Structure of the native myosin filament in the relaxed cardiac sarcomere

Myofibril preparation and vitrification

Demembranated left-ventricular mouse myofibrils were prepared as previously described²⁵. The myofibrils were collected by centrifugation at 3,000g for 2 min at 4 °C, followed by two washes with pre-relaxing buffer (100 mM TES pH 7.1, 70 mM KCl, 10 mM reduced gluthatione, 7 mM MgCl₂, 25 mM EGTA, 20 μM mavacamten, 5% dextran T500). To prepare for plunging, the pre-relaxing buffer was replaced with final relaxing buffer (pre-relaxing buffer plus 5.5 mM ATP). The relaxed myofibrils were then frozen onto Quantifoil Au R2/2 SiO₂ 200 mesh grids using a Vitrobot (Thermo Fisher Scientific). The myofibril suspension was incubated on the grid at 25 °C and 100% humidity for 30 s, blotted for 30 s from the opposite side of the carbon layer, and plunged into a liquid ethane–propane mixture.

Cryo-focused-ion-beam milling and cryo-ET

The preparation of lamellae for cryo-ET data acquisition was carried out by cryo-focused-ion-beam (cryo-FIB) milling using an Aquilos 2 cryo-FIB scanning electron microscopy system with a cryo-shield, according to previously described protocols^16,24,25 aiming for lamellae with a final thickness of 180 nm (range from 90 to 250 nm). The data acquisition was carried out with a Titan Krios transmission electron microscope (Thermo Fisher Scientific), fitted with a K3 detector and an energy filter (Gatan). The acquisition of overview images of myofibrils in the lamellae was carried out at a nominal magnification of ×6,700 to identify the C-zone regions.

The tilt series were acquired targeting the C zones at ×81,000 nominal magnification. The pixel size was calibrated to 1.146 Å using the 143.3 Å peak in the fast Fourier transform of the final thick filament reconstruction (from the M band to the C zone; Extended Data Fig. 4a,c). A dose-symmetric tilting scheme⁵² was applied during acquisition with a tilt range of −50° to 50° relative to the lamella plane at 2.5° increments. The sample was subjected to a total dose of 120 to 160 electrons per square ångström. Tilt series were acquired using a defocus between −3 and −6 µm. All images and a total of 89 tomograms were acquired using SerialEM⁵³.

Tomogram reconstruction and particle picking

Motion correction and contrast transfer function estimation were carried out in Warp⁵⁴; tilt series alignment was carried out in IMOD⁵⁵. Final tomogram reconstruction and subtomogram extraction were carried out in Warp. After binning the tomograms to a pixel size of 0.92 nm and low-pass filtering them at 60 Å, we used SPHIRE-crYOLO⁵⁶ to pick and trace both the thick and the thin filaments (Extended Data Fig. 1a,b). The average sarcomere length of the dataset was 2.326 µm (s.d. = 0.11 µm, N = 45), indicating that the sarcomeres are not hypercontracted. This is also supported by the 1.95:1 (102 thin filaments, 53 thick filaments) thin/thick filament ratio in the two tomograms we segmented.

Thin filament processing pipeline, model building and visualization

The traced thin filaments were resampled with an intersegment distance of 18 Å, leading to the extraction of 365,971 subtomograms with a box size of 293.5 Å (128 pixels, binning 2). Using customized scripts, each subtomogram was rotated to orient the thin filaments parallel to the X–Y plane using the previous angles from the tracing. Then, the central slab of 100 slices was projected and used as an input for 2D classification with ISAC^57,58,59. The classes that did not show a clear presence of thin filaments were discarded and the remaining segments were re-extracted as subtomograms and processed in RELION 3.1 (refs. ^60,61) via gold-standard dataset splitting. The initial helical reconstruction led to a 14.3-Å-resolution map (0.143 FSC criterion) with 27.4 Å rise and −167.2° twist (Extended Data Fig. 1). After removing the duplicated particles using a customized script, 100,447 subtomograms were further refined with two different masks that either covered the entire density of the thin filament, or included only the F-actin density. The full thin filament (F-actin and tropomyosin) was refined with helical reconstruction and reached a resolution of 8.2 Å while the refinement of F-actin alone resulted in an 8.3-Å-resolution map. For the thin filament map and the F-actin map, we applied a B-factor of −200 and −100, respectively. The two maps were aligned in ChimeraX⁶² and the individual chains from Protein Data Bank model 6KN7 (ref. ⁶³) were placed in the density with rigid body fitting. The model of the coiled coils of tropomyosin was improved with Namdinator, using automatic molecular dynamic flexible fitting⁶⁴. The final composite map (Extended Data Fig. 3a) was created by combining F-actin from the reconstruction of F-actin alone and tropomyosin from the full thin filament reconstruction using the color zone and splitbyzone functions in ChimeraX.

