Sequence-dependent mechanics of collagen reflect its structural and functional organization

Abstract

Extracellular matrix mechanics influence diverse cellular functions, yet surprisingly little is known about the mechanical properties of their constituent collagen proteins. In particular, network-forming collagen IV, an integral component of basement membranes, has been far less studied than fibril-forming collagens. A key feature of collagen IV is the presence of interruptions in the triple-helix-defining (Gly-X-Y) sequence along its collagenous domain. Here, we used atomic force microscopy to determine the impact of sequence heterogeneity on the local flexibility of collagen IV and of the fibril-forming collagen III. Our extracted flexibility profile of collagen IV reveals that it possesses highly heterogeneous mechanics, ranging from semiflexible regions as found for fibril-forming collagens to a lengthy region of high flexibility toward its N-terminus. A simple model in which flexibility is dictated only by the presence of interruptions fit the extracted profile reasonably well, providing insight into the alignment of chains and demonstrating that interruptions, particularly when coinciding in multiple chains, significantly enhance local flexibility. To a lesser extent, sequence variations within the triple helix lead to variable flexibility, as seen along the continuously triple-helical collagen III. We found this fibril-forming collagen to possess a high-flexibility region around its matrix-metalloprotease binding site, suggesting a unique mechanical fingerprint of this region that is key for matrix remodeling. Surprisingly, proline content did not correlate with local flexibility in either collagen type. We also found that physiologically relevant changes in pH and chloride concentration did not alter the flexibility of collagen IV, indicating such environmental changes are unlikely to control its compaction during secretion. Although extracellular chloride ions play a role in triggering collagen IV network formation, they do not appear to modulate the structure of its collagenous domain.

Significance

Collagens are the predominant proteins in vertebrates, forming diverse hierarchical structures to support cells and form connective tissues. Despite their mechanical importance, surprisingly little is established about the molecular encoding of mechanics. Here, we image single collagen proteins and find that they exhibit variable flexibility along their backbones. By comparing collagens with continuous and discontinuous triple-helix-forming sequences, we find that the type of helix interruption correlates with local flexibility, providing the first steps toward a much-needed map between sequence, structure, and mechanics in these large proteins. Our results inform our understanding of collagen’s ability to adopt compact conformations during cellular secretion and suggest a physical mechanism by which higher-order structure may be regulated by the distinct molecular properties of different collagens.

Introduction

Collagen is the most abundant protein in the animal kingdom and represents one third of the total protein in the human body (1,2). It is a major structural component of the extracellular matrix, contributing to the mechanical stability, organization, and shape of a wide variety of tissues. Twenty-eight distinct collagen types have been reported in humans, with different higher-order organizational structures (1, 2, 3). The most prevalent are fibril-forming collagens; these have a unique hierarchical structure whereby collagen molecules assemble in a parallel, staggered fashion into long fibrillar nanostructures that form higher-order structures (Fig. 1). These fibers are the predominant load- and tension-bearing structures in connective tissues. Network-forming collagens such as collagen IV associate end on to form sheet-like networks rather than fibrils (Fig. 1) (4,5). These collagen IV networks played an integral role in the evolution of multicellular life (3,6). Collagen IV networks are predominantly found in basement membranes (BMs), where they provide mechanical support and anchorage for cells and tissues and serve as a filtration barrier to macromolecules in organs such as the kidney (5,7). The mechanics of fibril-forming collagens have been the subject of numerous studies at the fibrillar and molecular levels (8, 9, 10 and many references therein). By contrast, the mechanics of network-forming collagens are far less studied (11). Similarly, knowledge of the structure of collagen IV lags significantly behind fibril-forming collagens at both the supramolecular and molecular scales.

Figure 1.

Localization and supramolecular structures of fibril-forming and network-forming collagens. All collagen chains are synthesized in the ER. The folded proteins are secreted from the cell, where they can form a variety of supramolecular structures in different tissues. The fibril-forming collagen molecule, procollagen, is post-translationally processed, whereby its propeptides are cleaved off to yield tropocollagen. Tropocollagens align laterally with an offset that gives rise to the characteristic D-banding pattern of fibrils. On the other hand, the globular domain of the network-forming collagen IVis not post-translationally cleaved. Its end domains serve an important role in network formation. The N-terminal end aligns four molecules and forms a 7S tetramer. On the other end, two molecules form head-to-head assemblies forming an NC1 hexamer. Recent studies show that chloride ions are essential in forming an organized network. Lateral interactions between collagens may also contribute to network assembly (not indicated in this schematic). Schematics are not to scale. To see this figure in color, go online.

All collagens share a characteristic triple-helical structure, formed by a (Gly-X-Y)_n repeating sequence in each of the three composite α-chains. However, two key molecular features distinguish collagen IV and fibril-forming collagens that have been shown to impact higher-order assembly. One is the globular C-terminal domain, whose presence inhibits fibril formation and is proteolytically removed from fibril-forming collagens before assembly (12). By contrast, the C-terminal noncollagenous domain of collagen IV (NC1) is not removed and plays a central, chloride-directed role in forming networks (13,14). The second distinguishing feature of collagen IV is the presence of natural discontinuities (interruptions) in the (Gly-X-Y)_n repeating unit within its collagenous domain. The absence or replacement of glycine every third residue prevents continuously triple-helical collagenous domains and promotes local destabilization (15,16). Interruptions are also present in other nonfibrillar collagens such as the FACITs, MACITs, and MULTIPLEXINs (1). Enhanced flexibility attributed to the presence of interruptions in collagens has been seen using rotary shadowing electron microscopy (EM) (17, 18, 19, 20); however, a detailed study on characterizing the functionality of distinct interruptions is lacking. Because triple-helix interruptions can result from disease-associated mutations in fibril-forming collagens and because for collagen IV, they may play a biological role in providing recognition sites for other macromolecular components in BMs (21, 22, 23), the ability to interrogate the physical properties of collagen in a sequence-specific manner is needed.

The size of the most abundant mammalian collagen proteins (≳300 kDa; ≳300 nm length) has imposed practical limitations on the structural insight available from conventional approaches. Diffraction-based analysis of molecular structure has been limited to studies of periodic structure within collagen fibrils (24,25) and to investigations of short triple-helical peptides (~30 amino acids or 10 nm in length) that are amenable to crystallization (23,26). These and other approaches have provided insight into the molecular determinants of triple-helix structure and stability, suggesting, for example, that imino acids (proline and hydroxyproline) are important for structural stability and that local interruptions in the (Gly-X-Y)_n repeating sequence associated with disease in fibril-forming collagens can disrupt the local structure (16,23). NMR studies on peptides have also found interruptions to result in increased local chain dynamics (23). Because collagen IV exists as heterotrimers, an additional structural question that arises—beyond whether or not a local sequence possesses a triple-helical structure—is the relative arrangement or stagger among its composite chains (27). Alternative methods for investigating the sequence dependence of collagen structure in the context of the full-length protein are needed.

