Modulation of corneal tissue mechanics influences epithelial cell phenotype

Whilst the control of stem cell differentiation using substrates of differing compliance has been extensively explored in vitro, the significance of this mechanism at a physiological level is not known. Here we set to explore the role of corneal surface biomechanics in controlling epithelial cell proliferation and differentiation. Using non-contact high-resolution Brillouin spectro-microscopy we showed that the corneal outer edge (limbus) has significantly lower bulk modulus compared to the central cornea, and that this difference is precisely delimited in the organ. Furthermore, the areas of the limbus with distinctly softer properties were shown to be associated with limbal epithelial stem cell (LESC) residence. Based on these findings, we then provided the first demonstration of the capacity to modulate LESC phenotype, both in vivo and ex vivo, solely through the recreation/restoration of suitable biomechanical niches. These results thus confirm the fundamental role of corneal biomechanics in directing epithelial stem cell behavior.


Introduction
The function of the human cornea is largely dependent on the maintenance of a healthy stratified epithelium, which in turn relies upon a population of stem cells located in its periphery, the limbus (Lehrer et al., 1998). These limbal epithelial stem cells (LESCs) proliferate and differentiate to repopulate the central corneal epithelium, where cells constantly undergo maturation, stratification, and ultimately, shedding from the ocular surface. These events have been shown to be modulated by biochemical and biophysical factors (Di Girolamo, 2015;Yoon et al., 2014). However, the mechanisms underpinning the homeostatic process of LESC self-renewal and differentiation remain largely unclear (West et al., 2015). More recently, a number of studies have shown that the behavior of LESCs, like other stem cell types (Engler et al., 2006), is strongly influenced by their immediate mechanical environment. This notion is supported by the cellular stiffness of LESCs (Bongiorno et al., 2016), as well as by the distinct structure (Boote et al., 2005), composition (Mei et al., 2012), and compliance (Hjortdal, 1996;Last et al., 2012) of the extracellular matrix (ECM) across the cornea. In particular, the impact of substrate stiffness on corneal epithelial cell attachment and viability (Chen et al., 2012), proliferation (Jones et al., 2012), and mechanosensing (Foster et al., 2014) has been explored in vitro, using biomimetic surfaces with defined elastic moduli. These studies showed that corneal epithelial cells grown on relatively soft substrates are able to retain specific LESC markers whereas cells cultured on corresponding stiff substrates are disposed to differentiate (Foster et al., 2014;Jones et al., 2012). This body of work provides substantial evidence that, at least in vitro, substrate rigidity regulates LESC phenotype.
To understand if such regulation occurs in vivo, we presently built a high-resolution Brillouin spectro-microscope ( Figure 1a) and used it to characterize the mechanical properties of live human corneas with unprecedented detail. This technique is based on the interaction of light with spontaneous acoustic phonons in the GHz frequency range. By measuring the optical frequency shift of the scattered light, Brillouin scanning can probe the local spontaneous pressure waves within biological tissues, from which the longitudinal modulus can be evaluated (Antonacci et al., 2013). Relating the hypersonic (GHz) modulus directly to the quasi-static modulus and Young's modulus is complicated by acoustic dispersion that can increase moduli significantly at GHz frequencies. However, the technique does provide information about variations and changes in mechanical properties at submicron scale (Elsayad et al., 2016). Previously, Brillouin spectro-microscopy (BSM) has been used to evaluate mechanical properties of cells and tissues both in vivo (Scarcelli et al., 2015a) and in vitro (Antonacci and Braakman, 2016;Scarcelli et al., 2015b), particularly in the cornea (Lepert et al., 2016;Scarcelli et al., 2012). Our newly-developed instrument was designed with a single-stage virtually imaged phased array (VIPA) spectrometer in combination with an original wavefront division adaptive interferometric, and a piezoelectric actuator (Lepert et al., 2016) to give high extinction of the elastically-scattered light and allowing faster (0.01-1 s) Brillouin scattering measurements. As such, we were able to obtain organ-wide scans of the human cornea, with both unprecedented detail and speed. Crucially, this achievement allowed us to identify critical biomechanical differences between the (softer) limbus and the (stiffer) central cornea, which were then correlated to the phenotype of corneal epithelial cells. This data supported our hypothesis that epithelial cell differentiation across the corneal surface is controlled by changes in substrate stiffness. More importantly, it allowed us to develop new a method to control the phenotype of corneal epithelial cells solely via modulation of the mechanical properties of collagen-based substrate materials.

