The molecular basis for sarcomere organization in vertebrate skeletal muscle

1Skeletal muscle is an essential tissue required for efficient movement in vertebrates. In humans, dysfunctional muscle function leads to a broad variety of physiological disorders, ranging from common muscle cramps to severe myopathies. In muscle, specialized cells align and assemble during development into a fundamental unit known as the muscle fiber. Individual muscle fibers can be very large syncytia of up to several centimeters in length and are comprised of the force-generating and load-bearing devices of muscles called sarcomeres.

2Each sarcomere can be axially divided into different zones based on its ultrastructure: the I (isotropic)-band where only actin (thin) filaments are present, the A (anisotropic)-band that contains overlapping actin and myosin (thick) filaments, the Z-disc that borders the sarcomere at its ends, anchoring the barbed (+) ends of actin filaments, and the M-band in the center, where only myosin filaments are present and cross-linked. The Z-disc is formed by the ends of antiparallel thin filaments from adjacent sarcomeres. These filaments are laterally cross-linked by α-actinin, forming an intricate network with other proteins such as titin, nebulin, and myotilin. In the M-band, myosin filaments are anchored in a hexagonal array by a complex of the proteins myomesin, titin, and obscurin/obsl1. The network formed by these proteins and myosin filaments provide mechanical stability for the sarcomere by absorbing misbalanced longitudinal force produced by myosin heads. Deformation of the M-band by shear forces could allow it to act as a strain sensor and thus play an important role in signaling.

3The myosin filaments are stably linked to actin filaments by a third “connecting” filament, consisting of the gigantic titin protein. Titin’s N terminus is bound to α-actinin, telethonin, and possibly CapZ at the Z-disc. Its C terminus interacts with myomesin and obscurin at the M-band.

4The basis of the mechanism of skeletal muscle action is the cyclical interaction of actin and myosin, converting chemical into kinetic energy, which leads to relative movement between thin and thick filaments and is the primary driver of force generation within the sarcomere.

5Images of fixed, stained, and sectioned sarcomeres acquired using electron microscopy revealed a hexagonal arrangement of thick and thin filaments and cross-bridges between them formed by myosin head domains in the A-band. The composition, regulation, and function of arthropod and vertebrate muscles differ in key aspects. In vertebrate skeletal muscle, one thin filament interacts with three neighboring thick filaments instead of two, leading to a more complicated model of cross-bridge formation. In insect flight muscle sarcomeres, the three-dimensional structure of the cross-bridges were further analyzed using electron tomography of plastic-embedded samples and freeze-substituted samples, showing the distribution and overall shapes of cross-bridges in different states. However, these studies were limited in resolution and did not allow an understanding of the molecular detail of cross-bridges in the A-band. Similarly, a three-dimensional tomographic study of the vertebrate sarcomere showing the variable lengths of thin filaments in different vertebrate muscle types was also limited due to the applied method. Therefore, the molecular details underpinning the organization of the vertebrate skeletal muscle sarcomere remain elusive.

6Z-discs are complex structures despite their fundamentally “simple” function as stable yet flexible crosslinkers of antiparallel actin filaments. The core unit is the actin-α-actinin link, crosslinking actin filaments at roughly 90° turns in variable layers of between 2 to 7. Images of the transverse section of Z-discs of chemically fixed mouse soleus and cardiac muscle showed two types of appearance in response to, and absence of, stimulation: a relaxed “small square” form and an active “basket-weave” form. Both states exhibit a square arrangement of thin filaments. Due to the lack of a 3D visualization of the I-band thin filaments, it remains obscure as to how the thin filaments progress from a hexagonal pattern in the A-band to a square pattern in the Z-disc.

7The actin-α-actinin crosslinks are known to be conformationally flexible and can be affected by the activation state of the thick filament as well as temperature and ionic strength. The conformational plasticity of the Z-disc is believed not only to provide resistance to mechanical forces during contraction and relaxation but also to translate mechanical stress into biochemical signals via a number of signaling pathways involving transiently associated Z-disc proteins.

8The recently presented structure from resin-embedded sections of the highly atypical Z-disc of the midshipman sonic muscle, structures that are at least 10 times thicker than the Z-discs of representative vertebrate skeletal muscles, revealed a fundamental arrangement of transverse crosslinks with an axial displacement of 19.2 nm, in line with previous work at lower resolution. A further study of the Z-disc from isolated branches of cardiac myofibrils using electron cryo-tomography (cryo-ET) determined averaged structures of actin-α-actinin complexes in two states (putatively, the relaxed and activated states) at ~23 Å resolution. These two structures highlight changes within the Z-disc in response to torque applied by the myosin heads on thin filaments, whereby α-actinin can pivot via linker domains between the central “rod” domain and the N-terminal actin binding domain. Nevertheless, any inherent strain within a given sarcomere was not ascertained within these data and, as the structures are averaged from many individual sub-volumes, any heterogeneity of α-actinin arrangement within the same Z-disc or across different Z-discs remains undiscovered. The current molecular understanding of actin-myosin interactions and the function and mechanism of other essential sarcomeric proteins, such as troponin, titin, and α-actinin has been derived from atomic structures of in vitro reconstituted components by X-ray crystallography, single particle cryo-EM or ensemble information, obtained from X-ray diffraction of muscle fibers. Although these structures serve as a foundation for determining interactions among different sarcomeric components, high-resolution information on their biological arrangement in the context of a sarcomere is lacking. As such, an in situ approach that allows a molecular and structural analysis of a functional sarcomere is necessary for understanding the native interactions and localization of sarcomeric proteins.

9The major obstacles in obtaining molecular structural details of the sarcomere using conventional tomographic methods lie in the artifacts arising from fixation and staining methods, dehydration of the sample, the diamond blade during sectioning, or from the shrinkage and deformation of plastic sections during imaging. All these factors impede the determination of high-resolution features of macromolecules within the sample. In the past 15 years, investigations of the structure of muscle fibers with conventional electron tomography have provided important insights into the sarcomeric actomyosin cross-bridge architecture. However, chemical fixation, freeze-substitution and EM hardware limited the obtainable resolution.