Here, we use atomic force microscopy (AFM) imaging of individual full-length collagen proteins to provide insight into the local determinants of structure. Image analysis provides a map of local flexibility as a function of position along the collagen chain. By applying this technique to collagen IV and to collagen III, we contrast the mechanics of network-forming and fibril-forming collagens at the molecular scale. By mapping the position-dependent flexibility, we relate the local sequence within full-length collagen proteins to the local bending mechanics. Because of the role of chemical environment in regulating collagen supramolecular assembly, we apply this AFM-based flexibility mapping to investigate the effects of distinct chemical environments on flexibility of collagens, globally and in a sequence-specific manner. We interpret these results in light of outstanding physiological questions regarding control of collagen conformations during secretion and its supramolecular assembly in the extracellular space.

Materials and methods

Collagen sources

Heterotrimeric [α1(IV)]₂-α2(IV)] collagen IV (113 μg/mL in 0.5 M acetic acid) was purified from Engelbreth-Holm-Swarn (EHS) tumor in lathrytic mouse and was a gift of Albert Ries (28,29). Bovine pro-N collagen III (pN-III) was extracted from fetal bovine skin following a previously published protocol (30). Reducing SDS-PAGE analysis showed both samples to be predominantly free of intramolecular cross-links. Rat tail tendon-derived collagen I was purchased from Trevigen (Cultrex 3440-100-01; Gaithersburg, MD) and is pepsin-treated collagen with a stock concentration of 5 mg/mL in 20 mM acetic acid.

Sample preparation for AFM

Collagen flexibility was mapped after deposition from three different solution conditions: 1 mM HCl + 100 mM KCl (pH ~3) (10), sodium acetate buffer (150 mM NaOAc, 25 mM Tris-OAc (pH 5.5)), and Tris-buffered saline (TBS) (150 mM NaCl, 25 mM Tris-Cl (pH 7.4)); the pH of these solutions were confirmed to remain stable over months to years. Collagen was diluted into the desired solution conditions at a final concentration of 0.2 μg/mL, for which 50 μL of the diluted sample was deposited and allowed to incubate for 20 s on freshly cleaved mica (Highest Grade V1 AFM Mica Discs, 10 mm; Ted Pella, Redding, CA). The excess unbound proteins were removed by rinsing with ultrapure water, and the mica was then dried using filtered air. All proteins were imaged under dry conditions, and the solution conditions of the samples refer to the conditions in which they were deposited onto mica. Previous work suggests that collagen conformations on the mica surface reflect their deposition conditions (10). Imaging was done with an Asylum Research MFP-3D atomic force microscope (Goleta, CA) using tapping mode in air. AFM tips with a 160-kHz resonance frequency and 5 N/m force constant (MikroMasch, HQ: NSC14/AL BS; Sofia, Bulgaria) were used.

Chain tracing and analysis

SmarTrace, a custom-built MATLAB code (The MathWorks, Natick, MA), was used to determine the bending flexibility of collagen chains (10). Persistence length determination by SmarTrace has been extensively validated (10,31). Chains were traced to subpixel resolution starting from the C-terminus (NC1-chain boundary for collagen IV; nonglobular end for collagen III). Homogeneous chain analysis assumes chains to have uniform flexibility (persistence length). The persistence length was found from the dependence of mean-square end-to-end distance, $⟨R^2\left(\Delta s\right)⟩$ (Eq. 1), and tangent vector correlation, $⟨\cos \theta \left(\Delta s\right)⟩$ (Eq. 2), on the segment length Δs. This model treats the chains as worm-like chains (WLCs) equilibrated in two dimensions and has previously been shown to describe fibril-forming collagens well in 100 mM KCl, 1 mM HCl (10).

To investigate the sequence dependence of collagen’s flexibility, the local effective persistence length was determined using Eq. 4 (described in more detail in Supporting materials and methods, Text S1; Fig. S1). 95% confidence intervals for the estimates of p(s) were determined using the cumulative probability function of the scaled inverse χ² distribution (Fig. S2), as described in Supporting materials and methods, Text S1. The dependence of this estimate on the number of chains is provided in Fig. S3. Analysis of simulated chain images showed that a window of length Δs = 30 nm was the minimal Δs that reliably produced the expected angular distributions for a wide range of persistence lengths tested (Fig. S4). Thus, for a chain of contour length L, persistence lengths could be extracted for segments centered from s = 15 nm through s = L − 15 nm.

Simulated images of chains with inhomogeneous flexibility were generated in MATLAB (32) and analyzed as described for the experimental images. Chains were generated as described in (10), but with a bending stiffness that varies along the chain and incorporating a “knob” at one end as a directional reference point. Specifically, the chains we simulated here had a total contour length of L = 400 nm, interspersing long, stiff regions with short flexible regions (Fig. 3 B): p = 85 nm (Δs = 79 nm), p = 5 nm (Δs = 1 nm), p = 85 nm (Δs = 78 nm), p = 5 nm (Δs = 2 nm), p = 85 nm (Δs = 77 nm), p = 5 nm (Δs = 3 nm), p = 85 nm (Δs = 76 nm), p = 5 nm (Δs = 4 nm), and p = 85 nm (Δs = 80 nm). The knob was introduced as a uniform-intensity disk of radius 7 nm, centered at the starting point of the simulated chain. Experimentally realistic background noise was included in the images, as described previously (10).

Figure 3.

Extracted persistence length profiles from simulated inhomogeneous chains. (A) Images of WLCs with inhomogeneous flexibility profiles were simulated to validate position-dependent flexibility analysis. For segments of length Δs = 30 nm centered at position s along the backbone, the effective persistence length is determined from the variance of angular changes along these segments θ_s(Δs) (Eq. 4). Error bars reflect a √n counting error. (B) The extracted persistence length profile p^∗(s) obtained by tracing and analyzing n = 297 simulated chains (thick black line) shows inhomogeneous flexibility along the simulated chains. The shaded regions represent 95% confidence intervals on the estimates of persistence length. p^∗(s) attains the expected persistence lengths in plateaus longer than Δs = 30 nm and exhibits minima at locations of enhanced flexibility. Minima are broadened with respect to the input p(s) profile (thin black line), in agreement with the expected effective persistence length profile obtained by convolving a Δs = 30 nm filter with the input p(s) (result: dotted line). The two-class model fit (red) recovers distinct persistence lengths of the rigid and flexible regions, returning p_r = 92 nm and p_f = 6 nm, respectively (χ²_r = 1.06). n = 297 simulated chains. To see this figure in color, go online.

Sequence alignment

The α1 (P02463) and α2 (P08122) amino acid sequences were obtained from UniProt (33). All chain alignments were initiated at the edge of NC1 domain, starting at the first Gly-X-Y unit of collagenous domain of both α chains. Specifics of the alignments are provided in a Supporting material. Graphical alignment of the amino acid representation and persistence length profile assumes a length of 0.29 nm/aa in the collagenous domain.

