For over a decade, researchers all over the world have used the conventional bulk sequencing methods on RNA extracted from a population of cells to study gene expression changes in different tissues (Cloonan et al., 2008; Lister et al., 2008; Mortazavi et al., 2008; Nagalakshmi et al., 2008; Wilhelm et al., 2008). Since then, the system has been optimized for different types of RNAs and qualities of starting material. In general, bulk RNA-Seq refers to every sequencing approach that relies on averaged gene expression from a population of cells to reveal RNA presence as well as quantity in a sample of cells during the time of measurement. Therefore, bulk-based approaches can identify differences between sample conditions. Bulk RNA-Seq is not particularly limited by technical applications to the heart, nevertheless there are many criteria to consider when choosing bulk sequencing to ensure high-quality data.

In 2013, Nature Methods titled scRNA-Seq as one of the most anticipated technologies of the year (Method of the Year 2013, 2014) with follow up nomination as technology of the year in 2019 due to its key role in the identification of cell types and functions, in addition to possible simultaneously multi-omics approaches in the future, highlighting its extensive role in genomic research.

Over the last couple of years, scRNA-Seq has had a massive effect on research. The reason is simple - while bulk RNA-Seq can measure the average gene expression across cells in a sample, identify differences between sample conditions and give a representation of highly regulated pathways, it fails to demonstrate the individual complexity of each cell and heterogeneity of tissues. Furthermore, some cell populations have high degrees of cellular and transcriptomic heterogeneity due to different cell types or indiscriminate states. The advent of scRNA-Seq technologies has addressed most of these limitations by facilitating the analysis of the transcriptome of every cell in each sample (Shapiro et al., 2013; Islam et al., 2014). This has enabled the creation of cell atlases (Regev et al., 2017; Cusanovich et al., 2018; Han et al., 2018; Tabula Muris Consortium., 2018; Cui et al., 2019; Litviňuková et al., 2020; Suryawanshi et al., 2020) at unprecedented resolution, the analysis of thousands of cells in parallel and even allows integration of chromatin status and multimodal analysis.

As a result, scRNA-Seq has shown to be a helpful technique for identifying cell subpopulations and elucidating dynamic cellular transitions during development and differentiation with an unparalleled level of detail and accuracy (Lafzi et al., 2018; Potter, 2018).

Several scRNA-Seq methods and technologies have been developed in recent years, all having different criteria regarding RNA transcript length, number of captured cells and read depth per cell (Svensson et al., 2018). Therefore, a specific scRNA-seq technology may be more beneficial for certain types of material than others, but all of them share similar workflows: samples preparation, single-cell capture, transcription and amplification, library preparation, sequencing and analysis (Potter, 2018). In general, there are three main technologies of scRNA-Seq: plate-, microfluidic- and droplet-based. In this section we will first explain the basic concepts of these currently used platforms and then their applications in cardiovascular science.

Per definition single-nuclei RNA sequencing (snRNA-Seq) is not a technique by itself, but more a modification of the scRNA-Seq methodology. However, due to its unique contributions and applications to the cardiac field, the technique has earned an exclusive overview of its own.

Single-cell sequencing is a powerful tool allowing in-depth characterization of cell populations within complex tissues. However, as discussed in the previous sub-section, scRNA-Seq has two major limitations when applied to cardiovascular tissues: dissociation of material and cell size. ScRNA-Seq systems require dissociation of tissue material and especially gaining high-quality single-cell resolution of the adult mammalian heart is rather difficult. Secondly, technical limitations regarding capturing techniques, leading to an under representation of individual cell types (i.e., CM) due to their large cell size and irregular shape (Ackers-Johnson et al., 2018). There are few platforms for sequencing individual adult CMs and given that all of them are plate-based systems, the number of cells they can analyze is limited.

