DRG–FLS co-culture gels (n = 36) were fabricated and allocated to an equibiaxial stretch under normal culture conditions (naïve n = 12); equibiaxial stretch after five days of incubation with an ilomastat inhibitor (inhibitor n = 14) or equibiaxial stretch after five days of incubation with dimethyl sulfoxide (DMSO) was taken as a vehicle control group since the ilomastat was dissolved in DMSO (vehicle n = 10). Half of the co-cultures from each group underwent an equibiaxial stretch, and the other half served as unstretched controls to account for the effect of each of the stretch and the incubation procedures at five days before the testing.

The culture media of co-culture gels receiving inhibitor were treated at each of day-in-vitro (DIV) 3, DIV6, and DIV7 with a 25 nM dose of ilomastat (Figure 1). Doses added to the culture media on DIV7 before the sub-failure stretch were given one hour before testing (Figure 1). The 25 nM ilomastat dose (GM6001; Millipore Sigma; Burlington, MA) was prepared from a 2.5 mM stock solution in DMSO (5 μL) dissolved into sterile cell culture H₂O to a concentration of 0.5 μM (Conant et al., 2004; Rogers et al., 2014); 25 μL of the 0.5 μM dilution was then added to the media to achieve a final concentration of 25 nM. The matched vehicle group was prepared accordingly with DMSO since the purchased stock ilomastat was stored in DMSO. The DMSO vehicle was prepared identically to the ilomastat with sterile DMSO (Invitrogen; Waltham, MA) dissolved in sterile H₂O to a concentration of 25 nM.

All cells were harvested from Sprague Dawley male rats under University of Pennsylvania IACUC-approved conditions and using sterile procedures as previously described (Zhang et al., 2018a; Ita and Winkelstein, 2019a). In brief, DRGs from all spinal levels were taken from embryonic day 18 Sprague Dawley rats obtained from the CNS Cell Culture Service Center of the Mahoney Institute of Neuroscience at the University of Pennsylvania. DRGs were dissected using fine forceps to remove individual DRG explants after exposing the rat’s spine and removing the spinal cord (Ita and Winkelstein, 2019a). In a monocellular, isolated culture, a DRG feeding medium consisted of Neurobasal feeding medium supplemented with 1% GlutaMAX, 2% B-27, 1% fetal bovine serum (FBS), and 10 ng/ml 2.5S nerve growth factor (all from Thermo Fisher; Waltham, MA) and 2 mg/ml glucose, 10 mM FdU, and 10 mM uridine (all from Sigma-Aldrich) (Zhang et al., 2018a).

FLS were harvested from both hind knees of a sexually mature adult rat by finely dissecting the capsular tissue surrounding the knee joints and dicing the isolated tissue (Saravanan et al., 2014; Ita and Winkelstein, 2019a). The diced tissue from both knee capsules was incubated together in Dulbecco’s Modified Eagle Medium (DMEM; Thermo Fisher) with 10% fetal bovine serum (FBS), 1% penicillin–streptomycin (P–S) (Thermo Fisher), and 2 mg/ml crude bacterial collagenase (C0130; Sigma-Aldrich) for six hours at 37°C under gentle agitation (Saravanan et al., 2014; Ita and Winkelstein, 2019a). Digested tissue was filtered with a 70 μm cell strainer, spun down at 300 g for five minutes, and resuspended in DMEM with 10% FBS and 1% P–S. The culture medium was changed every other day, and cells were passaged at 90% confluence.

Collagen gels were fabricated with rat tail type-I collagen (2 mg/ml; Corning Inc.; Corning, NY) cast in 12-well plates (1ml/well) with re-suspended FLS from passages three–five (∼5 × 10⁴ cells/mL) (Ita and Winkelstein, 2019a) and cultured for one day in DMEM media. At DIV1, DRGs (6-10/gel) were seeded onto the gel surface in 100 μL in supplemented Neurobasal medium. After allowing 24 h for DRGs to attach, the media were changed to supplemented Neurobasal media with 5% FBS that was optimized for the DRG-FLS co-culture system (Ita and Winkelstein, 2019a). Co-cultures underwent a full media change on DIV3. On DIV6, gels were triple-rinsed with 1X phosphate-buffered saline (PBS) and an additional layer of 2 mg/ml collagen (125 μL) was added to encapsulate the DRGs (Ita and Winkelstein, 2019a). After the PBS washes, the media were changed to supplemented Neurobasal media without serum (Attia et al., 2014). On DIV7, gels were cut into a cruciform shape with each arm having the dimensions of 6.25 mm by 8 mm (Zhang et al., 2017) and marked with a grid of fiducial markers defining elements using black India ink (Koh-I-Noor) to enable mechanical testing (Figure 2A).

