Single

昵称37064826 2020-07-17

展开全文

Introduction

Clinical Perspective

What Is New?

Single-cell sequencing can be used to identify subpopulation-specific and novel cell type-specific markers in the healthy and diseased heart.
Identification of new disease-driven cell populations provides insights into gene expression changes that are triggered by ischemic injury.
Myozenin2-enriched cardiomyocytes form a distinct subpopulation of cardiomyocytes in the healthy heart.
Cytoskeleton-associated protein 4 is a novel marker for activated fibroblasts that positively correlates with known myofibroblast markers in both murine and human diseased hearts.
In vitro experiments suggest a modulating function for cytoskeleton-associated protein 4 in myofibroblast activation.

What Are the Clinical Implications?

Single-cell sequencing of the adult heart allows us to examine molecular mechanisms that drive the cellular processes underlying heart disease.
New biology discovered by single-cell sequencing can ultimately lead to the development of novel therapeutic strategies.
We identified cytoskeleton-associated protein 4 as a new marker for activated cardiac fibroblasts during ischemic injury that appears to attenuate myofibroblast activation.

The heart consists of a collection of different cell types that coordinately regulate cardiac function.¹ Changes in cellular composition and function are mechanistically responsible for cardiac remodeling and repair during disease. Identifying the underlying differences in gene expression between cell types or transcriptome heterogeneity across cells of the same type will greatly help to improve our understanding of cellular changes under both healthy and diseased conditions.² RNA amplification strategies have provided the opportunity to use small amounts of input RNA for genome-wide gene expression analysis at single-cell resolution. The analysis of individual cells randomly drawn from a sample allows for an unbiased view of mRNAs present in different cell types of an organ, which will provide a more accurate classification of cellular diversity through differences in gene expression.^2–5 Recently, transcriptional profiling of cardiac cell populations during heart development by single-cell sequencing has led to the identification of lineage-specific gene programs that underlie early cardiac development.^4,6 However, so far, no studies have focused on single-cell transcriptomics of the adult heart.

Here we present a method for obtaining high-quality RNA from digested adult cardiac tissue for automated single-cell sequencing of both the healthy and diseased heart. Using this approach, we were able to collect reliable gene expression data for all main cardiac cell types. Clustering analysis uncovered both known and novel markers of certain cell populations and led to the identification of multiple subpopulations within a certain cell type. Single-cell sequencing analysis of both healthy hearts and hearts suffering from ischemic injury indicated that cardiac damage gives rise to new subpopulations of known cell types. Using our single-cell sequencing data, we were able to identify cytoskeleton-associated protein 4 (CKAP4) as a novel marker for activated fibroblasts, which we were able to validate in cardiac samples from patients suffering from ischemic heart disease. Additionally, in a set of in vitro experiments, we were able to show that CKAP4 is involved in myofibroblast activation.

Altogether, our data for the first time show the feasibility of using single-cell sequencing on the adult heart to study transcriptomic differences between cardiac cell types and the heterogeneity in gene expression within 1 cell population. This method will greatly advance the molecular insights into cellular mechanisms that are relevant for cardiac remodeling and function.

Methods

An expanded methods section, a step-by-step protocol for the single-cell sequencing approach, and any associated references are available in the online-only Data Supplement. The data, analytic methods, and study materials will be made available by the authors to other researchers for purposes of reproducing the results or replicating the procedure.

Experimental Animals

All animal studies were performed in accordance with institutional guidelines and regulations of the Animal Welfare Committee of the Royal Netherlands Academy of Arts and Sciences. C57BL/6J mice were subjected to sham (control) or ischemia reperfusion (IR) surgery as previously reported.⁷ Hearts were collected and analyzed 3 days after surgery.

Digestion of the Heart

After collecting the infarcted areas (infarct and border zone region) or the corresponding region of control hearts, the tissue was digested for 15 minutes and used for subsequent RNA isolation or single-cell sorting and sequencing.

Flow Cytometry

The freshly collected cardiac cell lysates were resuspended in DMEM, and living single cells were sorted into 384-well plates based on multiple scatter properties and DAPI exclusion. After cell sorting, the plates were immediately centrifuged and stored at −80°C.

Library Preparation and Sequencing of Single Cells

The SORT-seq procedure was applied as described previously.⁵ Illumina sequencing libraries were prepared using the TruSeq small RNA primers (Illumina) and sequenced paired-end at 75-bp read length with Illumina NextSeq.

Data Analysis of Single-Cell RNA Sequencing

Paired-end reads from Illumina sequencing were mapped with BWA-ALN⁸ to the reference genome GRCm38/mm10. For quantification of transcript abundance, the number of transcripts containing unique molecular identifiers per cell-specific barcode was counted for each gene. Next, the RaceID2 algorithm was used to cluster cells based on K-medoids clustering, visualize cell clusters using t-distributed Stochastic Neighbor Embedding (t-SNE), and compute genes up- or downregulated in all cells within the cluster compared with cells not in the cluster.^2,3

Bulk Sequencing and Data Analysis

For bulk sequencing, RNA from cardiac tissue was isolated using Trizol. Subsequently, libraries were prepared and sequenced using a similar protocol as described for single-cell RNA sequencing.

Pathway Analysis and Gene Ontology

To investigate whether differentially expressed genes in subgroups of cells share a similar biological function, enrichment analyses on these genes were performed using the gene ontology enrichment tool and the Kyoto Encyclopedia of Genes and Genomes pathway analysis using DAVID.⁹ Significant enrichment of genes in gene ontology terms and Kyoto Encyclopedia of Genes and Genomes pathway analyses are shown as P values corrected for multiple testing using the Benjamini–Hochberg method.

Human Heart Samples

Approval for studies on human tissue samples was obtained from the Medical Ethics Committee of the University Medical Center Utrecht, The Netherlands (12#387). In this study, we included tissue from the left ventricular free wall of patients with ischemic heart disease (infarct, border zone, and remote) and left ventricular free wall of nonfailing donor hearts.

