Distributed synthesis of sarcolemmal and sarcoplasmic reticulum membrane proteins in cardiac myocytes

It is widely assumed that synthesis of membrane proteins, particularly in the heart, follows the classical secretory pathway with mRNA translation occurring in perinuclear regions followed by protein trafficking to sites of deployment. However, this view is based on studies conducted in less-specialized cells, and has not been experimentally addressed in cardiac myocytes. Therefore, we undertook direct experimental investigation of protein synthesis in cardiac tissue and isolated myocytes using single-molecule visualization techniques and a novel proximity-ligated in situ hybridization approach for visualizing ribosome-associated mRNA molecules for a specific protein species, indicative of translation sites. We identify here, for the first time, that the molecular machinery for membrane protein synthesis occurs throughout the cardiac myocyte, and enables distributed synthesis of membrane proteins within sub-cellular niches where the synthesized protein functions using local mRNA pools trafficked, in part, by microtubules. We also observed cell-wide distribution of membrane protein mRNA in myocardial tissue from both non-failing and hypertrophied (failing) human hearts, demonstrating an evolutionarily conserved distributed mechanism from mouse to human. Our results identify previously unanticipated aspects of local control of cardiac myocyte biology and highlight local protein synthesis in cardiac myocytes as an important potential determinant of the heart’s biology in health and disease. Supplementary Information The online version contains supplementary material available at 10.1007/s00395-021-00895-3.

ribosomal RNA), we produced 27 double probes in total, and for Atp2a2 (encoding Serca2a; NM_009722.3) we produced 43 double probes in total. The mRNA probes were targeted to protein coding sequences. These probe sequences are provided in the enclosed spreadsheet.
Proximity Ligation: Anti-DIG and anti-DNP antibodies were diluted with 2% BSA in PBST at concentration of 1:100. Antibodies mix was added to cells and incubated overnight at 4C. Cells were washed 3 times per 5 minutes in PBST, then DuoLink kit was applied. Briefly: probes Plus and Minus were mixed with 2% BSA in PBST and incubated for 1h at 37C. Samples were washed 3 times per 5 minutes in Buffer A, ligase in ligation buffer was added, and cells were incubated for 30 minutes at 37C. After that cells were washed 2 times per 5 minutes in Buffer A, polymerase in Green buffer was added to cells and incubated for 3h at 37C. Then samples were washed 2 times for 5 minutes and mounted with DuoLink mounting media with DAPI.
Confocal Microscopy: Confocal microscopy was performed as previously described [3,6]. Samples were imaged using an A1R-HD laser scanning confocal microscope equipped with four solid-state lasers (405 nm, 488 nm, 560 nm, 640 nm, 30 mW each), a 60x/1.4 numerical aperture oil immersion objective, two GaAsP detectors, and two high sensitivity photomultiplier tube detectors (Nikon, Melville, NY). Where multicolor imaging was performed, individual fluorophores were imaged sequentially with the excitation wavelength switching at the end of each frame. Additionally, in a subset of cases, a differential interference contrast image was collected concurrently via a transmitted light detector.
Image Analysis: Images were analyzed using morphological object localization (MOL), a custom algorithm implemented in Matlab (Mathworks Inc, Natick, MA). Briefly, the cell body, nuclei, and signal puncta were identified using object-based segmentation and an exact Euclidean distance transform was applied to calculate distances from mRNA signals to the outer perimeter of the closest nucleus. Distances thus measured were plotted as cumulative distribution functions (CDFs). Additionally, signal abundance was assessed as the ratio of the integral of normalized voxel intensities to the volume of the relevant compartment (nucleus, cytosol, or whole cell).
Image Segmentation: The cell body mask was generated by threshold all channels with high sensitivity then combining the resulting binary images using a pointwise logical AND operation and finally selecting the largest connected component as the cell body. The nuclear mask was generated by thresholding the nuclear channel with low sensitivity and selecting all connected components with a volume greater than 0.1% the volume of the cell body as nuclei. A morphological closing with a 30-voxel radius disk was applied to the body and nuclear mask for edge smoothing. The mRNA masks were generated by thresholding their respective channels with high sensitivity and selecting connected components with volumes greater than 20 voxels to help exclude noise. Cell end sites (consistent with intercalated disks) were segmented using morphological filtering with a line structure element aligned perpendicular to the long axis of myocytes. Fluorescent Signal Localization: Fluorescent signals (RNAScope, MR-PLISH) were localized relative to cell nuclei using the distance transformation (DT) of the nuclear mask. The 3D exact Euclidean DT of the nuclear mask, an image where the value of each voxel is equal to that voxel's exact Euclidean distance from the nearest nuclear-positive voxel, was generated using a linear-time algorithm. 1 For a given cell, the set of mRNA-voxel distances was derived by extracting the subset of the nuclear DT which intersects the mRNA mask, thus outputting a distance measurement for each mRNA-positive voxel. Signal fraction as a function of distance from the nuclei was then plotted to produce the probability density function of an mRNA signal for an individual cell and cumulatively summed to produce the cumulative distribution function (CDF) for ease of interpretation and statistical analysis. To combine the CDF data from multiple cells, the cumulative signal fraction (CSF) was quantized from 0 to 1 in increments of 0.001 and distances measurements were linearly interpolated to create common CSF data points between all cells which were mean-averaged to create a single plot. In these plots, lines represent the mean CSF for a group of cells as a function of distance from the nuclei and the shaded region around a line represents the CSF's standard deviation at each nuclear distance.
In order to further probe the distribution of fluorescent signals within the cytosol, we examined normalized signal concentration (% voxels occupied) as a function of normalized distance from the nucleus. Briefly, CDFs ( Supplementary Fig. 4A) generated as described above were numerically differentiated to yield probability density functions (PDFs; Supplementary Fig. 4B) and the value of the PDF for each normalized distance value (0 being the edge of the nuclei and 1 being the cell periphery) was divided by the number of voxels available at that distance to yield normalized signal concentration ( Supplementary Fig. 4C). In one approach, we performed linear regression fitting of signal concentration vs. distance from nuclei between normalized distances of 0.2 and 0.8 ( Supplementary Fig. 4C -dashed lines) and evaluated the slope (Supplementary Fig. 4E). In a second approach, we numerically differentiated signal concentration as a function of normalized distance from the nuclei ( Supplementary Fig. 4D) and evaluated its mean value between normalized distance of 0.2 and 0.8 ( Supplementary Fig. 4F).
Other Plots: Fluorescent signal density is measured as the sum of all signal-positive voxel intensities in a given cellular compartment divided by the volume of that cellular compartment. Density measurements are normalized to the greatest density for a given plot. The signal histogram presents the voxel intensities (x-axis) of the segmented mRNA voxels normalized to the maximum voxel intensity of all gathered images. The signal fraction (y-axis) is presented as the mean ± the standard deviation.
Statistics: Pair-wise differences between localization CDFs was evaluated using the two-sample Kolmogorov-Smirnov test. Signal density is presented as the mean ± the standard deviation, and pair-wise differences were evaluated using the two-sample Wilcoxon rank sum test. An α value of 0.05 was used for all statistical tests. To test whether measured values different significantly from zero, the sign test was used with an α value of 0.05.

