Correlative super-resolution analysis of cardiac calcium sparks and their molecular origins in health and disease

2.1. Pipeline for correlating RyR2 underlying peripheral Ca²⁺ sparks

Ca²⁺ sparks are commonly recorded as x-t line-scan confocal kymographs. However, the two-dimensional spatial patterns of spontaneous Ca²⁺ sparks in the sub-plasmalemmal cytoplasmic space of rat right ventricular myocytes were observable with TIRF microscopy. In TIRF image frames, sparks appear as a transient (approx. 30–100 ms) and localized event of fluorescence with a full-width-at-half-maximum (FWHM) of 1–6 µm (figure 1a). The current benchmark for mapping RyR2 channels in cardiomyocytes is DNA-PAINT. With approximately 10 nm of resolution, DNA-PAINT represents a significant (approx. 25-fold) improvement in optical resolution over standard diffraction-limited methods such as TIRF (figure 1b). With DNA-PAINT, it is now possible to resolve individual RyR2s, observed as punctate labelling densities, that make up each RyR2 cluster (figure 1c). In this study, we sought to selectively map the two-dimensional spatial properties of sub-plasmalemmal sparks against the local, two-dimensional organization of RyR2 located within peripheral couplons (schematically illustrated figure 1d and asterisked region) in healthy and failing RV myocytes using TIRF microscopy.

• Download figure
• Open in new tab
• Download PowerPoint

To this end, we developed a novel experimental pipeline detailed recently [39] (figure 1e). In this protocol, RV cardiomyocytes were isolated freshly from rat hearts and transferred to a glass dish with a gridded bottom, allowing the cells to adhere to the grid. Cytoplasmic fluorescent [Ca²⁺]_i indicator Fluo-4 AM was loaded into cells and de-esterified before cells were placed within the TIRF field. Sub-plasmalemmal Ca²⁺ sparks were recorded from cells at specific grid coordinates onto two-dimensional time series. Cells were then fixed in situ and labelled with anti-RyR2 antibodies towards DNA-PAINT imaging and then returned to the TIRF microscope, tracked based on their grid coordinates and DNA-PAINT series recorded. Post-hoc analysis included rescaling and a semi-automated alignment of a time-averaged facsimile of the Ca²⁺ time series against the DNA-PAINT image, primarily using the unique outline of the myocyte to overlay the Ca²⁺ sparks with the underlying RyR2 map (figure 1f,g). The coordinates of both the Ca²⁺ sparks and the individual RyR2 puncta were detected and subsequently registered against each other (using the scaling and alignment vectors established for the whole image). This allowed the RyR2 channels and clusters within the ‘spark footprint’ (determined by a circular window centred around the spark's centroid and diameter equal to the spark FWHM) to be counted and sampled on a spark-by-spark basis (figure 1h). The percentage histograms of the total number of RyR2 puncta (figure 1i; main panel) and the number of unique RyR2 clusters detected with this method in healthy RV myocytes (i.e. clusters with ≥4 RyR2 puncta; inset) within the footprint of each spark demonstrate that spontaneous Ca²⁺ sparks can arise from broadly varying ensembles of RyR2 channels (mean of apporx. 18.1 RyR2 per spark) and unique clusters (mean approx. 5.2 clusters per spark; see electronic supplementary material, figure S1A,C).