10To circumvent these artifacts, cryo-focused ion beam-milling (cryo-FIB) of vitrified samples has been introduced, facilitating direct imaging via cryo-ET of cellular sections with minimal artifacts and in a frozen-hydrated state (e.g.,;; ). As these techniques preserve the hydrogen bond network and fine ultrastructure of samples, they allow the subsequent structure determination using sub-volume averaging after cryo-ET. Recent developments in the field have led to a number of structures at resolutions approaching the atomic scale derived from in situ environments and have demonstrated the ability to inform on the molecular and structural details of proteins in their native state and context. Cryo-FIB and cryo-ET were recently employed to investigate neonatal rat cardiomyocytes and provided an initial insight into the organization of muscular filaments inside immature sarcomeres, in which the thick filaments and thin filaments could be observed in a semi-striated arrangement. However, the details of a mature sarcomere from a fully developed vertebrate myofibril, wherein different zones and structures can be distinguished, remain elusive.

11Combining cryo-FIB, cryo-ET, and sub-volume averaging, we present the three-dimensional structure of a mouse psoas sarcomere in the rigor state with a comprehensive description of the organization and interaction of key sarcomeric proteins in molecular detail.

12In order to investigate the native organization of an intact sarcomere, we vitrified mouse psoas myofibrils in the rigor state (i.e., without ATP) by plunge-freezing in liquid ethane and subsequently prepared thin lamellae using cryo-FIB (; STAR Methods). The lamellae were 30–150 nm thin and ideally suited for cryo-ET (Figure S1).

13Vitrified myofibrils show an intact sarcomere ultrastructure, including apparent Z-discs, I-bands, A-bands, M-bands, and organelles such as mitochondria and sarcoplasmic reticulum localized between laterally adjacent sarcomeres (Figures 1A and S1). Importantly, the tomograms include structures that could not previously be visualized in situ. Of particular note are myosin double heads forming regular cross-bridges that resemble an “arrow-head,” thin myosin tails mainly composed of coiled-coil structures protruding from the thick filament, and details of the Z-disc and troponin complexes (Figures 1B–1D). The M-band consists of numerous structural proteins other than myosin, such as myomesin, titin, and obscurin as well as accessory proteins with transient localization, such as FHL-2, Nbr1, MURFs, or calpain-3. Together, they form pleomorphic densities decorating and cross-linking the thick filaments (Figures 1B and 1D). This heterogeneity hampers identification of individual proteins in this crowded region. Therefore, we focused on the details of the A-band, I-band, and Z-disc in the following sections.

14It is well-known from cross-sectional views of myofibrils that thin and thick filaments are hexagonally arranged in the A-band. Interestingly, in immature sarcomeres found within neonatal cardiomyocytes, the hexameric pattern is limited to thick filaments, suggesting that thin filament organization is a later process of sarcomeric maturation in vivo. In orthogonal views of our tomograms, we can clearly discern this hexagonal pattern in the A-band, demonstrating that our preparations are consistent with previous studies (Figures S1D–S1F). One thin filament is surrounded by three thick filaments and three other thin filaments at a distance of around 26 nm (Figures S1D–S1F). Thick filaments form triangular patterns with an average inter-thick-filament distance of 45 nm (Figures S1G–S1I). This arrangement of thick filaments is maintained from the A-band to the M-band where thin filaments are missing.

15Segmentation of the tomograms enabled us to unequivocally characterize the 3D arrangement of thin and thick filaments and the cross-bridges within a sarcomere (Video S1). To increase the level of observable detail that we could resolve with our data, we employed sub-volume averaging to determine the structure of the thin and thick filament and cross-bridges. The reconstruction of the thick filament was proved to be difficult due to its heterogeneity, flexibility, and an absence of obvious symmetry. Nevertheless, we obtained a low-resolution reconstruction showing the architecture of the core of thick filaments that we could use for the subsequent analyses (Figure S2).

16The three-dimensional reconstruction of the thin filament and cross-bridges could be resolved to a higher resolution. Segmentation of the tomograms and 3D classification of all sub-volumes of the cross-bridges indicated that the two heads from the same myosin molecule, known as double-heads, mostly bind to two neighboring actin subunits (Figure S3D). Hence, we merged multiple classes after modifying alignment parameters to obtain a structure of a native thin filament with the myosin double-head bound (Figure 2A; STAR Methods). The reconstruction has a resolution of 10.2 Å and we could clearly assign tropomyosin and the domains of actin and of the myosin heads and necks, including the essential chains (ELC) and part of the regulatory light chains (RLC) (Figures 2A and S3; STAR Methods).

17We calculated a homology model using the rigor cytoplasmic actomyosin cryo-EM structure (PDB: 5JLH) and the crystal structure of the S1 fragment from squid muscle (PDB: 3I5G) as a template and fitted it into the density map using rigid body fitting (Figures 2B and 2C). We chose this myosin structure because there are no crystal structures of a complete vertebrate myosin S1 fragment in the rigor state available. The model fits well into the density, demonstrating that the in vitro structures resemble well the structures determined in situ. At this resolution, both myosin heads bind to the actin filament in a structurally identical manner with close interactions formed between actin subunits and myosin as well as tropomyosin (Figures 2B and 2C). Both ELCs bind to the upper parts of the lever arms (residues L788–L800) as is seen in the crystal structure. The density threshold corresponding to the RLCs and RLC binding domains after averaging is considerably lower than the rest of the structure, indicating a higher flexibility or variability in this region.

18To improve the quality of the reconstruction in the region of the RLC, we shifted the center of sub-volume averaging from the thin filament to the myosin head. Through further classification, we determined a structure of the complete double-head myosin S1 fragment, including the two heavy chains that are bound to actin, two ELCs and two RLCs (Figure 2D). The S2 fragment that tethers myosin heads to the thick filament was not resolved in this structure, likely due to the different positions and angles it takes in the sarcomere.