Gene expression values obtained by quantitative polymerase chain reaction (qPCR) were plotted for correlation analysis.

Statistical Analysis (qPCR)

The number of samples (n) used in each experiment is indicated in the legend or shown in the figures. The results are presented as mean±standard error of the mean. For qPCR analysis, statistical analyses were performed using PRISM (GraphPad Software Inc). If 2 groups were compared, a Student’s t test was used.

Results

Isolation of High-Quality RNA From Digested Hearts From Adult Mice

To create a reliable gene expression atlas of all cardiac cell types, we first aimed to determine the optimal method for tissue digestion and RNA extraction to obtain high-quality RNA from single-cell suspensions of the adult heart. To do so, mice were euthanized, after which the hearts were perfused with cold perfusion buffer. After collecting the anterior wall of the left ventricle, the tissue was washed in ice-cold perfusion buffer, kept on ice, minced into small pieces, and transferred into a glass vial with cold digestion buffer. After digestion of the tissue, the cell suspension was gently pipetted up and down, after which the lysate was passed through a 100-μm cell strainer. The strainer was then rinsed with DMEM, and the cells were collected for subsequent RNA extraction or flow cytometry (Figure 1A and 1B).

**Figure 1.** **Sorting single cells from the adult heart. A**, Schematic representation of the heart and all main cardiac cell types. B, Work flow of the protocol. C, Gating strategy to sort single cells based on different scatter properties. D, Schematic image of the heart highlighting the area selected for enzymatic digestion; images of cells before D’ and after sorting D”. E, Representative bioanalyzer results for RNA quality for the indicated number of sorted cells from control heart. This quality step was performed on each heart used for digestion and downstream single-cell analysis. FSC indicates forward scatter; N/A, not applicable; RIN, RNA integrity number; and SSC, sideward scatter.

To optimize our digestion protocol, we started out by testing 4 different digestion solutions that are commonly used to digest muscle tissue containing liberase, collagenase II, pancreatin, or trypsin.^10–13 Based on cellular imaging and assessment of RNA quality by RNA integrity number (RIN), the digestion solution containing Liberase appeared most optimal for dispersing adult cardiac tissue while maintaining intact RNA (Figure IA and IB in the online-only Data Supplement).

To compare both the influence of the solution containing the single-cell suspension and time the samples were kept on ice before further processing, we tested both DPBS with 5% FBS¹⁴ and DMEM and collected the cells for RNA analysis either immediately or after having been on ice for 30 or 60 minutes. Although the time on ice did not seem to influence the RIN, using DMEM appeared to provide higher quality RNA when lysing cardiac tissue (Figure IC in the online-only Data Supplement).

Next, we examined whether digestion in a 37°C incubator would be better than a 37°C shaking water bath using 2 different concentrations of the digestion enzyme Liberase. Based on RIN, 0.5 mg liberase yielded good-quality RNA, and visually it was evident that using the shaking water bath was better at digesting the pieces of cardiac tissue into suspension (Figure ID in the online-only Data Supplement). RNA isolated from mechanically homogenized cardiac tissue was taken along as a positive control (control heart).

To test whether the method of RNA isolation would influence the RNA quality, we used both the mirVana RNA isolation kit (Thermo Fisher Scientific) and Trizol (Invitrogen) to isolate RNA from the digested cardiac cells. RNA integrity was comparable for samples isolated with Trizol or mirVana (Figure IE in the online-only Data Supplement).

Together these data indicated that, based on cell morphology and RNA quality, using liberase to digest adult cardiac tissue for 15 minutes in a 37°C shaking water bath before Trizol RNA isolation, was optimal for obtaining a single-cell suspension of the heart (Figure IE in the online-only Data Supplement, indicated by an arrow).

Single-Cell Sorting Strategy

After enzymatically dispersing cardiac tissue, flow cytometry was used to separate the cells. Empirically, we found that using a large nozzle size (130 µm) allowed for sorting all range of cardiac cells without damaging larger cardiomyocytes. We based our gating strategy on multiple scatter properties, including DAPI to sort for living cells and green autofluorescence to sort for more complex cells that contain cytoskeletal filaments (Figure 1C).^15,16 Our results indicated that on average 89.4% of the cells from control hearts were viable after sorting, of which 92% showed an autofluorescent signal (Figure 1C). Additionally, to enrich for bigger cells, we selected for cells with a higher forward scatter width. Imaging the cells after sorting indicated that the cells remained intact and suitable for further sequencing applications (Figure 1D and 1E). Because cardiomyocytes are notoriously difficult to isolate by sorting strategies, we wanted to confirm the quality of the isolated cardiomyocytes after sorting. To do so, we crossed αmyosin-heavy chain-Cre transgenic mice with R26 lox-Stop-lox tdTomato mice to mark the cardiomyocyte population (Figure IIA in the online-only Data Supplement). After sorting we obtained 88.8% living cells (Figure IIB in the online-only Data Supplement), in line with data obtained in Figure IC in the online-only Data Supplement. To confirm that these were living cells, we resorted the single-cell lysate, which indicated 99.5% of these sorted cells were alive. (Figure IIC in the online-only Data Supplement). Imaging of the Tomato signals showed the cardiomyocytes to be intact after the sorting procedure (Figure IID in the online-only Data Supplement). To ensure that the quality of RNA remained intact even after sorting, we collected different quantities of cardiac cells and isolated RNA. Bioanalyzer results indicate that after cell separation, the RNA isolated from the dispersed and sorted cells remained of good quality as indicated by RIN (Figure IE in the online-only Data Supplement).

Single-Cell Sequencing to Identify Gene Expression Signatures in All Main Cardiac Cell Types

Using the SORT-Seq protocol,⁵ on average we detected 16 874 raw unique reads per cell. The distribution of the readcounts across cells indicated that the reads come from single cells because we did not observe the bi- or multimodal distribution of reads one would expect when detecting a transcript from doublets or multiplets (Figure IIE in the online-only Data Supplement). After applying filtering procedures for quality and input (see online-only Data Supplement), 426 cells from 3 different control hearts were used for downstream in silico analysis.