BSM shows clear transition between softer limbus and stiffer central anterior cornea
The new technological development of the BSM allowed whole-organ imaging of immersed human corneas ( presenting stiff nuclei, as previously predicted (Foster et al., 2014), followed by a 10-15 µmthick Bowman's layer and the anterior-most stroma with 6.66±0.04 and 6.53±0.04 GHz shifts, respectively ( Figure 1e). In contrast, the limbal epithelium ranging 40-60 µm depth ( Figure 1d) and 6.34±0.14 GHz was followed by matrix with significantly lower frequency shift (6.24±0.09 GHz), which in turn was followed by tissue exhibiting slightly higher shifts (6.40±0.14 GHz, respectively; Figure 1e), probably corresponding to the continuation of the corneal stroma under the limbus (Figure 1b) (Van Buskirk, 1989).
Consistently, the distinction in frequency shifts between epithelium and the sub-epithelial matrix immediately underneath was less clear in the limbus than in the central cornea (compare Figure 1c and d). The limbus also showed a broader distribution range of Brillouin frequency shift values in all its layers, whereas the central corneal was characterized by uniform biomechanical properties within each layer (Figure 1e), with a profile consistent to that from previous studies (Scarcelli et al., 2012). The clear distinction between the mechanically-heterogeneous limbus and the more regular central cornea was more evident in very high-resolution scans (Supplementary Figure 1c) and under alternative vantage points (Video 1 and 2). This distribution pattern was probably due to the furrowed topography of the limbus, and the presence of numerous structures such as the Palisades of Vogt, focal stromal projections, and limbal crypts and/or pits (Video 1), particularly abundant in the superior and inferior limbus but less so in the temporal and nasal side of the cornea (Di Girolamo, 2015;Lagali et al., 2013;Miri et al., 2012;Shortt et al., 2007) (Supplementary Figure 1d). Moreover, the very high-resolution Brillouin scans also identified multiple pocket regions within the limbal epithelium that exhibited significantly lower shifts compared to their immediate surrounds (Figure 1d; Supplementary Figure 1e). Three-dimension Brillouin scans showed that these pockets comprise multiple spherical units 10-12 µm in diameter (Video 2) that are compatible, dimension (Romano et al., 2003) and biomechanical-wise (Bongiorno et al., 2016), with LESCs surrounded by stiffer cells and matrix.

BSM and immunofluorescence analysis shows LESCs residing on soft limbal matrix
Importantly, these limbal regions identified by BSM also represented the location of cells expressing ABCG2, CK-15, nuclear  (Van Buskirk, 1989), was well correlated to the collagen-Ipositive/collagen-V-negative connective tissue immediately underlying the resident LESCs ( Figure 2a-c). Moreover, the focal distribution of laminin-γ3, a characteristic marker of limbal basement membrane (Schlotzer-Schrehardt et al., 2007;Torricelli et al., 2013) along with the strong expression of integrin-α9, a transient amplifying cell/LESC marker (Schlotzer-Schrehardt and Kruse, 2005;Stepp, 2006), showed that this region corresponded to the corneal limbus ( Figure 2d). The low-Brillouin shift values were also well correlated with the tissue's structural features, previously characterized as comprising less compact collagen lamellae with irregularly arranged, branched, and intertwined collagen bundles (Boote et al., 2003;Komai and Ushiki, 1991). In contrast, the location of CK3-positive epithelial cells  Figure 2). This distinctive composition plays an important role on collagen fibril diameter and lamellar organization of the corneal stoma (Birk, 2001), which in turn has a critical influence in tissue transparency and elastic modulus (Boote et al., 2005;Hjortdal, 1996). Signal quantification also supported the distinctive expression pattern of markers in both limbus and central cornea ( Figure 2f). Taken together, these results showed that, in the human cornea, LESCs populate tissues that are significantly softer compared to those supporting differentiated epithelial cells, thus constituting a niche with distinct biomechanical, as well as a biochemical/biomolecular profiles.

