Lower number of detected genes in pathological ARs
To evaluate possible effects of cell shape on gene expression, we isolated single NRCMs and patterned them on CYTOOchips with fibronectin patches of defined ARs, namely morphotypes AR1, AR7, and AR11 (Fig. 1; Table 1 and Supplementary Fig. S1). In accordance with previously published reports, CMs with shapes corresponding to AR7 were chosen to resemble the morphotypic dimensions of normal CMs in a healthy heart [2]. All cells were monitored by light microscopy and morphotypes were characterized using 3D measurements of confocal image stacks (Supplementary Fig. S2). The chips were designed, so that the surface areas of the different ARs remained the same and this resulted in similar volumes (Table 1 and Supplementary Fig. S2b).
Similar to Kuo et al. [21], cells were harvested after 72 h of plating. Only those single cells that fully covered the fibronectin islands of micropatterns were selected using software (CellSorter) for picking, and immediately lysed in individual PCR tubes containing 2 μl of Smart-Seq2 cell lysis buffer [30]. The quality of the cDNA was checked using the Agilent Bioanalyzer (Agilent) and RNA-Seq libraries were prepared using in-house compatible Tn5 and Nextera index primers (Illumina). After a final clean-up, the size distribution of the sequence libraries was checked using the Agilent high-sensitivity chip and the concentration of each library was measured using the Qubit 3 Fluorometer (Invitrogen).
Single-cell RNA sequencing was performed at the in-house sequencing facility at the Karolinska Institutet using the Genome Analyzer HiSeq2500 (Illumina) for single-end sequencing of 56 bps.
We sequenced a total of 213 single cells with different morphotypes (Table 1) and detected an average of 4852 genes per cell (range 2603–7060). A gene with a read per kilobase million (RPKM) value of > 1 was considered as detected (Fig. 2a). The average number of detected genes in AR7 was 5332 and the number of detected genes was significantly lower in both AR1 (4391, p < 0.001) and AR11 (4835, p < 0.05), when compared to AR7.
Gene expression heterogeneity due to different cell shapes
The next step was to investigate whether cell shapes per se drove gene expression heterogeneity. We did this by performing unsupervised clustering based on all the expressed genes and we observed that the cells with similar morphotypes clustered together in three-dimensional t-SNE plots (Fig. 2b). In line with this, and to find out to what extent our morphotypes possibly changed the transcriptome, we compared patterned cells with unpatterned and pre-patterned CMs. The unpatterned cells were cultured cells that were allowed to take any shape, as detailed in the Methods section, and the pre-patterned cells were those where the cells lysed immediately after isolation. PCA plots were constructed based on the log-normalized expression values of all the expressed genes (Fig. 2c). Patterned cells tended to cluster together, depending on their ARs, but unpatterned and pre-patterned cells were more diverse, suggesting that arbitrary cellular shapes lead to diverse gene expression. Although the unpatterned cells showed more divergence, they partially overlapped the patterned cells, which probably reflected the same length of cell culture. However, the pre-patterned cells, which were never cultured, were even more diverse and distant from the other patterned and unpatterned cells (Fig. 2c). When we compared PC1 and PC3, we observed that the pre-patterned cells overlapped with the different patterned cells.
Downregulation of genes in pathological morphotypes
To analyze the consequences of the morphotypic heterogeneity in greater detail, we identified significant DEGs by comparing different morphotypes and presenting them as a Venn diagrams (Fig. 2d). The complete list of DEGs is provided in Supplementary Table S1. We observed that 734 and 280 genes were significantly downregulated in AR1 and AR11, respectively, compared to AR7. In addition, AR1 and AR11 shared 73 genes in common. On the other hand, only 343 and 85 genes were upregulated in AR1 and AR11, respectively, compared to AR7, with 29 common genes (Fig. 2d). These results suggest that fewer genes were upregulated in pathologic morphotypes. Together with the overall loss of gene expression observed in AR1 and AR11, these data support the notion that elongation of CMs or squaring of CMs had powerful effects on the number of expressed genes. Moreover, the differential gene expression analysis was performed by comparing AR7 versus unpatterned and pre-patterned cells. The complete list of DEGs can be found in Supplementary Table S2.
