Cryo-ET comparison of the hierarchical ultrastructure of silkworm, spider, and artificial silk fibers

Abstract

Spider and silkworm silks are renowned for their exceptional mechanical properties, which arises from their ultrastructural organization. However, this architecture remains incompletely understood. Here, we apply cryo-electron tomography to examine the hierarchical organization of silkworm, spider, and artificial silks. In silkworm silk, we observe nanofibrils of ~3.6 nm in diameter, interconnected by abundant bridges and representing the smallest fibrillar features currently accessible by cryo-ET. These nanofibrils align with the fiber axis and are organized into a herringbone pattern, with stacked layers building the micron-scale filament. Spider silk displays densely packed nanofibrils with near-perfect axial alignment and minimal voids. In contrast, silkworm silk shows regionally heterogeneous gaps, whereas artificial silk lacks the ordered packing characteristic of natural materials. These observations provide a structural basis for understanding silk formation and may guide future biomimetic fiber design.

Silk mechanics depend on how proteins are arranged inside fibers, yet this organization is not fully resolved. Here, authors apply cryo-electron tomography to silkworm, spider and artificial silks, uncovering aligned nanofibrils and distinct differences in packing density

Introduction

Natural silk, primarily derived from silkworms and the spider major ampullate gland, exhibits excellent mechanical properties, including high tensile strength and toughness. This gives them not only tensile strength comparable to that of high-performance steel or Kevlar but also exceptional toughness^1–3. For millennia, silk has been prized in textiles for its smooth texture, luster, and strength. More recently, due to its extraordinary mechanical properties, as well as biocompatibility, biodegradability, and eco-friendly manufacturing methods, silk fibers are becoming a candidate material for biomedical engineering^4–7, flexible electronic wearables^8,9 and data storage medium¹⁰, attracting increasing attention^11–20.

The superior mechanical properties of silkworm silk and spider silk are closely tied to their ultrastructure, specifically the morphology, size, and hierarchical assembly of the fundamental building blocks within the fibers. Decades of research have focused on understanding how silk proteins, stored in a soluble state in the silk glands, transform into insoluble solid fibers with distinct mechanical properties. Two primary models, liquid crystal spinning^21–25 and micellar spinning²⁶, have sought to explain this process. The liquid crystalline model proposes that spinning dopes in the spider gland and duct adopt a nematic phase, characterized by rod-like structures that are considered aggregates of spherical silk proteins^21,23. In contrast, the micelle model suggests that natural silk fibroin (NSF) forms micelles, approximately 100–200 nm in diameter, in solution due to its amphiphilic primary sequence^26,27. As the NSF concentration increases, these micelles coalesce into globules ranging from 0.8 to 15 µm in diameter, which subsequently align under shear forces to form fibers^26,28.

Despite these models, the precise structural units and assembly process of silk proteins in silk glands remain largely elusive. Our recent studies²⁹ have shown a multi-scale controllable self-assembly spinning mechanism based on the NSF in silkworm spinning dope, suggesting that flexible nanofibrils may represent key building elements, approximately 4 nm in diameter, rather than the previously suggested micellar structure^26,28,30,31 or rod-like structure formed by aggregation of globular fibroin molecules²³. Within the silk glands, these fibroin nanofibrils initially exhibit random orientation, but intriguingly, they align into a herringbone pattern in the anterior silk gland near the spinneret. This raises questions about whether this ordered structure persists in the final silk fibers or undergoes further transformation during the spinning process. Thus, it is essential to investigate the ultrastructure of silk fibers for a comprehensive understanding of the spinning process.

In addition to natural silk fibers, artificial silks have attracted increasing interest due to their potential for scalable production and customizable properties. Artificial silk is typically produced by regenerated silk fibroin^16,32 or recombinant silk proteins^{13–15,18,20} using techniques such as wet spinning^{13,15,16,18–20,32}, dry spinning^31,33, electrospinning¹⁴, interface wiredrawing³⁴ or microfluidic spinning^35,36. These processes aim to fabricate artificial silk fibers with structures and properties comparable to those of natural silk, but the properties of resulting artificial silk fibers are particularly influenced by factors such as protein sequence, spinning dope concentration, flow rate, humidity, post-spin stretching, buffer composition, etc^11,12,33. The ultrastructure within artificial silk is even less understood. Understanding these structural discrepancies is crucial for improving artificial silk production and bridging the performance gap between synthetic and natural materials.

Investigating the ultrastructure of silk fibers from the atomic to the macroscopic scales remains a major challenge. At the atomic level, previous studies using X-ray diffraction (XRD), Raman spectroscopy, neutron scattering, and solid-state nuclear magnetic resonance have established that silk is a semicrystalline polymer. Its crystalline regions are primarily composed of β-sheets formed by polyalanine, contributing to silk’s strength, while the amorphous regions, rich in glycine, provide the toughness of the silk fibers^37–41. However, a consensus on the hierarchical structure of silk from the nanoscale to the macroscale is still lacking. Various models based on atomic force microscope^42–47, scanning electron microscope (SEM)⁴⁸, and Anderson light localization⁴⁹ suggest the presence of nanofibrils with diameters ranging from 20 to 220 nm. Nevertheless, techniques like transmission X-ray microscopy⁵⁰ and ultramicrotomy⁵¹ have failed to detect nanofibrils, with some studies indicating silk fibers consist of spherical units, 10–200 nm in diameter, the size of which varies depending on the type of silk^{26,30,51–53}. Neutron scattering and small-angle XRD^40,54 suggest a long-range organization in silk fibers, although its nature and relationship to the protein sequence remain unclear.