For the identification and analysis of all main cardiac cell types, we used the RaceID2 algorithm.^3,17 K-medoids clustering of 1-Pearson correlation coefficients revealed 14 distinct cell clusters in the adult heart (Figure 2A). The separation between the different cell clusters was further validated by a t-distributed Stochastic Neighbor Embedding (t-SNE) map showing lower intracluster cell-to-cell distance compared with the intercluster distances (Figure 2B). Next, we assessed which genes were differentially expressed within each cluster compared to all other clusters (Database I in the online-only Data Supplement). We used the abundance of known marker genes to determine the cell identity of the different cardiac cell clusters (Figure 2C and Database I in the online-only Data Supplement). We were able to identify clusters belonging to all major cell populations in the heart (Figure IIIA in the online-only Data Supplement). The t-SNE maps showing the expression of some of these marker genes indicated the presence of cardiomyocytes (Figure 2D), fibroblasts (Figure 2E), endothelial cells (Figure 2F), and macrophages (Figure 2G).

**Figure 2.** **Clustering of cardiac cells based on gene expression differences. A**, Heatmap showing distances in cell-to-cell transcriptomes of 426 cells obtained from 3 hearts. Distances are measured by 1-Pearson’s correlation coefficient. K-medoids clustering identified 14 different cell clusters depicted on the x and y axes of the heatmap. B, t-SNE map indicating transcriptome similarities among all individual cells. Different numbers and colors highlight the different clusters identified by K-medoids clustering shown in A. C, Tables showing a list of known marker genes of main cardiac cell types used to identify the subpopulations of cells identified in A. **D through G**, t-SNE maps indicating the expression of selected, well-established cellular markers in cell populations identified as cardiomyocytes (D), fibroblasts (E), endothelial cells (F), and macrophages (G). Data are shown as normalized transcript counts on a color-coded logarithmic scale. t-SNE indicates t-distributed Stochastic Neighbor Embedding.

Taken together, these findings demonstrate that by using our method, we are able to generate a single-cell gene expression profiling of the adult heart, which can identify all major cell types.

Contribution of Mitochondrial Transcripts Differ for Individual Cell Types

While investigating gene expression signatures in all main cardiac cell types (Figure IIIA and Database I in the online-only Data Supplement), we observed that the contribution of mitochondrial and genomic transcripts varied among different cells. All cardiomyocytes have higher percentages of mitochondrial transcripts (58% to 86% of total transcripts) when compared with other cardiac cell types (Figure IIIB through IIIE in the online-only Data Supplement). Because mitochondrial transcripts are so abundant, we excluded them from the clustering analysis and focused only on the differential expression of genomic genes.

Single-Cell Sequencing Identifies Cell Type-Specific Subpopulations in the Healthy Heart

Single-cell sequencing of the adult heart revealed multiple clusters within the same cell type (Figure 2C and Figure III in the online-only Data Supplement). These cells are bioinformatically clustered based on the differential expression of genes, while they also express marker genes that identify them as a specific cell type. According to the gene expression of marker genes, we were able to identify 4 different clusters of cardiomyocytes, 2 clusters of endothelial cells, 2 clusters of fibroblasts, 2 clusters of macrophages, 1 cluster of smooth muscle cells, and 1 cluster of erythrocytes (Figure 2C and Figure III in the online-only Data Supplement).

To explore these clusters in more detail, we focused on the cardiomyocyte clusters. These clusters are defined as cardiomyocytes based on the enriched expression of cardiomyocyte marker genes compared with other cells (Figure 3A, indicated in red, and Figure 3B). The enriched expression of a divergent set of genes classifies them as separate cell clusters based on the RaceID2 parameters (Figure 3A, indicated in black, and Figure 3B). It is interesting to note that the relative expression of the well-known cardiomyocyte markers genes showed a high level of variation between the different cardiomyocyte clusters, and as expected we observed them to be enriched compared with the fibroblasts (Figure 3C).

**Figure 3.** **Gene expression differences across CM subpopulations. A**, Table showing the top 25 highest expressed and enriched genes per cluster. Genes highlighted in red are known as cardiomyocyte markers. B, t-SNE maps showing the distribution in the expression of selected cardiomyocyte markers that are enriched in indicated cardiomyocyte clusters. Expression is shown as normalized transcript counts on a color-coded scale. C, Heatmap of average normalized number of reads across the 4 cardiomyocyte clusters and a fibroblast cluster as a control. t-SNE indicates t-distributed stochastic neighbor embedding. CM indicates cardiomyocyte.

On the basis of the differential expression of cardiomyocyte marker genes, cluster 4 appeared to be the most divergent from the other cardiomyocyte clusters (Figure 3C). A t-SNE map confirmed that this cluster does express cardiomyocyte markers, such as cardiac troponin T, Tnnt2 (Figure 4A). However, it is the only subpopulation of cardiomyocytes that is enriched for Myozenin2 (Myoz2) expression (Figures 3 and 4B). To determine whether the clustering of Myoz2 expression is a product of stoichastic gene expression within the cardiomyocytes of the heart, we aimed to validate this cluster in an independent mouse model. To do so, we specifically sorted and sequenced cardiomyocytes from cardiac tissue from mice in which we labeled the cardiomyocytes with tdTomato (α myosin-heavy chain -Cre transgenic mice crossed with R26 lox-Stop-lox tdTomato mice) (Figure IIA in the online-only Data Supplement). Similar to the control hearts (Figure IVA in the online-only Data Supplement), we also detected a Myoz2-enriched cardiomyocyte cluster (Figure IVB in the online-only Data Supplement). In comparing the highly expressed genes in the Myoz2-enriched cardiomyocyte clusters from either C57BL/6J or tdTomato mice, we identified a large overlap in enriched genes from both clusters (Figure IVC in the online-only Data Supplement). The overlap in gene enrichment strongly suggests the clustering data to be reliable and that the Myoz2-enriched cardiomyocyte cluster is indeed different from the other cardiomyocyte clusters.