Variable flexibility model

Each amino acid of an α-chain was assigned a value of “0” if it was within a (Gly-X-Y) sequence or “1” otherwise (within an interruption). Distinct alignments of the three α-chain sequences were tested (supplied as a Supporting material), including linear alignment, loop(s) within α2 to better align its (Gly-X-Y) sequence blocks with α1 (including a disulfide-bridged loop (34)), and sequence-similarity alignments (based on a Clustal alignment of the sequences (35), adapted to ensure chain continuity). A flexibility class was assigned at each amino acid step along the aligned chains based on whether it was a 0-, 1-, 2-, or 3-chain interruption in the triple helix (Table 1). Each of these classes was assigned a local bending stiffness (via a local persistence length), and thus, the model assigned a local flexibility to each position along the contour. This model was then fitted to the determined effective persistence length profile p^∗(s) to determine the persistence lengths p₀, p₁, p₂, and p₃ that best describe each flexibility class. Conversion factors were included in the fitting routine, to convert between primary (aligned) sequence, measured in amino acids, and position along the contour from image analysis, measured in nanometers. Because of variability in identifying the starting position of the chains (found also when tracing simulated chains), a chain-stagger parameter was included, which represented the standard deviation of model chain starting positions. Finally, because we do not know where the collagenous domains start (for example, their starting position may be obscured by the NC1 domain), an offset parameter was included to linearly shift the modeled chains with respect to the traced chains. The fitting procedure is described in detail in the Supporting materials and methods, Text S2.

Table 1.

Physical classes of interruption

Flexibility class	α1 interruption?	α2 interruption?
Triple helix (p0)	0	0
One-chain interruption (p1)	0	1
Two-chain interruption (p2)	1	0
Three-chain interruption (p3)	1	1

Modeling the simulated chains (Fig. 3) required a simpler physical model containing only two classes: rigid p_r and flexible p_f, located along the chains as defined by the inputs to the simulations. No length conversion factors were necessary because the simulated chain profile positions were defined in nanometers. Chain-stagger and offset parameters were necessary to align the model results with the traced and analyzed simulated chain p^∗(s) profile.

Data availability

Primary and analyzed data are available from the corresponding author upon request. Sequence alignments used in this work are provided in a Supporting material.

Results and discussion

Most studies of collagen mechanics have focused on fibril-forming collagens such as collagen I, which associate laterally to form higher-order structures as shown in Fig. 1 (8). Considerably less is known about the mechanics of collagen IV, which forms the scaffold of basement membranes. We first quantified flexibility differences between these types of collagens by analyzing images of many individual collagen molecules.

AFM images of collagen

We imaged mouse heterotrimeric [α1(IV)]₂-α2(IV)] collagen IV molecules (also referred to as protomers) derived from the matrix produced from EHS tumor using AFM. Multiple forms of heterotrimeric collagen IV exist (α112, α345, and α556), coming from six different gene products (7); because our study here is limited to the α112 form, we henceforth refer to it simply as “collagen IV.” AFM directly images molecules deposited on a surface, so is free from potential metal replication artifacts of rotary shadowing EM, which has previously been used for imaging conformations and binding interactions of collagen IV (4,17,18,36, 37, 38, 39, 40, 41, 42). With appropriate imaging conditions, AFM offers superior spatial resolution than the platinum nanocrystallite size of 1–2 nm in EM (17), and our studies avoid the use of glycerol, which has been shown to affect collagen IV assembly (36). Although imaging in liquid conditions is possible with AFM, here we deposited collagens from the specified solution conditions and dried them before imaging. This follows previous work that used AFM to image and quantify the flexibility of collagen types I, II, and III (10,31,43), to observe interactions between the NC1 domains of collagen IV (44,45), and to observe binding sites of laminin on collagen IV (46). For the conditions used in this study (>100 mM ionic strength deposition before drying), numerous statistical measures demonstrate the collagen to be equilibrated on the mica surface (10).

An example AFM image of collagen IV deposited from acidic solution (100 mM KCl, 1 mM HCl) is shown in Fig. 2 A, along with an image of collagen I for comparison. The acidic conditions preclude lateral assembly of both fibril-forming and collagen IV molecules (10,36). Both proteins possess long chains, representing the collagenous domains. The contour length for EHS-derived collagen IV protomers is 360 ± 20 nm (N = 262), in agreement with previous reports (36) and, as expected, longer than the ~300 nm contour length of fibril-forming collagens (17). Similar to previous images recorded using rotary shadowing EM (17,36), we observe by AFM that the collagen IV protomer is capped by a globular domain, known as the C-terminal NC1 domain. Evident in the AFM image are two kinds of molecules: single protomers and dimers of protomers linked end-on by NC1 hexamers.

Figure 2.

Homogeneous WLC analysis comparing flexibility of collagen IV to fibril-forming collagens. AFM images of (A) mouse collagen IV and (B) rat collagen I are shown, with schematics of their corresponding molecular structures placed above the AFM images. The black arrowhead and arrow point at an NC1 hexamer and an NC1 trimer of a molecule, respectively. Scale bars, 200 nm. All collagen types were deposited from room temperature onto mica from a solution of 100 mM KCl and 1 mM HCl. (C) The persistence lengths were obtained using $⟨R^2\left(\Delta s\right)⟩$ (dark gray; Eq. 1) and $⟨\cos \theta \left(\Delta s\right)⟩$ (light gray; Eq. 2) analyses, as shown for the collagen type IV data here. These data are fitted well by the predictions of the WLC model. (D) Bar plot comparing the persistence lengths of fibril-forming collages to collagen IV. Values of p were obtained using the $⟨R^2\left(\Delta s\right)⟩$ (dark gray) and $⟨\cos \theta \left(\Delta s\right)⟩$ (light gray) WLC fits. Errors represent 95% confidence intervals on p. The fibril-forming collagens possess similar persistence lengths of p ≈ 90 nm (data reproduced from (10)), whereas collagen IV exhibits a substantially lower effective persistence length of p = 39 nm. n = 262 collagen IV chains. To see this figure in color, go online.

Visual comparison of collagen I and collagen IV images suggests that the triple helix of collagen IV is more flexible; collagen I bends on a long length scale, whereas collagen IV exhibits more frequent and shorter-range bending fluctuations.

Conformational analysis of collagen IV as a homogenous polymer

To quantify the flexibility of the collagenous domain of collagen IV, we applied our previously developed chain-tracing and analysis algorithm, SmarTrace (10). SmarTrace traces and provides conformational analysis of imaged chains, allowing for the implementation of polymer physics tools to determine their mechanical properties. To analyze collagen as a homogeneous polymer, we traced chains collected from AFM images and randomly segmented the contours into nonoverlapping pieces of different segment lengths. We calculated the mean-square end-to-end distance $⟨R^2\left(\text{Δ}s\right)⟩$ and the mean correlation of the beginning and ending unit tangent vectors (given by their dot product) $⟨\hat{t}\left(s\right)\cdot \hat{t}\left(s+\Delta s\right)⟩$ for all segments of length Δs within the pool of collected chains. These were fitted with the predictions of the inextensible worm-like chain (WLC) model to estimate persistence length (10,47):

$$⟨R^2\left(\text{Δ}s\right)⟩=4p\text{Δ}s\left[1-\frac{2p}{\text{Δ}s}\left(1-e^{-\frac{\text{Δ}s}{2p}}\right)\right],$$ (1)

$$⟨\hat{t}\left(s\right)\cdot \hat{t}\left(s+\Delta s\right)⟩=⟨\cos \theta \left(\text{Δ}s\right)⟩=e^{-\frac{\text{Δ}s}{2p}}.$$ (2)

These equations assume the polymers to be equilibrated in two dimensions, which has been shown to be the case for collagens deposited under the solution conditions used here (10). Although we cannot test this explicitly, we assume that the persistence lengths determined from these two-dimensional images represent the bending flexibility of collagen in three dimensions, an assumption that holds reasonable agreement with previous collagen flexibility measurements (see discussion in (10)) and that has been clearly demonstrated for other systems like DNA (47,48). The collagen IV results for both $⟨R^2\left(\text{Δ}s\right)⟩$ and $⟨\cos \theta \left(\text{Δ}s\right)⟩$ appear to be well described by the WLC model (Fig. 2 C), from which we find a persistence length of p = 39 ± 2 nm. To our knowledge, this is the first report of a net persistence length of collagen IV.