Despite significant progress, many aspects of the supramolecular structure and assembly process of silk remain unresolved. One major challenge is the lack of suitable techniques for studying the native structure of silk fiber in vivo. Furthermore, conventional sample preparation methods, such as chemical fixation, dehydration, resin-embedding, and heavy metal staining, often introduce artifacts that result in overly condensed structures, obscuring the fundamental units of silk. The intrinsically dense structure of silk further complicates efforts to achieve its high-resolution ultrastructural characterization.

In this study, we combined cryo-focused ion beam (cryo-FIB) milling with state-of-the-art cryo-electron tomography (cryo-ET) to investigate the ultrastructure of force-drawn silk fibers from silkworms, spiders, and that of artificial silk. Cryo-FIB is a cutting-edge technique used to prepare high-quality samples for structural and ultrastructural analysis, particularly in the context of cryo-ET. This method preserves the native structure, enables high-resolution imaging by producing ultra-thin sections, and minimizes artifacts typically associated with conventional sample preparation methods. Our study uncovered the morphological and hierarchical structure of fibroin within silk fibers with unprecedented level. The results revealed that the fundamental structural unit of silk is nanofibrils approximately 3.6 nm in diameter, which share the same morphology as those found in the silk gland. These nanofibrils are arranged in parallel along the fiber axis, forming a herringbone pattern. Multiple layers of this herringbone pattern stack to form the final micron-scale fiber, indicating that the herringbone pattern formed near the spinneret remains stable during fiber formation. In spider silk, a more condensed structure adorned with nanofibrils is clearly observed, with their long axes tightly aligned in parallel along the flow direction, leaving no visible gaps. In contrast, silkworm silk exhibits larger, regionally varying gaps between nanofibrils, while artificial silk lacks the highly ordered arrangement found in natural silks. In summary, gaining a deeper understanding of silk’s ultrastructure provides valuable insights into replicating and improving upon the natural spinning processes of silkworms and spiders, guiding the development of polymer-based materials with tunable properties.

Results

Structure of fibroin extracted from the silk glands of the silkworm

Fibroin, a key component of silk, is understood to exist either as a spherical macromolecular complex with a molecular weight of up to 2.4 MDa, composed of a fibroin heavy chain, light chain, and P25 in a molar ratio of 6:6:1⁵⁵, or as a rod-like structure formed by the elongation and aggregation of globular fibroin^23,30,52. In this study, NSFs were extracted from the lumen of the posterior silk gland of silkworms. Gel filtration chromatography and 4–16% SDS-PAGE analysis showed that NSFs are a macromolecular complex formed by the fibroin heavy chain, light chain, and P25 subunits (Supplementary Fig. 1a). Interestingly, 3–16% BN-PAGE analysis showed two distinct bands for freshly extracted NSFs, with higher molecular weight bands gradually transforming into lower molecular weight bands upon the addition of amphipol. This transformation displayed a dose-dependent relationship with amphipol concentration (Supplementary Fig. 1b). Metal shadowing analysis confirms that (Supplementary Fig. 1c–e).

For improved sample uniformity, NSFs were treated at a 1:6 molar ratio for 72 h (Supplementary Fig. 1b) and rapidly frozen via plunge-freezing. Cryo-ET revealed that NSFs exhibit typical fibrous protein characteristics, displaying high flexibility with a diameter of 4.0 ± 0.6 nm and a length of 32.7 ± 16.4 nm (mean ± SD) (Fig. 1a–c). While the diameters of individual nanofibers were highly uniform, the lengths varied significantly (Fig. 1c). Subtomogram averaging further revealed that fibroin nanofibers appear as a string of beads structure (Fig. 1d, Supplementary Fig. 2), and metal ions are essential for maintaining this structure²⁹.

Fig. 1. Natural silk fibroin is a beads-on-string structural nanofibril with a diameter of about 4 nm.

a Tomographic slice of NSF extracted from posterior silk glands. The yellow arrow indicates a typical nanofibril (top), with nanofibrils of varying curvatures shown below (yellow lines depict curved trajectories). b Tomographic slice of segmentation from the same region as in (a). c Nanofibril diameter and length measurements (n = 3 tomograms). d The 3D structure of the averaged NSF nanofibril in longitudinal view. e Predicted 3D structure of the fibroin heavy chain (Δ2303–2621) by AlphaFold 3.