**Figure 4.** **Myozenin 2 expression is restricted to a subset of cardiomyocytes. A** and B, t-SNE maps showing expression of *Tnnt2* (A) and *Myoz2* (B) across cells from control hearts. Expression is shown as normalized transcript counts on a color-coded scale. C, Representative confocal images of control hearts stained for TNNT2 (green), MYOZ2 (red), and DAPI (blue). Immunohistochemistry was performed on 3 control hearts. D and E, The Kyoto Encyclopedia of Genes and Genomes pathway analysis of top 200 upregulated (D) or downregulated (E) genes in the *Myoz2*–expressing cardiomyocyte cluster (cluster 4) compared with all other cardiomyocyte clusters (clusters 1, 3, and 9). t-SNE indicates t-distributed stochastic neighbor embedding.

Immunohistochemistry indicated MYOZ2 to be enriched in a layer of cardiomyocytes (costained with TNNT2) located at the epicardial surface of the heart (Figure 4C). Gene ontology analysis of the most differentially regulated genes in the Myoz2-enriched cluster versus other cardiomyocytes indicated the highly detected genes to be involved in degenerative disorders of the central nervous system (Figure 4D), whereas the lowly expressed genes appeared to be involved in cardiac diseases (Figure 4E). This finding is interesting because Myoz2, also known as Calsarcin-1, is an inhibitor of the pathological, prohypertrophic phosphatase Calcineurin.^18,19

In conclusion, our data show that single-cell sequencing analysis on cardiac cells can serve to identify subpopulations of a certain cell type, thus indicating a large heterogeneity among cells from the same cell type. Additionally, this approach allows for the detection of novel cell type-enriched gene expression, providing a basis for discovering new gene functions in certain cell types.

Single-Cell Sequencing of the Injured Heart

Ischemic heart disease is the most common form of cardiovascular disease inducing a remodeling response across the damaged area that involves fibroblast activation, immune cell infiltration, neoangiogenesis, and a change in cardiomyocyte function.^20,21 The identification of new regulators, transcription factors, and molecular pathways that are relevant for these cellular processes could eventually lead to the development of new therapeutic strategies for patients suffering from this disease.

To determine whether our method would allow for studying the influence of disease on inter- and intracellular changes in gene expression, we exposed mice to ischemic injury (IR) and collected samples 3 days after IR.⁷ Using the same isolation procedure as described earlier, we isolated 509 individual cells from the infarcted region of heart (Figures V and VI in the online-only Data Supplement). We pooled these cells with the 426 cells obtained previously from the corresponding region of control hearts for in silico analysis with RaceID2. Using our optimized protocol, we were able to obtain good-quality RNA from single-cell suspensions from the infarct region, as assessed by RIN (Figure VIB in the online-only Data Supplement). K-medoids clustering of 1-Pearson correlation revealed 17 different cell clusters in all pooled cells from control and 3 days after IR hearts (Figure VIC in the online-only Data Supplement). Using the abundance of known marker genes to determine the cell identity of the different cardiac cell clusters, we were able to identify clusters belonging to all major cell populations in the heart (Figure VIIA in the online-only Data Supplement), which was validated by t-SNE maps showing enriched marker gene expression in these individual clusters (Figure VIIB through VIIE in the online-only Data Supplement). To see how ischemic injury would affect the expression of mitochondrial and genomic genes within all cell types, we compared the average relative expression of genomic and mitochondrial genes within each cluster obtained from either healthy or diseased hearts (Figure VIIIA through VIIID in the online-only Data Supplement). On average, the expression of mitochondrial genes decreased after ischemic injury within all cell clusters.

Inter- and Intracellular Gene Expression Changes Induced by Ischemic Injury

By generating a t-SNE map to indicate transcriptome similarities between all individual cells, we were able to see that the majority of clusters contained cells from both control and injured hearts (Figure 5A and 5B). However, injury triggered the appearance of disease-enriched cell clusters (Figure 5A through 5D and Figure VIIIE in the online-only Data Supplement).

Focusing on the fibroblasts, we were able to identify 3 different fibroblast clusters (7, 13, and 15) (Figure 5B through 5D). By examining the library origin, we observed that the cells in clusters 13 and 15 predominantly stemmed from the injured hearts, whereas cluster 7 contained cells from both conditions (Figure 5C and 5D). Cells from clusters 13 and 15 were characterized by the relatively high expression of Postn, Wisp1, and Tnc, previously associated with fibroblast activation,^22–24 and t-SNE maps again confirmed the enriched expression of these genes in the disease-enriched fibroblast clusters (Figure 5E and Figure IX in the online-only Data Supplement). The disease-specific induction of these genes was validated by qPCR on cardiac tissue from either control or injured hearts (Figure 5F).

Ckap4 Expression Is Specifically Increased in Activated Fibroblasts

Having identified a population of fibroblasts stemming from both healthy and diseased hearts (cluster 7) and a population stemming predominantly from diseased hearts (cluster 13 and 15), we next determined the differentially expressed genes between these populations. Using a log2-fold change of 1.5 or −1.5 or more with a P value of 0.05, we found 11 genes upregulated and 20 genes downregulated in the disease-enriched fibroblasts population (Figure 5G). In using the same parameters for defining differentially expressed genes in larger cell clusters, the number of differentially expressed genes between cells from the healthy and ischemic hearts became much larger. For example, we found 205 genes differentially expressed between healthy and ischemic cardiomyocytes coming from clusters 1, 3, 4, and 8, whereas 191 genes were differentially expressed in macrophages coming from either the healthy or ischemic hearts forming cluster 5, 9, 14, and 16 (data not shown). The upregulated genes in these diseased-enriched clusters were related to various processes associated with extracellular matrix deposition and collagen deposition (Figure 5H), a hallmark of the fibrotic response after ischemic injury of the heart.²⁵ In line with this observation, among the enriched genes, we detected various collagen genes and markers known to be expressed in activated fibroblasts, such as Postn(Figure 5I).²² The identification of clusters that are enriched for disease-related genes from ischemic hearts further underscores the validity of our clustering approach.