The arms of the gels were loaded into grips attached to actuators with 500 g load cells immersed in a 37°C 1X PBS bio-bath (Figure 2A). Gels were pre-loaded until taut (less than 2 mN/arm) and then stretched in equibiaxial tension for one loading cycle at 4 mm/s to a displacement of 1.5 mm for each arm in a planar testing machine (574LE2; Test Resources; Shakopee, MN) to impose strains at magnitudes that induce pain in vivo and also increase nociceptive modulators in vitro (Dong et al., 2012; Zhang et al., 2017). The mechanical system was integrated with a polarized light imaging system and high-speed cameras (Phantom-v9.1; Vision Research Inc.; Wayne, NJ) that acquired collagen alignment maps and tracked fiducial marker coordinates during loading (Tower et al., 2002; Quinn and Winkelstein, 2008; Quinn et al., 2010a). Force and displacement data were acquired at 200 Hz from each arm, and high-speed cameras (500 Hz) tracked the grid of fiducial markers to enable strain calculations (Figure 2) (Ita and Winkelstein, 2019a).

Immediately after loading, the gels were released from the grips, washed with 1X PBS with 1% P–S, and transferred to a pre-warmed serum-free supplemented Neurobasal culture media with 1% P–S (Zhang et al., 2017). After 24 h, the gels were washed with 1X PBS and fixed in 4% paraformaldehyde (PFA) for immunolabeling. Protein expression was assayed 24 h after the stretch to allow for the MMP transcriptional and/or post-translational regulation that occurs hours-to-days after the initiation of mechanical loading in cell-seeded constructs (Kessler et al., 2001; Yang et al., 2005; Petersen et al., 2012). Matched gels for each group (naïve n = 6; inhibitor n = 7; vehicle n = 5) underwent the same protocol but did not undergo any stretch and served as unloaded controls.

Force data were filtered using a 10-point moving average filter (Zhang et al., 2018a; Ita and Winkelstein, 2019a); the maximum force sustained during a gel stretch that was recorded by any of the four load cells (i.e., arms; X1, X2, Y1, or Y2; Figure 2A) was taken as the peak force for that gel (Table 1). The marker locations in the unloaded reference image and the image immediately after the peak force were digitized with FIJI software (NIH) (Schindelin et al., 2012). Positional data for the markers were processed in LS-DYNA (Livermore Software Technology Corp.; Livermore, CA) to calculate the maximum principal strain (MPS) for each element of each gel (Figure 2C) (Ita and Winkelstein, 2019a); each group of four nodes designated by fiducial markers was designated as an element. The largest magnitude MPS sustained in any of the elements for each gel was taken as the peak MPS for that gel, and the average MPS of all elements was also calculated (Table 1).

Pixel-wise fiber alignment maps were created using 20 consecutive high-speed images acquired both before stretch (at unloaded reference) and immediately prior to the time of peak force using a custom script based on a harmonic equation in MATLAB (R2018; MathWorks Inc., Natick, MA) (Tower et al., 2002; Quinn and Winkelstein, 2008; Quinn et al., 2010a). Using fiber angle distributions, collagen organization was quantified by the circular variance (CV) of the spread of fiber angles, with a higher CV indicating a larger spread in the distribution of collagen fiber angles (Zhang et al., 2018a).