Most studies suggest that the spherical structures in nanofibrils represent either fibroin monomers^23,56 or fibroin aggregates^{26,30,31,52,57–59}. The fibroin heavy chain consists of 12 repeat sequences and 11 linker sequences, with the repeat sequences containing a high proportion of GAGAGS and GAGAGY motifs that govern the mechanical properties of silk fibers, and the linker sequences showing high conservation (Supplementary Fig. 3), which provides evidence for the concerted evolution of silk protein genes^60,61. Since AlphaFold 3 can only predict polypeptide structures up to 5000 amino acids⁶², we deleted one repeat and one linker sequence from the fibroin heavy chain (Δ2303–2621) to predict its structure. The results indicated that the fibroin heavy chain (Δ2303–2621) can fold into a spherical structure measuring approximately 10.6 nm × 10.7 nm × 15.1 nm (Fig. 1e), which is significantly larger than the 4 nm × 4 nm × 4 nm spherical structures observed in nanofibrils (Fig. 1d). Due to the absence of homologous sequences, the three-dimensional (3D) structure of fibroin heavy chain predicted by AlphaFold 3 lacked accuracy and confidence. Therefore, while this model does not accurately represent the true structure of the fibroin heavy chain, it provides a useful reference for estimating the size of fibroin. That is, if the fibroin heavy chain is folded into a tight spherical structure, its size is approximately 10.6 nm × 10.7 nm × 15.1 nm, while the actual spherical structure in individual nanofibrils is much smaller than the predicted size. Therefore, we hypothesize that the smaller globular structures observed in nanofibrils represent individual domains formed by segments of the fibroin molecule. Unfortunately, due to the flexibility of NSF nanofibrils, obtaining high-resolution 3D structures through single-particle reconstruction remains challenging. Consequently, it is still unclear how many spherical domains an individual fibroin nanofibril consists of.

Structure of fibroin in silkworm silk

To determine whether the morphology of NSF nanofibrils changes during silk formation, we reconstructed the morphology and organization of silk proteins within silk fibers from third-instar silkworms. Using cryo-FIB milling, the lamella with a thickness ≤100 nm was qualified for decreasing the overlapping signal in the 2D image during the data collection (Supplementary Fig. 4a, b). The lamella was transferred to a Titan Krios cryo-electron microscope (cryo-EM) for data collection (Supplementary Fig. 4c). In the fibroin region inside the silk fibers, nanofibrils are oriented parallel to the fiber axis, with consistent widths, whereas the sericin region lacked a distinct nanofibril structure (Fig. 2a, b, Supplementary Fig. 4d). Interestingly, an unknown macromolecular structure was observed in the sericin region (Fig. 2b), but this macromolecule is absent in the sericin region of cocoon silk. It is therefore speculated that this structure might be a specific macromolecule expressed during the early larval stage of silkworms. Despite the overall parallel arrangement of nanofibrils in the fibroin regions, we found that the nanofibril distribution was uneven, with tightly packed areas and looser regions (Fig. 2a, b). The imperfect parallel arrangement of nanofibrils resulted in gaps within the silk fibers, which could be a contributing factor to fiber fractures during stretching^63,64. In the densely packed areas, nanofibrils were laterally associated, forming bridges (Fig. 2c–f) that may be relevant to mechanical coupling between neighboring nanofibrils.

Fig. 2. The overall structure of fibroin in silkworm silk remains unchanged.

a Tomographic slices showing the morphology and arrangement of fibroin and sericin in the XY and YZ views of silkworm silk. The purple box: parallel fibroin nanofibrils; red box: non-parallel fibroin nanofibrils; blue box: loosely arranged fibroin nanofibrils; green box: tightly packed fibroin nanofibrils. b The 3D model of fibroin nanofibrils from the same region as in (a). Purple: fibroin nanofibrils; yellow: unknown biomolecules. c, d Tomographic slices showing typical cross-bridges between fibroin nanofibrils. e, f 3D segmentation from the same region as in (c, d), illustrating bridges (blue, silver gray) between fibroin nanofibrils (green, yellow, purple). g The 3D structure of the averaged fibroin nanofibrils is the fundamental building block in silks. h Nanofibril diameter and length measurements (n = 6 tomograms).

A small set of nanofibrils were picked from the tomogram, aligned, and averaged (Supplementary Fig. 5a), We found that the fibroins in silk also appeared as nanofibrils with diameters of 3.6 ± 0.8 nm and lengths of 81.9 ± 38.5 nm (mean ± SD) (Fig. 2g, h), morphologically similar to those extracted from silk glands in vitro (Fig. 1d). The results indicate that the morphology of fibroin nanofibrils appears broadly similar as they move through the lumen of silk gland to form silk fibers. Our measurements likely overestimate nanofibril length, as tightly connected nanofibrils cannot always be distinguished. Notably, such small diameters for nanofibrils have not been previously reported for silk.

Structure of silk protein in spider silk and artificial silk

There are significant differences in mechanical properties between silkworm silk and spider silk, particularly spider dragline silk, which surpasses silkworm silk and even outperforms the best synthetic fibers like Kevlar⁶⁵. Using metal shadowing detection, we found that the morphology of natural spidroin (NSP) in the lumen of the spider’s major ampullate gland (Araneus ventricosus) closely resembled the fibroin of silkworms, appearing as a string of beads (Fig. 3a–c). However, the molecular weight of NSP was lower than that of silkworm NSF (Fig. 3d). Recent studies have shown that the silk from major ampullate glands of spiders contains 18 proteins⁶⁶.