As the name suggests, snRNA-Seq utilizes only the nuclear component of single cells as input material, which is most appropriate when viable intact cells cannot be harvested from fresh tissue. Furthermore, there are other advantages of snRNA-Seq, when scRNA-Seq capturing methodologies cannot be applied. For instance, it allows the extraction of nuclei from frozen samples, which is extremely valuable for rare materials like the heart. Moreover, scRNA-Seq involves a dissociation step, which might damage sensitive cells while failing to release cells more tightly bound or those surrounded by a matrix. Although sampling a much larger number of cells can partially overcome this limitation, it may not always be financially feasible. Additionally, as mentioned before current enzymatic and mechanical methods for single-cell dissociation tend to introduce stress-induced transcriptional artefacts. Even with commercial nuclei extraction kits, the solution should be examined to ensure no big cells are present anymore. Afterwards, single-nuclei can be prepared to be captured with any of the above-described methodologies.

Nevertheless, there are some strategic limitations to be considered. While some studies have shown similar comparable transcriptomic results acquired from sc- and snRNA-Seq (Bakken et al., 2018), by utilizing RNA from within the nucleus only, a fair amount of the total RNA is excluded from the analysis, which could constitute a limiting factor in identifying dynamic cellular states.

Furthermore, taking single nuclei from frozen samples might result in false transcriptomic representation due to RNA degradation by freezing. Interestingly, CM can have a diploid or even polyploid as well as multinucleate nature (Landim-Vieira et al., 2020), which is not fully understood yet and therefore very valuable for future studies. In addition, extracting the nuclei removes the highly abundant mitochondria in CMs leading to possible alterations in gene expression or missing information when looking into network-based approaches.

While scRNA-Seq is currently the most widespread RNA-Seq technology, its prowess in deciphering the individual make-up of a single cell is just beginning to evolve. In recent years, many suppliers were able to integrate multi-omic approaches on the single cell level. Immunological studies have benefitted from paired B-cell or T-cell receptors, surface protein expression together with gene expression from a single cell (DeBerge et al., 2021). Single nuclei sequencing can be combined with ATAC (Assay for Transposase Accessible Chromatin) allowing for the analysis of chromatin accessibility at the single cell level, thus providing insights into cell types and states, and deeper understanding of gene regulatory mechanisms (Zhang et al., 2021). Further applications include proteomics (Cheung et al., 2021) and DNA methylation (Galvão and Kelsey, 2021). However, as most of these are still in their early stages including protocol optimization and integrational data analysis tools, it needs to be individually assessed whether specific research questions can be answered using these approaches (Jansen et al., 2019; Argelaguet et al., 2020). On the other side, computational approaches tried to leverage cell-type specific gene expression profiles from multiple scRNA-seq reference datasets to bulk RNA-Seq with the intention to apply cell compositions even to bulk RNA-Seq data. To date, multiple tools have been suggested, such as DESCEND (Wang et al., 2018), SCDC (Dong et al., 2021) and some others, but to our knowledge not one has been yet considered to be “state-of-art”.

Identifying gene expression profiles at single-cell resolution has revolutionized the field of transcriptomics. Using single-cell or single-nuclei based approaches has made it possible to look at individual cells within complex tissues, gaining valuable biological insights into rare cellular properties, cell-to-cell variability, and tissue identity. Despite such advantages, these approaches are limited by technical criteria to a certain degree. In this review, it has already been demonstrated how dissociating tissue can be a difficulty. In addition, this process causes the loss of information regarding the spatial organization of cells and cell populations throughout a tissue causing limited or incomplete interpretation. In in vivo situations though, spatial-specific expression has an enormous impact on biological networks. Gaining knowledge of spatial information in addition to timing, and level of developmental gene expression allows the description of interactive biological networks, where each element is influenced by its surrounding microenvironment (Zheng et al., 2017). Therefore, the persevation of spatial information holds promising approaches of combining genomic, transcriptomic, and proteomic features while retaining positional information.

Various scRNA-Seq methodologies have been discussed in the section above, but spatial approaches also have made extraordinary progress in the recent years (Ståhl et al., 2016; Qian et al., 2020) and gained increasing popularity. In general, spatial techniques vary from in situ hybridization (e.g., smFISH, seqFISH) over in situ sequencing (e.g., BaristaSeq, STARmap) to in situ capturing technologies (e.g., ST, Slide-Seq). All of them have their designated place in research, in situ spatial methodologies hold the prospect of possible combination with scRNA-Seq technologies. While some of these technologies and their drawbacks and advantages have been reviewed before (Asp et al., 2020), this review will highlight a few of the spatial methodologies and their applicability to cardiovascular research.