To visualize the protein expression of substance P, MMP-1, and MMP-9, fixed co-culture gels were blocked in 1X PBS with 10% goat serum and 0.03% Triton-X and incubated overnight at 4°C with primary antibodies to substance P (1:250, Neuromics Inc.; Minneapolis, MN; Cat# GP14103; RRID:AB_2629484), MMP-1 (1:400, Proteintech; Rosemont, IL; Cat# 10371-2-AP; RRID:AB_2297741), and MMP-9 (1:250, Thermo Fisher; Cat# MA5-15886; RRID:AB_11157246). Gels were then washed four times with 1X PBS and incubated with the secondary antibodies goat anti-guinea pig 633, goat anti-rabbit 488, and goat anti-mouse 568 (all Alexa Fluor 1:1000; Invitrogen) and DAPI (1:750, Thermo Fisher). Following incubation in secondary antibodies, gels were washed three times with 1X PBS and one time with deionized water and then cover-slipped with a fluoro-gel mounting medium (Electron Microscopy Sciences; Hatfield, MA). All washes were performed for 10 min each on a rocker plate. Gels undergoing the same protocol except with no primary antibodies were included as negative controls to verify the specificity of each antibody throughout the imaging process.

Confocal images (1024 × 1024 pixels) were acquired on a Leica TCS SP8 confocal microscope (Leica Microsystems; Wetzlar, Germany) using a ×20 objective (HC PL APO CS2 20x/0.75 dry; Leica Microsystems). All images were acquired at 400 Hz with a 2.5 airy unit pinhole size resulting in a total pixel dwell time of 600 ns. Lasers at excitation lines of 405, 488, 552, and 638 nm were used to excite the DAPI stain and secondary antibodies labeling MMP-1 (488 nm), MMP-9 (568 nm), and substance P (633 nm), respectively. Images were acquired for each gel (6 images/gel) in regions containing DRG soma and/or axons (Ita and Winkelstein, 2019a). Each image was acquired from a distinct region, with no DRG soma or its axons imaged twice.

The amount of positive labeling was quantified separately for each channel for each image using a custom MATLAB script that computes the number of pixels with fluorescence exceeding a threshold on a scale from 0–255 for 8-bit images (Ita and Winkelstein, 2019a). Thresholds for positive labeling were determined from samples with no primary antibodies to verify antibody specificity. The number of positive pixels for each of substance P, MMP-1, and MMP-9 was normalized to the number of positive pixels for DAPI to account for different cell densities in each image; this calculation was performed for all sub-failure stretched and unstretched gels, separately. Then, each of the DAPI-normalized positive pixels for substance P, MMP-1, and MMP-9 of gels undergoing sub-failure stretch were divided by the positive pixels in the respective unloaded control gel, separately for each label. As such, protein expression is represented as the fold-change of stretched co-culture gels over that for unstretched control gels and accounts for differences in cell density in the acquired images.

All statistical analyses were performed with α = 0.05 using JMP-Prov15.2.0 (SAS Institute Inc.; Cary, NC). Peak forces and maximum principal strains (peak and average) were compared across groups (naïve, inhibitor, vehicle) with a one-way analysis of variance (ANOVA) by group. The hypothesis that the circular variance increases with equibiaxial stretch was tested between the reference (unstretched) state and the sub-failure peak stretch state using a matched paired t-test for each group, separately. The effect of sub-failure stretch on the immunolabeling outcomes of substance P, MMP-1, and MMP-9 was tested using a multivariate ANOVA (MANOVA) separately for each group (naïve, inhibitor treated, and vehicle treated) to account for possible dependence between the proteins themselves (Tabachnick and Fidell, 2007; Shin et al., 2015). To further explore the relationships between MMP-1 and each of the MMP-9 and substance P after sub-failure stretch, the linear regressions and strength of correlations were separately tested for co-cultures that did or did not receive inhibitor treatment; MANOVA analyses on groups that did or did not receive inhibitor were also performed. Naïve and vehicle-treated stretched gels were included in that analysis as controls representing co-cultures that did not receive ilomastat.

The significant change in collagen kinematics observed with stretch likely underlies the elevated substance P in co-cultures without an inhibitor (Figure 8). This notion is supported by a body of work demonstrating that, in collagen networks, load-induced cellular responses are triggered by reorganization of the local fiber network surrounding its embedded cells (Sander et al., 2009; Bottini and Firestein, 2013; Zhang et al., 2016; Zarei et al., 2017). Such cellular responses include nociceptive signaling (Zhang et al., 2016; Zhang et al., 2017; Zhang et al., 2018b; Zarei et al., 2017). In particular, stretch-induced kinematic rearrangement of collagen fibers disrupts the α ₂ß₁-integrin adhesion between collagen molecules and the neuronal cell surface (Jokinen et al., 2004). This disruption of adhesion sites in stretched collagen networks mediates substance P expression in DRGs (Zhang et al., 2017). As such, it is likely that α ₂ß₁-integrin receptors are involved in the interactions between DRGs and the collagen network under the sub-failure stretch imposed here (Figure 8). Of note, the collagen fibers are not expected to be directly affected by ilomastat treatment even though ilomastat inhibits the catabolic functionality of MMP-1 to cleave type-I collagen (Grobelny et al., 1992).