Fig. 3. Tightly packed spidroin nanofibrils in spider silk.

a Spiders (Araneus ventricosus) captured from the wild in Sichuan, China. b Spider silk glands; arrowheads indicate the major ampullate gland. c Metal shadowing of spidroin extracted from the major ampullate gland of spiders, indicating spidroins are nanofibrils. d 4–16% gradient SDS-PAGE analysis of fibroin from silkworm and spidroin from the major ampullate gland of the spider. All experiments were independently repeated three times with similar results. e The low-magnification projection image of densely packed spidroin nanofibrils in spider silks. f Tomographic slice of spidroin nanofibrils enlarged in the position boxed in (e) (not the same silk). The inset shows the arrangement of silk protein nanofibrils. g The low-magnification projection image of loosely packed fibroin nanofibrils in artificial silk. h Tomographic slices of fibroin nanofibrils enlarged in the position boxed in (g). The inset shows the possible arrangement of silk protein nanofibrils.

To further investigate the relationship between the ultrastructure of animal silk and its mechanical properties, we manually pulled fresh silks from the spider and used cryo-FIB milling to prepare lamella ≤100 nm thick (Supplementary Fig. 6a–c). Then, the samples were transferred to a cryo-EM for high-resolution 3D imaging (Fig. 3e, Supplementary Fig. 6d). Cryo-EM revealed numerous nanofibrils arranged parallel to the long axis of the spider silk fibers. Unlike the loose arrangement of nanofibrils in silkworm silk, the nanofibrils in spider silk were tightly packed and well-oriented with no visible gaps (Fig. 3f), resulting in a much denser fiber structure with higher nanofibril density compared to silkworm silk (Supplementary Fig. 7a). This tight packing may explain why spider silk exhibits superior mechanical properties over silkworm silk.

We also prepared artificial silk fibers using the interface wire-drawing method³⁴ (Supplementary Fig. 8a, b) and obtained lamella ≤100 nm thick via cryo-FIB milling (Supplementary Fig. 8c, d). In striking contrast, although both natural and artificial silk are composed of NSF, artificial silk exhibited minimal nanofibril orientation and a non-uniform phase distribution (Fig. 3h, Supplementary Fig. 8e). These findings suggest that the buffer composition during spinning significantly affects the orientation degree and density of the nanofibrils within silk fibers, and these factors play a crucial role in determining their mechanical properties.

Structural organization of silkworm silk

To investigate whether silk fibers contain ordered structures similar to or higher than the herringbone pattern²⁹ observed near the spinneret in the anterior silk gland, we introduced a deep learning strategy-based method (REST) to establish the relationship between low-quality and high-quality density, applying this knowledge to restore signals in cryo-ET (Supplementary Fig. 5b)⁶⁷. The results show that this method improves the signal-to-noise ratio of tomographic slices and reduces some artifacts caused by the excessive density of silks and the missing data region in the 3D reconstruction of the sample (missing wedge) (Fig. 4a, b). Notably, we observed numerous regularly arranged dot-like structures in the XZ view of the silk fibers (Fig. 4a, b, bottom). By tracking each dot along the y-axis, we determined that each dot corresponds to a nanofibril, which was densely stacked on the YZ view (Fig. 4c, d). This arrangement is consistent with the herringbone patterns found near the spinneret in the anterior silk gland (Fig. 4e)²⁹. Previous studies^29,68 identified an axis within the herringbone pattern, possibly a pseudo-axis formed by interactions involving the N-terminal domain of the fibroin heavy chain of fibroin. In our experiments, we observed both tightly connected regions and blank regions within the same herringbone pattern. Interestingly, a distinct axis perpendicular to the nanofibrils was also evident in the blank regions of the herringbone pattern (Supplementary Fig. 9a), indicating that there may be other proteins involved in the formation of the herringbone pattern.

Fig. 4. Fibroin nanofibrils form an anisotropic herringbone pattern in silk.

a Tomographic slices of fibroin in XY and YZ views of silkworm silk. b Tomographic slices from the same region as in (a) using the REST method, which significantly removes noise and improves signal-to-noise ratio of fibroin nanofibrils, particularly in the XZ view, where point-like structures along the z-axis are evident. c A 3D model of fibroin nanofibril organization in the same slab as in (b), showing a herringbone-like pattern. d Tilted view of the 3D model in (c). e NSFs self-assembled into herringbone patterns at the distal region of the anterior silk gland near the spinneret. f Cryo-ET-based 3D model of the herringbone pattern, showing parallel nanofibrils along the z-axis.