19Although a homology model of the myosin motor domain and ELC of the squid myosin crystal structure (PDB: 3I5G) fits well into our density, the RLCs and the lower part of the lever arm (residues R811–L844) had to be fitted separately (Figure 2D). Compared with its conformation in the crystal structure, the lever arm is strongly kinked between the ELC and RLC binding regions (residues M801–E810), resulting in a bent conformation (Figure 2E). Interestingly, the kink in the upper head is in the opposite direction compared to the one in the lower head, bringing the RLCs of the two heads in close proximity (Figure 2F). This crucial observation reveals the ability of the myosin double head to accommodate force production, and has not been validated structurally in any previous study in any system. It is also consistent with previous cross-linking and FRET experiments, which suggested a small distance between the two RLCs in a double head.

20This arrangement, which is probably stabilized by interactions between the ELCs and RLCs, allows for the simultaneous binding of the two heads to actin and at the same time brings their necks close enough together to continue with the coiled-coil structure of the joint S2 fragment. In addition, the opposite direction of the kinks on the two lever arms leads to a similar length of the two arms. This is important because the length of the lever arm determines the step size of myosin heads on actin.

21In this context, it is interesting that the hinge in the lever arm together with a hook (residues W824–W826 in molluscan myosin) have previously been shown to be important for regulated myosins. In this case, the light chains control myosin activity even in the absence of actin and the presence of Ca2+ determines the stability of the ELC/RLC interface. Different ELC/RLC interfaces were also suggested from the latest model of myosin in the relaxed state, where the heads fold back to the thick filament. Notably, the arrangement of the ELC/RLC of the lower head in our model resembles that in the “blocked” head of the off-state myosin while the upper head ELC/RLC is similar to the “free” head (Figures S3I and S3J). This suggests that there are only two possible angles for the kink in the lever arm, which are likely to be determined by the interaction of the ELCs and RLCs. This interaction probably stabilizes the two conformations to rigidify the lever arm, which is needed for the proper transmission of the force of the power stroke. It will be interesting to visualize how the phosphorylation state of the RLC can modulate the interaction of the ELCs and RLCs and thus modulate muscle contractility.

22Various structures for the single isolated myosin head have been determined in different states of the ATP hydrolysis cycle (e.g.,;;;;; ). Furthermore, whereas the myosin double-head has also been determined in structures of isolated thick filaments and in 2D crystals of heavy meromyosin, our in situ actomyosin complex is a structure of the myosin double-head interacting with the thin filament determined directly within fully organized myofibrils. Our analysis of this structure can be used to further inform other states of the myosin double-head, such as the pre-power stroke state and the relaxed state, to further understand the cross-bridge cycle within a sarcomere in a native context.

23In order to identify the proportion of myosin heads that bind to thin filaments in a rigor state sarcomere, we annotated and isolated the densities corresponding to the cross-bridges as well as thick and thin filaments in the A-band (Figures 3A and S4). Combining annotated filaments with structures determined via sub-volume averaging allowed us to produce a molecular map of the thin and thick filaments and attached myosin heads (Figures 3B and S4).

24Of the 30 thin filaments annotated containing 5,664 actin subunits, 2,734 myosin heads were fitted into the cross-bridge densities, corresponding to 82.5% of the total number of myosin heads in this region calculated from the theoretical number of myosin heads per given length of the thick filament (6 heads per 14.3 nm, 294 heads per thick filament, total heads in our analysis = 3,315). This suggests that not all myosin heads are attached to thin filaments, even in the rigor state.

25Similar to sarcomeres from insect muscle, all thin filaments are in helical register (Figure 3B) and cross-bridges appear at regular target regions every ~37 nm on a thin filament between each pair of thick and thin filaments (Figures 3A and 3B; ). However, in contrast to the “double-chevron” arrangement of cross-bridges in an insect sarcomere, the cross-bridges in a vertebrate sarcomere show a higher variability regarding the distance between each other and tend to appear in clusters (Figures 3A, 3B, and 5). Although the most common composition of a cross-bridge is only one myosin double-head (Figure 3C), two consecutive myosin double-heads also appear at many target regions. Occasionally, there are also single heads bound to the thin filament (8.3% of all myosin heads) with their partners not identified within the tomogram. In addition, a rare “split-head” conformation also occurs when the two heads from one myosin binds to two different adjacent thin filaments (10 pairs observed; 0.6% of all heads) (Figures 3D and S4). This conformation has been previously suggested and vaguely indicated by 2D projection images. Our observation provides direct proof of this conformation in three-dimensions in the rigor state vertebrate sarcomere. The ability of myosin to split its two heads and bind to different thin filaments enhances the plasticity of the sarcomere and allows proper contraction under slight structural deformation due to external forces.

26Notably, although we annotated mostly the C-zone of the A-band, where myosin binding protein C (MyBP-C) is present, we did not identify any corresponding densities along the thick filaments with a regularity of 43 nm as seen in previous studies. In the rigor state sarcomere, the density corresponding to MyBP-C is likely to be overwhelmed by the large number of myosin heads bridging between the thick and thin filaments. Further structural investigation on relaxed myofibrils will shed light on the MyBP-C organization as retracted myosin heads abolish the cross-bridges, leaving only cross-links formed by MyBP-C between the thick and thin filaments.

27To further investigate the variability in myosin binding, we focused on the influence of actin orientation and the distance between thin and thick filaments (Figure 4). We plotted the orientation of the actin subunits bound by myosin heads in terms of their relative angles with respect to a fixed orientation perpendicular to the filament axis. The distribution showed three distinctive groups with a Gaussian distribution corresponding to the three thick filaments from which the myosin heads originate. For a specific thick filament, the orientations of myosin-bound actin subunits are mostly confined within a range of ~120° (Figure 4A). This indicates that myosin heads tend to bind to actin subunits oriented toward the direction of the thick filament from which it originates. This is also demonstrated by the combined myosin binding profile of all 30 thin filaments (Figures 4B and S5). In addition, this footprint map of myosin binding on a thin filament also exhibits the 37-nm hotspot (~13 actin subunits) of cross-bridge clusters as a consequence of this preferred binding orientation of actin subunits.