2.2. Correlation of RyR2 organization with Ca²⁺ sparks in RV failure

To study the structure–function relation of Ca²⁺ sparks in the failing RV, we examined myocytes acutely isolated from the RVs of rats administered with monocrotaline (MCT-RV) and age and sex-matched controls (Ctrl-RV). In the correlative, spark-by-spark analyses in MCT-RV, we observed a∼30% increase in the total number of RyR2 (mean of approx. 23.5 RyR2 per spark) and a approximately 60% increase in the number of segmented clusters (mean of∼8.1 clusters per spark; see electronic supplementary material, figure S1A–D) compared to Ctrl-RV. This was consistent with the likely recruitment of a greater number of RyR2 in the genesis of a spontaneous Ca²⁺ spark. Spontaneous sparks recorded using TIRF in MCT-RV cells also featured, on average, approximately 38% greater ‘spark mass’ (i.e. the integral of the fluorescence intensity of the spark; see electronic supplementary material, methods section for details) compared to Ctrl-RV (means approx. 38.4 compared to approx. 27.7, respectively). While the FWHM was unchanged between MCT-RV and Ctrl-RV, the spark frequency was >90% higher in MCT-RV (see plots and statistics in electronic supplementary material, figure S1E–G).

Close examination of DNA-PAINT images of sub-plasmalemmal RyR2 from correlative experiments showed a noticeably more dissipated morphology in MCT-RV cells compared to Ctrl-RV (figure 2a,b; see arrowheads in panel b). With the use of correlative sparks recorded over approximately 20 s, we generated ‘heat maps’ of the local nanoscale RyR2 morphology (represented by pseudocoloured binary masks of the DNA-PAINT RyR2 image; see electronic supplementary material, figure S2 for details) to indicate the average estimated spark mass of the locally recorded spontaneous Ca²⁺ sparks in Ctrl-RV and MCT-RV cells (figure 2c,d; insets show the greyscale DNA-PAINT images of the shown regions of interest). In the healthy control, the colour-gradations in the spark mass maps were relatively evenly distributed across clusters of similar size while smaller, fragmented clusters were sparse. In the failing cells, we observed a large majority of the fragmented (smaller and dispersed) RyR2 clusters in darker colours, suggesting that they alone are unlikely to produce sparks with greater spark mass.

• Download figure
• Open in new tab
• Download PowerPoint

The scattergrams between the spark mass of each Ca²⁺ spark recorded and the total number of RyR2 counted locally revealed a broader degree of scatter in MCT-RV compared to Ctrl-RV (figure 2e,f). In Ctrl-RV, sparks with spark mass <1.0 correlated with almost exclusively with regions consisting of no more than 25 RyR2 puncta. By contrast, sparks with spark mass <1.0 were correlated with a broader range of RyR2 counts (extending up to approx. 50 RyR2 puncta per spark; see the black arrows in figure 2f). This heterogeneity in MCT-RV is further illustrated in the frequency histograms of the total RyR2 count and the number of unique RyR2 clusters underneath each spark (figure 2g,h), represented by the wider distributions of MCT-RV compared to Ctrl-RV. Further, the direct ratio between the recorded spark mass and the locally counted RyR2 showed a lower mean and median for MCT-RV (mean ± s.d. 1.8 ± 6.7; median 0.53) compared to Ctrl-RV (mean ± s.d. 2.1 ± 3.1; median 0.91). The increased heterogeneity in MCT-RV was represented by an∼2.2-fold higher standard deviation compared to the control (p = 0.014, B = 6.099, d.f. = 1220 sparks; Bartlett's test; see violin plot in figure 2i).