Recent studies propose that the herringbone patterns of fibroin may actually resemble a 3D bottlebrush structure⁶⁸. Statistical analyses of the distances between points along the z- and x-axes revealed that the spacing between points along the z-axis is significantly smaller and more uniform than that along the x-axis (Supplementary Fig. 7b). This finding indicates that the nanofibrils within the silk are anisotropically arranged rather than simply stacked in parallel. Such an arrangement contributes to enhancing the mechanical properties of the material⁶⁹. Moreover, they are organized into a herringbone pattern, where multiple layers of this herringbone pattern are stacked along the x-axis to ultimately form micro-sized silk fibers (Fig. 4f). Obviously, the final micro-scale fiber does not exist as a compressed bundle of aligned nanofibrils, and no higher-order organization was observed beyond the herringbone pattern. This finding suggests that the herringbone pattern, formed near the spinneret in the anterior silk gland, remains intact throughout silk fiber formation and is not replaced by more complex arrangements. Notably, the anisotropic arrangement likely affects the mechanical properties of silk fibers in the x and z directions.

Discussion

In this work, we combined cryo-FIB milling with cryo-ET 3D reconstruction to investigate the hierarchical structure of force-drawn silk fibers from silkworms, spiders, and that of artificial silk. These findings provide insights into the spinning mechanisms and the relationship between ultrastructure and mechanical properties, which offer potential pathways for the manufacture of high-performance artificial fibers. Our cryo-ET data support a model in which nanofibrils represent the primary structural units resolved in silk, approximately 3.6 nm in diameter. These nanofibrils are anisotropically arranged, forming a herringbone pattern rather than being merely stacked in parallel. Multiple layers of this herringbone arrangement align in the same direction, ultimately giving rise to micro-scale silk fibers. While spider silk shares this parallel nanofibril organization, it features a notably denser structure compared to silkworm silk (Fig. 5).

Fig. 5. Structure, composition, and organization of silk.

a The fundamental structural unit of silk is a nanofibril approximately 3.6 nm in diameter. The fibroin is a fibrous protein rather than a globular protein, while sericin does not exhibit fibrous characteristics. These nanofibrils within the silk are anisotropically arranged, forming a herringbone pattern rather than being merely stacked in parallel. Multiple layers of this herringbone arrangement align in the same direction, ultimately giving rise to micro-scale silk fibers. b Comparative study of silkworm, spider, and artificial silk. In spider silk, a dense structure featuring nanofibrils is clearly visible, with their long axes tightly aligned parallel to the flow direction, leaving no discernible gaps. By contrast, silkworm silk exhibits larger, regionally varying gaps between nanofibrils, while artificial silk lacks the highly ordered arrangement characteristic of natural silks. Scale bar, 20 nm.

The size, shape, and distribution of fundamental elements are critical to a material’s performance^69,70. There has long been debate about the fundamental elements of animal silk, particularly whether silk proteins are globular or fibrous and what size micelles or nanofibrils form the silk fiber. In this study, we demonstrate that the fundamental unit of silk fiber is a beaded nanofibril with a diameter of approximately 3.6 nm, identified through cryo-ET reconstruction combined with AlphaFold 3 prediction. This is the smallest structural unit currently observable in silk fibers. By avoiding chemical fixation and staining, which risk altering macromolecular organization, we obtained unprecedented detail of silk’s internal structure. However, due to the flexibility of fibroin nanofibrils, we were unable to achieve high-resolution 3D structures through single-particle analysis or subtomogram averaging. Consequently, the exact number of globular structural domains within each nanofibril remains unclear. It is also important to note that the cryo-ET analyses in this study were performed on fibroin samples extracted from whole silkworms that had been frozen prior to dissection, a choice made for practical accessibility of the material. Although electrophoretic analysis, metal shadowing, and circular dichroism indicated that fibroin obtained from frozen animals is largely comparable to that extracted from living individuals²⁹, freezing may still introduce structural alterations that are not detectable by the assays currently available. Such potential influences should therefore be considered when interpreting the molecular organization reported here.

Moreover, sericin does not form a regular arrangement during the same spinning process (Fig. 2a, b), suggesting that protein sequence is a major factor, acting together with the spinning environment, in shaping fiber performance. The protein sequence dictates the material’s mechanical properties by shaping the morphology, size, and interactions of its fundamental elements. Recent studies have shown that introducing spider silk protein genes into silkworms can produce fibers with enhanced performance, highlighting the significance of the protein sequence⁷¹. Thus, designing protein fibers with specific functional properties through AlphaFold predictions combined with gene editing is another promising direction for future research in silk modification^{14,18,19,72,73}.

The herringbone pattern has increasingly been recognized as a higher-order arrangement^29,68,74 with significant implications for the mechanical properties of silk fibers. Whether this pattern is aligned parallel or perpendicular to the fiber axis is crucial for understanding silk fiber assembly. Our study found that the herringbone pattern is perpendicular to the fiber axis, while the nanofibrils within it are arranged parallel to the axis. The exact mechanism of herringbone pattern formation remains unclear. Previous studies suggested that this pattern might be related to interactions between N-terminal domains triggered by pH reduction^58,68. However, our experiments revealed that even in the blank regions of the herringbone pattern, where no nanofibrils were present, a vertical axis perpendicular to the nanofibrils remained (Supplementary Fig. 9a). This suggests that the formation of herringbone patterns may involve some unknown interactions or the participation of other proteins, emphasizing the importance of investigating the transformation of disordered nanofibrils in the silk gland into ordered herringbone patterns for understanding the spinning mechanism and simulating silk fiber formation in vitro. Additionally, we observed variations in the length of the nanofibrils within the herringbone patterns, possibly indicating another type of interaction between the nanofibrils.