All above mentioned procedures have a tremendous effect on data quality. However, in order to draw meaningful (biological) conclusions, downstream data procession and analysis is necessary. In fact, data analysis should be taken into consideration when designing a study in the first place, to allow for a qualitative interpretation of the data. Therefore, here we discuss the basic steps of data processing, analysis and opportunities for downstream analysis (Figure 1). Of note, this section will not be separated into bulk and single-cell sequencing as both follow similar pipelines. In addition, many single-cell advances are actually adapted from existing bulk approaches. At this stage, we would like to refer to other publications addressing data analysis steps and challenges (Koch et al., 2018; Dal Molin and Di Camillo, 2019; Lähnemann et al., 2020) in single cell vs. bulk RNA approaches.

After sequencing the libraries, the raw files need to be pre-processed. This step includes demultiplexing, quality control, trimming and alignment/de novo assembly. Even though existing in both bulk and single cell analysis, demultiplexing refers to slightly different processes. In bulk RNA-Seq, a sequencing lane typically will contain a pool of barcoded samples, which therefore require demultiplexing. In scRNA-Seq, each cell is considered a sample and, in most protocols, defined by a cell-specific barcode tag to each read. Therefore, each library normally contains multiple cell barcodes, which need to be distinguished as a sample. This process is called demultiplexing and gives each cell in each sample its specific RNA content. From here on, the steps of processing are in common between approaches, which are meant to cut off primer/adapters or poor-quality nucleotides, alignment of reads to a reference genome and counting reads per region/gene, creating a gene count matrix. Due to the overall high input amount for bulk sequencing, a high coverage can be achieved, but unfortunately scRNA-Seq normally deals with to the low concentration of input RNA and results in a high number of zero reads counts. However, these “false” zero counts, such as dropout events or sparsity, may mask the biological zero counts due to the non-expression of a group of genes in specific cells (Eraslan et al., 2019). Moreover, this high number of zero read counts in the gene expression matrix ultimately results in the reconstruction of an incomplete transcriptome, which renders the study of the different cellular phenotypes very challenging (Wang J. et al., 2019). Sparsity can impede downstream analysis and remains a challenge. To overcome this loss of information, computational imputation or denoising was developed, i.e., MAGIC and scIMPUTE (Lafzi et al., 2018; van Dijk et al., 2018). The last step is normalization. There are different approaches to normalization depending on the software used, but most of them are equally accepted and their individual usage mostly depends on the intented downstream analysis. However, the methods developed for bulk RNA-Seq such as RPKM, FPKM and TMM were not able to successfully remove the effects of different sampling from scRNA-Seq datasets, due to the considerable technical noise present in such data. Therefore, new normalization methods, which are able to handle noisy data and take advantage of the presence of UMIs were developed (Townes and Irizarry, 2020).

Afterwards, the data can be processed in various ways. Dimensional reduction is one of the most commonly used features and helps to visualize data and to identify relevant genes. For bulk-based studies, principal component analysis (PCA) is the most common used feature. Although this also applies to single cell approaches, because of the high dimensionality of the data other methodologies, such as t-SNE and UMAP, are superior to PCA. Other limitations include errors introduced by the inevitable technical variability during sample processing steps, sequencing depth or pipetting accuracy, also known as batch effect.

These are not unique to one methodology and can be found in the datasets generated by both, bulk and scRNA-Seq. However, these confounding factors are more amplified in scRNA-Seq, which may lead to a mix of technical variations with biological variables. In bulk RNA-Seq, instead, these batch effects are smaller and do not result as systematic errors, as bigger sample sizes usually lead to large-scale sample preparation and sequencing in parallel. Therefore, batch effect correction in scRNA-Seq requires other features capable of handling these variations, for example nearest neighbor integration. Further typical analyses include clustering and differential expression gene (DEG) detection. Unsupervised clustering further helps to identify groups and subgroups within single cell datasets and to pinpoint relevant cell groups for further analysis. DEG analysis has been established as a powerful tool to discover different regulation between groups, which can reveal regulatory effects. Although DEG can be also used to make between-sample comparisons, it is usually applied to identify transcriptome signatures that differentiate between cell groups. However, cell group assignment is still constrained by ambiguities, which could mask biological differences. Therefore, an improved reference set for cell type assignment is necessary to make DEG analysis in single-cells more reliable.