Substrate stiffness controls corneal epithelial cell phenotype in vitro
The surprisingly distinct properties of the limbus observed by BSM reinforced our hypothesis that biomechanics plays a role in controlling the phenotype of LESCs (Foster et al., 2014).
Conversely, the biomechanical features of the central cornea sustained the notion of a 8 mechanical differential (i.e., gradient or step-change in substrate stiffness) driving corneal epithelial cell differentiation (Foster et al., 2014;Shin et al., 2013). It is then reasonable to assume that the measured differences in substrate stiffness across the cornea play an important role in the tissue's function/homeostasis, by regulating the phenotype of corneal epithelial cells via mechanotransduction (i.e., maintaining undifferentiated cells in the limbus, and promoting their differentiation in the central cornea).
In this perspective, we aimed at further investigating the response of LESCs to surface compliance, first using an in vitro model. Previously, high-density collagen-I gels (tissue mimics) with different levels of stiffness via plastic compression have been used to that purpose (Gouveia et al., 2014). Now we explored the premise that collagen gel stiffness could be affected through its partial digestion using a type-I collagenase. This approach was chosen as it would also facilitate the topical application on corneal tissues in vivo, using a formulation currently approved for connective tissue softening (Brunengraber et al., 2014;Gelbard et al., 2013;Hurst et al., 2009). Thus, we used plastic-compressed collagen gel discs 2.5 cm in Ø, to which a solution of collagenase was applied in discrete areas for up to  Figure   3b). These results indicated that the enzymatic digestion can produce precisely controlled, localized differences in the stiffness of high-density collagen gels, resulting in the removal of up to one third of the original collagen content from within the gels without altering their volume or micro-topography, and subsequently maintain hLECs undifferentiated. As such, we then used the collagenase treatment to create a ring of softer collagen (representing the limbus, having shown by BSM that this tissue is naturally significantly softer) surrounding a stiffer area (representing the central cornea). Cells subsequently grown on the treated, outer-ring area (pseudo-limbus) expressed higher levels of CK15 up to 4 weeks, whereas hLECs on the central, untreated region of the gels displayed higher CK3 expression ( Figure   3e). Furthermore, the maintenance of CK15 expression by hLECs on collagen gels was, overall, significantly higher than on tissue culture plastic at equivalent stages in culture (Supplementary Figure 3c). These results suggested that cells residing upon this collagenase-created pseudo-limbus retained a more undifferentiated, LESC-type phenotype.
Together, these results demonstrate that the application of collagenase to high-density collagen gels creates localized areas that act as a niche for hLECs to maintain LESC properties such as those observed in the corneal limbus. The notion that LESC phenotype maintenance depended on the mechanical properties of softened gels and not on interactions with new cues created after collagenase treatment was supported by previous work using hLECs on semi-compressed collagen gels (Foster et al., 2014). The topographical, mechanical, and phenotype-modulating properties of semi-compressed collagen gels, despite being comparable to those of enzyme-softened gels (Supplementary 1 0  al., 2012), and therefore independent of exposed cryptic epitopes. Collagenase presented also the advantage to allow softening of collagen-based tissues both in vivo and ex vivo, and could thus be applied to evaluate the 'cell phenotype-through-biomechanical modulation' strategy demonstrated above, but using live, natural tissues (e.g., the cornea).

Softening the human central cornea creates a limbus-like mechanical and phenotypic milieu
In order to investigate the ability to control cell phenotype via modulation of the biomechanical properties of natural tissues, collagenase was used to soften limited and well- Taken together, these results showed that collagenase treatment has the potential to soften the matrix of the anterior layers of the central cornea, and thus create a substrate that is mechanically similar to the natural limbus, and biologically capable of maintaining a LESClike phenotype.

Softening the rabbit central cornea modulates epithelial cell phenotype
Next, we used a rabbit corneal model to demonstrate that phenotype-through-biomechanical modulation of corneal epithelial cells can be controlled in vivo through precise application of collagenase ( Figure 5). The outcome of collagenase treatment on intact and debrided (i.e., epithelium removed) rabbit corneas was followed by clinical observation, slit-lamp biomicroscopy, and immunohistochemistry analyses 1 and 5 days post-intervention (  Table S1). Moreover, no conspicuous neo-vascularization was observed following collagenase treatment of the cornea either on a macroscopic ( Figure 5a The fact that collagenase-treated tissues were able to retain their shape and structure without eliciting a major pro-inflammatory response (Supplementary Figure 3b) is of great importance for the validity of this method. Specifically, it showed that the collagen matrix can be softened without compromising the integrity of the cornea. Moreover, it supported the notion that the modulation in corneal epithelial cell phenotype was due to alterations in tissue biomechanics, and not the result of inflammation-related signaling. This is important, as severe inflammation of the ocular surface has been associated with corneal injury and disease leading to LESC loss (Deng et al., 2012), namely through mechanotransduction pathways (Nowell et al., 2016). These results thus indicated that the collagenase treatment was innocuous, if not beneficial, to the culture of corneal epithelial cells on collagen-rich matrices.