28Furthermore, the distances between the thin and thick filaments at both myosin-bound and myosin-free actin subunits show a Gaussian distribution with no obvious skewness (Figure 4C). The myosin-bound actin subunits are on average 0.4 nm closer to the thick filaments (95% confidence interval 0.2–0.5 nm) compared to myosin-free actin subunits. This small difference implies only a minor preference for myosin-binding when there is a shorter distance between the thin and thick filaments.

29Our molecular map of myosin head distribution enabled us to determine the arrangement of myosin heads emanating from thick filaments within the rigor state sarcomere (Figure 5A). In the OFF state, when myosin heads are retracted to the thick filament, they are arranged in a 3-start helical pattern with 14.3 nm between layers of crowns as illustrated by cryo-EM and X-ray diffraction studies. This regular organization is not maintained in sarcomeres in the rigor state, when thick filaments are in the ON state. We could observe that the distance between axially adjacent myosin heads varies (Figure 5B) and found a different binding of myosin heads in different sections of a thick filament (Figures 5D–5G). The S2 fragment, which forms a convex surface with the S1 fragment, provides enough flexibility for a myosin head to bind to a random actin subunit within the range of allowed distance and orientation (Figure 5C). This flexibility makes allowance for the mismatch of the 37-nm actin repeat in thin filaments and the 14.3-nm myosin repeat in thick filaments and results in the pseudo-regular arrangement of cross-bridges.

30Overall, the analysis of myosin heads in the A-band suggests that myosin binding is a stochastic process regulated by physical limitations such as the orientation of actin subunits. Myosin prefers to bind to regions on thin filaments that orient toward the thick filament, creating ~37 nm regular periodic cross-bridge clusters, consistent with previous observations in insect muscle and single-molecule experiments. In insect flight muscle, cross-bridges are formed in distinct lead and rear bridges. However, cross-bridges in vertebrate sarcomere are composed of various myosin double-heads in an irregular pattern. This pseudo-regular organization indicates a certain plasticity within a vertebrate sarcomere, whereby a strictly ordered myosin binding network is unnecessary.

31One advantage of our approach is that we can trace single filaments through the sarcomere in 3D, allowing us to follow thin filaments from the A-band through the I-band into the Z-disc. The ordered hexagonal pattern of filaments in the A-band breaks down in the I-band, where the thick filaments end (Figure 6D). Consistent with previous studies presenting cross-section images of the I-band and contrary to a previous model stating an equally distanced arrangement of thin filaments along the A-band to Z-disc transition, thin filaments show a transition in the I-band from a hexagonal pattern to an irregular pattern caused by the lack of cross-bridges in this region. The extent of displacement from the hexagonal arrangement varies among filaments (Figure 6D).

32Due to the absence of myosin in the I-band, troponin complexes are clearly visible with a periodicity of ~37 nm on the thin filament (Figures 1C, 6D, and S6). We applied sub-volume averaging to determine the structures of the actin-tropomyosin-troponin complex and the actin-tropomyosin complex in the I-band at resolutions of 19.8 Å and 10.6 Å (Figures 6A, 6B, and S6), respectively. We then calculated a homology model using an in vitro cardiac muscle thin filament cryo-EM structure of the Ca2+-bound state as a template (PDB: 6KN8) and fitted it into the density maps using rigid body fitting. The model fits well into the reconstruction of both structures, demonstrating that the thin filament is in the on-state with tropomyosin in the C (Ca2+ induced) state position (Figures S6I and S6J). This suggests that there is structural similarity between skeletal and cardiac troponin complexes when bound to actin filaments at this resolution, despite differences in respective crystal structures. Interestingly, the position of tropomyosin differs from that in the A-band, where tropomyosin is in the M (myosin-bound) state (Figures 6C, S6A, and S6B;; ). Thus, the binding of myosin to the thin filament shifts tropomyosin from the C-state to the M-state position in the A-band, whereas tropomyosin remains in the C-state in the I-band. Although different tropomyosin states have been observed separately in vitro, our in situ structures of thin filaments suggest that the tropomyosin position can vary on a local scale within the same sarcomere and even the same thin filament.

33To determine the location of this transition, we subdivided the tomograms in 37-nm sections around the A-band/I-band transition and averaged the thin filaments separately (Figure S6D). The averages revealed that tropomyosin is in the M-state in A-band sections. In the first I-band section, tropomyosin is in an intermediate state between the M-state and C-state and from the second I-band section, it occupies the C-state position (Figure S6F). This suggests that, contrary to previous predictions, the M- to C-state transition happens mostly within one tropomyosin unit and occurs at the section immediately after the A-band/I-band transition.

34The irregular pattern of thin filaments resumes an ordered state when the filaments approach the Z-disc (Figure 6G). In contrast to the exact square patterns observed in Z-discs from midshipman fish sonic muscle and rat soleus and cardiac muscle, the Z-discs in our reconstructions are slightly less well ordered and form squared to more rhomboid patterns (Figures 6G, 6H, and 7). This is likely due to the fact that we used single myofibrils instead of complete muscles, where Z-discs are laterally anchored and stabilized. Antiparallel thin filaments in the Z-disc with opposing polarities are connected by 33-nm long cross-links, which we attribute to α-actinin (;; Figures 7G–7I and S7D).

35By comparing Z-discs from different reconstructions, we found different Z-discs varied in thickness, ranging from ~80 nm to ~100 nm (Figures 7A–7F, S7A, and S7B; STAR Methods). Thicker Z-discs are more compact than thinner ones (thin filaments are closer laterally) (Figures 7A–7F and S7B). Interestingly, although the mean length of α-actinin is similar in all Z-discs (Figure S7D), the average angle between α-actinin and actin filaments in direction to the pointed end differs considerably, ranging from ~158° in the thickest and ~128° in the thinnest form (Figure S7A).