Mathematical Modeling:
We modeled the transport of mRNA in the cytoplasm by the following 1D advectiondiffusion partial differential equation: where is the concentration of mRNA i, D is the diffusion coefficient, a i is a uniform velocity and is the degradation rate. The equation is defined on the non-dimensional domain [0,1], for which the boundary = 0 represents the edge of the nucleus and = 1 represents the cell membrane. Accordingly, we have the following boundary conditions: where is the mRNA in-flux into the domain.
Parameters were fit to experimental control data in two steps: 1. We first fit the mRNA in-flux to the total amount of mRNA in the domain. We next fit , the scalar uniform velocity, based on the experimental data for the signal based on distance from the nucleus.
For the colchicine case, we started by fitting a new inward flux, as total mRNA signal amount differs slightly between experiments. We simulate the spatial steady state (reached at time ) based on the prior fit velocity . Then, for > , we set = 0 (simulating the effects of colchicine on microtubule-assisted transport) and simulate an additional 8 hours.
The diffusion coefficient D is equal for all mRNA species. It was obtained by fitting the equation with = 0 on the data for Gja1, which is least affected by the advection term (i.e. the fit is perfect with 1 = 0). We assumed that the half-life of mRNA in the cytoplasm is 1 day for all species, hence = ln 2 1 = 0.693/ .
The partial differential equation was discretized using the backward Euler implicit scheme, with the diffusion term discretized with a centered stencil and the advection term discretized with the upwind stencil. All fitting steps were performed using nonlinear least squares curve fitting.