Three-Dimensional Organization of Troponin on Cardiac Muscle Thin Filaments in the Relaxed State

Abstract

Muscle contraction is regulated by troponin-tropomyosin, which blocks and unblocks myosin binding sites on actin. To elucidate this regulatory mechanism, the three-dimensional organization of troponin and tropomyosin on the thin filament must be determined. Although tropomyosin is well defined in electron microscopy helical reconstructions of thin filaments, troponin density is mostly lost. Here, we determined troponin organization on native relaxed cardiac muscle thin filaments by applying single particle reconstruction procedures to negatively stained specimens. Multiple reference models led to the same final structure, indicating absence of model bias in the procedure. The new reconstructions clearly showed F-actin, tropomyosin, and troponin densities. At the 25 Å resolution achieved, troponin was considerably better defined than in previous reconstructions. The troponin density closely resembled the shape of troponin crystallographic structures, facilitating detailed interpretation of the electron microscopy density map. The orientation of troponin-T and the troponin core domain established troponin polarity. Density attributable to the troponin-I mobile regulatory domain was positioned where it could hold tropomyosin in its blocking position on actin, thus suggesting the underlying structural basis of thin filament regulation. Our previous understanding of thin filament regulation had been limited to known movements of tropomyosin that sterically block and unblock myosin binding sites on actin. We now show how troponin, the Ca²⁺ sensor, may control these movements, ultimately determining whether muscle contracts or relaxes.

Introduction

Muscles contract by a mechanism in which myosin cross-bridges (heads) from the thick filament move by cyclic attachment and detachment along actin-based thin filaments. To function normally, muscles must possess on-off switching mechanisms to regulate cross-bridge / thin-filament interactions and hence contraction. The regulatory systems that control contraction vary in different muscles. In vertebrate skeletal and cardiac muscles, the thin filament binding proteins, troponin and tropomyosin, form the regulatory switch, which blocks cross-bridge attachment and cycling at low intracellular Ca²⁺-concentration, leading to muscle relaxation. Upon excitation and the consequent rise in free Ca²⁺, this constraint is released and contraction follows (reviewed in (1)).

Current understanding of the structural mechanism of troponin-tropomyosin regulation comes largely from x-ray diffraction of intact muscle (2–4) and from electron microscopy (EM) and three-dimensional (3D) reconstructions of isolated and reconstituted thin filaments (5–7). These and other studies show that end-to-end bonded tropomyosin molecules form strands that run along the helically arranged actin subunits of the thin filaments (8). Each tropomyosin molecule contacts seven successive actin monomers every 385 Å and binds one troponin complex, giving a 1:1:7 stoichiometry to actin (1,7). Troponin functions to couple Ca²⁺-concentration changes to azimuthal movement of tropomyosin on the thin filament. Tropomyosin’s location on actin controls the access of cross-bridges to the thin filaments and thus regulates the cross-bridge cycling that drives contraction. At low Ca²⁺, tropomyosin is held by troponin at a location that sterically blocks myosin binding sites on actin, thus producing relaxation; this is the blocked or B-state of the thin filament (5–9). Thin filaments are switched on when Ca²⁺ binds to troponin, which moves tropomyosin to the closed or C-state position, where the myosin binding sites are partly uncovered. Myosin binding to the thin filament also alters the position of tropomyosin, and full activation of the thin filament requires binding of both calcium and myosin (10–12).

Troponin is a complex of three subunits. Troponin-I (TnI) inhibits actomyosin ATPase; troponin-C (TnC) binds Ca²⁺; and troponin-T (TnT) links the complex to tropomyosin (1). The narrow 180 Å-long N-terminal tail of TnT (TnT1) runs along roughly half of the tropomyosin’s length (i.e., over three actin subunits), whereas its C-terminal region (TnT2) is incorporated into the more globular end of troponin, where TnI and TnC are located (1, 7, 8; Fig. 1). Crystal structures (13,14) of this core-domain of troponin show a W-like structure (the TnIT arm) where each of the four segments of the W represents a different TnI or TnT helix, with the central two segments forming a TnT-TnI coiled-coil (the IT-helix). TnC is a bilobed structure with globular N- and C-domains connected by a single helix. In core-domain crystal structures, the C-lobe of TnC is mounted on top of the TnIT scaffold (Fig. 1); while, in a Ca²⁺-dependent manner, the N-lobe interacts with regulatory domains of TnI (13,14). Different troponin crystal structures indicate that the bulk of the core domain, particularly the TnIT arm, is largely conserved, whereas the relative location of the N-lobe of TnC to the TnIT arm varies, possibly due to differences in crystal packing or effects of Ca²⁺ binding (13,14). Tropomyosin acts as a molecular ruler, localizing troponin at discrete sites on the thin filament surface. This arrangement is responsible for the 385 Å periodicity of troponin core domains detected on thin filaments, a periodicity that is reinforced by the in-register alignment of tropomyosins (and thus troponins) on the two actin helices on opposite sides of the thin filament.