In addition to known markers for activated fibroblasts, we identified Ckap4 to be upregulated in this population of activated fibroblasts. Ckap4 is a transmembrane protein and can act as a receptor for various ligands in different cells.^26,27 However, its function in cardiac fibroblasts remains unknown. We could confirm the upregulation of Ckap4 after io on ischemic injury in the whole heart by bulk RNA sequencing (Figure 5J). However, our single-cell sequencing data allowed us to detect the upregulation of Ckap4 specifically in activated fibroblasts and not in any other cell types, which was confirmed by t-SNE maps (Figure 5K and 5L). By immunohistochemistry, we validated the induction of CKAP4 in hearts after ischemic injury, which overlapped with cells expressing the fibroblasts marker vimentin (Figure 6A). To investigate the functional relevance of Ckap4 expression in activated fibroblasts, we inhibited Ckap4 expression in fibroblasts, after which we exposed them to TGFβ, a well-known inducer of myofibroblasts formation (Figure 6B).²⁸ TGFβ was able to induce Ckap4 expression when compared with control. Small interfering RNA-mediated inhibition of Ckap4 resulted in a dose-dependent reduction of Ckap4 (Figure 6C) on the mRNA and protein level (Figure 6C and 6D). The qPCR analysis showed that TGFβ treatment led to an increase in the expression of markers for activated fibroblasts and that inhibition of Ckap4 further potentiated this effect under such conditions (Figure 6E). Although this finding warrants further investigation, this seems to imply that CKAP4 in activated fibroblast functions to dampen the expression of genes related to activated fibroblast.

**Figure 6.** *Ckap4* is specifically expressed in fibroblasts after IR and in activated fibroblasts in vitro. A, Representative confocal images of control (**left**) and 3 dpIR (**right**) mouse heart stained for CKAP4 (red) and known markers for different cardiac cell types (green): endothelial cells (PECAM1, **upper**), cardiomyocytes (ACTN2, **middle**), and fibroblasts (VIM, **lower**). Immunohistochemistry was performed on 3 hearts per condition. B, Schematic overview of *Ckap4* knockdown experiments in activated NIH/3T3 fibroblasts. C, qPCR for *Ckap4* after TGFβ stimulation in NIH/3T3 fibroblasts after *Ckap4* siRNA treatment or scrambled siRNA treatment. Expression levels are relative to vehicle-treated NIH/3T3 cells transfected with scrambled siRNA (n=3 to 10; 2-sample t test; #P<0.005 versus scrambled siRNA vehicle treated; *P<0.005 versus scrambled siRNA vehicle treated; $P<0.005 versus scrambled siRNA TGFβ treated). D, Representative confocal images of NIH/3T3 cells on vehicle or TGFβ treatment without or with *Ckap4* knockdown. E, qPCR for marker genes of activated fibroblasts in NIH/3T3 cells on vehicle or TGFβ treatment with or without *Ckap4* knockdown. Expression levels are relative to vehicle-treated NIH/3T3 cells transfected with scrambled siRNA (n=5 to 9; 2-sample t test; #P<0.05 versus scrambled siRNA vehicle treated; *P<0.05 versus scrambled siRNA vehicle treated; $P<0.005 versus scrambled siRNA vehicle treated). 3 dpIR indicates 3 days afer IR; *Ckap4*, cytoskeleton-associated protein 4; IR, ischemia reperfusion; NIH, National Institutes of Health; qPCR, quantitative polymerase chain reaction; siRNA, small interfering RNA; TGFβ, transforming growth factor β; and VIM, vimentin.

Also in cardiac samples from patients suffering from ischemic heart disease, we were able to confirm a positive correlation in expression between CKAP4 and genes known to be induced in activated cardiac fibroblasts (Figure 7A through 7C). In addition, immunohistochemistry showed a strong overlap between CKAP4 expression and various well-established fibroblast markers in human ischemic hearts (Figure 8D), suggesting that CKAP4 also has a role in activated fibroblasts in humans.

**Figure 7.** **CKAP4 is coexpressed with markers of activated fibroblasts in human ischemic hearts. A** through C, Pearson correlation between the expression of *CKAP4* and markers for activated fibroblasts *POSTN*(A), *CTHRC1* (B), and *FN1* (C) determined by qPCR analysis on human cardiac tissue (n=30, including control hearts and 3 different parts of the ischemic heart: remote, border zone, and infarct, n=35). D, Representative confocal images from human ischemic hearts stained for CKAP4 (red) and markers for different cardiac cell types (green): endothelial cells (PECAM1, **upper left**), cardiomyocytes (ACTN2, **upper right**), and fibroblasts (ACTA2, **middle left**; DDR2, **middle right**; VIM, **lower left**; PDGFR, **lower right**). Immunohistochemistry was performed on 3 hearts per condition. CKAP4, cytoskeleton-associated protein 4; qPCR, quantitative polymerase chain reaction; andVIM, vimentin.

Taken together, our single-cell RNA sequencing data on healthy and diseased hearts demonstrate that we are able to identify disease-specific subpopulations of various cell types. Comparing gene expression patterns between healthy and diseased subpopulations within cell types allowed us to detect cell type-specific upregulation of various genes. Using this approach, we identified Ckap4 as a marker specifically upregulated in activated fibroblasts.