2.3. Relationship between non-random, sub-plasmalemmal Ca²⁺ spark pattern and multi-scale RyR2 organization

A novel visualization unlocked by the correlative approach is the direct overlay of the centroids of all spontaneous Ca²⁺ sparks recorded using TIRF over a window of time onto the two-dimensional super-resolution image of the intricate rows of sub-plasmalemmal RyR2 clusters (figure 3a). We investigated the non-random distribution of the spark centroids visually observed across the field of view by performing a two-dimensional quadtree segmentation of domains with higher density of aggregation of spark centroids over a 15 s time window (see electronic supplementary material, §S1.9 and figure S3). This segmentation revealed two-dimensional, sub-plasmalemmal regions, typically a few hundred nanometres in width, throughout a cell's footprint which we classified as ‘recurring spark sites'. Figure 3b,c illustrate the overlays between these recurring spark sites and the super-resolution RyR2 images in Ctrl-RV and MCT-RV respectively (magnified views shown in figure 3d,e). Strikingly, we observe that the recurring spark sites rarely overlap with larger RyR2 clusters directly. Instead, they appear to extend between adjacent clusters in both Ctrl-RV and MCT-RV. There was also no observable, consistent co-location of the fragmented RyR2 with the recurring spark sites in MCT-RV. The occupancy of the recurring spark sites between RyR2 clusters, often within hundreds of nanometres, resembled the shared ‘RyR2 super cluster domains’ proposed previously as a structural correlate of clusters that are likely to co-activate [40,41]. It was also conceivable that they aligned with the putative super-clusters of the Ca²⁺ release trigger, L-type Ca²⁺ channels with Ca_v1.2 observed recently [42].

• Download figure
• Open in new tab
• Download PowerPoint

Both super-resolution and electron tomography image data have confirmed that clustering patterns of RyR2 channels in healthy cardiomyocytes, albeit not always crystalline, follow a distinct nearest-neighbour distance and the average spacings relative to the surrounding RyR2 [24,30,31]. We sought to investigate whether the spark locations are unique to regions with RyR2 clustering. To this end, we examined the above variables across experimental datasets (Ctrl-RV and MCT-RV) as well as synthetic images based on control cells where either RyR2 or Ca²⁺ sparks were positioned according to random uniform distributions (see electronic supplementary material §S1.10.1 and figure S4). A modest rightward shift and a concurrent broadening in the percentage histograms of the total RyR2 count (figure 3f) within each Ca²⁺ spark footprint was observed in both randomized datasets. This reinforces the gain in Ca²⁺ release that is achieved by the clustering of a relatively small number of RyR2 compared to randomly and sparsely organized RyR2 (see synthetic image in electronic supplementary material, figure S4). By contrast, considerable rightward shifts are observed in both percentage histograms of the nearest neighbour distance of each RyR2 located under each Ca²⁺ spark, and the average neighbour distance (i.e. average spacing of a given RyR2 punctum relative to the surrounding RyRs) in clusters detected underneath the spark footprint. This suggests the closeness and uniformity of the intra-cluster organization of RyR2 are intrinsic features critical to producing a spark (figure 3g,h). Similar to a previous report based on ExM [31], a modest shift is observed in the rightward tail in the average RyR2 neighbour distance in MCT-RV (red in figure 3h) which may reflect that clusters undergo loosening or ‘fraying’ of the RyR2 positions prior to the fragmentation during RV failure [32]. Generally, we note the similarity in each of these distributions between Ctrl-RV and MCT-RV datasets in comparison to the randomized datasets. This supports the RyR2 remaining within unfragmented clusters as the functionally dominant sub-population of channels driving the genesis of Ca²⁺ spark during disease.

3.1. A novel approach to correlate the molecular-scale RyR2 cluster architecture with local Ca²⁺ sparks

Presented in this paper is an experimental interrogation of the spatial relationship between properties of the elementary events of Ca²⁺ release in primary adult cardiomyocytes and the intrinsic molecular-scale organization of RyR2. Hiess et al. have previously achieved a similar correlation, albeit with TIRF microscopy where GFP-RyR2 clusters were unresolved [43], hence a statistical analysis of the single-channel level RyR2 organization was not possible. Our approach was enabled by the development of a correlative imaging protocol [39] that overcomes the broad unavailability of animal models with endogenous fluorescent reporters of either RyR2 and/or [Ca²⁺]_i that can simultaneously model the cellular and molecular basis of cardiomyopathy. To our knowledge, only DNA-PAINT [30] and ExM [32] currently offer optical resolution sufficient to localize individual RyR2 which motivated our choice for correlative DNA-PAINT. The ability to preferentially sample the nanoscale patterns of RyR2 organization based on a functional correlate has unlocked a powerful interrogation of the nano/micron-scale structure/function relationship.