Figure 1.

Thin filament cartoon showing the troponin core domain crystal structure on actin-tropomyosin (adapted from (15)). Note: bilobed TnC, particularly its C-lobe, is mounted on top of the W-like TnIT arm; also note the central location of the TnIT-helix, a coiled-coiled conserved in crystal structures and formed from TnI and TnT (13,14). One end of TnI (the TnI mobile domain) emerges from the core domain, and is depicted as a spiral crossing the cleft between the two actin helices to bind to the closest azimuthally neighboring actin subunit (15). The first part of the tail of TnT (TnT1) is drawn as an arrow beginning at the N-terminal end of TnT2 and is shown running parallel to tropomyosin.

Although the organization and regulatory movements of tropomyosin are well known from x-ray and EM studies, the location and orientation of troponin, and thus its interactions with tropomyosin and actin, are poorly understood. Attempts to determine the organization of the core domain and TnT1 by EM and 3D reconstruction have yielded inconclusive and contradictory results (16–19): such reconstructions did not retain features that were easily recognizable as troponin, making identification and docking of troponin subunits ambiguous (16,17,19). There were several reasons for these difficulties (1). The symmetry of troponin organization on thin filaments differs from the helical symmetry of actin and tropomyosin. Thus, standard helical reconstruction protocols used for actin filaments (20,21) are inappropriate for revealing troponin, as they are based on actin helical symmetry. The helical symmetrization used in these procedures treats all actin monomers and any associated densities as equivalent. Because troponin is not equally distributed on each actin subunit along the thin filament, it is averaged out and not reconstructed (2). In principle, single particle reconstruction of thin filament segments comprising a full repeating unit of the thin filament (14 actin monomers and two tropomyosin-troponin complexes) overcomes these problems. However, poor visualization of the relatively weak troponin density (molecular mass ∼80 kDa) in thin filament electron micrographs has made accurate identification of its position impossible. Thus, the selection of identical segments from thin filaments that is required for single particle reconstruction was compromised. Moreover, the conformational dynamics of troponin and plasticity of its interactions with tropomyosin and actin also probably hamper a high resolution reconstruction of its density on the thin filament. In previous work, we approached the problem by creating different models and positions of troponin on thin filaments and scoring the models by cross correlating to thin filament EM data. This method located the general position and polarity of the troponin core domain on actin-tropomyosin. However, when the best model was used as an initial reference for thin filament single-particle reconstruction, the troponin signal degraded during iteration and then disappeared, suggesting that either the modeling or the EM data was not optimal (18).

In the current extension and refinement of our previous work, we have developed methods to overcome the obstacles to single-particle EM reconstruction of thin filaments, enabling us to reveal key features of troponin organization. We first developed improved methods to isolate and negatively stain native cardiac thin filaments that display much more robust and regular troponin densities than seen previously. These native cardiac filaments did not suffer from artifacts sometimes encountered with filaments reconstituted from engineered or tissue purified proteins. We also established new, to our knowledge, protocols to preprocess EM images to determine the axial position of troponin and allow us to acquire the near identical filament segments necessary for reconstruction. Following these preliminary steps, we have applied single particle methods (without imposing actin helical symmetry) enabling us to reconstruct thin filaments showing clear and recognizable troponin density.

Single particle methods normally require a starting reference model for initial image alignment and reconstruction, and the final result can suffer from bias toward this reference. We have avoided this problem by comparing reconstructions obtained using multiple different starting models. In one, we used a featureless sphere to represent troponin. In others, we used variously oriented and positioned core domain structures. All of the models produced the same final structure, demonstrating that the reconstructions had little or no model bias. Each of the independently generated reconstructions, as well as their average, showed strong actin and tropomyosin densities, together with consistent troponin density and polarity relative to actin. These features made possible a low resolution docking of troponin crystal structures to the reconstruction, enabling an evaluation of the molecular orientation of troponin and assessment of previous conclusions based on in vitro regulatory protein interactions (22,23) and fluorescence polarization studies on muscle fibers (24). Of importance, our reconstructions also showed regions of troponin that are not present in crystal structures and that were not a feature in the initial reference models. These include densities attributable to the TnT1 tail running adjacent to tropomyosin, and the C-terminal mobile domain of TnI, which is considered responsible for pinning tropomyosin in the blocked state. By defining the geometry of troponin on thin filaments, our results suggest the first, to our knowledge, detailed structural mechanism for understanding muscle relaxation and activation.

