In Vivo Femtosecond Laser Machined Transepithelial Nonlinear Optical Corneal Crosslinking Compared to Ultraviolet Corneal Crosslinking

Abstract

Purpose: This study assessed the safety and efficacy of transepithelial crosslinking (CXL) using femtosecond (FS) laser-machined epithelial microchannels (MCs) followed by UVA CXL compared to FS laser (NLO CXL) in rabbits.

Methods: The epithelium of 36 rabbits was machined to create 2- by 25-µm MCs at 400 MCs/mm². Eyes were treated with 1% riboflavin (Rf) solution for 30 minutes, rinsed, and then crosslinked using UVA or NLO CXL. Rabbits were monitored by epithelial staining, optical coherence tomography (OCT) imaging, and esthesiometry. After sacrifice at 2, 4, or 8 weeks, corneas were examined for collagen autofluorescence and immunohistochemistry.

Results: NLO CXL showed no epithelial damage compared to UVA CXL, which produced on average 23.89 ± 5.6 mm² epithelial defects that healed by day 3. UVA CXL also produced loss of corneal sensitivity averaging 0.83 ± 0.24 cm force to elicit a blink response that persisted for 28 days and remained significantly lower than control or NLO CXL. OCT imaging detected the presence of a demarcation line only following UVA CXL but not NLO CXL.

Conclusions: Even with improved transepithelial Rf penetration, UVA CXL resulted in severe epithelial damage, loss of corneal sensitivity, and delayed wound healing persisting for a month. When MCs were paired with NLO CXL, however, these issues were mostly negated. This suggests that MC NLO CXL can achieve a faster visual recovery without postoperative pain or risk of infection.

Translational Relevance: UVA CXL is a successful procedure, but there is a need for a transepithelial protocol. The combination of MCs and NLO CXL is able to keep the benefits of UVA CXL without causing epithelial damage.

Introduction

Corneal collagen crosslinking using ultraviolet A light (UVA CXL) to photoactivate riboflavin (Rf) solution, a procedure that biomechanically stiffens the corneal stroma, has been shown to be an effective treatment for keratoconus as well as low refractive errors.^1,2 However, the most effective UVA CXL treatment, the Dresden protocol, requires epithelial debridement for penetration of Rf into the stroma. Since epithelial debridement leads to postoperative pain, delayed visual recovery, and increased risk of bacterial keratitis and corneal ulceration,^3–8 recent research has focused on the development of epi-on procedures that enhance transepithelial Rf penetration. 

While various epi-on procedures have been proposed, they are arguably less effective, reporting stabilization of KMax in only 43% compared to 93% of patients receiving traditional UVA CXL.^9–13 Interestingly, the line of demarcation in the stroma was also observed to be shallower when the epithelium remained intact.⁹ Research has also shown extensive epithelial damage following epi-on procedures as well as persistent postoperative pain.^13–18 Specifically, the commonly used excipient, benzalkonium chloride (BAK), has been shown to damage the epithelium even without subsequent UVA exposure.^14,19 Furthermore, Taneri et al.¹⁷ reported that epithelial defects and perceived pain were common following transepithelial UVA CXL techniques. Finally, more extreme epithelial damage was observed following high-intensity, accelerated transepithelial UVA CXL.¹³ All of these observations together indicate that epithelial damage is not only a result of varying Rf penetration techniques but also due to UVA exposure to the epithelium, making the pursuit of a truly transepithelial crosslinking technique a two-part task: one, achieving maximal Rf penetration through epithelium and, two, protecting the epithelium from photoactivation. 