The persistence length of collagen IV is significantly less than that of fibril-forming collagens deposited under the same solution conditions (Fig. 2 D). Our earlier work revealed that the persistence lengths of different fibril-forming collagens (I, II, and III) are very similar, all falling within ~10% of p = 90 nm under these solution conditions and being well described as semiflexible polymers (10). By contrast, the persistence length of collagen IV is less than half of this value, reflecting a more flexible protein. This stark difference in bending flexibility between collagen IV and fibril-forming collagens contrasts with their very similar thermal stabilities (49). (This is achieved in part via more extensive proline hydroxylation in collagen IV, which compensates for the presence of interruptions (49).)

This striking difference in flexibility likely relates to the interruptions that characterize collagen IV molecules. The triple helix of collagens is defined by a repetitive (Gly-X-Y)_n sequence, in which the Gly is obligatory to form a stable triple-helical structure. The collagenous domain of collagen I is 96% triple helical (with nonhelical regions confined to its telopeptide ends (50)), whereas in collagen IV it is ~80% triple helical, with interruptions to the (Gly-X-Y)_n sequence occurring throughout the chain (Fig. S5). This distinction is shown schematically above the AFM images in Fig. 2. Because unstructured polypeptide chains are extremely flexible with p < 1 nm (51), we expect interruptions in the triple-helix-defining sequence to significantly enhance flexibility. This expectation is consistent with our finding of a lower persistence length of collagen IV and with the visual comparison of images of collagen IV and collagen I chains (Fig. 2, A and B).

From the example chain images in Fig. 2 A, it appears that the flexibility of the collagen IV molecule varies along its contour and is more flexible away from the NC1 domain, i.e., toward the N-terminus. In a study using rotary shadowing EM to examine collagen IV, Hofmann et al. found that it exhibits regions of variable flexibility along its contour (17). Thus, although the homogeneous WLC model provides a good fit to our experimental data (Fig. 2 C), we sought to assess the heterogeneity of flexibility along the collagen IV triple helix.

Position-dependent flexibility analysis

To address structural variability and to evaluate the contributions of interrupted regions to the overall flexibility of collagen IV, we extended the SmarTrace analysis to include determination of the sequence-dependent chain flexibility. To do so, we use the variance of tangent angles around each position s along the chain backbone to quantify the bending stiffness at that location. This approach was developed by Hofmann et al. to analyze the position-dependent flexibility of collagens (17). This method complements other AFM image analysis methods developed to quantify discontinuous mechanical properties of DNA (52) or to map sequence-dependent properties of individual proteoglycans (53). It is expected to be more appropriate for determining the flexibility of collagen molecules than an alternative position-dependent stiffness analysis developed for more rigid biological filaments (54).

The flexibility of a filament at a position s along its contour is described by its local bending rigidity α(s). The bending rigidity is related to the persistence length via the thermal energy, k_BT:

$$p\left(s\right)=\frac{\alpha \left(s\right)}{k_{\mathrm{B}}T}.$$ (3)

The persistence length thus describes the flexibility of a chain with a given bending stiffness at a given temperature. Practically, we are limited to examining the flexibility of a chain over a segment of finite length Δs, which we define centered at position s; we use the term “effective persistence length” to denote this averaged nature of the response, p^∗(s; Δs). The effective persistence length of a segment of length Δs is determined from the variance of the tangent angles θ_s between the ends of the segment centered at s:

$$p^{\ast }\left(s;\Delta s\right)=\frac{\text{Δ}s}{\text{Var}\left({\theta }_{\mathrm{s}};\text{Δ}s\right)}.$$ (4)

The analysis methodology was validated through tests on simulated chains. We generated AFM images of chains with inhomogeneous flexibility profiles, traced the chains using SmarTrace, and analyzed the resulting contours using Eq. 4. The chains were simulated as previously described (10), here incorporating a position-dependent bending stiffness and a “knob” at one end (which served as a directional reference point, analogous to the NC1 domain of collagen IV). Fig. 3 A shows an example of a simulated image, along with a schematic of the inhomogeneous flexibility profile imposed in the simulations (long regions of relatively stiff chains with p = 85 nm, interspersed with very short regions with p = 5 nm). The short regions of substantially increased flexibility are not visually apparent in the simulated images but are clearly seen in the effective persistence length profile determined from the traced chains (Fig. 3 B). A notable difference between the input and extracted persistence length profiles is the apparent broadening of the flexible regions: p^∗(s; Δs) exhibits much broader wells than the input p(s) profile. This is the expected result of the analysis: the Δs = 30 nm segment length is substantially longer than the regions of flexibility in our simulated chains and acts to some degree as a low-pass filter of flexibility along the chain (though we stress that p^∗(s) is not simply the mean of p(s) over a 30-nm window; see Supporting materials and methods, Text S1; Fig. S1). Fig. 3 B includes the expected effective persistence length profile, which considers this convolution between the ideal chain and the 30-nm filter. This expected profile agrees well with p^∗(s; Δs) obtained from analyzing the simulated images. Although a shorter segment length should provide more localized information on sequence-dependent flexibility, our validation tests found Δs = 30 nm to be the minimal length that returned reliable estimates of p^∗(s) from simulated chain images (Fig. S4). Thus, Δs = 30 nm was used for the analysis presented herein.

Having validated the persistence length mapping algorithm, we applied it to experimental images of collagen IV. The effective persistence length profile of collagen IV is shown in Fig. 4. From this result, it is obvious that the flexibility of collagen IV varies markedly along its contour, with the N-terminal half of the molecule being significantly more flexible than the region closer to the NC1 domain. Similar flexibility trends are observed whether collagen chains are traced starting at the NC1 domain (Fig. 4) or in the reverse direction from the 7S domain (Fig. S6). Our flexibility profile agrees well with that of Hofmann et al., who also observed increased flexibility toward the N-terminus of collagen IV (17).

Figure 4.