36We propose that the different forms of Z-discs represent conformational states of the sarcomeric unit. We hypothesize that post-isolation mechanical factors, such as different on-grid blotting forces before freezing, increase the stress/strain in some myofibrils. The angle between actin and α-actinin becomes more acute and the Z-discs fold together like a parallel hinge, resulting in the thicker appearance of the Z-disc (Figures 7D–7F, S7A, and S7B). This situation is likely similar to activated myofibrils under tension. Thinner forms possibly represent the Z-disc under low strain/stress, where the angle between actin and α-actinin is more obtuse (Figures 7A–7C, S7A, and S7B). Considering a stable interaction interface between the actin-binding domain of α-actinin and actin filaments and the rigidity of the rod region of α-actinin, the flexibility in the neck between the rod and the actin-binding domain (ABD) of an α-actinin probably serves as the central hinge of this pivot-and-rod structure. In order to experimentally validate that stress/strain induces different conformations of the Z-disc, activated myofibers will have to be vitrified under defined strain/stress conditions prior to cryo-ET analysis in the future.

37We annotated all clearly visible densities in the thinner Z-disc where single α-actinin densities were more apparent to understand the α-actinin organization between antiparallel thin filaments (Figure 7G; Video S2). Top and side views of the resulting three-dimensional model reveal that α-actinin forms a mesh by cross-linking the actin filaments, which is less well ordered than anticipated from the projection (Figures 7G–7I) or from idealized helical models in previous work. A prominent and unexpected feature is a doublet of α-actinin cross-links with 6 nm spacing, which is only possible when two α-actinins bind to longitudinally adjacent actin subunits along the actin filament (Figures 7J, 7K, and S7F). We observe up to three of these α-actinin pairs connecting two antiparallel thin filaments over ~37 nm, at ~18.5-nm intervals. Thus, although the spacing of these doublets follows the 37-nm actin helical repeat, it places individual α-actinin molecules closer together (maximally 12 nm rather than 18.5 nm) than previously presumed. Doublets have also been observed in an in vitro reconstituted α-actinin-F-actin raft. Our analysis shows that this arrangement was not an in vitro artifact but a feature of native sarcomeres. Interestingly, these doublets were not apparent in the recent reconstruction of the F-actin-α-actinin complex from branched cardiac myofibrils obtained through sub-volume averaging, in which the angles between α-actinin and actin resemble the thick-form Z-disc. Importantly, as single α-actinins were not distinguishable in these tomograms, this study relied strongly on subtomogram averaging at certain positions, assuming a highly symmetric Z-disc. However, our visualization of the Z-disc illustrates that it is a heterogeneous, non-symmetrical structure. The doublets, which are not present at regular intervals, were therefore probably lost during sub-volume averaging of the Z-disc or are not present in cardiac myofibrils.

38Consistent with previous studies, we observed several ~37 nm spaced α-actinin cross-links (Figures S7E and S7G), which corresponds roughly to the half-helical pitch of actin filaments. However, we also observed α-actinins that do not follow this pattern (Figures 7J, 7K, and S7E) suggesting a less ordered organization of the Z-disc than previously assumed or inferred from atypical actin-α-actinin arrays in nemaline rods or midshipman muscle. By performing sub-volume averaging of the α-actinin positions, we could show that it connects opposite sides of antiparallel actin filaments, forming a slight “S” shape (Figure S7H). The actin binding domains and the central rod region of an α-actinin from the crystal structure fit well into the density (Figure S7H), and the four spectrin-like repeats can be clearly distinguished in a projection of the reconstruction (Figure 7L). This “S” shape fits well to the α-actinin crystal structure, considering the flexibility in the neck region, but differs from the conventional “basket-weave” model that is derived from observations of transverse sections of Z-discs. In order to form the S shape, the two ends of an α-actinin, where it interacts with actin filaments, need to have different conformations; one end resembling that of the basket-weave model with the other end in an alternate conformation that interacts with actin differently at the flexible neck region (Figure S7I). Future studies at a higher resolution will elucidate the molecular basis of this interaction and shed light on the organization of other proteins in the Z-disc, such as accessory smaller proteins like ZASP, myotilin or FATZ that may play a role in α -actinin dimerization, or to visualize the position of the α-actinin-binding Z-repeats of titin or the actin-binding repeats of nebulin. The potential contribution of these proteins to the arrangement of α-actinin doublets would also be of fundamental interest for understanding diseases of the Z-disc with aberrant arrangement of alpha-actinin like actininopathies, zaspopathies, nemaline myopathies, or myofibrillar myopathies.

39Cryo-FIB milling combined with cryo-ET enabled us to visualize native vertebrate sarcomeres in three-dimensions with minimal artifacts or damage. Our structure of the actomyosin complex shows the complete myosin double head in the rigor state in molecular detail and pinpoints the ELC/RLC interface as the position within myosin crucial for the structural rearrangement of the double head when bound to the thin filament. In addition, our model of the sarcomere shows a pseudo-regular distribution of myosin cross-bridges in the A-band. An irregular pattern of the I-band thin filaments illustrates the importance of the cross-bridge and the Z-disc in maintaining the regularity of thin filaments within the sarcomere. Our observation that the thin filaments and cross-links of α-actinin within the Z-disc can accommodate different arrangements demonstrates the need for the sarcomere to respond to and accommodate the collective forces exerted on it during muscle contraction. Collectively, our high-resolution structure of the sarcomere, although not yet a complete atomic model of the entire structure, identifies previously unresolved details of the molecular architecture of A-bands, I-bands, and Z-discs and highlights the molecular plasticity of its components. It allows improved dynamic modeling of muscle contraction and will serve to test new concepts on the molecular, cellular, and physiological level, not least for the development of new approaches for investigating muscle diseases.

40Although this study provided an overall picture of the skeletal sarcomere, subsequent work needs to be performed to address a number of missing pieces arising from this research. In this study, we focused on sarcomeres from skeletal muscle in rigor, where the myosin cross-bridges are locked in a strong-binding state. Other states of the cross-bridge cycle need to be investigated in order to provide a dynamic understanding of muscle contraction. This study could not include an analysis of M-band proteins, MyBP-C, or other less regularly arranged proteins that also play an essential role in sarcomere regulation, function, and plasticity. Additional in situ labeling methods in future studies will help to elucidate the organization of these less common and less organized proteins. In the analysis of Z-discs of different thicknesses, our hypothesis that the different Z-disc conformations may correspond to the different stain/stress states of the sarcomere should be validated experimentally using custom approaches where muscle can be examined under defined load. Future studies using greater numbers of tomograms acquired on myofibrils should help to improve the resolution of structures obtained in this study to achieve insights on the atomic level.

41Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Stefan Raunser (stefan.raunser@mpi-dortmund.mpg.de).

42This study did not generate new unique reagents.

43The EM structures generated through this study have been deposited in the Electron Microscopy Data Bank (https://www.ebi.ac.uk/pdbe/emdb) under accession numbers 12289, 12291, 12292, 12293. The in situ actomyosin model is deposited in the Protein Data Bank (https://www.rcsb.org) under accession number 7NEP.

44This study used muscle isolated from the psoas of 3-month female BALB/c mice. Animals were euthanised in a schedule-1 procedure by cervical dislocation following licensed procedures approved by King’s College London ethics committee and the Home Office UK. The isolation of muscle is described below in the Method details.

45Myofibrils were isolated from BALB/c mouse psoas muscle essentially as previously described. Briefly, bundles of the freshly excised psoas major muscle were tied to plastic supports and their sarcomere length adjusted as judged by laser diffraction to 2.2 - 2.4 μm. They were allowed to equilibrate in rigor buffer (20 mM HEPES pH 7, 140 mM KCl, 2 mM MgCl2, 1 mM EGTA, 1 mM DTT, Roche complete protease inhibitor) over night at 4°C. The following day, the central section of the bundles was dissected into ~2 mm pieces and homogenized 3 to 4 times at 7,700 rpm for 15 s, and 3 times at 10,000 rpm for 5 s (IKA TC10 basic ULTRA-TURRAX® homogenizer with S10N-5G dispersing element, IKA England) with intermittent washes in rigor buffer, with myofibril separation monitored by light microscopy of small samples at each step. Myofibril suspensions were stored in complete rigor buffer at 0°C for 1-3 days until vitrification. Grids with myofibrils were vitrified using a Vitrobot Mark IV plunger (Thermo Fisher Scientific, USA). 2 μl of myofibril-containing solution was applied to glow-discharged Quantifoil R1.2/1.3 Cu 200 grids. After incubation on the grid for 60 s at 13°C, excess solution was blotted for 15 s from the opposite side from the sample with a Teflon sheet attached to the front pad of the apparatus. The grids were then vitrified by plunge-freezing into liquid ethane and afterward clipped into cryo-FIB-specific autogrid rings (Thermo Fisher Scientific) with marks for orientation alignment and a cut-out for subsequent milling at a shallow angle.

46Clipped autogrids were loaded into a shuttle and transferred into an Aquilos cryo-FIB/SEM dual-beam microscope (Thermo Fisher Scientific). Overview tile sets for the grids were acquired at 256 x magnification with the scanning electron beam using MAPS software (Thermo Fisher Scientific). The grids were then sputter coated with platinum for 15 s to minimize charging effects arising from the electron beam, allowing better recognition of myofibrils (Figure 1). Vitrified myofibrils were localized and labeled as lamella sites. Prior to FIB-milling, organometallic platinum was deposited onto the grids through a gas-injection-system to prevent damage to the sample at the leading edge from the gallium ion beam. For each lamella site, the coincident point between electron beam and ion beam was determined by adjusting stage Z height. The stage was tilted to allow a 6-10° incidence angle of ion beam. FIB-milling using gallium ions was then performed in four steps as described in Table S1. During each step, all lamellae were milled before proceeding to the next step. All lamellae were polished within one hour to minimize water deposition onto the surface of lamellae. During polishing, the lamella was monitored with the electron beam at 5 kV, 25 pA to help estimate its actual thickness via charging propensity.

47Autogrids were rotated by 90° after milling before being inserted into an autogrid cassette so as to align the longitudinal axis of lamellae perpendicular to the tilt axis of the transmission electron microscope (TEM) stage later. This ensures autofocus and record positions are aligned along the tilt axis on a lamella and have close to the same eucentric height. Grids were then loaded into a Titan Krios (Thermo Fisher Scientific) equipped with a zero-loss energy filter and a K2 Summit direct electron detector (Gatan Inc., USA). SerialEM software was used to acquire images. Lamellae overview images were acquired at 6,300x or 8,400x magnification. High magnification tilt series were acquired at a calibrated post-energy-filter TEM magnification of 28409x (nominal magnification of 81,000x, pixel size 1.76 Å) with a dose-symmetric tilt scheme and a total dose ranging from approximately 130 to 155 e-/Å2. The stage was tilted from −60° to +60° relative to the lamella tilt angle (6 - 10°) with an increment of 3° during data acquisition. The defocus at which acquisition was performed ranged from 2.4 μm to 4 μm (Table S2). Collected tilt movies were subsequently motion-corrected and stacked in batch using a custom python script with MotionCorr2 and EMAN2. Subsequent tomogram reconstruction steps including tilt series alignment, CTF estimation and correction, and weighted back projection were performed using the IMOD software package. Platinum particles deposited onto the surface of lamellae were used as fiducial markers for alignment of tilt series where possible whereas in their absence patch tracking was used to align the tilt series.