Collagenase treatment restores limbus capacity to support LESCs after stiffening
Finally, we explored the effect of collagenase on the phenotype of corneal epithelial stem cells grown on artificially stiffened corneas. An alkali burn technique was thus developed, as this method allowed the experiments to be performed in small, well-defined, localized areas.
First, human corneas subjected to alkali burn ex vivo were treated with collagenase and tested for their mechanical properties, as well as their ability to support the LESC phenotype ( Figure 6). Not unexpectedly, the alkali burn led to major alterations in the corneal limbus, with clearance of most of its non-fibrillar elements and significant stiffening of its matrix (Figure 6a and b). These effects were probably due to proteoglycan removal following alkali exposure (Scott and Thomlinson, 1998). Subsequently, alkali-burned corneas also failed to maintain the LESC phenotype of repopulating cells. Specifically, hLECs seeded on the limbus of alkali-burned corneas showed significantly reduced expression of LESC markers ABCG2, CK15, Δ Np63, and integrin-α9, while expressing higher levels of differentiation marker CK3 compared to cells on non-burned, control tissue (Figure 6c and d). However, these effects were shown to be reversed by application of collagenase on alkali-burned corneas, with treatment leading to reduced collagen fiber density and significant decrease in the stiffness of the limbus matrix (Figure 6a and b). But most importantly, the recovery of the original tissue compliance was accompanied by the ability to support re-epithelialization by hLECs expressing LESC markers (Figure 6c and d).
Similar results were observed in an in vivo chemical burn model. Rabbit corneas subjected to a localized alkali burn had the temporal half of the limbus damaged while maintaining its nasal half intact, to allow re-epithelialization (day 0, Supplementary Figure 9). On the second day post-burn, the affected areas of the limbus were either treated with collagenase (treated) or remained untreated (alkali), and were subsequently compared to undamaged limbus tissue (control). Immediately after the burn, the affected area of the limbus became opaque ( Figure 7a). In untreated corneas, this haze showed little or no reduction in size (Figure 7b, alkali). In contrast, collagenase-treated tissues showed a significant (p = 0.002) reduction of haze area compared with alkali-burn tissues (p = 0.0023) and with time (90±3 to 56±5% of 1 4 initial size from day 2 to 7, respectively) ( Figure 7b). Furthermore, treating the burned limbus with collagenase allowed the tissue's native cellular profile to be restored ( Figure 7c).
Specifically, the new epithelium populating the collagenase-treated burned limbus tissue showed significantly higher expression of LESC markers CK15, Δ Np63, ABCG2, and integrin-α9, and lower expression of differentiation markers CK3 and α 3β1 compared with their untreated (alkali) tissue counterparts, and an expression profile similar to the undamaged (control) limbus (Figure 7c and d).
Overall, these results showed that collagenase effectively reversed the stiffening of the corneal limbus following alkali burn and allowed the recovery of its natural LESC-supporting capacity. Moreover, the restorative effect of collagenase did not elicit substantial neovascularization, nor did it compromise the central cornea epithelium (Supplementary Figure   10). As such, this data demonstrated that the phenotype of corneal epithelial cells, and particularly of LESCs, can be influenced in vivo via the manipulation of their mechanical environment.

Conclusions
In summary, this study demonstrates the strong correlation between compliance of the corneal tissue and epithelial cell phenotype. Specifically, the results show that compliant substrates support the growth of undifferentiated LESCs, whereas stiffer substrates promote their differentiation. Moreover, we applied this phenomenon to the development of a novel collagenase-based method to affect LESC phenotype-through-biomechanical modulation both in vivo and ex vivo. This work opens multiple areas of research, particularly regarding the use of BSM to study the mechanobiology of stem cell niches, in real time and with unparalleled detail and resolution, as well as the use of enzymes to regulate stem cell phenotype though modulation of the mechanical properties of substrate tissues in situ. Funding Acquisition and Supervision, R.R.M, C.P. and C.J.C.

