to natural embryonic structures corresponding to E7.5 (Figure 2C). Furthermore, by day 11 the 4F-iPSC did not have distinct similarity to any of the assayed embryonic stages, whereas 3F-iPSC and ESC acquired an advanced expressome more similar to E12-newborn stages (Figure 2E, right) indicating progression of cardiac differentiation in vitro. Despite simultaneous downregulation of pluripotency-related genes in all three groups by day 11, expression of early and late cardiac markers in 4F-iPSC failed to reach levels shown for 3F-iPS or R1-ESC by the end of the differentiation process (Figure 2E, right). Upon unsupervised hierarchical clustering of the 72-gene set, inconsistent modulation of gene expression resulted in segregation of 4F-iPSC from both ESCs and 3F-iPSC (Figure 2F). 4F-iPSC samples clustered separately at days 5 and 11, in agreement with their inconsistent biological phenotype during spontaneous in vitro cardiac differentiation that was incompetent to produce functional cardiac tissue with visible beating activity. This anomalous profile of quality controlled 4F-iPSC cells was independently confirmed when samples were analyzed according to a genome-wide expressome representing overall expression profiles (Supplementary Figure 2).

Cytoskeleton and extracellular matrix remodeling together with epithelial-to-mesenchymal transition pathways are dysregulated in 4F-iPSC

Based on the selected differences in gene expression according to the natural cardiogenesis-derived cardiogenic panel, 4F-reprogrammed cells may contain specific abnormal transcription profiles compared to 3F-iPSC and reference ESCs that could identify initial developmental corruption. Genome-wide pairwise comparison on microarray data of pluripotent cells bioengineered in the presence or absence of c-Myc (4F-iPSC vs. 3F-iPSC) demonstrated an increasing numbers of differentially expressed transcripts (DETs) with subsequent in vitro differentiation (Figure 3A and Supplementary Figure 3). 82 transcripts were differentially expressed between 3F- and 4F-iPSC at day 0, with p values after FDR control at 0.01 and log2 FC at 2 (Figure 3A, red dots and Supplementary Table 3). These genes correspond to prioritized cytoskeleton and extracellular matrix (ECM) remodeling pathways (Figure 3B). Downstream effects of the baseline expression profile in 4F cells resulted in magnified differences as in vitro differentiation proceeds through subsequent developmental milestones, with 399 DETs at day 5 and 726 DETs by day 11 (Supplementary Figure 3). Functional differences between 3F-iPSC and 4F-iPSC were noted in ECM and cytoskeleton remodeling that were persistent throughout differentiation and consistent with dysregulation of epithelial-to-mesenchymal transition (EMT)-related functions and cardiac-related signaling pathways (Supplementary Figure 3). In order to select the genes uniquely regulated in undifferentiated 4F-iPSC that could be responsible for the non-beating phenotype, pairwise comparisons between the three pluripotent lines (3F vs R1, 3F vs 4F and R1 vs 4F) were performed. Resulting lists of differentially expressed transcripts in beating vs non-beating comparisons (3F vs 4F and R1 vs 4F) were overlapped to identify genes differentially regulated in non-beating 4F-iPSC. Finally, genes from this comparison that were also contained in the 3F vs R1 group (both lines were beating) were removed as background noise not involved in regulating the defined outcome (beating), resulting in a list of 27 transcripts, corresponding to 23 genes, differentially regulated in non-beating 4F-iPSC (Figure 3C). Thus, genome-wide transcriptome analysis enabled dissection of early mechanisms that are distinctive in cardiogenic dysfunction within 4F-iPSC lines and highlights cardiogenic pathways that may be responsible for individual variability between bioengineered stem cells.

Figure 3.

