Real-time single-molecule observation of incipient collagen fibrillogenesis and remodeling

Significance

Collagen is the principal constituent of the extracellular matrix and is critical for providing structure to organs and tissues. Although collagen has been vigorously studied, the assembly process remains poorly understood. Utilizing video rate scanning atomic force microscopy, we visualize, in real-time, dynamic time-resolved assembly, growth, and remodeling of incipient fibril structures. We show that collagen fibrils, even at extremely early timepoints of development, display rope-like characteristics and undergo remodeling events, such as bird-caging. These findings serve as a framework to better understand collagen growth in disease-related systems.

Abstract

The hierarchic assembly of fibrillar collagen into an extensive and ordered supramolecular protein fibril is critical for extracellular matrix function and tissue mechanics. Despite decades of study, we still know very little about the complex process of fibrillogenesis, particularly at the earliest stages where observation of rapidly forming, nanoscale intermediates challenges the spatial and temporal resolution of most existing microscopy methods. Using video rate scanning atomic force microscopy (VRS-AFM), we can observe details of the first few minutes of collagen fibril formation and growth on a mica surface in solution. A defining feature of fibrillar collagens is a 67-nm periodic banding along the fibril driven by the organized assembly of individual monomers over multiple length scales. VRS-AFM videos show the concurrent growth and maturation of small fibrils from an initial uniform height to structures that display the canonical banding within seconds. Fibrils grow in a primarily unidirectional manner, with frayed ends of the growing tip latching onto adjacent fibrils. We find that, even at extremely early time points, remodeling of growing fibrils proceeds through bird-caging intermediates and propose that these dynamics may provide a pathway to mature hierarchic assembly. VRS-AFM provides a unique glimpse into the early emergence of banding and pathways for remodeling of the supramolecular assembly of collagen during the inception of fibrillogenesis.

An essential component of collagen function is time-dependent changes in collagen structure at multiple scales. Studying the temporal evolution of hierarchic self-assembly can provide valuable insight into how supramolecular organization is achieved and how it is perturbed in connective tissue disorders that arise from collagen mutations. After assembly, collagen in the extracellular matrix continues to undergo structural remodeling in response to mechanical stress (1, 2). Dynamic remodeling is an essential property of collagen at all physical scales. Individual collagen monomers, defined here as the triple-helix formed by three polypeptide chains, are metastable at physiological temperatures, and likely require supramolecular interactions to provide stability (3, 4). A dynamic, time-resolved picture of collagen assembly, growth, and remodeling is critical to understand matrix homeostasis and disease.

There are 28 types of collagens in humans and other animals (5–7), of which Type I fibrillar collagen is the most abundant. Type I collagen is predominantly expressed in skin and bone. It has multiple biomechanical and biological functions, including dissipating deformation stress of tissues (8–10), inducing cell polarity during cell-matrix adhesion (11), and directing bone mineralization (12–14). Type I collagen adopts organized structures on multiple scales spanning seven orders of magnitude from nanometer-sized triple helices, described as the collagen monomer, to over millimeter-sized fascicles in connective tissue (9, 15, 16). At the smallest scale, the monomer is approximately 300 nm long and 1.5 nm in diameter. These monomers self-assemble into fibrils, which can reach diameters of several hundred nanometers and lengths of microns (7). The mature fibril is a highly ordered structure with a 67 nm periodic banded structure, referred to as “D-banding” (17) (Fig. 1A). D-banding arises from the staggered packing of monomers within the fibril, which are divided into five subdomains, D1 to D5. D1 to D4 have lengths of 67 nm and D5 is 31 nm, resulting in distinct regions of differing protein density, referred to as gap and overlap regions (18). The mechanical and functional behavior of collagen, along with higher levels of spatial organization in tissue, critically depends on this banding (19).

Fig. 1.