Declaration of Interests
The authors declare no competing interests. The limbus showed a discontinuous distribution of laminin-1 compared to the central cornea, and the specific expression of laminin-γ3 and integrin-α9 (d). Conversely, the central corneal epithelium was positive for CK3+12, as well as for collagen-VII (e). Cell nuclei were detected 2 4 using DAPI. f) Marker expression was quantified and represented as average ± S.D. from three independent experiments (n = 3; *** corresponds to p< 0.001). of alkali-burned corneas receiving collagenase treatment (treated), before (baseline) and after burn (day 0), as well as at day 2 and 7 post-burn. The haze resulting from the burn was delimited (traced lines), and its area quantified (b) at day 0 (initial wound), 2 (dark blue) and 7 (light blue bars). Values corresponded to haze area average ± S.D. from three independent experiments (n = 3; ** corresponds to p < 0.01. c) Representative confocal immunofluorescence micrographs (3D reconstruction) of chemically-burned (alkali) and collagenase-treated burned limbus (treated) at day 7, with (d) corresponding marker expression quantification. Epithelial cells repopulating the alkali-burned limbus in vivo 3 3 expressed significantly lower levels of LESC markers CK15, Δ Np63, ABCG2, and integrinα 9 while showing higher expression of the CK3 differentiation marker compared to the undamaged limbus (control) tissue. However, collagenase treatment successfully restored the ability of the burned limbus to support cells as in control tissues. Cell nuclei were detected using DAPI. Marker expression was represented as average ± S.D. from three independent experiments (n = 3; *, **, and *** corresponds to p<0.05, 0.01, and 0.001, respectively). Scale bars, 50 µm.

Brillouin spectro-microscopy
The design and setup of the Brillouin spectro-microscope and VIPA spectrometer, as well the evaluation method for the Brillouin scattering from fresh human corneas were performed as previously described (Lepert et al., 2016)

Collagen gel production
Plastic-compressed high-density collagen gels were produced as previously described (Brown et al., 2005;Foster et al., 2014;Jones et al., 2012). Briefly, collagen gels were made by neutralizing sterile rat-tail type-I collagen (2.2 mg.ml -1 in 0.6 % acetic acid; First Link Ltd, UK) in 10× Modified Eagle's Medium (MEM; Thermo Scientific, RRID:SCR_008452) and 1 M sodium hydroxide (Sigma-Aldrich, RRID:SCR_008988) at a 8:1:1 volume ratio. The solution was gently mixed and cast into circular molds (2 mL per 1.9 cm 2 ø wells in 12-well plates) prior to gelling for 30 min at 37 °C. Collagen gels were then compressed between two layers 3 5 of nylon mesh (50 µm mesh size) under a fixed load of 134 g for 5 min, or semi-compressed with 64 g for 2.5 min at room temperature. Compressed collagen gels were also coated with the matrix extracted from human corneas. Briefly, stromal tissues isolated from either the central cornea or limbus regions were minced and digested with 1 × 10 4 activity units of collagenase per L -1 of PBS for 60 min at 37°C. Stroma and limbus protein extracts were then precipitated by incubation with 4× volume of ice-cold ethanol overnight at -20°C, followed by centrifugation at 10,000 ×g for 10 min at 4°C, and resuspended in PBS. The different corneal extracts were drop-spotted onto compressed collagen gels, and allowed to dry overnight to create a continuous coating.

Collagenase treatments
Tissue softening was performed by applying collagenase type-I isolated from Clostridium histolyticum (#17018-029, Thermo Scientific). Dose and time of incubation were optimized to reproduce the biomechanical and functional properties of semi-compressed collagen gels 12 .
Briefly, lyophilized collagenase powder was solubilized in phosphate buffered saline (PBS) at 5 × 10 -2 g.L -1 (1 × 10 4 activity units.L -1 ) and then applied onto sterilized Whatman Grade 1 filter paper until saturation, i.e., approximately 0.2 mL.cm -2 of paper. The collagenasesoaked paper cut-outs were then gently applied onto the surface of dense collagen gels and incubated for up to 60 min at 37°C. Treatment of whole human corneas was performed for 60 min at 37°C (2 units.cm -2 .h -1 of total collagenase activity). The collagenase-soaked filter paper cut-outs were subsequently discarded, and the gels or corneas were washed thrice with PBS for 15 min on a rocker agitator to remove remaining collagenase enzyme and products from digestion. PBS-soaked filter paper cut-outs were incubated for 1 h using a similar method, in order to produce mock-treated specimens (treatment time = 0 min).
Collagen gels treated for different durations were imaged by bright-field photography using a Nikon DSRL digital camera (Nikon, Japan) immediately after washes.