Thick filament processing pipeline

The traced thick filaments were resampled with an intersegment distance of 130 Å, leading to the extraction of 67,492 subtomograms with a box size of 1,280 Å (160 pixels, pixel size 8 Å). 2D classification was carried out similarly to the thin filament processing, using a central slab of 400 Å, resulting in 37,118 high-quality particles that were re-extracted as subtomograms. 3D classification with refinement and helical reconstruction (430 Å, 0° twist) resolved four classes that showed different orientation of crown 2. Class A showed ‘projected’ IHMs, class B had a mixture of conformations resulting in fuzzy density for crown 2 IHMs, and class C showed ‘retracted’ IHMs (Extended Data Fig. 1b). The refined coordinates were used to individually re-extract the three classes from Warp, using a box size of 144 pixels and a pixel size of 4 Å. The individual classes were refined in RELION 3.1 via gold-standard dataset splitting using a featureless cylinder as an initial reference and their coordinates were mapped back into the tomograms using ArtiaX⁶⁵. Classes B and C showed particles organized in filaments but randomly distributed away from the M line. Class A, which was later resolved as the segment from crown A8 to A12 (cMyBP-C stripe No. 2; Extended Data Fig. 1g), showed a unique distribution within the sarcomere, localizing as an array of particles parallel to the M line, roughly 200 nm away from it. We later understood that the crowns 2 contributing to the helical average of class A were crowns A5 and A8. As it turned out, whereas all crowns 2 have a β-angle of about +30°, crowns A5 and A8 have a β-angle of about −25°, explaining why the ‘projected’ class stood out during the refinement with imposed helical symmetry (Extended Data Fig. 7b,c). As we could identify the location of class A along the thick filaments, we wrote a customized script to calculate the ‘axially shifted coordinates’: starting from a known anchor point, we calculated the 3D coordinates of the next thick filament segments, shifting the position of class A coordinates 43 nm Z-wards and 43 nm M-wards (Extended Data Fig. 1f,h). This allowed us to resolve distinct reconstructions for each thick filament segment. With this strategy, we progressively resolved eight structures of the thick filament, spanning from the M band to titin C-type super-repeat 2, in the C zone (Extended Data Fig. 1). The resulting 3D maps showed high variability in resolution within different regions of the same map leading to oversharpening of the more flexible regions; we therefore used LocSpiral⁶⁶ to improve map interpretability. All reconstructions showed a three-fold rotational axis and were therefore refined with C₃ symmetry. The M-band reconstruction further revealed two orthogonal two-fold rotational symmetry axes that intersect at the three-fold axis at an angle of 60° and was later refined applying D₃ symmetry.

To build the final composite map, the power spectra of the reconstructions were normalized with relion_image_handler⁶¹ and a soft cylindrical mask of 15 px (about 60 Å) was applied to all maps. The filtered reconstructions were aligned in ChimeraX (fit in map) and the final map was created by merging the densities of the different segment, using the maximum value at each voxel (volume add, volume maximum in ChimeraX). To obtain a continuous and homogeneous density that we could use to trace the myosin tails, we took our 18-Å reconstruction of the last five stripes of cMyBP-C and extrapolated a 200-nm-long helix with relion_image_handler (430 Å, 0° twist)⁶¹.