Position-dependent flexibility profile of collagen IV. (A) Schematic representation of the α1 and α2 amino acid sequences from mouse collagen type IV. Rectangles indicate regions of the sequence containing triple-helix-competent sequences with (Gly-X-Y)_nG, n ≥ 4. Interruptions of this repetitive sequence are indicated by thinner lines and occur more frequently toward the N-terminus of the chains (right side of the schematic). (B) Position-dependent persistence length map of collagen type IV deposited from 100 mM KCl, 1 mM HCl. The dashed line indicates the persistence length of continuously triple-helical fibril-forming collagen molecules (10). Shaded curves represent 95% confidence intervals on the effective persistence length estimate p^∗(s; Δs). The profile was calculated from n = 262 chains. The effective persistence length map begins ~15 nm into the collagen sequence because of the use of centered 30-nm windows for determination of p^∗. (C) Four-class flexibility model using two-loop (blue) and Clustal (red) alignments. Alignments differ in the number of amino acids that loop out and thus do not participate in the main backbone, as shown schematically above and below the plot. The persistence length profile is aligned with the amino acid sequence representations using model outputs for best-fit offset (nm) and nanometer/amino acid conversion. The two-loop (Clustal) alignment is best fitted with persistence lengths of p₀ = 105 (109) nm, p₁ = 83 (86) nm, p₂ = 21 (42) nm, and p₃ = 1.9 (2.0) nm, with χ²_r = 12.6 (12.8). Best-fit persistence lengths of these different structural classes, particularly p₁ and p₂, differ for other chain alignments (Table S1). To see this figure in color, go online.

To gain sequence-dependent insight into the flexibility profile requires aligning this map with the collagen IV sequence, which was unavailable to Hofmann et al. (17). An amino acid representation of the full sequence of mouse α1α1α2 collagen IV is shown in Fig. 4 A. The α-chain polypeptide sequences have been aligned from the beginning of the collagenous domain beside the NC1 domain (from the first Gly-X-Y). The blue and red boxes correspond to segments of α1(IV) and α2(IV), respectively, that have the (Gly-X-Y)_n sequence required to form a triple helix. In this representation, we have assumed a minimal length of (Gly-X-Y)₄G to be sufficient to form a triple helix. This assumption of n = 4 does not strongly affect the representation (Fig. S7). The thinner lines in Fig. 4 A represent interruptions in the triple-helix-competent (Gly-X-Y)_n repetitive sequence.

The position-dependent flexibility profile of collagen IV aligned with the amino acid sequences is shown in Fig. 4, A and B. An examination of the sequences of the α-chains reveals that most of the interruptions lie in the N-terminal half of the protein, where we also see an increase in flexibility (lower persistence length). Maxima in local persistence length align well with extended triple-helical stretches in all three chains, found ~30, 90, and 160 nm from the NC1 domain. Within these plateau regions, collagen IV’s effective persistence length is p^∗ ≈ 95 nm, the persistence length found for continuously triple-helical collagens (see Fig. 2) (10). Between these maxima, minima in persistence length are found ~60 and 130 nm from the NC1 domain. These two locations align well with regions of the sequence in which both α1 and α2 chains possess interruptions in the (Gly-X-Y)_n sequence. This suggests that interruptions of the triple-helix-forming sequence in all three α-chains of α1(IV)₂–α2(IV) strongly impact the flexibility of the molecule, apparently more so than having interruptions in only one or two of the constituent chains.

Model for interpreting position-dependent flexibility

To gain a deeper structural interpretation of the persistence length profile, we implemented a simple physical model to describe the sequence-based flexibility of a collagen containing interruptions (Supporting materials and methods, Text S2). This model assumes that the local flexibility of collagen can be completely described by how many of the three component chains contain a triple-helix-compatible sequence at that location. To implement the model, each amino acid in each of the α1 and α2 chains was assigned a value of either “0” (contained within triple-helix-compatible sequence (Gly-X-Y)_n) or “1” (within an interruption). The three α-chain sequences were aligned starting from the edge of the NC1 domain, and different chain alignments were tested. These include a linear, sequence-based alignment (Figs. 4 A and S8); nonlinear alignments that include previously proposed intrachain loops within α2 (34,55,56) and improve the registration of triple-helix-forming sequences; and alignments arising from a Clustal-based analysis of sequence similarity between α1 and α2 chains (35), which require the incorporation of copious loops and bulges to confer backbone continuity.

For each chain alignment, a flexibility class was assigned at each amino acid step along the backbone based on whether it was a 0-, 1-, 2-, or 3-chain interruption in the triple helix (Table 1). Each of these classes was assigned a local bending stiffness (via a local persistence length) and thus the model assigned a local flexibility to each position along the contour. This model was then fitted to the determined persistence length profile p^∗(s; Δs) to determine the persistence lengths p₀, p₁, p₂, and p₃ that best describe each flexibility class. The approach was validated by application to the simulated chains; here, a two-class model (rigid or flexible) recovered the distinct persistence lengths of the rigid and flexible regions and described the data well (p_r = 92 nm; p_f = 6 nm; χ²_r = 1.06). A slight overestimate of persistence lengths is expected for finite samples (Supporting materials and methods, Text S1). Agreement between the four-class flexibility model and the measured persistence length profile p^∗(s; Δs) of collagen IV would indicate that this physical classification is a valid approach to describing the variable flexibility of collagen IV. By contrast, disagreement would indicate that other factors (such as the influence of X and Y identity on flexibility, length of interruption, and location with respect to interruptions (extended triple-helix tracts) may be of key importance for quantifying the structural properties of this protein.

This model of a physically heterogeneous collagen IV captured positional variations in flexibility along the chain, comparing favorably to the measured p^∗(s) profile (Fig. 4 C). The model returned a value for triple-helix local persistence length of p₀ ~100–160 nm for all distinct α-chain alignments tested (Table S1). This estimate of p₀ is somewhat larger than the net persistence length of continuously triple-helical collagens found by treating the chains as homogeneous (Fig. 2 D) but is consistent with local values of p^∗ determined along their backbones (see below). Also consistent among chain alignments was the finding that p₂ < p₁. In other words, having an interruption in two of the chains leads to a more flexible structure than having a one-chain interruption.

Some of the chain alignments resulted in fits of the variable flexibility model that were not minimized within the specified parameter range: for example, a linear chain alignment (aligned based only on primary sequence) produced for an overlapping interaction p₃ = 200 nm, the maximal value allowed for fitting, a value that represents an unphysically rigid structure of three polypeptide chains that lack any triple-helix-forming sequence. Alignments that returned this high value, however, were the only ones that captured the rigidity in the collagenous region adjacent to the NC1 domain (e.g., Fig. S8).

Alignments that optimized the fit within their parameter bounds returned an optimal length scaling of 0.29 nm/amino acid, commensurate with triple-helix lengths from collagen diffraction analysis (23,57). These alignments also returned the smallest value for p₃ ≈ 2 nm, i.e., that an overlapping interruption is the most locally flexible structure within the collagenous domain of collagen IV. These optimal alignments require at least some outward loops in the longer α2 chain (e.g., “two-loop” alignments) and perhaps extensive minor loops and bulges along the length of each of the three α-chains (“Clustal” alignments).