Canonical pathways and biological processes were influenced by different cell shapes
While the analysis of a single DEG can provide important information, this analysis needs to be complemented by unsupervised bioinformatic approaches. In this study, we applied pairwise comparisons of morphotypes (i.e., AR1 versus AR7, AR11 versus AR7, and AR1 versus AR11) using IPA software [20]. This enabled us to identify a variety of enriched canonical pathways, as well as diseases and biological functions (Fig. 3). Oxidative phosphorylation, eukaryotic initiation factor 2, protein kinase A, and cardiac beta-adrenergic signaling were inhibited in both AR1 and AR11, when compared to AR7, suggesting that common pathways are affected in pathological conditions. On the other hand, Sirtuin signaling was activated in both AR1 and AR11 compared to AR7 (Fig. 3a). We observed upregulation of vitamin D receptor/retinoid X receptor (VDR/RXR) activation and downregulation of cyclic adenosine monophosphate-(cAMP) mediated signaling, in AR11 compared to AR1 (Fig. 3a). We also investigated the cellular processes that are affected by DEGs due to morphotypic variations. Degeneration of the heart, apoptosis of CMs, and necrosis of muscle were predicted to be activated in AR1 versus AR7. However, we observed lower flux of ion and recruitment of cells in AR11 versus AR7 and less migration of cells in AR11 versus AR1 (Fig. 3b). When we took these differences into account, it was interesting to note that changes in cell shape not only caused changes in function [21], but they also caused profound alterations in gene expression.
Sarcomeric and mitochondrial structure of patterned NRCMs
Structural aspects of patterned CMs were examined by staining sarcomeric structures using an alpha-actinin antibody (Fig. 4a). Based on this analysis, we did not observe overall changes in sarcomere structure, but we did observe changes in the orientation of sarcomeres, which appeared more radial in AR1.
Taking into account the fact that mitochondrial pathways were significantly affected, we used the MitoTracker (Cell Signalling Technology) to visualize mitochondria in patterned NRCMs with various ARs (Fig. 4b). We did not observe major mitochondrial network changes: sarcomeric and mitochondrial staining of an unpatterned CM is also shown in Fig. 4c, d. An in-depth analysis of the effects of cell shape on overall cell structure and mitochondrial function is in progress, but that is probably beyond the scope of this paper.
Effects of mechanotransduction pathways on the functional consequences of various ARs
A wide variety of different signal transduction cascades are involved in cardiac mechanotransduction, including nuclear lamina proteins, ion channels, mitogen-activated protein kinases, membrane receptors such as ATR1, and integrins that include Src kinase [5]. As the integrin/Src kinase pathway is probably one of the best-known factors involved in mechanotransduction, but not the only one, we determined the effects of this pathway first [32].
We started by inhibiting β1-integrin in all three morphotypes with blocking anti-β1-integrin antibody (10 µg/ml, cat. no. 555002) (BD Transduction Laboratories). After quality control, we analyzed a total of 87 sequenced β1-integrin-inhibited single cells with different morphotypes (Table 1) and detected an average of 3908 genes per cell (range 2596–5754). The average number of detected genes was 3884, 3978, and 3840 in β1-integrin-inhibited AR1, AR7, and AR11 morphotypes, respectively. Moreover, the number of detected genes was not significantly different between morphotypes of β1-integrin-inhibited condition (Fig. 5a, left panel).
We inhibited Src with saracatinib, a known Src inhibitor [11], in all three morphotypes and sequenced 14 Src-inhibited single cells from each AR, making a total of 42 cells. We performed intra-condition studies, by comparing the number of genes detected between treated morphotypes in each condition. A gene with a read per kilobase million (RPKM) value of ≥ 1 was considered as detected. The average of 4235 genes per cell (range 2294–7012) was detected among all Src-inhibited single cells, while the average number of 3938, 3996, and 4771 was detected, respectively, in Src-inhibited AR1, AR7, and AR11 morphotypes. Moreover, the number of detected genes was not significantly different between all Src-inhibited morphotypes (Fig. 5a, middle panel). In a third, related set of experiments, we overexpressed Src by transducing patterned CMs with the adenovirus–Kras–Src vector in all three morphotypes [34]. After quality control, we analyzed a total of 74 sequenced Src-overexpressed single cells with different morphotypes (Table 1) and detected an average of 4360 genes per cell (range 2251–6382). The average number of detected genes in Src-overexpressed AR7 was 4963 and the number of detected genes was significantly lower in both Src-overexpressed AR1 (4293, p < 0.05) and Src-overexpressed AR11 (3838, p < 0.0001), when compared to Src-overexpressed AR7 (Fig. 5a, right panel).