Several key conditions and limitations in our approach are however noteworthy. The evanescent TIRF field of illumination offered us assurance that only 100–200 nm of fluorescence was being axially integrated in both the DNA-PAINT (RyR2) images and Ca²⁺ spark series. Examination of previous dSTORM [41] and 3D ExM data [32] of rat ventricular muscle that capture RyR2 clusters in both sub-plasmalemmal and interior regions offer reassurance that less than 5% of the sub-plasmalemmal clusters extend beyond this depth. While the Ca²⁺ sequences were also recorded within an evanescent field, the diffusion of both the spark Ca²⁺ and the Ca²⁺ indicator dye over micron-scale distances means that sparks originating deeper in the cell interior may in principle be visible within the TIRF field. This may limit the accuracy of the spatial correlation of the sparks and local RyR2 counts; however, our protocol to exclude sparks larger FWHM and R²-value (a measure of fitting accuracy and signal-to-noise ratio; SNR) from the correlative analysis allowed us to minimize the impact of out-of-plane sparks.

Deviations in the assumed flatness of these peripheral RyR2 clusters, asymmetries and heterogeneities in the two-dimensional spread of Ca²⁺ in the recorded sparks and any acute remodelling RyR2 clusters within the window of spark imaging (inside 3–4 h from the point of isolation of cells), albeit minor, are possible sources of error. While Hiess et al. have previously observed with confocal microscopy the movement of some peripheral RyR2 clusters in quiescent cells maintained under high [Ca²⁺]_o, this fraction was reported to be small [43]. Although they could reverse this behaviour entirely with exposure to tetracaine, this strategy was incompatible with our objective of observing intrinsically spontaneous sparks. Factors that may further limit the precision of the spark-by-spark sampling of local RyR2 patterns include incomplete RyR2 labelling with the anti-RyR2 antibody chosen (used widely by many groups, and labelling efficiency previously [26] estimated to be at least approx. 90%), and the alignment error between the averaged (diffraction-limited) Ca²⁺ image and the super-resolution RyR2 image.

The key determinants of the latter are the resolution and the SNR of the averaged Ca²⁺ image. electronic supplementary material §S1.10.2 and figure S5 outline a series of simulations by which we demonstrate an alignment error under 100 nm for each iteration given the typical TIRF resolution of approximately 250 nm and SNR of approximately 17.0 in our averaged Ca²⁺ image data. For context, an alignment error of 100 nm represents the omission or addition of approximately 2.5 RyR2 channels in a tightly organized cluster in Ctrl-RV (mean nearest neighbour distance of approx. 38.23 nm). Finally, the correlative approach also prevented us from calibrating the pixel values in the Ca²⁺ images against [Ca²⁺]_i with a Ca²⁺ ionophore due to the requirement of preserving the cellular structure for DNA-PAINT. As a standardization, all experiments were performed under identical excitation intensities and camera settings which allowed use to quantitatively compare the intensity information Ca²⁺ images between cells and samples.

3.2. Spatial encoding of sparks in the RyR2 organization

The non-random dependence of the spark locations on the close RyR2 clustering particularly reinforces numerous simulations demonstrating how close (and filled) RyR2 lattice structures promote the ignition of a spark [31,36,38]. However, similar to previous DNA-PAINT [30] and ExM [31] data, our correlative datasets feature a large majority of clusters underneath the spark footprint with looser RyR2 arrangement (typical neighbour distances >38 nm) than the fully filled lattices. As confirmed by a recent computational study, this feature is likely to contribute to the considerable variability that we see in the integrated signal of Ca²⁺ sparks [44] (e.g. Figure 2e,f) and certainly the broad range of ‘spark mass/RyR2s’ ratios that we observe even in the control cells.