48Tomograms were initially 4x binned and lowpass-filtered to 60 Å using EMAN2 for visualization. To pick filaments accurately without the interference from signals from cross-bridges, an equatorial mask in the Fourier transform of XY slices of tomograms was applied. CrYOLO was afterward employed to detect the shape of thick filaments in the XZ planes of the filtered tomogram. The detected points were then traced by a python script, which generated the coordinates for the thick filaments. In total, 13,700 segments of thick filaments (sub-volumes) were picked from 8 tomograms from multiple myofibrils with an inter-segment distance of 105 Å. These segments were extracted from unbinned and 2x binned tomograms with a box size of 128 pixels (450 Å) using RELION. Prior information of the orientations of the thick filaments in a tomogram was used to generate a featureless cylinder-like reference using PEET. The sub-volumes were then aligned and averaged using RELION. Three different strategies were used during alignment. First, the sub-volumes were aligned without any symmetry imposed and a global refinement was enabled. This resulted in aligning the missing wedge and a map with anisotropic resolution was generated (Figure S2B). Then, a C3 symmetry was imposed during refinement according to a previous structural study of the relaxed human cardiac myosin filament. Although the averaged map had an isotropic resolution, the quality of the map did not improve (Figure S2C). In the end, without any symmetry imposed, only local refinement was enabled to prevent aligning the missing wedges (Figure S2D). This averaged structure has an estimated resolution of 30.4 Å based on the “gold-standard” FSC with 0.143 criterion. The map was used as a model for the thick filament in Figures 3 and 5.

49Thin filaments were picked automatically in a similar method as described above. Instead of using crYOLO, thin filaments, which appeared as dense dots from the XZ view (Figure S1E), were recognized and traced by the TrackMate plug-in in Fiji. In total, 32,421 segments (sub-volumes) were picked from 8 tomograms from multiple myofibrils with an inter-segment distance of 63 Å. These segments were extracted from unbinned and unfiltered tomograms with a box size of 200 pixels (351 Å) using RELION. In order to exclude low-quality sub-volumes using 2D classification, the extracted sub-volumes were first rotated to orient thin filaments parallel the XY plane based on filament orientations calculated from particle positions. Projection images along Z axis were then generated from the central 100 slices (175 Å) of each particle and subsequently classified using 2D classification in the ISAC software from the SPHIRE package. This approach, compared to using the entire sub-volume for projection, excluded signals from other filaments at the corner of the sub-volume and thus enabled a more reliable assessment of particle quality (Figure S3B). 21,130 selected good sub-volumes were aligned, averaged, and classified in 3D using RELION, based on a cylinder-like reference generated from averaging all sub-volumes after aligning the longitudinal axis of segments using PEET (Figure S3C). In each 3D class, translation and rotation parameters were modified to align the most prominent double-head of myosin (Figure S3D). After combining the modified good classes and removing duplicate particles, 18,090 sub-volumes were refined locally with a mask including the thin filament and one pair of myosin double-head. The final average has an estimated resolution of 10.2 Å based on the “gold-standard” FSC with 0.143 criterion. Local resolution was estimated in SPHIRE using the two half-maps and the mask used during averaging. The final map was sharpened with a B-factor of −250 and filtered to the nominal resolution (Figure S3F).

50In order to resolve the light chain domains of myosin heads, sub-volumes centered on myosin the double-heads (shifted 90 pixels along x axis compared to original sub-volumes) were extracted. The positions of these new sub-volumes in tomograms were calculated by a python script using the original sub-volume coordinates and their alignment parameters. The new sub-volumes were sorted through 3D classification in RELION. 4,519 particles from the two good classes were refined locally with masking out the thin filament and myosin double-head (Figures S3E and S3G).

51An initial atomic model of the in situ actomyosin complex was first built by rigid-body fitting of actin subunits and myosin motor domains from the atomic model of the in vitro actomyosin complex (PDB: 5JLH), tropomyosin from the atomic model of isolated cardiac thin filament (PDB: 6KN8) and myosin lever arms from the crystal structure of rigor-like squid myosin S1 (PDB: 3I5G) using ‘Fit in Map’ in Chimera. Essential light chains (ELCs) and the regulatory light chain (RLCs) were fitted separately into the averaged map after removing 5 amino acids (M796-Y800) at the hinge on the α-helix between ELC and RLC. The model of RLC together with part of the heavy chain lever arm (K801-L839) was rigid-body fitted into a segmented map generated by “Color Zone” in Chimera. Based on this initial atomic model and the sequences of actin, myosin and tropomyosin in mouse fast skeletal muscle, a homology model was calculated using SWISS-MODEL (Figure 2). Comparison of lever arms in the in situ rigor upper head, lower head, squid myosin S1 fragment, and the blocked and free heads from relaxed insect flight muscle (PDB: 6SO3) (Figures S3I and S3J) was performed by aligning the RLC of the heads using the “MatchMaker” function in Chimera.

52In order to visualize the organization of cross-bridges between thin and thick filaments, the densities of a complete A-band were annotated manually using Amira (Thermo Fischer Scientific) on a slice-by-slice basis. A user-defined mask based on gray values was used during manual annotation to prevent the picking of weaker densities. The 2D annotation stacked to form a 3D segmented volume depicting the arrangement of thin and thick filaments as well as the cross-bridges formed by myosin heads (Figure 3A). During sub-volume averaging, a structure of the thin filament fully decorated with myosin heads (Figure S3C) was obtained prior to 3D classification. This was low-pass filtered to 17 Å and extended to a length of 1,755 Å (~63 actin subunits) based on the helical symmetry detected in the structure (−166.6° turn and 27.9 Å rise) using RELION helix toolbox. This long fully-decorated thin filament model was then fitted into each segmented volume corresponding to a thin filament with bound myosin heads using “Fit in Map” in Chimera. The myosin heads that matched segmented volume were kept and colored depending on which thick filament they originated from while the other myosin heads were removed. This generated a complete map of myosin heads bound to thin filaments in the A-band (Figures 3B and 5; Video S1).

53The models of double-head and split-head conformations were made from initially fitting our actomyosin structure into the segmented volume (Figures 3C, 3D, and S4). For the split-head conformation, the upper head from the double-head conformation was fitted into densities for both heads. The S2 domain from segments of a crystal structure (PDB: 2FXM) was manually placed at the interface between the two regulatory light chains for its speculated position and orientation. Surface models were generated with the “molmap” command in Chimera at a resolution of 15 Å.

54In order to investigate the relation between myosin binding and the orientation of their bound actin subunit relative to the neighboring thick filament, the corresponding angle of each actin position relative to the first actin subunit in the thin filament was calculated based on actin helical parameters determined from the averaged structure. The angles where there was a myosin bound were divided into three groups depending the original thick filaments and plotted as a circular histogram. Values of angles from different filaments were combined by aligning the average of one myosin group (represented as red color in Figure 4), resulting a final histogram representing all myosin-bound actin positions (Figure 4).