Next, we investigated the heterogeneity in whole-genome expression profiles between morphotypes of each treated condition. We constructed intra-condition PCA plots, using all expressed genes in the samples of each condition (Fig. 5b). Importantly, we observed that Src-overexpressed morphotypes clustered according to their morphotypes (Fig. 5b, bottom panel), whereas β1-integrin-inhibited and Src-inhibited morphotypes were intermingled (Fig. 5b, top and middle panels). This indicates that inhibition of the integrin/Src kinase pathway at different levels significantly interfered with shape-dependent effects on gene expression and hence links mechano-sensation to regulation of the transcriptome.
To further explore the role of integrin-mediated and Src-mediated mechanosensing pathways on the outcomes of various cell shapes, we identified DEGs, comparing the intra-condition morphotypes (Supplementary Tables S3–5). We derived the enriched IPA canonical pathways and biological functions when we pairwisely compared morphotypes within each treated condition (Fig. 5c).
Interestingly, no canonical pathway or biological function was significantly enriched in any pairwise comparison of Src-inhibited morphotypes.
These results emphasize that inhibition of β1-integrin and of Src abate the effects of cell shape in terms of number of detected genes, heterogeneity in expression profiles, and enriched canonical pathways.
On the other hand, intra-condition comparison of Src-overexpressed morphotypes showed that the number of detected genes was significantly lower in AR1 and AR11 compared to AR7, an effect which was seen in intra-condition comparisons of untreated condition. Moreover, a distinct heterogeneity between expression profiles of different Src-overexpressed morphotypes was revealed in PCA plots, while similar feature was represented in the gene expression PCA plot of untreated morphotypes. Furthermore, cAMP-mediated signaling, protein kinase A signaling, and sirtuin signaling pathway were enriched when we compared AR11 vs AR7 morphotypes in both Src-overexpressed and untreated conditions. Likewise, cardiac hypertrophy signaling, protein kinase A signaling, calcium signaling, sirtuin signaling, and oxidative phosphorylation canonical pathway were enriched, comparing AR1 vs AR7 morphotypes in both Src-overexpressed and untreated conditions.
To evaluate the features of the DEGs, we additionally performed Gene Ontology (GO) enrichment analysis, using PANTHER classification system. Significantly enriched GO biological processes were identified using the Fisher’s exact test type, corrected by false discovery rate. Particularly, “regulation of signaling”, “cellular response to chemical stress”, “regulation of signal transduction”, “response to external stimulus”, and “response to stress” were among significantly enriched GO terms when we compared β1-integrin-inhibited AR11 vs β1-integrin-inhibited AR7. Furthermore, the “cell surface receptor signaling pathway” was enriched when we compared β1-integrin-inhibited AR7 with both β1-integrin-inhibited AR1 and AR11 (Supplementary Fig. S4).
On the other hand, “cardiac muscle hypertrophy in response to stress”, “regulation of cardiac muscle cell contraction”, “sarcomere organization”, and “regulation of the force of heart contraction” were significantly enriched when we compared Src-overexpressed AR7 with both Src-overexpressed AR1 and AR11 (Supplementary Fig. S5). Src-inhibited treatment is not presented, since no GO term was significantly enriched in Src-inhibited morphotypic comparisons.
These findings suggest possible effects of mechanotransduction on morphotype-dependent gene expression via the integrin/Src kinase system as inhibition of this pathway abates features of morphotype-dependent gene expression, whereas overexpression of Src pronounced similar attributes of cell shape.