Cardiogenic roadmap establishes procardiogenic profile at pluripotent stage. A. Volcano plot depicting the differentially expressed transcripts (DETs) between 3F and 4F-iPSC at day 0, with log 2 fold change on x-axis and negative log 10 p-values on y-axis. Cut-offs used were log 2 fold change of 2 (real fold change 4) and p-value <= 0.01. B. Biological functions prioritized as significantly different between 3F and 4F-iPSC at day 0. Red bar marks statistical significance p-value= 0.05. C. Venn diagram representing DETs at day 0 from multiple comparison analysis. Highlighted in yellow, DETs specific to 4F-iPSC. D. Expression profile in pluripotent ESCs, 3F-iPSC and 4F-iPSC at pluripotent stage for 23 genes extracted from the list of transcripts (in yellow) uniquely patterned in 4F-iPSC. E. Expression matrix through natural cardiac development for the selected 23 genes differentially expressed in 4F-iPSC.

Cardiogenic roadmap predicts pro-cardiogenic outcome at pluripotent stage for individual iPSC lines

We hypothesized that disruption of cardiac differentiation properties of iPSC cells may be linked to dysregulated expression of 23 genes uniquely patterned in less-cardiogenic cells at the pluripotent stem cell stage (Figure 3D and Supplementary Figure 4). Gene expression profile for these 23 genes extracted from the comprehensive roadmap, revealed a high degree of correlation between early stages of natural embryonic development and cardiogenic cells (3F and R1-ESC) (Figure 3E and Supplementary Figure 5), with 4F-iPSC clustered separately from all other samples and showing low similarity to any other samples. PCR verification of expression levels for these genes reduced the list to a total of 12 markers independently confirmed as uniquely patterned in less-cardiogenic cells (Supplementary Figure 6), providing a validated list of candidates linked to the anomalous cardiogenic behavior of 4F-reprogrammed cells. Discrepant regulation of this group of genes in 4F cells highlights pre-cardiogenic programs enabling downstream heart development from the pluripotent stage (Supplementary Figure 6). Therefore, the cardiogenic roadmap established from natural developmental processes may predict corrupt gene expression patterns and subsequent dysfunctional heart developmental outcomes from iPSC lines.

Composite index predicts cardiac potential of induced pluripotent stem cell lines

To determine if the 12 genes identified through unbiased analysis are randomly associated to high/low cardiogenic lines or are mechanistically linked as regulators of cardiogenic variability, an empiric assay was developed. The predictive power of the profile was tested in four independent iPSC lines with distinct cardiogenic potential according to the onset of beating activity as a functional surrogate marker (Figure 4A) from previously validated cardiogenic (3F-iPSC) and less-cardiogenic (4F-iPSC) lines as test case examples (ref?). The observed beating activity correlated with expression levels of mesendoderm marker Cxcr4 during gastrulation, showing increased expression in early beating iPSC lines. Furthermore, endoderm and ectoderm markers followed opposite expression trends, with Sox1 and Hnf4 being upregulated in early differentiation stages of less cardiogenic lines (Supplementary Figure 7). iPSC derived from both adult and embryonic sources were included, as were two different reprogramming strategies (lentiviral infection and stable integrated dox-inducible transgenes). The panel of confirmed 12 genes was profiled within each sample at pluripotency stage using quantitative RT-PCR. Genes were categorized as positive predictors (upregulated in cardiogenic 3F-iPSC/ESCs; Figure 4B) and negative predictors (upregulated in less-cardiogenic 4F-iPSC; Figure 4C). Contribution of each gene was calculated for each assayed line as described in the methods section (Supplementary Figure 8). Average percentage contribution of positive (Figure 4D) and negative predictors (Figure 4E) was obtained for each iPSC line revealing higher levels of positive predictor genes in the more cardiogenic lines as well as upregulated negative predictor values in less-cardiogenic 4F-iPSC. A composite predictor including contribution from all genes in the predictive panel was then calculated (positive – negative predictors), exposing the final cardiogenic predictive score for each studied line. This cardiogenic score, from pluripotent cells, matched the cardiac differentiation potential according to the functional marker of beating activity for each of the studied lines, and was therefore able to anticipate cardiogenicity of independent iPSC lines. In order to validate this cardiogenic score, values were obtained for a total of 18 reprogrammed pluripotent lines derived from three somatic sources (embryonic fibroblasts-MEFs, tail-tip fibroblasts-TTFs and cardiac fibroblasts-CFs) using either lentiviral or Dox-inducible reprogramming strategies and previously characterized for cardiogenic potential in terms of beating activity as a functional surrogate marker (Figure 4G). When plotted according to the onset of beating activity two distinctive groups were apparent, with early beating cell lines displaying higher cardiogenic score values and late beating cells linked to lower scores. Based on the ROC curve analysis the optimal threshold was set at 13.8, which retained ~80% sensitivity while still maintaining the specificity and precision at 100% level (Figure 4H and Supplementary Figure 9). In summary, this PCR platform based on developmentally curated genes discriminates between reprogrammed lines with high and low cardiogenic potential within cells at the pluripotent state.