6
Collagen gels were washed in PBS, blocked in PBS supplemented with 5% bovine serum albumin (BSA) and 0.1% Triton X-100 for 1 h, incubated overnight at 4°C with rabbit anticollagen type-I primary antibody (Table 1)   immediately after. Gel hydration was subsequently calculated as the ratio of (W f -W d ) / W f and expressed as a percentage. The experiment was performed using 10 independent gels for each duration of treatment (n = 10).

Atomic Force Microscopy
Analysis of surface topography of collagen gels treated with collagenase for different durations was performed in static force (contact) mode using an Easyscan 2-controlled passaged using Accutase cell detachment solution (Thermo Scientific) for 10 min at 37°C, and re-plated at 3 × 10 4 cells.cm 2 , up to passage 4.

Proliferation of corneal limbal epithelial cells on collagen gels
Collagen gels treated with collagenase (treated) or PBS (untreated) for 1 h were transferred into Transwell culture inserts (Corning, NY, USA) where they were coated with laminin (1.5 × 10 -6 g.cm -2 ; Thermo Scientific) for 2 h prior to cell seeding. Human limbal epithelial cells (hLECs) were suspended at 3 × 10 5 cells per mL of CnT-7 medium and then transferred onto collagen gels (1 mL per insert). All cultures were incubated at 37 °C and 5 % CO 2 for 2 weeks, with cell culture medium being replaced every 2 days. The AlamarBlue assay was used to assess the proliferation and viability of hLECs grown on treated and untreated collagen gels after 1 week. Briefly, gels were incubated with resazurin reagent (#R7017, Sigma-Aldrich) diluted 1:10 in fresh culture medium and incubated for 4 hours at 37°C, after which 100 µL of culture supernatants were sampled, in triplicate, for fluorescence emission analysis at 590 nm using a Fluoroskan Ascent plate fluorometer (Thermo Scientific). Cell number was calculated by interpolation using a standard curve for fluorescence values of 1, 5, 10, 20, 50, and 100 x 10 4 cells, with values corresponded to average ± SD of 3 independent experiments (n = 3).

Cell migration assay
Collagen gels previously treated with collagenase (treated) or PBS (untreated) 1 h were laminin-coated, seeded with 5 × 10 5 hLECs in 1 mL of CnT-7 medium, and incubated at 37°C for 6 hours with the substrate held at an initial 45° tilt to ensure that cells only attached to the lower half-surface of the gel, and with the upper half being above the air-liquid interface. Subsequently, cultures were washed three times with PBS to remove any remaining unattached cells, and incubated at 37°C for 24 h in CnT-7 medium while being imaged at 10 min intervals by time-lapse bright-field microscopy using a Lumascope 500 inverted microscope (Etaluma, CA, USA). Micrographs were binarised using the ImageJ v1.7 4 0 software to better determine the position of individual cells in each image frame. Cell speed (µm.h -1 ) was evaluated by determining the position of 100 individual cells during the initial 24 h of migration, and tracing total distance covered by moving cells using the standard parameters of the wrMTrck plugin for ImageJ v1.7. Data was expressed as the average ± S.D. from three independent experiments using cells from independent donors.