In some regions, the variable flexibility model underestimates the rigidity, all located in the C-terminal half of the protein (Fig. 4 C). The region adjacent to the NC1 domain exhibits a rigidity similar to the other extended triple-helical regions, yet the model significantly underestimates the persistence length, both of the plateau at s ≈ 30 nm and of the adjacent minimum at s ≈ 50 nm. These model results imply that the overlapping interruption closest to the NC1 domain possesses a distinctly enhanced stability not captured by our simple physical model. Enhanced stability in this region could be conferred by post-translational glycosylation of hydroxylysines within this overlapping interruption (58, 59, 60). Unfortunately, a glycosylation map of collagen IV lacked sequence coverage in this region (61), so this remains a speculative, yet testable, prediction.

The alignment of the three chains of the α1α1α2 mouse collagen IV collagenous domain is unknown. A disulfide-bridged loop within α2 has been proposed, which helps better to align triple-helix-competent sequences in the chains (34,55); this is included as the larger, N-terminal loop in our “two-loop” alignment (Fig. 4; Supporting materials and methods). Most of our chain alignments also agree with the determined alignment of α-chains at an integrin-binding site in human collagen IV (Table S1), the only location for which chain alignment in collagen IV has been determined (27). To validate or further constrain chain alignments will require improvements in spatial resolution from this AFM imaging and chain-tracing approach and/or complementary approaches such as proteolysis to identify unstably structured regions within the collagenous domain (62).

In our alignments, we ignored the register of the three chains, as the register has been identified at only one location and only in human collagen IV (27). We tested a staggered alignment of the three chains that included this identified local alignment, but it failed to produce an optimized fit (Table S1). It is highly likely that some of the interruptions alter the local register of the three chains (such that a leading chain at some locations may become the middle or trailing chain at others) (16) and that one “universal” registration does not apply along the entire collagenous domain.

It is also possible that the variable flexibility model is insufficient to completely describe the measured flexibility profile because of its many simplifying assumptions. Beyond stabilization by post-translational modifications, potential intramolecular cross-links (though these were few in our sample; Materials and methods), and local variations in chain stagger, there are other attributes that may be of direct relevance to local stability and flexibility. For example, we have ignored the length dependence of interruptions when defining flexibility class. Short one-chain interruptions may act as “stammers,” serving to alter the local register among chains instead of breaking the triple-helix structure (16). Single glycine substitutions may introduce local kinks in the structure (63). Longer interruptions may destabilize flanking triple-helical regions, leading to increased breathing of the adjoining triple helix and a lower persistence length (64). The local sequence within and flanking the interruption also may alter its stability (23). And finally, this physical model ignores all sequence-dependent attributes of not only the interruptions but also the triple helix itself by classifying each amino acid purely by whether or not it is found within a (Gly-X-Y) sequence. We next sought to test the assumption of mechanical homogeneity within the triple helix by mapping the flexibility profile of a fibril-forming collagen.

Triple-helix mechanical inhomogeneity: collagen III

To probe the variability of bending stiffness along the triple helix, we imaged the pN-III molecular variant of bovine collagen III using AFM (Fig. 5 A). Collagen III is a homotrimeric [α1(III)]₃] fibril-forming collagen devoid of interruptions in its ~300-nm-long triple-helical domain. The pN-III variant retains its N-terminal propeptide (65), providing us with a directional reference point for position-dependent flexibility analysis.

Figure 5.

Position-dependent flexibility profile of collagen III. (A) AFM image of bovine collagen pN-III deposited from room temperature onto mica from a solution of 100 mM KCl and 1 mM HCl. Scale bars, 200 nm. (B) Position-dependent persistence length map of collagen III aligned with its amino acid sequence representation. This collagen is continuously triple helical and therefore is represented as a single bar with the MMP site marked. Shaded curves represent 95% confidence intervals on the effective persistence length. The profile was calculated from n = 267 chains. To see this figure in color, go online.

We found that the flexibility of a triple helix varies significantly along its contour, although less than along the interruption-containing collagen IV (Fig. 5 B). This provides direct evidence for distinct mechanical properties of different sequences within the collagen triple helix. Our results show considerably more variation in local flexibility along the backbone of collagen III than previous rotary shadowing EM measurements on this molecule (17), perhaps because of the differences in resolution provided by the two imaging and chain-tracing approaches and/or the choice of segment length used to determine the effective persistence length. Our flexibility map shows that the effective persistence length ranges from ~60 to 180 nm in different regions of collagen III. Notably, the larger values are similar to values returned for p₀ of the triple helix in our model fitting, above, which ranged from 100 to 160 nm, and are longer than the net persistence length of continuously triple-helical collagens (Fig. 2 D).

In the context of our physical model for collagen IV, this variability of persistence length within collagen III has additional implications. The range of p^∗ found for collagen III suggests that the assumption of a single persistence length p₀ to characterize all triple-helical regions of collagen IV is an oversimplification. However, the model-extracted values of p₂ and p₃ are significantly shorter than the lowest values of p^∗ along the collagen III chain, with values of p₁ similar to or less than the most flexible regions of collagen III. It is possible that two chains possessing a (Gly-X-Y)_n sequence may be able to induce a triple-helix-compatible structure in the third chain, leading to a value of p₁ similar to regions of the collagen III triple helix. Overall, we can conclude that the enhanced flexibility of collagen IV is dominated by the effects of interruptions, though variability of X, Y residues within a continuous (Gly-X-Y)_n sequence also influences bending flexibility. Previous single-molecule imaging has suggested that sequence may also control the local diameter of collagen (66).

The regions of greatest flexibility along pN-III collagen are found near the N-terminus (which contains conformationally distinct regions (67,68)) and, intriguingly, at the matrix-metalloprotease (MMP) site (Fig. 5 B). This site is the target of proteolysis by MMPs and is of key importance for extracellular matrix remodeling, e.g., during embryonic development and cancer metastasis. To our knowledge, ours is the first demonstration of enhanced mechanical flexibility at the MMP site. Our ability to detect this feature when previous studies using EM did not (17) may arise from the higher-resolution chain tracing in our AFM images compared to the particulate nature of rotary shadowed chains (31) and from our directional tracing from the C-terminus toward the N-terminus. When we traced the chains starting at the N-terminus, the contrast in flexibility around the MMP site was diminished (Fig. S9), perhaps resulting from challenges in identifying the start of the triple-helical domain in the presence of the N-propeptide (67).

The MMP site has previously been identified as possessing distinct characteristics within the collagenous domain. Intriguingly, the sequence in this MMP region of collagen I has been found, in peptides, to exhibit structural tautomerism, transitioning between a triple-helix and β1-bend structure (69). It is interesting to speculate that this structural flexibility may underlie the enhanced bending flexibility revealed here through AFM imaging. The MMP region also exhibits enhanced sensitivity to proteolysis by trypsin and has been suggested to have an enhanced propensity to unwind (70, 71, 72, 73, 74, 75, 76). We propose that this structural instability is what gives rise to the enhanced bending flexibility in this region; three weakly interacting chains are predicted to bend more easily than a tightly wrapped triple helix. This would produce a lower local persistence length.