Materials and Methods

Detailed methods are provided in the Supporting Material text. In brief, native thin filaments were isolated from porcine hearts (25,26), negatively stained, and EM performed as previously ((27–29), see Fig. S1 in the Supporting Material). Image processing was carried out on 755 filaments as outlined in Fig. 2. Following filament unbending using ImageJ (30,31)(Fig. 3, a–c), power spectra were computed as a first screen to select troponin-decorated filaments for use in the reconstruction and to eliminate filaments deficient in troponin. 72 filaments lacking a 385 Å meridional reflection were assumed to have lost their troponin and were not considered further (Fig. 3, d–f). Cross correlation against a model structure was used to determine up-down filament polarity (32), and filaments were oriented with pointed ends facing up in subsequent work. Estimation of troponin axial positions on filaments, essential for successful single-particle reconstruction, was carried out by autocorrelation and cross correlation steps (Fig. 4), and filaments then computationally divided into equal length segments with troponin centered in each segment. One-dimensional density profiles of troponin-containing segments showed extra density due to the presence of troponin (Fig. 4, i and j), and the absence of this indicator was used as a second screen to eliminate troponin-deficient filaments; 138 more filaments lacking troponin were eliminated by this step.

Figure 2.

Flow chart outlining the reconstruction protocol.

Figure 3.

EM and image analysis of thin filaments. Selected filaments following straightening (a, b, c) together with corresponding power spectra (d, e, f; average of 10 filaments of each type). In (a) periodic troponin densities (arrows) lie at the sides of filaments and give rise to an ∼385 Å meridional reflection (based on calibration against the 59 Å layer line) in the power spectrum (d, arrow). In (b), troponin is not seen on filaments, although its presence is inferred from the 380 ± 8 Å reflection in the power spectrum (e, arrow); in this case it is presumed to be located on the top and bottom of the filaments. Ten percent of the filaments showed no signs of troponin in either images or Fourier transforms (c; f, arrow); these had presumably lost most of their troponin and were not considered further. Scale bar represents 500 Å.

Figure 4.

Troponin axial positions determined by cross correlation. (a and e) Unbent filaments with troponin bound to the sides and top/bottom, respectively (cf. Fig. 3, a and d and Fig. 3, b and e). (b and f) Filaments in (a and e) filtered from 1/450 Å⁻¹ to 1/300 Å⁻¹. (c and g) Autocorrelation functions of (b and f); the autocorrelation function has a peak in the center as it is a correlation of a function with itself. (d) Cross correlation function between (c and b). The peaks in this function point to the troponin axial positions seen directly in the filament (a), as shown by the black arrows. (h) Cross correlation between (f and g). The troponin axial positions can be estimated from the peak positions, as in (d). (i and j) Average segment density profiles corresponding to the filaments in (a and e), respectively, after determining troponin axial positions. The two highest peaks, at the center and right (arrows), indicate troponin positions in a segment of 160 × 160 pixels. The asymmetry of these profiles (oriented with filament pointed end to left) confirms that troponin is present, and the central positions of the peaks show that troponin axial positions are correctly determined.

The relative rotation of filament segments about their long axis was determined by matching against two-dimensional projections of a reference structure (21,33,34), built from models of actin, tropomyosin, and troponin (14,35–39)(Fig. 5). Back projection using the rotation angles thus determined produced an initial 3D reconstruction. The reconstruction was used as a new reference and the process iterated 80 times. To increase signal/noise ratio in the reconstruction, the two troponin densities on the two sides of the filament were aligned and averaged in each cycle. Reconstructions were made beginning with each of 10 different reference models (Fig. 5) and these reconstructions averaged. Fig. S2 shows that all angles were represented. Alignment of atomic structures to the reconstruction was carried out using Chimera (38).

Figure 5.