Constructing the fibril. Understanding the fibril structure begins with the collagen triple-helix (monomer) which consists of five distinct regions, D1 to D5 (A). When multiple monomers aggregate together to form fibrils, they do so in a stacking-like manner, which is repeated and preserved throughout the entire fibril structure. The shortest repeating unit, the microfibril structure, has been determined from X-ray diffraction (18). The microfibril consists of five stacked triple helices color coded by the different D regions of the monomer. Since D5 is shorter than D1 to D4, it does not fully cover the D4 region, leading to an overlap region, where all the monomers are stacked, and a gap region, where only four are stacked. It is these gap and overlap regions that constitute one “D-band” and give collagen fibrils their characteristic banded appearance. To better visualize how individual monomers are nestled in the fibril structure, five microfibrils are shown stitched together adjacently. One individual monomer traced in black with its corresponding gap (G) and overlap (O) regions is labeled. The D1 to D5 labeling indicates which D-region of the monomer is incorporated into the labeled D-band. Last, in the mature fibril structure, these microfibrils are heavily preserved and are repeated along and around the fibril. In terms of collagen fibril organization, it has been proposed that fibrils possess a self-similar rope-like structure (20, 21) (B). Although it is still not well understood over what range of length scales self-similarity extends, work by Bozec (20) and Kadler (21) suggest that the nature of the fibril organization is fractal-like. Taken together with our data, we propose a model in which self-similarity extends across all length scales at the fibril level down to the monomer level.

Despite the central importance of collagen supramolecular structure to function, fundamental questions regarding its assembly remain. Collagen fibrils exhibit banding and supercoiling that are preserved throughout the entire fibril structure. The supercoiling results in a rope-like topology (20). While this rope-like organization is not evident in images of mature D-banded fibers, it can be revealed when they are perturbed either mechanically or using chemical denaturants (20, 22–24). Human skin collagen fibrils treated with urea unravel, revealing the underlying rope-like helical organization (22). Similarly, mechanically stressed collagen obtained from multiple tendon sources shows fraying and buckling, revealing an internal rope-like structure (20, 23, 24). This mechanism by which the internal rope-like structures in mature collagen are visualized in the studies described above is often via a “bird-caging” in which there is internal unraveling of collagen’s rope-like structure. Rope-like topology is found at all levels of structure starting with the triple-helix, the microfibril, and up to the mature fiber. Thus, the collagen structure has been described as self-similar across all length scales (20, 21), schematically shown in Fig. 1B.

The collagen fibril is a dynamic structure, where breathing and buckling are central to its function even in a static, equilibrium state (25–28). Biophysical studies have shown the folding transition temperature for the collagen monomer is just below 37 °C, facilitating local remodeling of a monomer within the fibril in response to cellular interactions (3). Collagen-binding proteins may exert local mechanical stress on the triple-helix, much like transcription factors when bound to DNA (29). In DNA, torsional stress from supercoiling forms bubbles, which can facilitate binding of transcription machinery. Similarly, local remodeling of the collagen fibril may present cryptic binding sites for integrins and other collagen binders that would be otherwise occluded (25–28).

In order to observe the dynamic, time-resolved picture of collagen assembly, growth, and remodeling, we use video rate scanning atomic force microscopy (VRS-AFM). We image the self-assembly of rat tail tendon Type I collagen on a mica surface at a rate of one frame (2 × 2 µm) per second, providing a unique window into incipient collagen fibrillogenesis. Although it is not well understood over what range of length scales self-similarity extends, here we show that rope-like, self-similar characteristics exist at the earliest timepoints of fibril assembly and therefore propose a model in which rope-like, self-similar characteristics extend across all fibril length scales. Locally reversible association and disassociation of fibrils through various remodeling events such as bird-caging or fraying indicate dynamic remodeling at the earliest stages of fibril growth. We propose this remodeling may correct off-pathway, nonnative intermediate fibrils and instead help to promote native rope-like, banded structures compatible with subsequent stages of mature collagen assembly.

Results

VRS-AFM Visualizes Real-Time Growth Events.

VRS-AFM was used to monitor the real-time self-assembly of rat tail tendon Type I collagen fibrils on a mica surface within 4 µm² field with an imaging rate of ~1 frame/s over several minutes. Stock collagen monomers were diluted to a concentration of ~30 µg/mL in a potassium chloride solution (KCl, pH = 10) to induce fibril formation and then immediately injected into the AFM through perfusion tubing. Fig. 2 shows a series of frames from a representative VRS video. The full videos are available in the SI Appendix as movie files. Monomers are seen nucleating, elongating, interacting with neighboring structures, and progressively covering the mica surface.