Moreover, MMP-1 and MMP-9 are likely involved in the cell–matrix interactions under load, despite remaining unchanged by sub-failure stretch across all groups (Figure 8). At the 24 h time point after a noxious stretch, the gene and protein expression levels of MMP-1 and MMP-9 may increase from transcriptional and/or post-translational regulation of MMPs (Bartok and Firestein 2010; Murphy and Nagase 2009; Petersen et al., 2012; Yang et al., 2005). Furthermore, MMP-1 can itself bind with the neuronal α ₂-integrin receptors and type-I collagen fibers to form a trimeric complex on the neuronal cell surface, activating intracellular ERK cascades and subsequent substance P release (Dumin et al., 2001; Campos et al., 2004; Conant et al., 2004). Although results suggest that MMPs do not increase on the neuronal cell surface (Figures 4, 7), it is possible that the FLS in the co-culture gels regulate MMPs via mechanisms such as MMP sequestration and/or retention (Bottini and Firestein, 2013). If this were the case, altered MMP expression might be localized to FLS cells and not in the neuronal regions that were assessed in this study (Figures 4, 6). Indeed, the possibility that FLS cells retain MMPs is supported by the finding that MMP-1 and MMP-9 localize to FLS cells after a uniaxial stretch to failure in this same co-culture system (Ita and Winkelstein, 2019b).

The fact that the kinematics of the collagen fibers are the same regardless of pre-treatment may be a result of the degradation changes not being captured by the CV measurements that were made using the full gel or could further amplify the non-ECM dependent roles of MMP-1 and MMP-9 in mediating neuronal dysregulation (Dumin et al., 2001; Conant et al., 2002; Conant et al., 2004; Visse and Nagase, 2003; Boire et al., 2005; Allen et al., 2016). Quantifying the regional localization of gel-entrapped MMPs would reveal if the matrix sites undergo MMP-mediated remodeling and/or degradation. If that were the case, regions with MMP-mediated alterations could have consequences on the bulk mechanics of the collagen gel and change microstructural collagen fiber kinematics under load. This suggestion is supported by the evidence of localized collagen degradation and anomalous collagen realignment events in the rat facet ligament after intra-articular MMP-1, both at its physiological configuration (no load) and in the sub-failure loading regime (Ita et al., submitted).

Increased substance P at a peak maximum principal strain of 18.64 ± 6.57% for naïve co-cultures and 30.62 ± 14.95% for co-cultures treated with the DMSO vehicle (Figure 8; Table 1) is consistent with prior reports that stretch at strain magnitudes between 8 and 40% induces pain in rats and increases substance P protein expression in neuron-seeded collagen gels (Dong et al., 2012; Zhang et al., 2017; Zhang et al., 2018a). The fact that substance P is significantly elevated when these two groups (naïve; vehicle) are pooled (Figure 8) indicates a limitation in sample size for each group individually (Figures 4, 6). The forces produced by the sub-failure stretch in this study (14.47 ± 6.43 mN) (Table 1) are lower in magnitude than those reported in collagen gels with only DRGs exposed to a uniaxial configuration (26.25 ± 13.00 mN) (Zhang et al., 2018a). This difference in peak forces between collagen gels with only DRGs (Zhang et al., 2018a) and DRG-FLS co-cultures (Table 1) may be due to the effects of the FLS cells on their surrounding collagen environment or may be due to the different boundary conditions and stretch modalities in the studies (equibiaxial vs. uniaxial). Indeed, fibroblasts embedded in collagen networks exert forces on and remodel collagen in a manner that has well-documented effects on collagen gel mechanics when loaded (Evans and Barocas, 2009; Sander et al., 2011; Kural and Billiar, 2013; Ita and Winkelstein, 2019a).