3D reconstructions were computed using different models to exclude model bias in single particle analysis. Models (top two rows) were generated by using actin-tropomyosin (magenta) and either a sphere for troponin, or by translocating, rotating, tilting, or truncating the troponin core domain (yellow) compared with the starting model of Pirani et al. (18). (a) Pirani model. (b) Sphere model. (c) Core domain rotated anticlockwise 30°. (d) N-lobe of TnC removed from core domain. (e) Lower part of core domain tilted 20° toward the filament axis. (f) Upper part of core domain tilted 20° toward the filament axis. (g) Core domain translocated −20 Å along the filament axis. (h) Core domain translocated 20 Å along the filament axis. (i) Core domain rotated clockwise 20° azimuthally around the filament axis. (j) Core domain rotated anticlockwise 20° azimuthally around the filament axis. (k–s) 3D reconstructions using models (a–i), respectively. (t) The reconstruction from model (j), shown in red, diverged quickly even for F-actin and tropomyosin. Of all the models shown, (j) was the only one that did not converge, viz., did not show densities attributable to troponin or a continuous tropomyosin strand. This is consistent with the core domain in this model being furthest away from the consensus converged position (∼70 Å at the lower end of the core domain), leading to rapid degradation of the reconstruction. Note the occasional appearance of spurious density in some of the reconstructions. We assume this to be noise because, using the same data set, its position, shape, and presence varies depending only on the reference model used. In addition, as expected of noise, this density disappears both at higher contour levels (whereas the troponin density remains) and in the averaged reconstruction (while the troponin signal is augmented; see Fig. 6b).

Results and Discussion

EM and 3D reconstruction of thin filaments

Porcine cardiac muscle thin filaments in low Ca²⁺ relaxing solution were well separated and showed minimal background material when observed by negative staining (Fig. S1). They displayed helically arranged actin-subunits and, in many cases, prominent densities (Fig. 3 a), projecting at 380 ± 8 Å intervals (Fig. 3 d), which are the core domain of the troponin complex.

To determine troponin organization, images of filaments were divided computationally into equal length segments and 3D reconstructions carried out by single particle analysis (cf. 21,32–34). To ensure preservation of troponin density in the reconstruction, helical averaging was avoided, and the segments treated as true, asymmetric single particles. Single particle analysis is normally performed on individual, identical macromolecular complexes. When it is carried out on thin filaments, the segments used must mimic true single particles, and ideally will meet the following requirements: 1), be completely or near completely decorated with troponin; 2), show multiple views of troponin; 3), have known polarity; 4), have troponin positioned identically in each segment. Our protocol (Figs. 2–4) allowed particles containing troponin to be selected, accurately aligned, and properly averaged (see Materials and Methods and Detailed Methods in the Supporting Material).

Single-particle reconstructions can suffer from bias toward the reference model used to initiate the reconstruction. To circumvent model bias, we used a variety of starting models to determine the angle of rotation of each troponin-containing segment about its own axis and to refine its axial position. These reference structures were built from an atomic model of F-actin-tropomyosin (37) together with various models of troponin (13,14) positioned equivalently on two azimuthally adjacent actin subunits on opposite sides of the actin double helix (Fig. 5, a–j). Each model was rotated about its long axis and projections made every 4°. The rotational angle of each experimental filament segment was then determined by finding the best match to the 90 reference projections, and these angles were used to calculate a backprojection. The reconstruction thus produced was used as a new reference, and the process iterated for 80 cycles. The position and appearance of actin, tropomyosin, and troponin in the reconstructions generally stabilized by the tenth round and remained stable through the 80 cycles, generating an ∼25 Å resolution map according to the 0.5 Fourier Shell Correlation threshold criterion (Fig. S3). Ten separate reconstructions, each initiated from a different starting model, were produced in this way (Fig. 5, k–t).

In the first model, troponin was represented by a sphere comparable in mass to the core domain (Fig. 5 b). This produced a reconstruction showing actin, tropomyosin, and extra densities on each side of the actin helix that are attributable to troponin (Fig. 5 l). The helically arranged actin subunits were well defined, with clear subdomain structure, and were associated with continuous tropomyosin strands following the two actin helices (located in the blocking position, over actin subdomains 1 and 2). Troponin features included a mass slanting at ∼50°, and extending over parts of two adjacent actin subunits, together with a well-separated smaller globular density. Although weak, these densities nevertheless resembled features expected of troponin, viz., the TnIT arm and the N-lobe of TnC (see Fitting crystal structures of the troponin core domain (below) and Figs. 6 and 7). These densities (and their refinements noted below) differed significantly (by ∼15 Å axially and ∼30° rotationally) from their location in our earlier work ((18), the model shown in Fig. 5 a).

Figure 6.