Although we were unable to reconstruct the higher-order structures in spider silk due to its dense nanofibril composition, we did observe similar herringbone patterns in the third limb of the spider’s spinning duct, akin to those found in the anterior silk gland of silkworms (Supplementary Fig. 9b), suggesting that this morphological feature is conserved across distinct silk-producing species. Notably, in both spiders and silkworms, the spinning dope does not completely fill the spinning duct (Supplementary Fig. 9c, d)⁷⁵, suggesting a unique liquid environment within these ducts beyond simple acid conditions⁷⁶. This specialized environment appears to facilitate the silk extrusion from the spinneret, resembling the process of interface wiredrawing³⁴, where continuous fibers are produced by placing protein solutions in suitable buffers. Furthermore, a comparison of the ultrastructure between artificial and natural silkworm silk reveals that, although both are derived from NSFs, differences in buffer composition during the spinning result in marked variations in fiber ultrastructure. These structural differences, in turn, significantly influence the mechanical properties of the resulting silk fibers. Thus, further investigation into the liquid environment within the anterior silk gland lumen is essential for advancing our understanding of silk spinning mechanisims¹⁵. It is important to note that the fibers analyzed in this study were obtained by force drawing. This approach was necessary because silk collected from cocoons is relatively dry and tends to exhibit poor conductivity, which makes the acquisition of high-quality tomographic data extremely challenging. Although we attempted to match the drawing speed to that of natural spinning as closely as possible, the internal molecular arrangement may still deviate to some extent from that in native fibers. Therefore, the organizational features visualized here likely represent structures formed under a certain degree of experimental intervention. Future analyses of material subjected to less extensive drawing will be important for dissecting how mechanical stretching contributes to the emergence of the final ultrastructure.

In summary, cryo-ET of cryo-FIB-milled silks has allowed us to examine the morphology and architecture of silk proteins within fibers, overcoming artifacts typically introduced during sample preparation with resin, heavy metals, and other agents. These significant advances enabled us to explore the relationship between silk’s ultrastructure and its properties more thoroughly. We provide direct structural evidence that within silkworm silk, fibroins exist in the form of nanofibrils. These fibroin nanofibrils, with a diameter of approximately 3.6 nm, are the fundamental structural units of silkworm silk, arranged in an anisotropic herringbone pattern. In contrast, the nanofibrils in spider silk are more densely packed. We anticipate that further studies combining cryo-ET of cryo-FIB-milled silkworm silk, artificial silk, and spider silk from seven different spider silk glands will enable us to better correlate mechanical properties with the ultrastructure of the silk, offering insights into the design and engineering of artificial silk. This study establishes a research paradigm for the characterization of both natural and synthetic silk fibers, while also providing potential molecular evaluation criteria for assessing the mechanical properties of artificial fibers. While our data provide direct visualization of nanofibrillar organization, we acknowledge that the limited resolution and potential preparation-related influences mean that complementary approaches will be required to fully establish molecular details. We note during preparation and peer review of our work, a complementary manuscript was published in a preprint repository and interested readers can find the work there⁷⁷.

Methods

Ethical statement

No ethical approval was required for this study, as silkworms (Bombyx mori) and spiders (Araneus ventricosus) are not subject to institutional animal care and use regulations.

Experimental animal preparation

Silkworms (Bombyx mori) were reared to the third larval stage at room temperature, fed a diet of mulberry leaves. Twenty adult Araneus ventricosus spiders (body length 1.1–1.7 cm) were collected from the wild in Sichuan and Beijing, China, between August and October. The spiders were housed individually in separate containers and provided with moths and water.

Preparation of natural silk fibroin, natural spidroin, and cryo-vitrification

Silkworm larvae were reared at room temperature until reaching the early wandering stage, after which they were frozen using liquid nitrogen and stored at −80 °C. Silk glands were dissected from the frozen silkworms, and the cells on the surface of the posterior of the middle silk gland (PMSG) were manually peeled off using forceps. Then, the spinning drop was transferred to the extraction buffer (50 mmol L⁻¹ NaCl, 50 mmol L⁻¹ Bis-Tris, 0.5 mmol L⁻¹ EDTA, pH 7.5) to dissolve the NSF, resulting in a NSF solution. NSF solution was mixed with amphipol in extraction buffer at a mass ratio of 1:6. After incubation at 4 °C for 72 h, 3 μL of the mixture was loaded onto a glow-discharged holey grid (GIG, 1.2/1.3, Au) and vitrified by flash-plunging into liquid ethane using a Vitrobot Mark IV (FEI, USA). Blotting time, force level and humidity were set to be 7 s, 0 and 100%, respectively.