Discussion

Here we show that our optimized technique to isolate and sort adult cardiac cells in combination with a high throughput method to sequence single cells with SORT-seq^2,5,29 provides a unique opportunity to reliably obtain single-cell gene expression data of the adult mammalian heart (Figure X in the online-only Data Supplement and step-by-step protocol in the online-only Data Supplement). Among the sequenced cells, we were able to identify all major cardiac cell types, including cardiomyocytes, fibroblasts, endothelial cells, and macrophages.^1,30

A major advantage of single-cell sequencing is the ability to detect heterogeneity within a certain cell type in the heart. Clustering analysis of our single-cell sequencing data shows differentially expressed genes in subsets of cells that are likely to contribute to the functional diversity within different cell types. For example, we found 4 clusters of cardiomyocytes that when compared to fibroblast, as expected, show a significant enrichment for cardiomyocyte marker gene expression. However, our data also show that there is substantial heterogeneity in the expression of well-established cardiomyocyte markers among the different cardiomyocyte subpopulation (Figure 3C). Although we currently do not know the biological relevance of this observation, it could imply that there are functionally different cardiomyocyte populations already in a healthy heart.

In addition, we show that only a subset of cardiomyocytes express Myoz2, a protein belonging to a family of sarcomeric proteins.¹⁸ Myoz2 has been shown to tether α-actinin to calcineurin, a well-known inducer of cardiomyocyte hypertrophy,³¹ thereby inhibiting the pathological hypertrophic response of cardiomyocytes.¹⁸ Myoz2 expression limited to only a subset of cardiomyocytes, predominantly located toward the epicardial region of the heart, raises the possibility that subpopulations of cardiomyocytes respond differently to calcineurin-mediated hypertrophy, with some being more resistant than others. A challenge for single-cell sequencing studies is to discriminate between difference based on stoichastic changes in gene expression or biology.³² To show that the differences in expression in the Myoz2-enriched cardiomyocte cluster are biologically relevant, we additionally clustered sequenced cardiomyocytes from an additional mouse model (Figure IV in the online-only Data Supplement). Also in this study, we were able to identify a distinct cluster of cardiomyocytes to be enriched for Myoz2 expression. When comparing the enriched genes in both Myoz2 clusters, we detected a large overlap, suggesting that these clusters of Myoz2-enriched cardiomyocytes indeed resemble each other and might be functionally different compared with the other cardiomyocytes. Determining the biological function of this subset of cardiomyocytes could yield valuable insights into cellular mechanisms responsible for cardiac function and pathologies.

Comparing single-cell sequencing datasets from both the healthy and injured heart revealed that disease gives rise to new subpopulations of known cell types that appear specific for or enriched with cells coming from the diseased heart. Although it is known that the cellular composition of the heart changes during pathological stress,³³ we are limited in our ability to detect genome-wide changes in expression specifically occurring within each cell type during cardiac stress. Our single-cell sequencing data now allow us to identify transcriptome-wide differences in all mayor cardiac cell types among different conditions. By defining disease-related cell subpopulations and the associated changes in gene expression, we expect to identify novel molecular mechanisms that are relevant for the cellular changes underlying heart disease.

The potential of the described approach is nicely exemplified by the identification Ckap4 as a novel marker for activated fibroblasts. Ckap4 was previously reported to have a function in cell proliferation during tumor progression,³⁴ but its function in the heart so far has been unstudied. In our dataset, we found Ckap4 to be expressed in the same cell population as Postn, Ctrc1, and Fn1, well-known markers for cardiac myofibroblasts.^22–24 Immunohistochemistry on both control and ischemic hearts confirmed the expression of CKAP4 to be specific for the stressed heart and to overlap with vimentin, a marker for fibroblasts in the ischemic heart. In vitro loss of function of Ckap4 showed an overactivation of myofibroblast-related genes before and after TGFβ stimulation, suggesting a role for Ckap4 during fibroblast activation. Moreover, we observed a positive correlation between CKAP4 and fibroblasts markers in cardiac biopsies from patients suffering from ischemic heart disease.

In an attempt to collect living cells for our single-cell analysis, we biased our flow cytometry gating toward larger cells. As a result, we preferentially isolated cardiomyocytes (71% for sequenced cells in control heart and 59% in diseased hearts) while it is estimated that 30% of cardiac cells are cardiomyocytes.³⁰ Nonetheless, we were still able to detect also other cell populations. Further optimization of our gating strategy is required to obtain a representative spectrum of cardiac cell types after cell sorting by flow cytometry.

Inherent to the technique, the low-sequencing efficiency prevents us from detecting lower expressed genes in a cell. Because the majority of reads are currently being spent on mitochondrial genes (≤23% to 84% of reads for cardiomyocytes clusters from both sham and IR), efforts to separate out transcripts derived from the mitochondria could greatly enhance the sequencing efficiency.

Although future developments will continue to optimize the single-cell sequencing technology on organs, our study for the first time indicates the feasibility of using this technique on adult cardiac tissue. Our data indicate the potential of this method to identify transcriptional differences between cardiac cell types and to study the heterogeneity in gene expression between the different subpopulations. Together these discoveries create major opportunities to unveil new gene functions for cellular biology and organ function.

Acknowledgments

The authors thank Judith Vivié, Susanne van den Brink, Jean-Charles Boisset, Anna Alemany Arias, and Mauro Muraro (Hubrecht Institute) for technical assistance and helpful discussions on single-cell sequencing. The authors gratefully acknowledge Thomas Cahill for providing expertise related to macrophages biology. M.M.G., B.M., G.P.A.L., and E.v.R. designed the experiments. M.M.G., B.M., H.d.R., H.T., D.V., and S.v.d.E performed all experiments. M.M.H.H. provided human biopsies. M.M.G. and B.M. analyzed the data. M.M.G., B.M., and E.v.R. wrote the article.

Sources of Funding

This work was supported by the Leducq Foundation (14CVD04) and the European Research Council under the European Union’s Seventh Framework Program (ERC Grant Agreement AdG 294325 GeneNoiseControl and CoG 615708 MICARUS). M.M.G. was supported by a Dr Dekker postdoctoral fellowship from the Dutch Heart Foundation (NHS2016T009).