However, scRNA-Seq also provided new bioinformatics tools, with RNA velocity and trajectory analysis being the most prominent features. Both aim to describe the dynamic changes among captured cell signatures. RNA Velocity takes advantage of the ratio between spliced and unspliced RNA to develop dynamical patterns with possible cell fate prediction, while the trajectory attempts to predict a path that certain cells undergo during specific conditions (i.e., differentiation). Especially, deconvolution of RNA seq data has gained increased recognition in the last year (Baron et al., 2016; Wang X. et al., 2019; Newman et al., 2019). Deconvolution attempts to use single cell datasets to deconvolute bulk datasets by leveraging cell-type specific gene expression profiles from multiple scRNA-seq reference datasets, and thus synergistically combines specific information gained from both, bulk and single cell datasets.

While extracting RNA from heart tissue has no particular limiting factor, human heart samples are highly valuable due to the many ethical restrictions, making them hard to obtain. Many cardiac studies are therefore performed using other mammalian models such as mice or human tissue-derived cell lines. Nevertheless, patients with heart disease can undergo surgery, allowing for ascertainment of cardiac biopsies, but the amount of tissue retrieved can be insufficient or compromised. However, healthy individuals normally do not usually tend to undergo biopsy, rendering the collection of healthy living heart tissue as controls close to impossible.

Nevertheless, RNA-based studies have provided important insights to cardiac research, especially helping to decipher ncRNA and their potential role as regulators of numerous cellular processes in the progression of cardiovascular diseases Transcriptomic studies using NGS enabled clear distinguishment between various cardiac cell types, such as CMs, endothelial cells, fibroblasts, and immune cells. Arguably one of the first studies that compared accuracy and speed of NGS with microarray analyses demonstrated that in the Gαq transgenic mouse cardiomyopathy model RNA sequencing was accurate and sensitive enough to detect abundant and even rare cardiac transcripts (Matkovich et al., 2010). Subsequent studies highlighted that even atrial and ventricular cells show clear significant differences in their transcriptomic profiles (Sehnert et al., 2002; Ng et al., 2010). While atrial CM are mostly described with Myh4/ALC1 and Myh6/αMHC, the ventricle expresses more Myh3/ELC, Myh7/βMHC and Myl2 myosin genes (Ng et al., 2010). In depth reports of the current knowledge on the different molecules in cardiac cell subtypes have been reviewed before (Zhou et al., 2018; Colpaert and Calore, 2019; Siasos et al., 2020). NGS was essential to reveal that >80% of the genome is transcribed in various classes of RNA with protein-coding transcripts only accounting for max. 3% of the genome, and the remainder of transcripts lacking coding ability (Dunham et al., 2012). Many of these noncoding transcripts are functionally active RNA molecules that can be subdivided into small noncoding RNAs (<200 nt), such as microRNAs (miRs), and longer noncoding RNAs (>200 nt) that include ribosomal RNAs, natural antisense transcripts and long noncoding RNAs (lncRNAs) (Engreitz et al., 2016). NGS has helped to discover cardiac microRNAs (Matkovich et al., 2010; Leptidis et al., 2013) and lncRNAs (Matkovich et al., 2014; Yang et al., 2014) in the developing and diseased human heart and how they can potentially function as therapeutic target molecules. While these studies contributed significantly to our current understanding of cardiac (patho)physiology, they failed to address the importance of cardiac cellular heterogeneity and cell-cell interactions.