Figure 4.

Composite index predicts cardiac potential of induced pluripotent stem cell lines. A. Beating activity for four iPSC lines differentiated in vitro, error bars represent standard error, p-value <2.2e-16. B and C. Relative expression of the 12-gene panel for 3F-iPSC, 4F-iPSC and four unrelated iPSC lines. Positive predictors refers to genes originally upregulated in cardiogenic cells, negative predictors refers to genes originally upregulated in less-cardiogenic 4F-iPSC. D. Average percentage relative expression of positive predictors for each studied line. E. Average percentage relative expression of negative predictors for each studied line. F. Composite score (average percentage relative expression for positive minus negative predictors) for each of the studied lines in arbitrary units. G. Cardiogenic score for 18 iPSC lines, arranged according to their beating activity (early to late). H. ROC curve for performance measurement on the sensitivity and specificity (1-specificity) for the composite score predicting beating phenotype. Right y-axis shows color scale for score values with red representing higher values and blue lower. The curve is colored according to the score.

Discussion

The integration of high-throughput technology and stage-specific approaches created herein a temporally regulated dataset of gene expression throughout natural embryonic cardiac development. While other non-genetic factors such as the epigenetic state⁴⁰ and stochastic phenomena⁴¹ may have a concomitant influence upon differentiation outcomes, we here focused on the transcriptional component of the differentiation process, providing a broadly applicable and technically accessible platform that may be leveraged for stage identification, guided differentiation and comparative bioinformatic applications.

The novel information contained in this comprehensive roadmap of normal cardiogenesis allowed us to build a translatable atlas of pathways and networks defining innate mechanisms of cardiogenesis. By isolating the expression matrix for a subset of cardiac-related genes, we established a discrete guide for cardiogenicity used here to gauge and predict the ability of reprogrammed progeny to give rise to cardiac tissue. Specifically, innately cardiogenic iPSC reprogrammed using Oct4, Sox2 and Klf4 lentiviruses were compared to less-spontaneously cardiogenic counterparts bioengineered in the presence of the oncogene c-Myc.¹⁶ This fourth component of the reprogramming cocktail has been previously shown to impair the in vitro cardiogenicity of mouse 4F-iPSC, while preserving their potential for cardiac contribution and repair in vivo,^{16, 39} underscoring the importance of context-dependency of the cardiac differentiation process. During early phases of in vitro differentiation of c-Myc-containing cells, delays were observed in the onset of the initial transcriptional programs required to trigger cardiac differentiation according to the developmental roadmap, as demonstrated by the lack of upregulation of precardiac genes after 5 days of in vitro differentiation as well as abnormal clustering of differentiated 4F samples with ESC and 3F-iPSC at the pluripotent state. Transcriptional amplification of already active genes caused by Myc may explain the impaired transition of 4F-iPSC to a pro-cardiogenic state,⁴² maintaining instead high levels of pluripotency markers despite being subjected to differentiation conditions. This unique bioinformatics analysis thus identifies failing cardiogenesis, suggesting the importance of early pathways that are sufficient to disrupt onset of canonical developmental processes.