Disclosures

None.

Footnotes

*Dr Gladka and B. Molenaar contributed equally.

https://www./journal/circ

The online-only Data Supplement is available with this article at https://www./journal/circ/doi/suppl/10.1161/CIRCULATIONAHA.117.030742.

Eva van Rooij, PhD, Hubrecht Institute, KNAW, Uppsalalaan 8, 3584 CT Utrecht, The Netherlands. E-mail e.vanrooij@hubrecht.eu

References

1. Pinto AR, Ilinykh A, Ivey MJ, Kuwabara JT, D’Antoni ML, Debuque R, Chandran A, Wang L, Arora K, Rosenthal NA, Tallquist MD. Revisiting cardiac cellular composition.Circ Res. 2016; 118:400–409. doi: 10.1161/CIRCRESAHA.115.307778.Link Google Scholar
2. Grün D, van Oudenaarden A. Design and analysis of single-cell sequencing experiments.Cell. 2015; 163:799–810. doi: 10.1016/j.cell.2015.10.039.Crossref Medline Google Scholar
3. Grün D, Muraro MJ, Boisset JC, Wiebrands K, Lyubimova A, Dharmadhikari G, van den Born M, van Es J, Jansen E, Clevers H, de Koning EJP, van Oudenaarden A. De novo prediction of stem cell identity using single-cell transcriptome data.Cell Stem Cell. 2016; 19:266–277. doi: 10.1016/j.stem.2016.05.010.Crossref Medline Google Scholar
4. DeLaughter DM, Bick AG, Wakimoto H, McKean D, Gorham JM, Kathiriya IS, Hinson JT, Homsy J, Gray J, Pu W, Bruneau BG, Seidman JG, Seidman CE. Single-cell resolution of temporal gene expression during heart development.Dev Cell. 2016; 39:480–490. doi: 10.1016/j.devcel.2016.10.001.Crossref Medline Google Scholar
5. Muraro MJ, Dharmadhikari G, Grün D, Groen N, Dielen T, Jansen E, van Gurp L, Engelse MA, Carlotti F, de Koning EJ, van Oudenaarden A. A single-cell transcriptome atlas of the human pancreas.Cell Syst. 2016; 3:385.e3–394.e3. doi: 10.1016/j.cels.2016.09.002.Google Scholar
6. Li G, Xu A, Sim S, Priest JR, Tian X, Khan T, Quertermous T, Zhou B, Tsao PS, Quake SR, Wu SM. Transcriptomic profiling maps anatomically patterned subpopulations among single embryonic cardiac cells.Dev Cell. 2016; 39:491–507. doi: 10.1016/j.devcel.2016.10.014.Crossref Medline Google Scholar
7. Hullinger TG, Montgomery RL, Seto AG, Dickinson BA, Semus HM, Lynch JM, Dalby CM, Robinson K, Stack C, Latimer PA, Hare JM, Olson EN, van Rooij E. Inhibition of miR-15 protects against cardiac ischemic injury.Circ Res. 2012; 110:71–81. doi: 10.1161/CIRCRESAHA.111.244442.Link Google Scholar
8. Li H, Durbin R. Fast and accurate short read alignment with Burrows-Wheeler transform.Bioinformatics. 2009; 25:1754–1760. doi: 10.1093/bioinformatics/btp324.Crossref Medline Google Scholar
9. Huang da W, Sherman BT, Lempicki RA. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources.Nat Protoc. 2009; 4:44–57. doi: 10.1038/nprot.2008.211.Crossref Medline Google Scholar
10. Masson-Pévet M, Jongsma HJ, De Bruijne J. Collagenase- and trypsin-dissociated heart cells: a comparative ultrastructural study.J Mol Cell Cardiol. 1976; 8:747–757.Crossref Medline Google Scholar
11. Spada F, Fuoco C, Pirrò S, Paoluzi S, Castagnoli L, Gargioli C, Cesareni G. Characterization by mass cytometry of different methods for the preparation of muscle mononuclear cells.N Biotechnol. 2016; 33(5 Pt A):514–523. doi: 10.1016/j.nbt.2015.12.007.Crossref Medline Google Scholar
12. Liu L, Cheung TH, Charville GW, Rando TA. Isolation of skeletal muscle stem cells by fluorescence-activated cell sorting.Nat Protoc. 2015; 10:1612–1624. doi: 10.1038/nprot.2015.110.Crossref Medline Google Scholar
13. Etchevers H. Primary culture of chick, mouse or human neural crest cells.Nat Protoc. 2011; 6:1568–1577. doi: 10.1038/nprot.2011.398.Crossref Medline Google Scholar
14. Yuelling LW, Du F, Li P, Muradimova RE, Yang ZJ. Isolation of distinct cell populations from the developing cerebellum by microdissection.J Vis Exp. 2014;91:52034.Medline Google Scholar
15. Chorvat D, Kirchnerova J, Cagalinec M, Smolka J, Mateasik A, Chorvatova A. Spectral unmixing of flavin autofluorescence components in cardiac myocytes.Biophys J. 2005; 89:L55–L57. doi: 10.1529/biophysj.105.073866.Crossref Medline Google Scholar
16. Gao G, Johansson U, Rundquist I, Ollinger K. Lipofuscin-induced autofluorescence of living neonatal rat cardiomyocytes in culture.Mech Ageing Dev. 1994; 73:79–86.Crossref Medline Google Scholar
17. Grün D, Lyubimova A, Kester L, Wiebrands K, Basak O, Sasaki N, Clevers H, van Oudenaarden A. Single-cell messenger RNA sequencing reveals rare intestinal cell types.Nature. 2015; 525:251–255. doi: 10.1038/nature14966.Crossref Medline Google Scholar
18. Frey N, Richardson JA, Olson EN. Calsarcins, a novel family of sarcomeric calcineurin-binding proteins.Proc Natl Acad Sci U S A. 2000; 97:14632–14637. doi: 10.1073/pnas.260501097.Crossref Medline Google Scholar