The finding that the equibiaxial stretch imposed here induces the same strains but lower forces than previous work without FLS (Zhang et al., 2018a) has implications for injury thresholds defined by strain and/or force magnitudes. Furthermore, the imposed forces and strains, although not significantly different, vary (Table 1). This spatial variability in strain fields produced by the equibiaxial loading paradigm translates to spatial variability in mechano-regulated cellular responses. The findings highlight the necessity that the complexities of in situ ligament boundaries be considered, and incorporated, when defining cellular responses to stretch using in vitro modeling.

The lack of effect of a sub-failure stretch on MMP expression levels (Figures 4, 7) may be due to technical considerations of the immunolabeling approach with respect to the pro- and active states of MMPs. For example, the primary antibodies used to visualize MMP-1 and MMP-9 bind to the pro- and active forms of the enzyme; furthermore, the antibody interaction is not expected to be altered if the MMP is in a bound state with the ilomastat inhibitor. As such, the MMP expression levels measured by immunolabeling cannot distinguish between inhibitor-bound MMP or between the pro- and active forms. As such, it is possible that an effect of ilomastat treatment on MMP enzyme activity may be missed using this approach. Nonetheless, ilomastat inhibition may lower the total MMP levels, regardless of their state, especially since ilomastat is present from DIV3 to DIV7 in the culture (Figure 1) when baseline FLS secretion of MMP-1 begins to plateau (Attia et al., 2014).

An alternative explanation to the lack of the effect of stretch on MMP levels in the ilomastat-treated gels could be that the catabolically inactive roles of MMP-1 and MMP-9 dominate in nociceptive-related signaling because ilomastat only inhibits the active forms of MMPs. Since ilomastat inhibits MMP activity by binding to the Zn²⁺ binding site of MMP-1 and MMP-9 (Grobelny et al., 1992; Galardy et al., 1994), ilomastat is expected to decrease the amount of active enzyme forms and to not affect the amount of pro-enzymes. As such, any inhibitory effect of ilomastat would likely only act on the collagenolytic functionality of MMP-1. This assumption that ilomastat only interferes with Zn²⁺-dependent MMP functionalities suggests that the myriad cell signaling roles of MMP-1 that are independent of its Zn²⁺ binding site (Dumin et al., 2001; Conant et al., 2002; Conant et al., 2004; Visse and Nagase, 2003; Boire et al., 2005; Allen et al., 2016) remain unaltered. This notion supports that ECM substrates of MMP-1 and/or MMP-9 (namely, type-I collagen as a substrate for MMP-1) are involved in nociceptive transmission and implies that any clinical intervention targeting MMP-1 and/or MMP-9 in pain from ligament injury or joint disorders must interfere with at least the active enzymatic form, although any role of the pro (zymogen)-enzyme forms cannot be ruled out.

Although the 25 nM ilomastat dose was chosen as the maximum dose expected to inhibit MMP-1 and MMP-9 without inhibiting MMP-3 (Grobelny et al., 1992; Galardy et al., 1994), it is possible that this concentration is too low to consistently inhibit MMP-1 and MMP-9. In fact, concentrations of ilomastat between 1 and 100 μM reduce the contractile function and collagen production of smooth muscle cells and fibroblasts in a dose-dependent manner (Daniels et al., 2003; Rogers et al., 2014). So, it is possible that higher doses of ilomastat would have more robust effects on MMP expression levels. Nevertheless, the 25 nM concentration and dosing beginning DIV3 (Figure 1) attenuates stretch-induced substance P levels (Figure 8), suggesting it to be sufficient to interfere with MMP pathways that regulate nociceptive pathways, at least to some extent.