The decreased triple-helix stability in the MMP region of fibril-forming collagens has been attributed to a low imino acid content (73,77,78). Thus, we examined how local imino acid content correlates with flexibility (Figs. S10 and S11). We saw no obvious correlations between imino acid content and flexibility along the collagen III contour (Fig. S10 A), which was confirmed by a linear correlation coefficient of R² = 0.017. Although an anticorrelation between imino acid content and flexibility in the C-terminal half of collagen IV might be inferred by visual comparison of these profiles (Fig. S11 B), quantitative analysis revealed no statistical correlation between these quantities, neither for the full length of the chain (R² = 0.084) nor for the first 200 nm from the NC1 domain (R² = 0.15). Thus, a simple picture of imino acids enhancing rigidity (17,79) does not apply here. Incidentally, recent work has suggested that imino acids may instead enhance the flexibility of triple-helical regions (80). Although the anticorrelation we observe in some regions supports this conclusion, we do not find this to be the case globally along the collagen backbone. Thus, although imino acids may contribute to local flexibility, at present our data suggest that they are not the main driver of bending flexibility of the triple helix.

We also looked for correlations between the locations of the imino acid within the (Gly-X-Y) triplet and the flexibility. This is because most prolines in the Y position are 4-hydroxylated (whereas most in the X position are not, with a small fraction 3-hydroxylated) (61,81, 82, 83). (We do not have specific information about the hydroxylation state of each proline in our samples.) Hydroxyproline increases the thermal stability of the triple helix (84,85), and we thus wished to determine whether its presence correlated with increased triple-helix bending rigidity. As seen in Fig. S10 B, we find no correlation between flexibility and Y-positioned prolines. X-positioned prolines can enhance conformational flexibility (within Gly-Pro-Hyp triplets) and have recently been correlated with enhanced thermal stability of collagens (80). However, we also find no correlation between local flexibility within collagen III and the local density of X-positioned prolines (Fig. S10 B).

The physical model we have developed to assess the structural relationships between interruptions and flexibility could be adapted to study sequence-dependent influences on flexibility in collagens lacking interruptions, such as collagen III. At present, our flexibility analysis is limited by accuracy in chain registry and has an ~100-amino-acid resolution (30-nm filter window) (Supporting materials and methods, Text S2), but improvements in imaging and tracing algorithms should be able to push this to finer length scales. Even with the current resolution, this approach should allow initial comparisons of bending flexibility with measured and predicted variations in local helical symmetry, thermal stability, sequence content, and post-translational modifications. Such comparisons would provide important links among the composition, structure, and mechanics of collagen and would offer a further means to connect high-resolution studies on triple-helical peptide constructs with their in situ properties in the full-length proteins (23,86). Such knowledge of the breadth of properties exhibited by continuously triple-helical structures would enable elucidation of the effects of glycine substitutions and of sequence interruptions on the molecular structure and mechanics of collagen (63,87,88).

Physiological implications of collagen IV flexibility

Thus far, our analysis has determined collagen IV properties only when deposited from acidic solution conditions. These conditions (pH ~3) are more acidic than the varying physiological conditions during biosynthesis and secretion. Previous studies of collagen IV flexibility also deposited the protein from very nonphysiological conditions (25 mM acetic acid, 50% glycerol) (17). Thus, we sought to investigate its flexibility in more physiologically relevant conditions.

Intracellularly, folded proteins are transported in secretory vesicles with low pH and chloride concentrations relative to the cellular exterior (89,90). Upon secretion, collagen’s chemical environment changes to one with a higher chloride concentration and neutral pH. This environmental switch is important for triggering assembly into higher-order structures; NC1 interactions strengthen at neutral pH and higher chloride concentrations (42,91,92), supporting collagen IV assembly into networks, and neutral pH and chloride favor lateral assembly of fibril-forming collagens (93,94). Environmental conditions such as pH and ionic strength have also been found to influence the conformations of fibril-forming collagen molecules (10,43). These observations raise the question of how much the conformations of the collagenous domain of collagen IV are controlled by chloride and pH. We posited that the pH and chloride gradient experienced during secretion might affect the flexibility of collagen IV, conferring a shorter persistence length and enhanced flexibility early in the secretory pathway, which would impart a lower free energy cost for compact configurations.

To test this hypothesis, we analyzed collagen IV under two additional solution conditions of pH and Cl concentrations. To represent vesicular solution conditions, we took an extreme limit by eliminating the chloride ions and used a sodium acetate buffer at a slightly acidic pH (150 mM NaOAc, 25 mM Tris-OAc (pH 5.5)). TBS (150 mM NaCl, 25 mM Tris-Cl (pH 7.4)) served as an extracellular proxy because of its relatively high chloride content and neutral pH.

AFM images of collagen IV molecules deposited from each of these two solution conditions are shown in Fig. 6, A and B. Visually, the collagen conformations look similar in both images and appear similar to those found in the previous condition of 100 mM KCl, 1 mM HCl (pH ~3) (Fig. 2 A). It is important to note that although collagen was deposited from a chloride-free environment in the AFM image in Fig. 6 B, there are still some head-to-head assemblies. This is due to the presence of sulfilimine cross-links in the NC1 hexamer, a post-translational modification that endows mechanical strength and stability to the collagen IV network (95). The presence of such cross-links protects the dimers from dissociation in a chloride-free environment. We do not observe any evidence of higher-order association or network formation in our images in these conditions, likely because of the low solution concentration of collagen IV used for deposition (0.2 μg/mL) (36).

Figure 6.

Effect of chloride and pH on the flexibility of collagen IV. Inspired by the changes in the chemical environment along collagen’s secretion pathway, we imaged collagen IV under two solution conditions, (A) TBS (150 mM NaCl, 25 mM Tris-Cl (pH 7.4)) and (B) sodium acetate buffer at a slightly acidic pH (150 mM NaOAc, 25 mM Tris-OAc (pH 5.5)). Scale bars, 200 nm. (C) Position-dependent flexibility map of collagen type IV. The shaded curves represent 95% confidence intervals on the effective persistence length estimates p^∗(s; Δs). The profiles were calculated from n = 272 (TBS; black) and n = 273 (NaOAc; red) chains. To see this figure in color, go online.

Surprisingly, we find similar overall flexibility among both buffered solution conditions (net persistence lengths p = 43 ± 3 nm for TBS and p = 41 ± 3 nm for NaOAc; Fig. S12) and the previous acidic solution (p = 39 ± 2 nm for 100 mM KCl + 1 mM HCl). Collagen IV exhibited no global curvature in any of these conditions (data not shown). Therefore, collagen IV flexibility is not tuned by changes in pH and chloride concentrations that mimic those experienced during secretion.

The net persistence length of a collagen molecule can be used to estimate its average size in solution, which we can compare with the size of vesicles that have been proposed to transport collagens from the endoplasmic reticulum (ER) to the Golgi. To do so, we assume that collagen molecules adopt a random coil conformation in solution whose extent is approximated as a sphere with a radius given by the radius of gyration, R_g. We find that collagen IV is predicted to be considerably more compact (R_g = $\sqrt{\frac{pL}{3}}$ = 74 nm, assuming a contour length L = 410 nm and p = 40 nm) than the interstitial, fibril-forming collagens (R_g = 95 nm, with L = 300 nm and p = 90 nm). Thus, despite its longer length, collagen IV would pose less of a size burden to the secretory machinery of the cell if it were to be transported in small vesicles. The 150–190 nm diameter of such random coils is significantly less than the 300-nm length dimension commonly stated as required for transport, determined by (incorrectly) assuming fibril-forming collagens to act as rigid rods. Nonetheless, both collagen IV and fibril-forming collagens would require additional compaction to fit in the standard COPII vesicles involved in ER-to-Golgi transport (60- to 90-nm diameter) (89,96).