Surface views of thin filament reconstructions (pointed end facing up). (a) Control F-actin with subdomains marked on one subunit. (b and c) native thin filaments. In (b), the reconstruction was made using the average of individual reconstructions (Fig. 5, k–s), as a reference. Densities resembling the TnIT arm and N- and C-lobes of TnC are labeled. Tropomyosin strands are marked with black arrows and the putative C-terminal TnI extension with a double-sided arrow. (c) Higher threshold version of (b) showing that the tropomyosin strand is wider at the barbed end of the filament than at the pointed end, an indication of the additional presence of TnT1 at the barbed end of the core domain; white arrows indicate the relatively narrow tropomyosin strand at the pointed end clearly bridging over actin subdomain 2, leaving a gap which, in the lower actins (yellow arrows), is filled with extra density. (d and e) cross sections through the top (d) and bottom (e) of the reconstruction in (c), again indicating the increased width of the strand in the region below the core domain (yellow arrows) compared with above (white arrows); actin subdomains numbered. (f) Difference map (gold), formed by subtracting actin-tropomyosin densities from the map in (b) and then superposing it on a map (purple) generated by imposing helical symmetry on (b), highlights densities attributable to troponin; in addition to the core domain, densities attributable to TnT (open arrow), the C-terminal mobile domain of TnI (double-sided arrow), and possibly the cardiac muscle-specific N-terminal TnI chain (bracket) are visible. (g) Reconstruction of F-actin-tropomyosin decorated with an 80 residue long fragment representing the C-terminal mobile domain of TnI (from (15)). In this helical reconstruction, the fragment binds equivalently to every actin subunit; arrows point to the TnI density on one actin subunit. (h) The troponin difference densities in (f) superposed on the map in (g) showing that the putative mobile domain in native filaments overlaps with the TnI peptide in the reconstituted filament. Note: TnI and TnT difference densities appear larger than actual relative to the TnIT arm density due to the low surface contour used to depict the differences, as well as possible mobility of these regions. The resolution of the full reconstruction in (b) was 25 Å, although those of the troponin and of the actin-tropomyosin parts of the map were 32 Å and 21 Å, respectively (Fig. S3). The latter compared well to previous 18 Å reconstructions of actin-tropomyosin (42).

Figure 7.

Fitting troponin core-domain crystal structures to the 3D reconstruction. (a and b) Two orthogonal views of the thin filament with the low Ca²⁺ core domain (14) fitted into corresponding densities, showing back and front views of the fitting. The core domain is shown in ribbon view with TnI, cyan, TnT, yellow, and TnC red (cf. Fig. 1). Note the ∼50° orientation of the TnIT helix of the troponin core domain relative to the filament long axis. (c) In addition to the core domain, the atomic model of actin-tropomyosin (37) was fitted to the reconstruction and substituted for corresponding densities shown in (b). Troponin difference densities made translucent are superposed. The tropomyosin structure is shown in ribbon view and actin subunits in space-filling view; actin residues 222, 226, and 311 are highlighted in red on one actin subunit lying under the putative TnI mobile domain.

Because troponin features in the first reconstruction were weak, we decided to test their validity by carrying out additional reconstructions using filaments with variously oriented and positioned core domains as reference models. Nine models were built with the low Ca²⁺ core domain structure (14) (filtered to 20 Å resolution) substituted for the sphere and located at different axial and azimuthal positions on actin, or with different rotations and tilts, or with part of the core domain missing (Fig. 5, a, c–j). The first used the arrangement of troponin we had determined in earlier work using reconstituted filaments and a different structural approach ((18), Fig. 5 a). The others differed from this model by movement of troponin by ±20 Å and 20° to 30° from its initial position. No attempt was made to include TnT1 or the C-terminal mobile regulatory domain of TnI in any of the reference models, as crystal structures for these domains are not available (13,14,40). Despite the substantial differences in the starting models (Fig. 5, a, c–j), troponin density in the final reconstructions converged to a new consistent shape, site, and orientation on actin (Fig. 5, k, m–s). This displayed clear and recognizable features of the core domain similar to those obtained with the sphere model (Fig. 5, b and l; note the evolution of the final structure from one of the starting models is shown at different iterations in Fig. S4). The consensus position of troponin in these reconstructions was up to ∼45 Å away from its position in the starting models. Rotation and tilt of the core domain relative to the filament axis also differed dramatically from that in the different models by up to 100° (e.g., Fig. 5, c and m, Fig. S4). The key conclusion from this common organization of troponin in the different reconstructions is that the initial model did not significantly bias the final structure. In one initial model (Fig. 5 j), where troponin was positioned furthest (∼70 Å) from this consensus position, troponin, actin, and tropomyosin densities were degraded after three cycles of iteration (Fig. 5 t), and thus did not even converge on a well-defined F-actin-tropomyosin structure. Similarly, when the sphere model was centered at a similar distance from the consensus site, the reconstruction rapidly degraded (data not shown). Failure of the latter reconstructions to converge (when the troponin model is far from its consensus position) implies that troponin contributes significantly to the alignment of segments, and further supports the validity of the consensus organization arrived at from the other nine models. The nonconvergent reconstructions were discarded.