As a proof-of-concept application, we herein demonstrate the utility of a developmental roadmap to gauge cardiac potential of pluripotent sources. By contrasting expression profiles of temporally regulated cardiac-related genes between stage-specific embryonic hearts and in vitro differentiated cells, various levels of cardiogenicity among pluripotent cell lines were apparent according to differential expression profiles shortly into initiation of spontaneous differentiation. These disparities further correlated with alterations in essential pathways such as EMT, required for endocardial cushion formation,⁴³ epicardium development,⁴⁴ reparative functions,^{45, 46} and cardiosphere generation from heart biopsies,⁴⁷ suggesting disruption of these processes that are linked to decreased functional beating activity in 4F-iPSC lines.

Differences in known pluripotency markers and tissue specific expression patterns at the undifferentiated state do not correlate with variable differentiation propensities of ESC lines.⁴⁸ Our design using genome-wide bioinformatic approach, allows the inclusion of potentially relevant genes whose regulatory roles might not be yet defined. As a result, we describe a panel of genes not previously linked to cardiogenicity and grouped here into a predictive score that is associated, in an impartial way, with variation in the cardiac differentiation ability of reprogrammed iPSC cells. This approach was able to identify a subset of heterogeneous identities with distinct patterning in less-cardiogenic samples. These genes include epithelial keratins (Krt8 and Krt18),⁴⁹ inflammation-related chemokine ligands (Cxcl3 and Cxcl5)^{50, 51} and interferon-induced genes (Ifi44, Ifit1, Ifit3 and Ifi203), a Golgi chaperon (Rtp4),⁵² interferon regulator Usp18,⁵³ metalloprotease Mmp9 and unknown function Klhk13. From these, only metalloproteinase Mmp9, upregulated within poorly cardiogenic cells at day 0, has been demonstrated to be activated in heart disease and remodeling, with low levels improving survival and differentiation of cardiac stem cells.^{54, 55} In contrast to previous studies using differentiated progeny to establish the capacities of pluripotent cells,⁵⁶ we focused our analysis on undifferentiated samples, allowing faster screening of candidate lines shortly after reprogramming and reducing potential variability due to the reprogramming procedures.⁵⁷

The prognostic potential of the composite score described in this work was tested in four iPSC lines reprogrammed by lentiviral infection or doxycycline induction in fibroblasts from embryonic or adult tissues. A relative value was obtained mapping the relative cardiogenic ability of each undifferentiated line in the framework of the original high and low cardiogenic lines. The delivered ranking faithfully recapitulated the observed cardiac potential, in terms of beating activity, across both cell sources and reprogramming strategies, confirming the linearity index-phenotype and therefore its predictive value. We validated these results by applying the cardiogenic score to a total of 18 reprogrammed lines with different origins and reprogrammed using two different strategies. Although the linearity between index and phenotype was not strictly conserved, the most cardiogenic lines (early beating) had higher scores than the less cardiogenic ones (late beating). Based on the ROC curve analysis, 60% percentile of the score, corresponding to 13.8 in our dataset, led to prediction of the cardiogenic potential of reprogrammed pluripotent lines, with ~80% sensitivity and 100% specificity and precision. By maximizing the specificity of the cut-off, we ensured selection of high cardiogenic potential in profiled cell lines at the risk of filtering out iPSC lines that demonstrate high levels of beating activity. Application of this score, therefore, would allow discrimination of the 40% most cardiogenic lines by applying a PCR panel, limiting time and resources necessary to reproducibly obtain full cardiac potential via lengthy in vitro differentiation.

Overall, we have established a comprehensive time-course of healthy cardiac development that offers a template to define natural differentiation stages, as well as disruptive changes in expression patterns responsible for heart disease or dysfunctional networks resulting in incompetent cardiogenicity in vitro. By combining this unbiased expressome platform with transcriptomes from iPSC lines characterized by high and low cardiac differentiation potential, we have developed a tool to forecast cardiogenicity of reprogrammed sources prior to differentiation. Furthermore, utilizing the genomic analysis strategies described herein may identify and prioritize candidate genes offering a novel approach to determine molecular targets that regulate cardiac phenotype of individual bioengineered stem cells. The novel bioinformatic platform described in this work allows prospective selection of cardiogenic lines for reproducible generation of cardiac derivatives required for theranostic applications and thus opens a new avenue for optimization of stem cell-based regenerative medicine platforms based on bioengineered stem cells.