The timepoint of 24 h after a stretch that was used for immunolabeling was chosen to allow for transcriptional and post-translational regulation to occur in the DRG and FLS cells; yet, it may not have captured the faster MMP regulatory processes. For example, a failure stretch quickly (within 3 min) increases neuronally localized MMP-1 and MMP-9 expressions (Ita and Winkelstein, 2019a; 2019b) and is likely due to an existing extracellular MMP rapidly localizing to the neuronal cell surface, endocytic/exocytotic processes, or even cell rupture caused by the severity of the mechanical insult (Visse and Nagase, 2003; Bottini and Firestein, 2013). Studies of cyclic loading in tendons and dermal fibroblasts, however, show that MMP-1 gene expression is elevated after 4–24 h (Kessler et al., 2001; Yang et al., 2005). Responses that indicate neuronal dysregulation have also been shown to be evident after 24 h; for example, an expression of neuronal substance P increases after a mechanical stretch and intracellular calcium peaks in response to exogenous MMP-1 exposure (Zhang et al., 2017; Ita et al., 2018). Although the timepoint chosen in this study has limitations in capturing rapid cellular responses, it is advantageous in detecting regulators of more sustained nociception (Figure 8).

Despite the limitations pointed out, the ability of ilomastat to abolish the relationships between MMP-1 with MMP-9 and substance P (Figure 8) supports these biochemical mediators having mechanistic relationships in the context of stretch-induced nociception in the ligament. The positive correlative relationships demonstrated here (Figure 8) are consistent with parallel increases in MMP-1, MMP-9, and substance P immediately after a gel stretch that induces its failure (Ita and Winkelstein, 2019a; 2019b). Since active MMP-1 catalyzes pro-MMP-9 by cleaving its bait region (Conant et al., 2002; Visse and Nagase, 2003) and MMP-9 cleaves substance P (Diekmann and Tschesche, 1994; Rawlings et al., 2014), all three proteins could be related in their response to noxious stimulus (i.e., stretch) here (Figure 8). However, since ilomastat also inhibits MMP-9 at a higher affinity than MMP-1 (Grobelny et al., 1992; Galardy et al., 1994), a definitive relationship between MMP-1 and MMP-9 cannot be concluded since ilomastat acts on both active forms of the enzymes. Studies utilizing siRNA silencing techniques in cell cultures could be exploited to target MMP-1 synthesis (Rogers et al., 2014). siRNA-induced post-transcriptional silencing would suppress the translation of MMP-1 (Agrawal et al., 2003), selectively removing MMP-1 production by cells without the off-target effects on other MMPs that is unavoidable with the ilomastat synthetic inhibitor. In fact, separately silencing MMP-1 synthesis in FLS cultures and/or DRG cultures before their co-culture could be helpful in fully defining the role of cell-specific MMP-1 in nociceptive responses. That approach would also directly determine if the complete obliteration of MMP-1 affects the pro- and/or active levels of MMP-9.

Collectively, the findings in this study implicate MMPs in painful ligamentous injuries and contribute to a growing body of work linking MMPs to nociceptive-related signaling pathways and/or pain (Kawasaki et al., 2008; Sbardella et al., 2012; Allen et al., 2016; Ita and Winkelstein, 2019a; Ita et al., 2020a; Ita et al., 2021). Although MMP-9 has a well-established role in nociception, particularly for neuropathic pain (Kawasaki et al., 2008), studies examining the role of MMP-1 in nociception are sparse by comparison. Since MMP-1 is generally only expressed in pathologic disease and/or injury states in in vivo systems (Sbardella et al., 2012), investigating techniques that selectively inhibit MMP-1 may be of use since less detrimental off-target effects are expected. Selective MMP inhibition is particularly important if inhibition studies are pursued in vivo, since the primary reason that broad-spectrum MMP inhibitors such as ilomastat have failed in clinical trials is due to off-target effects in part from their systemic administration (Coussens et al., 2002; Fingleton, 2008; Vandenbroucke and Libert, 2014). Although those same off-target effects are not expected to be at play in the co-culture model used in this study, ilomastat may also have unanticipated actions such as binding to non-MMP enzymes that contain metals or interfering with ADAMTS released by fibroblasts (Fingleton, 2008; Bottini and Firestein, 2013; Ernberg, 2017). No selective inhibitors for MMP-1 exist currently; yet, selective inhibition of MMP-13 has been successfully achieved by targeting regions other than the preserved Zn²⁺ binding site (Baragi et al., 2009; Jüngel et al., 2010; Gege et al., 2012). Pursuing MMP-1 inhibition by this same selective approach may provide a promising opportunity to define the effect of isolated blocking of MMP-1 activity on other MMPs, such as MMP-9, and to determine if selective MMP-1 inhibition in vivo has the potential to reduce pain without clinical side effects.