Analysis of troponin features in the thin filaments

To enhance shared features and decrease noise, reconstructions in Fig. 5, k–s, were averaged and the average used as a new reference to build a final reconstruction. The troponin mass in this reconstruction again had an outline resembling the troponin IT arm, the TnC C-lobe protrusion on the upper part of the arm, and the N-lobe of TnC (Fig. 6 b, cf. Fig. 1, Movie S1). The two troponin densities lay on the outer aspect of the central pair of azimuthally neighboring actin subunits, on subdomains 1 and 2, and were quite distinct from actin and tropomyosin densities at lower radius. The putative TnIT arm was the strongest feature, consistent with its mass in core-domain crystal structures, and was oriented obliquely and tilted away from the filament axis (cf. (24)). This density appeared to be attached to the tropomyosin strand and was centered over, but not obviously in contact with, actin subdomain 2 (Fig. 6 b, Fig. 7, a and b). A density most simply attributable to the N-lobe of TnC (see section on alignment below) was seen on the upper surface of actin subdomain 1, to the side of the TnIT arm. Density putatively coming from the TnT tail (TnT1), thought to be a single α-helix over much of its length (41), was suggested by a greater width of the tropomyosin strand over ∼3 actin subunits from the base of the core-domain in the direction of the filament’s barbed end (Fig. 6 c, yellow arrows), compared with the pointed end (white arrows). The widening can also be seen in filament cross sections (Fig. 6, d and e). This orientation of TnT1 from the base of the core domain toward the filament’s barbed end was supported by analysis of one-dimensional projection profiles of averaged filament segments (Fig. 4, i and j). In addition, extra mass is noted adjacent to the TnC N-lobe on actin subdomain 1, possibly showing the C-terminal region of TnI (Fig. 6 b, discussed below).

Difference density analysis

To illustrate more clearly the features of troponin suggested by the reconstruction, these features were analyzed further by computing a difference map between the reconstruction and a reconstruction lacking troponin (Fig. 6 f). In the difference map, the troponin core domain is positioned on the outer side of the upper part of the elongated TnT1-like density. This is seen to run alongside tropomyosin in the direction of the barbed end of the filament, confirming the position of the TnT-like density and the orientation of the troponin complex deduced in the original reconstruction. Based on the difference map, the entire complex is ∼200 Å long. This is ∼60 to 70 Å shorter than the length estimated in other studies (41,43), possibly because the N-terminal end of TnT1 is not resolved from actin and tropomyosin in the reconstruction. This part of TnT varies greatly between TnT isoforms (41) and may not form a discernible mass on the thin filament. In contrast, a short extra density runs above the core domain, and may be the C-terminus of TnT, which is absent from troponin crystal structures (13,14). We speculate that this end of the TnT2 might also interact with tropomyosin and/or TnI (cf. 44–46).

A second region of density, separate from actin and tropomyosin, appears to emerge from behind the putative N-lobe of TnC, on the extreme outer edge of actin subdomain 1, to form an obliquely oriented tangent to the neighboring actin monomer on the opposite F-actin helical strand (Fig. 6 f, double-sided arrow, cf. the original reconstruction in Fig. 6 b, where this density is seen to fill the cleft between two azimuthally adjacent actin subunits observed in Fig. 6 a). This density is present only on the actin subunits at the level of the troponin core domain and is absent from neighboring actins above and below (Fig. 6 b, Movie S1). It has the same location, dimensions, and orientation as density found in a previous reconstruction of F-actin-tropomyosin decorated with an 80 amino acid C-terminal construct of TnI, representing the TnI mobile regulatory domain (15) (Fig. 6 g). This construct inhibits actin-tropomyosin stimulated myosin ATPase by binding to troponin-free F-actin-tropomyosin in a 1:1 molar ratio to actin, constraining tropomyosin in the blocking position (15,47). Although the density attributed here to the C-terminal domain of TnI is weak, its coincidence with that in the C-terminal TnI-decorated filaments (Fig. 6 h) suggests that we are observing the native organization of the C-terminal TnI regulatory peptide in the current maps.