ScRNA-Seq is able to look at cell compositions in an unbiased fashion. In 2016, two labs first reported the use of scRNA-Seq on embryonic mouse hearts by investigating lineage-specific gene programs underling early cardiac development (DeLaughter et al., 2016; Li et al., 2016). To date, multiple researchers have used scRNA-based techniques on various cardiac materials allowing to look at cell populations at a single-cell resolution, and started a new age of transcriptomics including the generation of transcriptomic and epigenetic cell atlases of adult mice (Cusanovich et al., 2018; Han et al., 2018; Tabula Muris Consortium, 2018) and humans (Regev et al., 2017; Cui et al., 2019; Litviňuková et al., 2020; Suryawanshi et al., 2020). Investigations deciphering the heterogeneity and subpopulations of cell types define region- or disease-specific gene expression profiles not only benefit cell atlases but also help to identify novel molecular mechanisms relevant for cardiac disease and new therapies. For instance, Li et al. (2019) investigated cardiac endothelial cells following ischemic injury and were able to show clonal proliferation of resident endothelial cells post-myocardial infarction.

Single-cell approaches, however, do not stop here. The heart is a multi-cellular organ and the interaction between cell of ligands and receptors on target gene expression is not yet fully understood. Browaeys et al. recently developed a tool for modelling intercellular communication on the single-cell level by linking ligands to target genes. This as well as other promising tools enable investigation of possible signaling mediators in addition to gene interaction by expression profiling (Browaeys et al., 2020; Cang and Nie, 2020; Efremova et al., 2020; Gladka, 2020; Jin et al., 2021). Furthermore, network and trajectory analyses on single-cell data led to the identification of fibroblasts as a critical constituent in promoting cardiomyocyte maturity (See et al., 2017) and regulatory interactions between transcription factors and target genes (Jackson et al., 2020). Another recent study by Litviňuková et al. (2020) provided comprehensive transcriptomic data on six distinct cardiac regions of the adult human heart using both sc- and snRNA-Seq technologies. Similar to other reports, they observed the cellular heterogeneity of cardiomyocytes, pericytes, and fibroblasts, and additionally, they analyzed cell-to-cell interactions suggesting a direct interaction through NOTCH2. Another study using microfluidics approaches in healthy human donors was able to identify and study the cellular and transcriptional diversity in a healthy heart (Tucker et al., 2020). The ICELL8 platforms claims to be able to deal with a wide range of sizes, while still making high throughput sequencing possible. Wang et al. used this platform on 21,422 single cells-including CMs and non-CMs from normal, failed and partially recovered adult human hearts to reveal inter- and intra-compartmental heterogeneity in response to stress as well as CM contractility and metabolism as the source of changes in heart function (Wang et al., 2020).

Taken together, an increasing number of studies have used methodologies on single cell level to shed light on the roles of various cell populations within in the mammalian heart helping future cardiac research and advances with their findings.

Interpreting single-cell properties can be difficult (Asp et al., 2020; Stickels et al., 2020a), but by adding spatial context to the transcriptome it will greatly benefit the current understanding of many biological networks. However, the main hurdle for these spatial methods is restricted RNA capture efficiency, which becomes increasingly more challenging with higher resolution. Even though spatial transcriptomics has proved to be a valuable source of transcriptomic information, spatial transcriptomics in the heart, cardiac development, and regional changes in gene expression during heart maturation are strong areas of great interest for the future.

Two studies utilized spatial transcription offered by 10X Genomics and subsequently combined spatial transcriptomic data with scRNA-Seq through computational deconvolution methods in order to reconstitute detailed spatial transcriptomic maps of human myocardial infarction (Kuppe et al., 2020) and the development of the human heart (Asp et al., 2019). In 2017 the group of Asp et al. (2017) already used the ST method to study fetal markers and their localization within adult human cardiac biopsies and later on created a spatiotemporal atlas of the developing human heart in the first trimester (Asp et al., 2019) by combining scRNA-seq data with spatial transcriptomics and in situ sequencing. In 2020, Asp et al. (2020) used their approach of combing scRNA-Seq along with ST and in situ sequencing to uncover novel cell types, such as clusters of fibrosis-associated fibroblast-like cells and a subpopulation of cardiac muscle cells. Furthermore, they also visualized their result by integrating the spatial information into 3D transcriptional maps. A major caveat of this combined approach is that it maximizes the efficiency of using data information by providing both cell-type and regional information from tissues, e.g., human embryo tissue, that are not easily obtained.