55In addition, the molecular model of each thin filament was converted to a myosin-binding profile sequence consisting of R, G, B, and E representing actin bound by myosin from three different thick filaments and non-bound actin, respectively. The sequences of all 30 annotated filaments were aligned by multiple sequence alignment using the “msa” package in R with the MUSCLE algorithm utilizing a customized weighting matrix (Figure S5; Table S3). When a thin filament was treated as a single strand, adjacent actin subunits had a huge change in orientation (166.6°) and thus very different preference for myosin binding, resulting in nonoptimal multiple sequence alignment. Therefore, each thin filament was considered as two actin strands and only one strand was used for multiple sequence alignment. The other strands were aligned using the same alignment as the first strands. The occurrence of myosin binding at each actin position was summed up for myosin from each thick filament and then used to color actin filament models in Chimera, showing averaged hotspots for myosin binding on a thin filament (Figure S5B).

56To further investigate whether myosin prefers to bind at shorter distances between thin and thick filaments, the distances between thin and thick filaments at the positions where there is a myosin head bound and where myosin is absent were calculated using a python script. For each actin subunit not bound by a myosin head, three measurements were taken between the actin and three different neighboring thick filaments. For an actin subunit that is bound by a myosin head, the distance between the actin subunit and the corresponding thick filament is considered as the myosin-bound measurement while the distances between this actin subunit and the other two neighboring thick filaments are considered as the myosin-free measurements. All measurements were plotted in a histogram (Figure 4C).

57Thin filaments in the I-band were picked automatically as described in the previous sections with crYOLO. 15,153 sub-volumes from 4 tomograms from multiple myofibrils showing I-band in the field of view were aligned with calculated priors (phi and theta angles) and refined in 3D using RELION. Helical processing was used to minimize missing wedge artifacts using the helical parameters determined in the A-band thin filament structure. This resulted to a final structure of I-band thin filament without troponin with an estimated resolution of 10.6 Å using the 0.143 criterion. The map was sharpened with a B-factor of −600 and filtered to the nominal resolution. Thin filament segments decorated by troponin densities were picked manually in IMOD. In total, 704 sub-volumes were aligned to an initial reference generated by averaging all sub-volumes without alignment in PEET. The averaged structure of the I-band thin filament including troponin has an estimated resolution of 19.8 Å after masking. A homology atomic model of the I-band thin filament was calculated using SWISS-MODEL based on the atomic model of isolated cardiac thin filament and rigid-body fitted into the averaged maps (Figures 6 and S6).

58To investigate the location where the transition of tropomyosin from the M-state to C-state occurs, three tomograms containing both A-band and I-band were analyzed. Particles around the A/I junction in these tomograms were divided into 37-nm sections. Four sections (one in A-band and three in I-band) were selected using a customized script. Sections 1-4 contains 1206, 1126, 988 and 947 particles, respectively. The particles from the four sections were averaged separately using the same strategy as described above (Figures S6D and S6F). The averages were aligned to an undecorated actin filament structure for the comparison of tropomyosin positions. The particles colored by their tropomyosin states were visualized in the 3D view in IMOD to depict their distribution (Figure S6E).

59Different Z-disc thicknesses were observed in different sarcomeres from multiple myofibrils. The thickness was measured between the sites where α-actinins start to bind to the antiparallel thin filaments using IMOD (Figures 7A, 7D, and S7A). The relative angle between α-actinin and actin was measured as the obtuse angle between the pointed end of actin filament and α-actinin using Fiji (Figure S7A). From our observations, the Z-discs from the same myofibril have a similar thickness.

60A tomogram showing both I-band and Z-disc was 4x binned and filtered initially with a SIRT-like filter using IMOD, followed by a low-pass filter at 60 Å. Thin filaments in the tomogram were first picked automatically as described in the previous section. The segments in the Z-disc were further curated manually and merged into the originally picked filaments. The polarities of the filaments were determined based on the location of the ends of the filaments. 3D organization of these filaments was generated in Fiji and shown in Chimera (Figure 6F). To analyze the organization of α-actinin in the thin-form Z-disc, potential positions of α-actinin were sampled between ends of filaments with opposite polarities. The exact location of α-actinin was searched for by aligning sub-volumes extracted at these positions to a cylinder using a limited search range with PEET. The new positions and the corresponding orientation of sub-volumes representing the location of α-actinin were clustered based on the determined new position. Obvious false-positive and false-negative positions were then curated manually. In the thick-form and intermediate Z-discs, 12 α-actinins were picked in each tomogram manually in IMOD. The angles between the pointed end of actin and the long axis α-actinin and the lengths of α-actinins in both types of Z-discs were calculated and plotted as histograms using python scripts. The 3D organization of the thin-form Z-disc was generated in Fiji and shown in Chimera (Figures 7H and 7I). A local 3D model of one thin filament with its neighboring thin filaments and α-actinins was generated by fitting models of actin filament and plotting back a map from the crystal structure of α-actinin (PDB: 4D1E) according to their 3-dimensional positions and orientations using tools developed as part of the TEMPy software package (Figure 7K).

61Sub-volumes at the position of α-actinin were extracted from the 4x binned tomogram with a box size of 80 pixels (560 Å). An initial reference was obtained by averaging all 384 sub-volumes without refinement using PEET based on their orientation calculated from the coordinates of the α-actinin and the connected thin filaments. These sub-volumes were aligned to this reference during 3D refinement using RELION to generate a final average. The central rod domain from the crystal structure of α-actinin (PDB: 4D1E) was fitted into the averaged density manually. The actin binding domain, together with actin filaments was fitted into the density based on the atomic model of actin filament decorated by the first calponin homology (CH1) domain of filamin A (PDB: 6D8C), which shares a similar structure to the CH1 domain of α-actinin (Figures 5L and S7F).

62Statistical parameters are reported in figures and figure legends.