Adult spiders were anesthetized at 4 °C for 30 min and then dissected on ice to isolate the major ampullate glands. The silk glands were rinsed three times with phosphate buffered saline (PBS), and the outer epithelial cell layer was carefully removed using fine forceps. The gel-like spinning dope was gently extracted and transferred into PBS to prepare the NSP solution.

Biochemial analysis of natural silk fibroin and natural spidroin

The homogeneity and composition of NSF were assessed by size-exclusion chromatography (SEC) and sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE). SEC was performed using a Superose 6 increase 3.2/300 column (GE Healthcare, USA). Protein fractions were subsequently analyzed by SDS-PAGE and visualized by Coomassie Brilliant Blue staining.

The component of NSP was examined by SDS-PAGE. A homemade linear gradient SDS-PAGE gel (3–16%) was prepared using a Hoefer SG 30 gradient maker (Thermo Fisher Scientific, USA). The concentrations of NSF and NSP were determined by measuring absorbance at 280 nm (A₂₈₀) with a NanoDrop 2000C spectrophotometer (Thermo Fisher Scientific, USA).

NSF solution was incubated with amphipol (Anatrace, USA) at mass ratio of 1:3, 1:6, 1:9, respectively, at 4  °C. Following different incubation periods, the samples were mixed with native-PAGE loading buffer (Invitrogen, USA) and subjected to electrophoresis at 4  °C under the following conditions: 90 V for 30 min, followed by 175 V for 90 min. The gels were stained with Coomassie Brilliant Blue. A linear gradient blue native polyacrylamide gel electrophoresis (BN-PAGE) gel (3–16%) was cast following Witting’s protocol⁷⁸, with a Hoefer SG 30 gradient maker (Thermo Fisher Scientific, USA).

Metal shadowing

Metal shadowing was performed following established procedures^29,79. For metal shadowing imaging of NSP and NSF solutions extracted respectively from spider and silkworm, a 5 μL solution (30 μg mL⁻¹) was applied to glow-discharged copper grids for 2 min. Excess solution was carefully removed using blotting paper. For metal shadowing imaging of NSF in silkworm spinning dope and NSP in spider pinning dope, a glow-discharged copper grid was placed on the spinning dope and gently pressed with tweezers to avoid any artifact formation. NSF was dehydrated through a graded ethanol series (0, 25%, 50%, 75%, 100%) for 4 min each step. The samples were then shadowed with tungsten using a DV-502B high vacuum evaporator (Denton Vacuum, USA) and air-dried. A tungsten wire (8 cm in length) was clamped between two electrodes at a distance of 3.8 cm, with a separation of 9.3 cm from the center of the specimen platform. The angle between tungsten wire and the sample was approximately 10 °. The NSF samples were rotated during evaporation for 14.5 min. Transmission electron microscopy was performed using a Tecnai Spirit (FEI, USA).

Preparation of artificial silkworm silk, force-drawn silk, and cryo-vitrification of silk samples

Artificial silkworm silk was prepared by the interface wiredrawing as previously described³⁴ with modifications. Briefly, 3 μL buffer (50 mmol L⁻¹ Citric acid, 50 mmol L⁻¹ Bis-tris propane, 16% w/v Polyethylene glycol 3350, pH 5.0) was placed on a glass plate, followed by the addition of an equal volume of NSF solution (3 mg mL⁻¹). The solutions were quickly mixed with a 10 μL pipette tip, and the resulting artificial silks were immediately placed on the glow-discharged mesh grids for subsequent analysis.

Fresh silkworm and spider silks were obtained by the forcible silking method. Third-instar silkworms and size-matched spiders were gently immobilized using adhesive tape and placed under a stereomicroscope. Silk fibers were then carefully drawn from the spinnerets using fine-tipped tweezers and immediately transferred onto glow-discharged mesh grids. Third-instar larvae were selected due to their relatively small silk diameter (~3–5 μm), which facilitates cryo-vitrification.

To ensure the physiological relevance of the fibers studied, the reeling speed was manually controlled to match the natural spinning rates of silkworms (4–15 mm s⁻¹) and spiders (10–20 mm s⁻¹)⁸⁰. To prevent dehydration, 3 μL of distilled water was added to the grid prior to silk application. Excess liquid was blotted with Whatman filter paper for 6 s, and then the grids were rapidly plunge-frozen into liquid ethane using an EM GP system (Leica Microsystems, Germany). Vitrified grids were stored in liquid nitrogen until further use.

Cryo-focused ion beam milling and Cryo-ET data acquisition

Cryo-FIB milling was performed using a Helios NanoLab 600i Dual Beam SEM (Thermo Fisher Scientific, USA), as previously described⁸¹. The goal was to prepare lamellae with a final thickness of ≤100 nm. Frozen grids were transferred with the cryo-transfer shuttle into the SEM chamber via a Quorum pp3000T cryo-transfer system (Quorum Technologies, UK) at −180  °C. During milling, the angle between the FIB and the silk surface was maintained at 5 °–10 °, with milling performed from both sides to create vitrified silk lamellae. The ion beam accelerating voltage was set at 30 kV, and the current ranged from 0.43 nA (rough milling) to 10 pA (fine milling). The fine lamellae had a thickness of less than 100 nm.