Organ culture of whole human corneas
Collagenase-treated whole human corneas were transferred into Transwell culture inserts after Brillouin spectro-microscopy analysis, washed 3 × 5 min with an excess of sterile PBS to detach and remove existing epithelium, and then seeded with 5 × 10 5 hLECs in 2 mL of CnT-7 medium. Corneas mock-treated with PBS were used as non-softened controls. For the ex vivo alkali burn model, human corneas were similarly processed after treatment with strips of filter paper soaked with 0.5 M of NaOH (alkali) or sterile PBS (control) for 60 s, followed by 3 × 5 min washes with an excess of sterile PBS. A subgroup of alkali burn corneas were also subsequently subjected to collagenase treatment, as described. Organ culture was performed at 37 °C and 5 % CO 2 in CnT-7 medium for 2 weeks, followed by a 1 week period in SHEM and air-lifting (Jones et al., 2012), with media change every 2 days.
The central region of collagenase-treated and control corneas was monitored daily by brightfield microscopy to evaluate re-epithelialization, and then analyzed by confocal immunofluorescence microscopy, as described above. Experiments were performed three independent times (n = 3).

Quantitative RT-PCR analysis
Collagen gels treated with collagenase (treated) or PBS (control) for 60 min were laminincoated, seeded with 3 × 10 5 hLECs, and incubated in CnT-7 medium at 37°C for up to 1 week. Cells were then harvested using a flexible cell scraper (Sarstedt, Germany), and their mRNA isolated using standard extraction with Trizol (Thermo Scientific). The assessment of mRNA quality was performed using a Nanodrop 2000 spectrophotometer (Thermo Scientific) 1 to ensure the 260/280 ratio was within the 1.

Quantitative immunofluorescence analysis
Collagen gels treated with collagenase across half their surface or in a ring-shape area for 60 min were laminin-coated, seeded with 3×10 5 hLECs, and incubated in CnT-7 medium at 37°C for up to 4 weeks. Subsequently, gels were washed twice in PBS, fixed in 4% paraformaldehyde in water for 20 min at room temperature, and washed again with excess PBS. Cells were then permeabilised in PBS containing 0.1% Triton X-100 (PBS-T) for 5 min, blocked with blocking buffer (PBS-T containing 5% BSA) for 1 h, incubated with the rabbit anti-CK3 or anti-CK15 antibodies (Table 1) diluted 1:1000 in blocking buffer for 4 h at room temperature, washed again for 3 × 15 min in blocking buffer, and incubated with IRDye 800CW goat anti-rabbit secondary antibody and CellTag 700 stain diluted 1:1000 in blocking buffer for 2 h at room temperature. Gels were finally washed 3 × 15 min in PBS-T before being imaged by near-infrared quantitative analysis using an Odyssey CLx System (Li-Cor Biosciences, RRID:SCR_014579). Signal specificity was evaluated from gels incubated without either primary or secondary antibodies. The intensity of signal corresponding to cellular CK3 and CK15 detection was determined for the collagenase-treated and untreated 4 2 areas of the gels, and normalized to the number of cells present on the corresponding areas, determined by the CellTag stain signal. Marker expression was also compared to that from cells grown on tissue culture plastic for corresponding periods of time. Data was represented as protein expression relative to the average expression of untreated (control) areas of the gels from three independent experiments using cells from independent donors.
Subsequently, culture supernatants were collected on ice, centrifuged at 10,000 ×g for 10 min at 4°C, sampled, and analyzed for human IL-6 expression using the Quantikine enzymelinked immunosorbent assay (ELISA) kit (R&D Systems, RRID:SCR_006140) according to the manufacturer's instructions. The expression of IL-6 was then normalized for the amount of total protein present in the supernatants, and expressed as pg.mL -1 .µg -1 of total protein.
Both conditions were assayed in triplicate.

Collagenase treatment of rabbit corneas
Animal experiments were performed following the appropriate guidelines and

Statistical analysis
Error bars represent the standard deviation (S.D.) of the mean, analyzed a priori for homogeneity of variance. Replicates from each independent experiment were confirmed to follow a Gaussian distribution, and differences between groups were determined using oneor two-way analysis of variance (ANOVA) followed by Bonferroni's multiple comparison post hoc test. Significance between groups was established for p < 0.05, 0.01, and 0.001, and with a 95% confidence interval. The average ± S.D. Brillouin frequency shift was calculated using 100 points from discrete corneal tissues (epithelium, sub-epithelial lamina, anterior stroma, sub-limbal conjunctiva) from three independent donors (n = 3), and represented as a scatter plot to illustrate intra-tissue variations. The average ± S.D. frequency distribution of elastic modulus of both treated and untreated gels was tested for correlation, and fit to Gaussian curves using a non-linear least-squares regression model, which were then analyzed for independence using the Chi-square test.