Fig. 2.

VRS overview (Movie S1). (A) Progression of frames from a VRS video recording the growth of individual D-banded collagen fibrils on a mica surface. The video is obtained at an imaging rate of ~1 frame/s and a resolution of 256 × 256 points and lines (7.8125 nm/pixel). Here, representative time points are presented, with the entire video consisting of 520 frames. From these frames, the growth of individual collagen fibrils can be tracked. (B) displays the height profiles corresponding to the red line in the frames, while (C) illustrates the progression of the collagen’s maximum height tracked by the red line over the course of the full video.

D-banding is observed within the first 2 min of assembly. Fibrils are first seen when their heights are around 1 to 3 nm with some indications of periodic height variations. Within minutes, these grow in height (6 to 8 nm), thicken, and mature with much more defined D-banding. Fibril orientation and growth observed here is consistent with epitaxial interactions between collagen and the crystalline mica surface (30–32). The height of growing fibrils can be tracked (Fig. 2B) and often plateau within the duration of the experiment (Fig. 2C).

Unidirectional Growth and Fibril Fusion Preserve Fibril Polarity.

Time series images of individual fibrils, i.e., kymographs, were extracted from the VRS videos (Fig. 3A). In one example, a monomer at time t = 0, which matches the expected length (300 nm) and height (1.5 nm) of a monomer (33), begins accruing additional monomers and elongating on the surface by t = 16 s. Growth only occurs on one end of the fibril while the other remains dormant (t = 0 to 48 s). Based on previous imaging of collagen growth in larger fibrils, the active end likely corresponds to the N termini of monomers within the fibril (21, 34, 35). For the fibrils where both ends can be observed, unidirectional growth is a feature of the earliest stage of assembly. In the same kymograph, we also observe an early fusion of two fibrils. At t = 54 to 65 s, a growing fibril enters the frame and then fuses with the dormant terminus, resulting in a single fibrillar structure (Fig. 3B). Multiple instances of such fusion events between dormant and growing ends were observed (SI Appendix, Fig. S1). As previous studies on larger fibrils have indicated (34, 35), unidirectional growth is facilitated by the uniform N-C terminal polarity of monomers within a growing fibril. Merging of two fibrils preserves this polarity by favoring fusion of a growing N-terminal end with a dormant C terminus–presenting end. Notably, when two growing ends of fibrils meet (two N termini), we do not see resolution of a successful end-to-end junction (SI Appendix, Fig. S2).

Fig. 3.

One-sided fibril growth and fusion (Movie S1). Kymograph showing the one-sided growth (A) of collagen over time and its eventual fusion with an adjacent fibril (B). Initially (t = 0 s), both termini of a fibril in its extremely early stage of development are visualized. As time progresses (t = 0 to 70 s), only one side of the fibril is actively elongating while the other remains dormant. This dormancy persists until it encounters another collagen structure on the surface (t = 64 to 70 s, red arrow) and promptly fuses with it.

D-Banding Occurs within the First Minute of Assembly.

Collagen banding assumes the ordered assembly of individual monomers with periodic gap and overlap regions. It was previously proposed that growing fibrils adopt a disordered intermediate state prior to maturation into a banded morphology (36). We find that this disordered intermediate is very transient. In one VRS kymograph of growing collagen, there is a clear, extended series of bands after 2 min of assembly (Fig. 4A). We tracked the progression of D-band formation on these extremely early collagen structures. Even as early as t = 21 s, the short, 400 nm structure shows two maxima in the height profile (Fig. 4B). Within the remaining short time span, banding extends and becomes more prominent (t = 81 s). Fourier analysis of the height profiles indicates convergence to a periodicity of ~67 nm after the first minute (Fig. 4C).

Fig. 4.

Tracking fibril periodicity (Movie S1). (A) Kymograph of the development of the 67 nm D-banding periodicity of a fibril growing on mica. As time progresses, the fibril becomes a more ordered structure with the characteristic D-banding becoming evident in both the AFM image and the corresponding axial height traces (B). The periodicity is then determined by Fourier transform of these height profiles, and the results for every frame of the video are recorded. During the early frames, the periodicity changes sporadically until around frame 100, when the periodicity converges toward a value of ≈67 nm, the expected D-banding value (C).