Cryo-ET data acquisition was carried out using a Titan Krios transmission electron microscope fitted with a K2 detector. Overview tilt series of silks lamellae were acquired under a magnification of ×105,000, resulting in a physical pixel size of 1.36 Å, under counting mode. Before data collection, the pre-tilt of the sample was determined visually, and the pre-tilt was set to be 10 ° or -9 ° to match the pre-determined geometry caused by loading grids. And the tilt range was set to be between −60 ° and +40 ° for −10 ° pre-tilt or −40 to +60 ° for +10 ° pre-tilt, with a 3 ° step, resulting in 33 tilts and 99 electrons per tilt series. The slit width was set to be 20 eV, with the refinement of zero-loss peak after collection of each tilt series, and nominal defocus was set to be −3.5 to −4.5 μm. For the fibroin extracted from silk glands, the tilt range was set to be between −55 ° and +55 ° with a 3 ° step, resulting in 35 tilts and 115 electrons per tilt series. All tilt series used in this study were collected using a dose-symmetry strategy-based beam-image-shift facilitated acquisition scheme, by in-house developed scripts within SerialEM software. The rest of the collection details are consistent with those mentioned above and are summarized in Supplementary Table 1.

Tomogram reconstruction and sub-volume averaging of the fibroin

Motion correction and contrast transfer function estimation were carried out in Warp; tilt series alignment was carried out in IMOD 4.8.49 and AreTomo 2. For the 77 tomograms of fibroin extracted from silk glands, the data were 4× binned and denoised using the deconvolution method implemented in Warp to facilitate the particle picking. To pick filaments, CrYOLO 1.6.0⁸² was employed to detect the straight filaments in the XY planes of the tomograms. These segments were extracted with a box size of 40 pixels (218 Å) using Warp. The reference was a featureless cylinder using Dynamo 1.1.332. The sub-volumes were then aligned and classified using RELION 4.0. This averaged structure has an estimated resolution of 24 Å based on the “gold-standard” FSC with 0.143 criterion. The details see supplementary Fig. 2.

For the fibroin within fibers, due to the low signal-to-noise ratio (SNR) in lamella, tomograms were reconstructed using a SIRT-like filter and binned 2, as implemented in IMOD 4.8.49. Particle picking was performed using CrYOLO 1.6.0 on the filtered tomograms. These segments were extracted with a box size of 90 pixels (245 Å) using RELION 3.0. The reference was a featureless cylinder using Dynamo 1.1.332. The sub-volumes were then aligned and classified using RELION 3.0. This averaged structure has an estimated resolution of 22 Å based on the “gold-standard” FSC with 0.143 criterion. The details see supplementary Fig. 5a.

Restoring the tomograms by deep learning

Due to the dense arrangement of silk fibers and the presence of a significant missing wedge effect 3D, especially in the XZ plane, we employed the REST method⁶⁷ for tomogram restoration. First, we generated a ground truth by randomly distributing the averaged map (many fibers) to densely arrange in the particle box. Then, employing random rotations and shifts (high-quantity datasets), noise and missing wedge artifacts were added to the ground truth to form the low-quantity datasets. These data were paired and then used as training pairs for REST training. The trained model was subsequently utilized to recover the raw tomograms from the weighted back-projection (WBP) reconstruction. The details see supplementary Fig. 5b.

Measurement and visualization

Visualization and figure generation were performed in UCSF ChimeraX v1.3 and IMOD 4.8.49. The diameter and length of fibroin nanofibrils were measured manually using the measure tool in IMOD 4.8.49 and Fiji 2.32. To analyze the density of spider and silk fibers, we projected the same number of XY slices into a single 2D projection keep the same volume. The resulting projection was then binarized, and the proportion of the fibers in space was quantified by calculating the ratio of white pixels (representing fibers) to the total number of pixels in the projection.

Reporting summary

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

Author contributions

K.S., Y.L., and P.Z. conceived the experiments. K.S., Y.L., and X.Z. performed the cryo-EM sample preparation. Y.L. performed the FIB milling. H.Z. and Y.L. performed data collection. H.Z., K.S., and Y.L. analyzed the data. K.S. and H.Z. wrote the draft and composed the figures. K.S., H.Z., and Y.L. revised the manuscript and figures. P.Z. supervised the project, analyzed the data, and revised the manuscript and figures.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Data availability

The tomograms and subtomogram average maps shown in this study have been deposited in the Electron Microscopy Data Bank (EMDB) under accession codes: EMD-65708 (Fig. 1a, tomogram of natural silk fibroin extracted from posterior of middle silk glands), EMD-65681 (Fig. 1d, averaged map of natural silk fibroin nanofibril), EMD-65705 (Fig. 2a, tomogram of fibroin and sericin in silkworm silk), EMD-65682 (Fig. 2g, averaged map of fibroin in silkworm silk) and EMD-65706 (Fig. 4a, tomogram of fibroin in silkworm silk). The source data underlying Figs. 1c, 2h and 3a–d and Supplementary Figs. 1a–e, 7a, b and 9a–d are provided as a Source data file. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-70477-1.