Model building and visualization of the thick filament

The model of the thick filament was built using a combination of previously available models and AlphaFold2 predictions²⁸. For the model of myosin II, we started from the IHM of human β-cardiac heavy meromyosin (Protein Data Bank entry 5TBY; ref. ²⁷) while the tails were predicted in AlphaFold2 using the amino acid sequence of MYH7 from Mus musculus (5 segments of about 250 amino acids with about 20 amino acids overlap). Similarly, the C-terminal domain of cMyBP-C was predicted in AlphaFold2 using the last 590 amino acids of MYBPC3 from M. musculus. The region of titin from domain A101 to m3 (amino acids 24,760–32,350) was submitted for prediction as multiple entries (each ≈950 amino acids) with overlapping terminal domains. AlphaFold2 predictions of titin resulted in distinctive structural motives (for example, TK–m1, C-type super-repeat domains 1 to 3, C-type super-repeat domains 7 to 9 and cMyBP-C C8–C9) that unequivocally dictated the register and the position of titin domains. The model culminated in titin domains and the 9 cMyBP-C stripes with distances from the M line in agreement with previously published data from immunoelectron microscopy (for example, titin domains A77–78, A80–82, A153, A165 and cMYBP-C C7 domain)^15,18. The models were initially built in the map using rigid body fitting and their final organization was later adjusted with multiple rounds of molecular dynamic flexible fitting in Namdinator⁶⁴, starting with 40-Å low-pass-filtered densities and gradually using the higher-resolution maps. The models spanning from crown A18 to A28 were obtained by cloning the model spanning from crown A15 to A17. For the visualization in Fig. 2a, we used the ChimeraX functions colorbyzone, splitbyzone and Gaussian filter with standard deviation = 3. The depictions in Fig. 5c,e,f were obtained using the Chimera unroll function on the structure of all components individually. ChimeraX lighting was set as follows: soft intensity 0.1; direction 0.577, −0.577, −0.577; color 100, 100, 100; fillIntensity 0.5; fillDirection −0.81, −1,1; fillColor 100,100,100; ambientIntensity 1.4; ambientColor 100,100,100; shadow 1; qualityOfShadows finer; depthBias 0.01; multiShadow 64; msMapSize 2000; msDepthBias 0.004; moveWithCamera 1; depthCue 0. ChimeraX cartoon style was set as follows: width 2.5; thick 1; xsection oval strand width 3.2; xsection rect coil width 3.2; thickness 1.2. ChimeraX colour palette was as follows: thick filament core #707ec1; crown 1 #7ed5dc; crown 2 #7ea6dc; crown 3 #7edcb4; essential light chain #e154d8; regulatory light chain #9a41de; myosin blocked head #55667e; myosin free head #ace2ff; titin-α #d95d87; titin-β #dc7ea6; TK #844b63; m1-9 #eab1c9; M-band proteins-A #546845; M-band proteins-B #a8d18a; cMyBP-C #dec98f; F-actin #a8d18a; tropomyosin A #d1a8a8; tropomyosin B #d1a8bd; troponin #a8a8d1.

Position and orientation of myosin crowns

To quantitatively describe the 3D arrangement of each crown, the thick filament was modelled as a cylinder, and the myosin IHMs were represented as triangles. The vertices of the triangles were determined by the coordinates of three points: the ATP-binding site in the free head, the same site in the blocked head, and the head–tail junction site. For each IHM, the Euler angles are calculated in a 3D Euclidean space with the origin on the centroid of the triangle, the x axis parallel tangential to the cylinder, the y axis parallel to the radius, and the z axis parallel to the axis of the cylinder. The coordinate system was calculated for each IHM to obtain the Euler angles (α, β and γ) specific to each crown (Extended Data Fig. 7a). To obtain azimuthal angle, radius and z-axis height, we used the centroid of each IHM, calculated the cylindrical coordinate and inferred twist, radial distance and rise, respectively (Extended Data Fig. 7d).