Fitting crystal structures of the troponin core domain

Additional insights into the reconstruction were gained by fitting atomic structures of subunits into corresponding features of the EM density map (Fig. 7). An actin-tropomyosin atomic structure (37) was first fitted to corresponding densities in the reconstruction, as previously (37). An approximate fitting of the low Ca²⁺ troponin core-domain crystal structure (14) was then carried out, using the asymmetric geometry of the TnIT arm, which is invariant in crystal structures (13,14), as the key guide to orientation and position. Although the volume of troponin in the reconstruction was smaller than the atomic structure, its major feature was similar in appearance to the TnIT arm, which is in turn the main component of the crystal structure. Similar asymmetric features seen in the crystal structure and reconstruction include the tapering of the TnIT arm density at its lower end and a characteristic bulge due to confluence with the C-lobe of TnC at its upper end (Figs. 7, a and b, Fig. S5). This asymmetry made it possible to fit the TnIT arm in only one plausible orientation; the only other reasonable way of alignment (by reversing its orientation) led to a poor match. A closely related modification of the crystallographic model, based on fluorescence polarization (24) fit equally well. There are several possible reasons why the volume of the TnIT arm in the average reconstruction did not fully envelop the corresponding atomic structure. These include inherent or stain-induced variability or disorder of troponin on native filaments, partial troponin dissociation, penetration of stain into troponin, or imperfect alignment in the reconstruction procedure. Although this ambiguity precludes high resolution atomic docking, we are nevertheless able to characterize the location and orientation of the major domains of troponin, and thus describe structure/function relationships.

In the average and all nine individual convergent reconstructions, a small globular feature was consistently observed to the right of the TnIT arm, on actin subdomain 1 (Figs. 5, 6 b, and 7 a). When the above mentioned atomic models (14) were aligned to the reconstructions based solely on the TnIT arm, the TnC N-lobe in the models was found to coincide with this density, strongly suggesting that it shows this domain of TnC. The reliability of this feature is strongly supported by its appearance in reconstructions even when absent from starting models (Fig. 5, b, l, d, and n). The C- and N-lobes of TnC in the fitted structure are separated by ∼22 Å, and the line joining them is oriented roughly perpendicular to the F-actin axis, consistent with fluorescence polarization studies (24). Because the N-lobe of TnC is attached to the rest of the core domain by a single helix that may be disordered at low Ca²⁺ (14), the position of the N-lobe on the thin filament may be variable. Such local disorder of the TnC N-lobe may account for its relatively small mass in reconstructions compared to that in atomic models and to that of the C-lobe density in reconstructions, as the C-lobe is firmly linked to the TnIT arm (Fig. 7).

The location of the core domain next to tropomyosin in the fitted structure positions the N-terminal end of TnT2 in the atomic model (Fig. 1, NTerm. TnT2) close to the start of the putative TnT1 density (Fig. 7, b and c), with which it is continuous (41), which further supports the proposed alignment. Because the ends of tropomyosin and the troponin complex are not resolved in the reconstruction, alignment of the troponin core domain to any particular tropomyosin pseudorepeat on actin is ambiguous. However, the likely arrangement can be inferred from published cross-linking data, binding studies, and antibody competition analysis (22,23). Accordingly, the troponin core domain was placed next to tropomyosin pseudorepeats 3 and 4 on the actin-tropomyosin model (37) so that TnT residues in the low Ca²⁺ crystal structure (14) would be closest to their natural target (near residue 174 on tropomyosin (22,23)). This alignment orients the tip of TnT2 toward subdomain 1 of actin (Fig. 7 c, see models of the troponin core domain position in Fig. S6). This arrangement is in agreement with the model of Jin and Chong (22), in which troponin binding to tropomyosin occurs at two major sites, one via interactions with TnT1, as mentioned above, and the other by contacts with the N-terminal end of TnT2, as indicated here.

In the low Ca²⁺ crystal structure, the bulk of TnI ends behind the C-lobe of TnC, but the C-terminal TnI regulatory domain emerging from this point is mobile and not resolved (14). It is at this position in the reconstruction that the putative TnI extension discussed earlier is expected to begin. Mapping the extension onto the atomic model of actin-tropomyosin shows that it bridges the cleft between azimuthally adjacent actin subunits on opposite strands of the double helix, extending from subdomain 1 of the actin from which it originates to subdomain 4 of the neighboring actin in the direction of the filament barbed end (15) (Fig. 7 c). The extension, known to contain an excess of basic residues, then passes over the region of actin containing helix 222–230 (including acidic surface residues Asp-222 and Glu-226, shown as red spheres on actin subdomain 4 in Figs. 7 c, Fig. S6), and ends on actin subdomain 3 near Asp-311. Here, the tip of the extension abuts tropomyosin and/or TnT1 on the inner edge of actin subdomain 1, where it could constrain tropomyosin on the opposite side of the filament in the myosin blocking position (Fig. 7 c). We note that TnT1 lying alongside may augment the rigidity of tropomyosin in addition to strengthening its end-to-end bonding (48,49).