Structural Remodeling.

Some fibrils appear to undergo remodeling during growth. In some events, the end of a growing fibril bifurcates over the course of a few seconds (Fig. 5A). Both ends then continue to grow. Other remodeling events occur within the body of the fibril (Fig. 5B), resulting in the formation of internal bubbles. These remodeling events often preserve the D-banded fibril structure, producing smaller D-banded fibrils. In some cases, disruptions reanneal, indicating reversible structural transitions (SI Appendix, Fig. S3). Structural remodeling events also reveal the rope-like internal structure (Fig. 6) through a phenomenon known as bird-caging (20, 37). We have found examples of similar remodeling dynamics in previously published collagen growth studies (38–40), indicating this is not a phenomenon limited to our experimental design. However, not all previous AFM studies of collagen assembly show remodeling (41). We attribute these differences in remodeling propensity to the solution conditions and sources of collagen used.

Fig. 5.

Fibril remodeling (Movie S1). Two examples of collagen fibrils undergoing remodeling on a mica surface. At first, the fibrils are intact on the surface, but several seconds afterward, the fibrils come apart into separate individual strands (A) or form an internal bubble (B). Red boxes indicate major areas of remodeling on the fibrils.

Fig. 6.

Fibrils as ropes (Movie S2). 1 × 1 µm images of fibrils taken from the VRS videos showing distinct rope-like characteristics. A previously intact fibril on the surface uncoiling itself in a manner like the bird-caging of a rope, (Left, t = 213 s) and a fibril exhibiting frayed ends similar to the end of a rope (Right, t = 149 s).

When torsional forces on a rope are suddenly released or changed, the internal stresses can cause the individual strands of the rope to splay out, resembling the bars of a birdcage. The observation of in situ bird-caging in these experiments support the encoding of rope-like characteristics at the earliest stages of fibrillogenesis. The rapid timescale over which these remodeling events occur indicates cooperative transitions between metastable conformational states. Further, we hypothesize that remodeling serves to reorganize nonnative interactions within the growing fibril and promote on-path assembly toward a mature collagen structure.

Discussion

High-speed AFM provides a powerful approach for observing dynamics of collagen biological processes at the supramolecular level in real-time (39, 42–44). Using this technique, we track the fibril self-assembly process in real-time and show the unique early assembly of Type I collagen. We observe very rapid formation of ordered D-banded structure, fibril fusion, and remodeling. The remodeling events, bifurcation of the fibril end or internal bird-caging, occur rapidly, often within a single frame, suggesting a cooperative transition much like the folding/unfolding of a protein domain.

In vivo, the assembly of collagen occurs in the presence of both a cell surface and the extracellular matrix. The extent to which this complex environment affects collagen fibrillogenesis is not understood. For example, AFM experiments have shown that the chemical properties of the surface and solution can influence the structure of collagen fibrils (31, 32) and monomers (45, 46). Different mica mineral types—muscovite versus phlogopite—direct the assembling collagen into distinct morphologies, and furthermore these interactions are sensitive to ionic strength of the surrounding buffer, indicating electrostatic forces may be driving epitaxy. During the early stages of fibrillogenesis, collagen–collagen and collagen–surface interactions may be in competition, leading to potentially suboptimal assemblies. While it is below the detection resolution of VRS-AFM, we propose that another source of frustration may be mismatched interactions between D-period domains of adjacent collagen monomers or smaller fibrils during fibrillogenesis. In our dynamic, time-resolved AFM studies of collagen growth, we clearly observe remodeling intermediates such as bird-caging. These may demonstrate the innate ability for collagen fibrils to remodel and relieve stress caused by interactions with a complex environment.