19. Frey N, Barrientos T, Shelton JM, Frank D, Rütten H, Gehring D, Kuhn C, Lutz M, Rothermel B, Bassel-Duby R, Richardson JA, Katus HA, Hill JA, Olson EN. Mice lacking calsarcin-1 are sensitized to calcineurin signaling and show accelerated cardiomyopathy in response to pathological biomechanical stress.Nat Med. 2004; 10:1336–1343. doi: 10.1038/nm1132.Crossref Medline Google Scholar
20. Tarone G, Balligand JL, Bauersachs J, Clerk A, De Windt L, Heymans S, Hilfiker-Kleiner D, Hirsch E, Iaccarino G, Knöll R, Leite-Moreira AF, Lourenço AP, Mayr M, Thum T, Tocchetti CG. Targeting myocardial remodelling to develop novel therapies for heart failure: a position paper from the Working Group on Myocardial Function of the European Society of Cardiology.Eur J Heart Fail. 2014; 16:494–508. doi: 10.1002/ejhf.62.Crossref Medline Google Scholar
21. van Berlo JH, Maillet M, Molkentin JD. Signaling effectors underlying pathologic growth and remodeling of the heart.J Clin Invest. 2013; 123:37–45. doi: 10.1172/JCI62839.Crossref Medline Google Scholar
22. Li L, Fan D, Wang C, Wang JY, Cui XB, Wu D, Zhou Y, Wu LL. Angiotensin II increases periostin expression via Ras/p38 MAPK/CREB and ERK1/2/TGF-β1 pathways in cardiac fibroblasts.Cardiovasc Res. 2011; 91:80–89. doi: 10.1093/cvr/cvr067.Crossref Medline Google Scholar
23. Colston JT, de la Rosa SD, Koehler M, Gonzales K, Mestril R, Freeman GL, Bailey SR, Chandrasekar B. Wnt-induced secreted protein-1 is a prohypertrophic and profibrotic growth factor.Am J Physiol Heart Circ Physiol. 2007; 293:H1839–H1846. doi: 10.1152/ajpheart.00428.2007.Crossref Medline Google Scholar
24. Maqbool A, Spary EJ, Manfield IW, Ruhmann M, Zuliani-Alvarez L, Gamboa-Esteves FO, Porter KE, Drinkhill MJ, Midwood KS, Turner NA. Tenascin C upregulates interleukin-6 expression in human cardiac myofibroblasts via toll-like receptor 4.World J Cardiol. 2016; 8:340–350. doi: 10.4330/wjc.v8.i5.340.Crossref Medline Google Scholar
25. Kong P, Christia P, Frangogiannis NG. The pathogenesis of cardiac fibrosis.Cell Mol Life Sci. 2014; 71:549–574. doi: 10.1007/s00018-013-1349-6.Crossref Medline Google Scholar
26. Conrads TP, Tocci GM, Hood BL, Zhang CO, Guo L, Koch KR, Michejda CJ, Veenstra TD, Keay SK. CKAP4/p63 is a receptor for the frizzled-8 protein-related antiproliferative factor from interstitial cystitis patients.J Biol Chem. 2006; 281:37836–37843. doi: 10.1074/jbc.M604581200.Crossref Medline Google Scholar
27. Chavda B, Ling J, Majernick T, Planey SL. Antiproliferative factor (APF) binds specifically to sites within the cytoskeleton-associated protein 4 (CKAP4) extracellular domain.BMC Biochem. 2017; 18:13. doi: 10.1186/s12858-017-0088-y.Crossref Medline Google Scholar
28. Chen H, Yang WW, Wen QT, Xu L, Chen M. TGF-beta induces fibroblast activation protein expression; fibroblast activation protein expression increases the proliferation, adhesion, and migration of HO-8910PM [corrected].Exp Mol Pathol. 2009; 87:189–194. doi: 10.1016/j.yexmp.2009.09.001.Crossref Medline Google Scholar
29. Hashimshony T, Yanai I. Revealing developmental networks by comparative transcriptomics.Transcription. 2010; 1:154–158. doi: 10.4161/trns.1.3.13190.Crossref Medline Google Scholar
30. Bergmann O, Zdunek S, Felker A, Salehpour M, Alkass K, Bernard S, Sjostrom SL, Szewczykowska M, Jackowska T, Dos Remedios C, Malm T, Andrä M, Jashari R, Nyengaard JR, Possnert G, Jovinge S, Druid H, Frisén J. Dynamics of cell generation and turnover in the human heart.Cell. 2015; 161:1566–1575. doi: 10.1016/j.cell.2015.05.026.Crossref Medline Google Scholar
31. Panther F, Williams T, Ritter O. Inhibition of the calcineurin-NFAT signalling cascade in the treatment of heart failure.Recent Pat Cardiovasc Drug Discov. 2009; 4:180–186.Crossref Medline Google Scholar
32. Marinov GK, Williams BA, McCue K, Schroth GP, Gertz J, Myers RM, Wold BJ. From single-cell to cell-pool transcriptomes: stochasticity in gene expression and RNA splicing.Genome Res. 2014; 24:496–510. doi: 10.1101/gr.161034.113.Crossref Medline Google Scholar
33. Kehat I, Molkentin JD. Molecular pathways underlying cardiac remodeling during pathophysiological stimulation.Circulation. 2010; 122:2727–2735. doi: 10.1161/CIRCULATIONAHA.110.942268.Link Google Scholar
34. Kimura H, Fumoto K, Shojima K, Nojima S, Osugi Y, Tomihara H, Eguchi H, Shintani Y, Endo H, Inoue M, Doki Y, Okumura M, Morii E, Kikuchi A. CKAP4 is a Dickkopf1 receptor and is involved in tumor progression.J Clin Invest. 2016; 126:2689–2705. doi: 10.1172/JCI84658.Crossref Medline Google Scholar