The super-positioning of Ca²⁺ spark centroids and the RyR2 maps (e.g. figure 3a) provided us with an unprecedented view into the spatial encoding of the spark locations across the sarcomerically organized Ca²⁺ handling machinery. The localizations of the recurring spark sites between larger RyR2 clusters are, to our knowledge, the only experimental imaging data supporting the putative ‘triggered saltatory’ recruitment of neighbouring RyR2 clusters at the genesis of a Ca²⁺ spark [40] in both healthy and failing myocytes. While these recurring spark sites (figure 3b–e) may be accentuated by the possible constructive interference of concurrent sparks or waves, constraining this visualization to sparks with FWHM ≤ 6.0 µm, we have minimized the impact of such spurious localizations.

The scattergrams between spark mass and local RyR2 counts (figure 2e,f) demonstrate a broad correlation in both failing myocytes and the healthy controls. When the two-dimensional positions of RyR2 and sparks were randomized separately, the distribution of the total RyR2 count within the spark footprint was not entirely annulled (figure 3f). In fact, it became broader, suggesting that the local RyR2 ensemble is only a rough determinant of the Ca²⁺ spark. The high degree of scatter in both spark mass scattergrams of Ctrl-RV and MCT-RV (but exacerbated in sparks with higher spark mass in MCT-TV) further reflects how numerically similar pools of RyR2 can give rise to Ca²⁺ sparks whose integrated [Ca²⁺]_i can vary by 2–3 orders of magnitude.

3.3. Heterogeneities in the structure/function relation in RV failure

Increasing structural heterogeneity, including local t-tubule remodelling [45–47], local regions of RyR2 fragmentation [17,38,48] and diminishing regularity of the expression patterns Ca²⁺ handling proteins over nanometre/micron length-scales [49], are now understood to be hallmarks of wide-ranging cardiac pathologies. In MCT-RV, we have previously observed evidence of heterogeneous RyR2 cluster fragmentation [31] and the receding sub-domains of RyR2-modulators and molecular tethers, junctophilin-2 (JPH2) [32] and BIN-1 from some clusters [18]. In this context, the increasing heterogeneity in the ratio between spark mass and RyR2 count in MCT-RV (figure 2i) is the likely result of differently configured local Ca²⁺ handling machineries, not limited to the spatial organization of RyR2. Downregulation of the cytoplasmic Ca²⁺ removal mechanisms (the Na⁺/Ca²⁺ exchanger and SR Ca²⁺ ATPase) [18,50] along with the overload of SR luminal [Ca²⁺] need to be considered with the 30% increase in the RyR2 ensemble observed underneath Ca²⁺ sparks. While the individual cluster size of RyR2 is reduced in MCT-RV as a consequence of the cluster fragmentation or fraying [31,32], the modest increases in the RyR2-RyR2 neighbour distance measurements offer some initial clues of the evolving functional coupling in and around the dyads. An enhancement of the functional coupling of the loosely packed RyR2 is certainly conceivable with the reduced L-type Ca²⁺ channel expression [50] and higher SR [Ca²⁺]_i in MCT-RV myocytes, offering greater driving force during Ca²⁺ release. The super-resolution heatmaps encoding spark mass do not however reveal a clear correlation between fragmented RyR2 and regions with higher average spark mass. This observation may however be different if Ca²⁺ leak could be mapped with a more sensitive approach than diffraction-limited TIRF imaging with Fluo-4 Ca²⁺ indicator.

4.1. Microscope setup

All experiments were performed on a Nikon TE2000 (Nikon; Japan) modified to enable TIRF dual-colour imaging with 488 nm and 671 nm excitations on an approximately 15 µm × 15 µm illumination field (as detailed previously [39]). The full list of specifications and settings can be found in the electronic supplementary material, methods section. Emitted light was recorded onto a Zyla 5.5 scientific CMOS camera (sCMOS; Andor, Belfast). Raw image series for both Ca²⁺ imaging and DNA-PAINT were acquired using the open-sources Python Microscopy Environment [51] (PyME) software.