Our data support the hypothesis that structural self-similarity exists at the smallest, earliest stages of fibrillogenesis, resulting in periodic banding and rope-like structure (Fig. 1B). In the VRS-AFM videos presented here, we observe dynamic processes such as bird-caging in the very early stages of growth and assembly. The early bird-caging may mirror the bird-caging of mature fibers observed during mechanical stress (20, 23, 24), indicating self-similarity across all time scales and length scales in collagen hierarchical structures. Although it is still unclear what stresses collagen will experience during growth in the ECM, we have demonstrated that even at the earliest timepoints of development, fibrils may have the innate ability to undergo various forms of remodeling in response to stress. Dynamical self-similarity may provide a common pathway for the resolution of stress across all length scales of collagen, allowing for resilient deformation mechanics to occur. Overall, this work presents a framework to understand how mutations, post-translational modifications, and assembly conditions may affect the biology and pathology of collagen.

Methods

Materials.

High concentration collagen I from rat tail tendon in 0.02 N acetic acid with telopeptides still intact was purchased from Fisher Scientific. Buffers were prepared from analytical-grade chemical reagents and ultrapure water.

AFM/Sample Preparation.

VRS-AFM (tapping-mode) was performed on a Cypher ES AFM (Asylum Research) using BL-AC10DS tips with a nominal resonance frequency and spring constant of 1.5 MHz and 0.1 N/m, respectively. Q factor for Movies S1–S3 were 1.430, 1.334, and 4.5725 respectively. A mica disc with a 3 mm radius atop a sapphire post was epoxied to the VRS stage, allowed to set for 5 min, freshly cleaved, and had a drop of buffer (300 mM KCl, 10 mM phosphate buffer, pH = 10) placed on top of it. The mica was then imaged to ensure the imaging area was clean and devoid of debris. At this point, the stock collagen solution was diluted with buffer to a concentration of 30 µg/mL collagen and then immediately perfused into the imaging media. Imaging began immediately afterward at a rate of ≈1 frame/s (250 Hz) to capture the development of early collagen structures. For imaging, Bluedrive photothermal excitation was used to reliably excite the cantilever to its resonance frequency and facilitate high-resolution imaging. All imaging was performed at room temperature. To minimize the tip-sample force, the highest setpoint to free air amplitude ratio that would allow for stable high-resolution imaging was maintained for imaging.

Automated Fibril Detection and Quantitative Analysis.

To facilitate quantitative analysis and create high-fidelity outlines of the self-assembling collagen while eliminating background noise (SI Appendix, Fig. S4), the videos were processed through an image detection algorithm based on a version of the U-Net machine learning model (47), originally trained to detect cells in images, that was retrained to detect collagen fibrils. To do so, ~380 frames from separate VRS videos were hand annotated to differentiate collagen from the surrounding mica and then used to train the machine learning program. An example of the machine learning output is shown in SI Appendix, Fig. S4. Each frame from each video was then processed through the machine learning program to create high-fidelity outlines of the fibrils to facilitate quantitative analysis.

Image Analysis.

Each frame of each VRS video was converted into a table of values and imported into MATLAB (2020b) for analysis. Each frame was then overlayed with its corresponding fibril outlines produced by the U-Net image detection process to create high-fidelity outlines of the fibrils and reduce background noise. For determining fibril periodicity, a height profile was drawn along the axis of select fibrils and then Fourier Transformed to indicate if any underlying periodicity is present. This process was repeated frame by frame to track the evolution of a fibril’s height profile and its periodicity. Additional quantitative measurements revealing the height and area growth rates of fibrils are shown in SI Appendix, Fig. S5.

Supplementary Material

Appendix 01 (PDF)

Movie S1.

VRS-AFM video of collagen growing on a mica surface. Here monomers can be seen nucleating onto the surface and progressively elongate and grow into D-banded fibrils. The video was obtained with a scan rate of 1 frame per second, scan area of 2 x 2 μm, and 256 x 256 points and lines.

Movie S2.

VRS-AFM video of collagen growing on a mica surface showcasing a clear bird-caging event. The video was obtained with a scan rate of 1 frame per second, scan area of 2 x 2 μm, and 512 x 256 points and lines.

Movie S3.

VRS-AFM video of collagen growing on a mica surface with various forms of remodeling (internal bubble formation, fibrils splitting apart). The video was obtained with a scan rate of 1 frame per second, scan area of 2 x 2 μm, and 512 x 256 points and lines.

Data, Materials, and Software Availability

Raw VRS-AFM data have been deposited in Figshare (10.6084/m9.figshare.25848856.v1) (48).