Sinusoidal compression percentage

To quantify the curviness of the myosin tails, for each tail, we first obtained the atomic coordinates of the α-carbons for the two amino acid chains in the coiled coils. We then traced a new 3D curve running through the central points between each α-carbon couple. For each curve segment, we calculated the sinuosity S by the ratio of the length of the curve C to the Euclidean distance between the ends L:

The sinusoidal compression percentage (SCP) is then given by:

Tomogram segmentation and cMyBP-C links

To describe the 3D organization of the sarcomere components, we selected two representative tomograms (Figs. 1a and 4a and Supplementary Videos 1 and 2) and denoised them using cryo-CARE⁶⁷. With a customized script, we mapped back each subtomogram using a binary mask of their corresponding structure, matching the coordinates and orientations obtained from the 3D refinement. The resulting binary MRC files were imported in Dragonfly⁶⁸ and used as a template for pseudo-segmentation of the tomograms. The resulting label layers were manually validated by inspecting each tomographic slice for unassigned densities and further tracing the flexible components that were averaged out during the refinement (that is, the cMyBP-C links from thick to thin filament). After clearly identifying and segmenting 76 cMyBP-C links in our tomograms, we measured the angle that the link formed relative to the thick filament z axis, using the position of the C7 domain as a pivot point. The angular distribution was plotted in GraphPad Prism.

Antigen expression and purification

A fragment of the TK domain encompassing human TTN transcript variant-IC (NM_001267550.1), residues 33812–34076, was expressed in Escherichia coli BL21 [DE3] cells in fusion with an N-terminal His₆ tag. The insoluble fragment was extracted from inclusion bodies with 8 M urea, 50 mM potassium phosphate pH 8.0, 0.5% Tween 20 (buffer B) by sonication with a Branson sonifier microtip on ice. Insoluble material was pelleted at 15,000 r.p.m. for 20 min in an SA 600 rotor (Sorvall) and the soluble supernatant was applied to an Ni-NTA column equilibrated in buffer B. After washing the column as above in buffer B, bound protein was eluted with 250 mM imidazole in buffer B and equilibrated stepwise against 6, 4, 2 and 0 M urea in 40 mM HEPES buffer pH 7, 50 mM NaCl, 4 mM dithiothreitol and 0.1% Tween 20 (buffer C). The insoluble precipitate was spun down and the soluble protein was further purified by gel filtration purification on a Pharmacia Superose 12 column equilibrated in buffer C. The purified kinase fragment was used for commercial rabbit immunization, and serum was collected after three booster injections.

Cloning, expression and purification of rat titin A170-kinase

For affinity purification, a soluble TK construct, A170-kinase, was used. The sequence encompassing the A170 (FN3) and kinase domains of rat titin (XM_008775521.1 residues 31897–32344) was cloned into a modified pCDFDuet vector containing an N-terminal His-tag, expressed in E. coli strain BL21 [DE3] using standard protocols and purified by nickel affinity and size-exclusion chromatography according to ref. ⁶⁹.

Antigen coupling and affinity purification of anti-TK antibody

A 1 mg quantity of purified A170-kinase was dialysed into coupling buffer (100 mM sodium phosphate pH 8, 250 mM NaCl, 1 mM dithiothreitol), and then coupled to 2 ml NHS-activated Sepharose 4 Fast Flow slurry following the manufacturer’s instructions (Cytiva Life Sciences). Antibody affinity purification was carried out using standard procedures described previously⁷⁰. Following equilibration with 10 ml PBS with 0.05% Tween 20, 5 ml of the rabbit anti-kinase serum was applied to the A170-kinase–Sepharose column, which was then washed with 20 ml PBS containing 0.05% Tween 20, 4 ml PBS and finally 4 ml 50 mM NaH₂PO₄ pH 7.4, 500 mM NaCl to remove nonspecifically bound proteins. Bound antibodies were then eluted with fractions of 0.5 ml 0.1 M glycine HCl pH 3 into 1 ml 1 M Tris HCl pH 9, with those containing protein pooled, dialysed into PBS containing 5 mM NaN₃, concentrated to about 0.24 mg ml⁻¹, flash-frozen in 50-µl aliquots and stored at −80 °C. Specific reactivity of the purified immunoglobulins was confirmed by western blotting against various titin fragments containing the kinase as well as control fragments.

Super-resolution microscopy

Immunofluorescence labelling was carried out on mouse and rabbit psoas myofibrils as previously described¹⁵ using the affinity-purified TK antibody at 1 µg ml⁻¹ and Atto647N-labelled anti-rabbit IgG secondary antibody for visualization. Stimulated emission depletion microscopy was carried out on a STEDYCON (Abberior) attached to a Leica TCS SP5 ll confocal microscope. Images were recorded at a pixel size of 15 nm.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.