Although it is generally accepted that collagens are not trafficked in such small vesicles, there remains considerable uncertainty about the mechanism used for procollagen trafficking during secretion. To our knowledge, there is no evidence of any collagen-associated proteins such as Hsp47 acting to compact collagen to facilitate loading into ER exit vesicles. In fact, some studies have found collagens IV and VII to be transported from the ER to Golgi in larger vesicles with 400 nm diameter (97,98), into which fibril-forming and collagen IV could easily be accommodated. However, recent work has found that direct trafficking of procollagen I from the ER to Golgi occurs and that these large vesicular carriers are not required (99). In this case, collagen (in)flexibility would not pose a strong energetic cost on its transport to the Golgi, consistent with our finding that collagen flexibility is not tuned by chemical environmental changes experienced during secretion.

As shown above, collagen IV should not be viewed as a uniformly flexible structure, so again, we performed a position-dependent analysis. We sought to determine how both triple-helical and interrupted regions of the collagenous domain of collagen IV are affected by chloride and pH. These regions have been implicated in assembly into higher-order networks (40), and we posited that their stability and flexibility would vary between intracellular (lower Cl⁻ and pH; assembly-disfavoring) and extracellular (higher Cl⁻ and pH; assembly-favoring) conditions. Specifically, because the triple helix is thought to loosen and destabilize at lower pH and Cl⁻ (albeit at significantly lower ionic strengths than used here) (10,100), we anticipated that a similar destabilization would enhance microunfolding of the triple-helical regions of collagen IV. This would be seen by a shortening of the high persistence length plateaus in sodium acetate relative to TBS and a concomitant broadening of the minima.

Surprisingly, we find that the flexibility profile of collagen IV is affected very little by these changes in pH and chloride. Fig. 6 C shows the superposition of the flexibility profiles of collagen IV in sodium acetate and TBS buffers, including the respective 95% confidence intervals of the effective persistence lengths. Over the vast majority of the collagenous domain, the effective persistence length is statistically unaffected by this change in solution conditions upon deposition. This contrasts with the strong mechanical response of collagens to decreases in ionic strength (10). Over the range of pH from ~3 to 7.4, the net charge on collagen IV is expected to change significantly, from positively charged to approximately neutral (Fig. S13). Despite this change in charge, the net persistence length of collagen IV remains the same (p = 40 nm), and its position-dependent flexibility is essentially invariant (Fig. S14). This finding provides further evidence that the conformations we observe are not governed by electrostatic interactions with the surface. Furthermore, it implies that electrostatic interactions such as salt bridges (101) do not contribute substantially to the net bending stiffness of collagen at physiologically relevant ionic strengths.

There are only a few regions in the effective persistence length profile where the flexibility differs between these two buffered solution conditions. The region ~170 nm from the NC1 domain is of interest as a potential site for NC1 binding that may lead to higher-order assembly (36). The sensitivity of the structure and mechanics in this region of the collagenous domain to solution pH and Cl⁻ could provide a responsive element that prevents premature assembly inside the cell. AFM-based mapping of ligand binding sites along collagen IV could help to reveal how these interactions are modulated by pH and chemical environment, and to relate the binding to local collagen structure. Sequence-based predictions of binding sites on collagen IV have been made for various binding partners (35,102), which could be compared with AFM-based mapping experiments. Of particular interest would be proteins implicated in chaperoning collagen folding (e.g., Hsp47), secretion (e.g., TANGO1), and assembly (e.g., SPARC), as well as other proteins involved in BM assembly (103,104). For example, EM-based mapping has found laminin, a major BM component, to bind 140 nm from the NC1 domain along the collagen IV molecule (37). Analysis of our persistence length profile shows that this binding site is at a flexible region, which coincides with a large overlapping interruption. On the other hand, the glycoprotein nidogen was found to bind 80 nm from the NC1 domain (41), within a triple-helical region. These results suggest that there may be a broad diversity of binding modes to collagen IV, which can be modulated by its local structure and perhaps further tuned by the chemical environment. Studies of the temperature-dependent flexibility profile, binding interactions, and in vitro assembly morphologies could elucidate the impact of sequence-dependent mechanical and thermal stability on guiding BM assembly.

Our findings on the bending flexibility of collagen can be used to quantify energetic costs of various candidate compartments and constraints along its secretory and assembly pathways. For example, the enhanced flexibility of collagen IV imparted by its interruptions means that it is expected to adopt more compact conformations in solution compared with its shorter fibril-forming collagen counterparts. Thus, its secretory burden on the cell machinery may be less. However, this flexibility of collagen IV endows a greater entropic cost for lateral ordering into higher-order structures compared with the stiffer collagen III. The greater rigidity of continuously triple-helical collagens facilitates their lateral assembly into highly ordered thick fibrils. We speculate that these mechanical properties contributed to the development of collagen in evolution, in which collagen IV arose first but now is a minority component, at least by mass portion, compared with fibril-forming collagens that form the basis of most connective tissues in higher-level organisms (6).

Conclusions

AFM-based mapping offers the ability to measure local structural properties such as bending stiffness within the context of full-length collagen proteins and correlate these with sequence-dependent properties elucidated from studies on much shorter collagen-mimicking peptides. Our results reveal a heterogeneous mechanical response along the triple-helical domains of collagens. We found that the presence of interruptions provides significant flexibility to collagen IV and were able to describe the shape of its flexibility profile reasonably well with a simple physical model that classifies sequences solely based on the presence of interruptions in the (Gly-X-Y) pattern. In our complementary imaging and analysis of collagen III molecules, we found significant mechanical heterogeneity along their lengths, indicating that a more refined picture of collagen flexibility should include sequence-dependent flexibility within triple-helical regions. Notably, we found collagen III to possess a unique mechanical signature at the MMP site, a key sequence for collagen’s extracellular remodeling. Surprisingly, in light of past suggestions that proline content influences local triple-helix rigidity (17,67,68,70,71), we find no net correlation between local proline abundance and flexibility of collagenous domains, though some tracts in both collagens III and IV exhibit an anticorrelation between flexibility and imino acid content. In this study, our analysis was limited to determining effective persistence lengths over 30-nm segments and deconvolving the effect of local sequence; future improvements in imaging and image analysis that allow more refined backbone tracing and registry of chains should further enhance our understanding of how local sequence influences mechanical properties of collagens.

Author contributions

A.A.-S. performed all experiments and analyzed all experimental data. A.L. devised and validated the position-dependent analysis protocols. A.A.-S. and N.R.F. developed the variable flexibility model. All authors contributed to the project design and to writing the manuscript.

Editor: Markus Buehler.

Supporting citations

References (105, 106, 107) appear in the Supporting material.