Tissue- and species-specific peptide extensions frequently distinguish troponin subunits in different muscles. For example, mammalian cardiac muscle TnI contains an N-terminal extension absent in the skeletal isoform (50,51). Phosphorylation of this extension by protein kinase A modifies Ca²⁺ affinity of the N-lobe of TnC, presumably by its interactions with this domain (51). Narrow density emerging from the top of the core domain can be seen to approach the TnC N-lobe and may be this cardiac-specific TnI extension (bracket in Fig. 6 d).

Summary: Functional implications of troponin organization

Our single-particle reconstruction of native cardiac thin filaments clearly reveals troponin and shows greatly improved detail over previous models, including features not present in crystal structures. Multiple distinct reference models, with different troponin definition, starting positions, orientations, and structures all led to very similar final locations and orientations of troponin in the reconstructions, supporting the reliability of our model. Although alignment of crystal structures to the reconstruction (Fig. 7) is approximate due to the moderate resolution (±5° to 10° error in any direction), fitting of major domains is unique; when the core domain is reoriented, for example by inverting or rotating it to switch the positions of the TnIT arm and TnC, or by turning it back to front to reverse the location of the lobes of TnC, none of these variants fits sensibly into the EM volume. The appearance of troponin is qualitatively the same as that displayed by rotary shadowed molecules, which first demonstrated the characteristic structure of a globular head (the core domain) at the end of a narrow stalk (TnT) (43). Our reconstruction yields additional detail not resolved by rotary shadowing, or in any previous thin filament reconstruction, and is in good agreement with crystal structures of the core domain. The orientations of TnT1-like and core-domain densities provide clear evidence for the polarity of troponin on the thin filament consistent with that proposed originally (43,52), and in contrast to reports suggesting the opposite (17,19). In addition, the ∼50° angle of the TnIT arm (Fig. 7) relative to the filament axis supports the conclusion of Knowles et al. (24), based on spectroscopic data from intact muscle, relating orientations of specific probes on troponin to the axis of the filament. Although the spectroscopic methods do not reveal troponin polarity or azimuthal or axial positions on thin filaments, our reconstruction is most consistent with the orientation of the TnIT arm in Model 2 rather than Model 1 of Knowles et al. (24). Our reconstructions do not have sufficient resolution to distinguish differences in rotation of the N-lobe of TnC within the core domain, which is what discriminates the Vinogradova and Knowles models (14,24), nor to pin-point exact amino acid to amino acid contacts between troponin and actin or tropomyosin.

3D reconstruction from electron micrographs has yielded abundant evidence of steric control of actin-myosin interactions by tropomyosin (5–7). In contrast, 3D EM relating troponin structure to tropomyosin regulatory function has been ambiguous, and the molecular mechanism of troponin-tropomyosin regulation has remained unresolved. Here, we have visualized the troponin complex at ∼25 Å resolution, and, through this, have been able to infer its influence on tropomyosin position in the relaxed state, providing new, to our knowledge, insights into the structural basis of troponin function. Our results show that the troponin core domain is mounted on top of the TnT1 tail where it forms a scaffold for regulatory domains of TnC and TnI. Our results on native filaments support previous suggestions based on synthetic filament constructs (15) that, at low Ca²⁺, TnI bridges azimuthally adjacent actin subunits across the filament, interacting with tropomyosin or TnT1 on the opposite actin helix from its origin in the core domain. The end of the C-terminal mobile domain of TnI appears to target actin subdomain 3, where it is likely to constrain TnT1 and tropomyosin to positions on actin that sterically interfere with cross-bridge binding and cycling, thus leading to muscle relaxation. In this scheme, blocked-state tropomyosin would be wedged between two troponin complexes, on one side by the C-terminal TnI domain (and TnT) and on the other by the troponin core domain, suggesting a functional interaction between the two sides of the thin filament (Fig. 1). We propose that this structural inhibition is reversed during muscle activation, following Ca²⁺-binding to the N-lobe of TnC. The TnI mobile domain would then be attracted to the TnC N-lobe and less so to actin, favoring release of tropomyosin from its blocking position, leading to myosin cross-bridge cycling on actin (1,39,50,51). To our knowledge, these new insights into troponin organization and its relation to actin and tropomyosin provide a framework for a deeper understanding of the structural mechanics of the thin filament and its role in contractile regulation.