4.2. Animal models and isolation of ventricular myocytes

Experiments were performed according to the UK Animals (Scientific Procedures) Act of 1986 and with UK Home Office approval (license number 70/8399) and local ethical approval. As a model of right ventricular (RV) failure, adult male Wistar Crl rats aged approximately 5 weeks were given an intraperitoneal injection of crotaline (Merck, NJ) to induce pulmonary arterial hypertension. Age and sex-matched controls were given an equivalent bolus of saline. At 3–4 weeks, the MCT treated animals were monitored for signs of RV failure were euthanized along with age-matched Ctrl animals. Hearts were dissected acutely and right ventricular myocytes enzymatically isolated following Langendorff perfusion. For a detailed account of the animal model and cell isolation, see the electronic supplementary material, methods section.

4.3. Ca²⁺ sparks imaging

RV myocytes were loaded with Fluo-4 AM fluorescent Ca²⁺ indicator (ThermoFisher Scientific) and immersed in a Tyrode's solution containing 5 mM CaCl₂ at pH 7.4 and adhered to a gridded imaging dish with a #1.5H glass coverslip bottom (Ibidi, USA) for 90 mins before being immersed in fresh Tyrode's solution for Ca²⁺ spark imaging. The dishes were clamped securely onto the stage of the TIRF microscope system such that the grid was aligned with the straight edges of the camera's field of view, under brightfield illumination. Cells forming a substantial footprint were illuminated with a 488 nm laser. The local changes in the Fluo-4 fluorescence were recorded at a frame rate of 10 Hz with either Tetraspeck microspheres (ThermoFisher Scientific) attached to the imaging grid or the outline of the cell's contact patch with the grid included within the imaging frame. See detailed protocol in the electronic supplementary material, methods section.

4.4. Cell fixation and immunolabelling for DNA-PAINT imaging of RyR2

Immediately following Ca²⁺ imaging, cells were fixed in situ with 2% paraformaldehyde (Sigma-Aldrich; w/v in phosphate buffered saline; PBS) for 10 mins at room temperature (RT). Samples were washed and subjected to immunofluorescence labelling with a monoclonal mouse anti-RyR2 IgG (MA3–916; ThermoFisher) primary antibody and an anti-mouse IgG secondary antibody (Jackson Immunoresearch) conjugated to a DNA-PAINT P1 ‘docking’ strand (as designed by Jungmann et al. [28]) was applied, diluted at 1 : 100 in incubation solution. See details of the labelling protocol and the production of the secondary antibody and the base sequences of the P1 ‘docking’ strands in the electronic supplementary material, methods section.

4.5. DNA-PAINT imaging and primary processing

Following immunolabelling, samples were washed 3 times in ‘Buffer C’ formulated by Jungmann et al. [28], and the P1 ‘imager’ strands linked to Atto655 were applied at 1.2 nM in Buffer C. The dish was returned to the microscope stage and the grid coordinates recorded from the Ca²⁺ imaging were used to return the field of view to the corresponding cells imaged previously. For exciting the Atto655 on the imager strands marking the RyR2 targets at the very edge of the cell, the 671 nm laser was focused at a supra-critical angle onto the field of view to achieve evanescent TIRF illumination. Time series of the DNA-PAINT events consisting of 20 000–50 000 frames were acquired at 100 ms/frame integration time. The analysis included the detection of single molecule events and least-squares fitting of a two-dimensional Gaussian to localize their sub-pixel scale centroid. The event positions were then rendered into a 16-bit greyscale TIFF image with a pixel scaling of 5 nm per pixel using an algorithm based on Delaunay triangularization [52].

4.6. Correlative and quantitative image analysis

The full image analysis protocol consisting of the registration of the Ca²⁺ spark data to the DNA-PAINT data and the localization of Ca²⁺ sparks has been published elsewhere [39].