Regarding the first task, we have shown that femtosecond (FS) machining of the epithelium to form microchannels (MCs) can significantly enhance stromal Rf penetration, achieving concentrations equivalent to 50% of that obtained by epithelial debridement after creating 400 MCs/mm² or less than 0.3% of the surface area.¹⁴ Concerning the second task, we had previously established that a regeneratively amplified, 1-kHz FS laser induced photoactivation of Rf within the cornea capable of inducing precise nonlinear optical collagen crosslinking (NLO CXL) using powers below the American National Standards Institute (ANSI) limit of 46.1 mW.²⁰ The precise nature of this CXL method has many advantages over UVA CXL. First, two-photon photoactivation of Rf only occurs within the focal volume of the laser, leaving other tissue unaffected by Rf photoactivation. Second, the focal volume can be precisely placed below the epithelium and above the endothelium, avoiding damage to these critical tissues. If used in combination with our MC technique, transepithelial (TE) NLO CXL could hypothetically result in a procedure with no visible epithelial defects within 1 day, cause decreased postoperative pain, and permit faster visual recovery without risk of microbial infection. To test this hypothesis, we performed both MC UVA CXL and MC NLO CXL in live rabbits to compare the safety and efficacy of both transepithelial procedures. 

Methods

Riboflavin Delivery Optimization

Tissue Preparation

Intact rabbit eyes were shipped (Pel-Freez, Rogers, AR, USA) as previously reported.^14,21,22 Briefly, eyes were rinsed in Dulbecco’s modified Eagle's medium (Sigma Aldrich, St. Louis, MO, USA), inspected for epithelial damage using Lissamine green staining (10 mg/mL in phosphate-buffered saline [PBS]; Sigma Aldrich), and incubated prior to treatment. Eyes showing epithelial staining, indicating damage, were discarded or used as control corneas treated by epithelial debridement. After incubation, eyes were imbibed with Rf solution using one of two techniques: epithelial debridement followed by 30 minutes of dripping with Rf solution in 20% high fraction dextran or epithelial micromachining followed by 30 minutes dripping of Rf solution of varying osmolarities between 200 and 400 mOsm (osmolarities were varied by varying the concentration of NaCl in the solution). Methylcellulose solution was used in place of dextran with MC treatment because it is iso-osmotic and better suited for use with intact corneal epithelium than dextran that can dehydrate the cornea. 

The creation of MCs was achieved using our previously reported protocol.¹⁴ Briefly, a 1030-nm, 1-kHz, amplified FS laser beam (One Five Origami, NKT Photonics, Birkerod, Denmark) was directed into previously described delivery optics.²¹ Pulse energy was adjusted to 5 µJ, and the maximum MC depth was set to 25 microns to ensure that the MCs did not extend into the corneal stroma. The pattern chosen for this experiment was 50-µm spaced MCs within a 6-mm diameter circular region, equivalent to 400 MCs/mm². This pattern was chosen because it had previously been shown to produce sufficient levels of stromal Rf without detectable epithelial damage.¹⁴ 

Measuring Stromal Riboflavin Concentration

Measurements of stromal Rf concentration were done as previously described.¹⁴ After Rf treatment, the epithelium was removed, followed by removal of the entire cornea from the globe around the limbus, and a central corneal button was cut using a 6-mm trephine. The button was then placed in 1 mL PBS solution and left to soak overnight. Care was taken to make sure that the corneal button was removed and placed in fluid within 4 minutes or less to avoid loss of Rf into the anterior chamber. 

The concentration of Rf in the eluate, in mg/mL, was then measured using a Spectramax Gemini XPS fluorescence plate reader (Molecular Devices, Sunnyvale, CA, USA) with 380 nm excitation and 540 nm emission as previously published.¹⁴ Serial dilutions of the stock Rf solution were used to create a standard curve, which was then used to convert the measured intensity values of each into a measure of µg/mL. Standard curve calculations were performed each time the experiment was repeated to avoid any day-to-day errors. 

Optimizing Rf Photoactivation

Since UVA single-photon photoactivation of Rf using TE Rf delivery requires UVA light to pass through the corneal epithelium before photoactivation of stromal Rf, we assessed the ability of the standard UVA CXL procedure to achieve stromal CXL through this layer, as well as the layer of Rf remaining on the surface of the epithelium. Specifically, we tested the hypothesis that UVA focused through the Rf-treated corneal epithelium would lead to absorption of UVA light and reduce stromal CXL compared to NLO CXL, which focuses infrared 1030-nm FS light through the epithelium that is not absorbed by Rf. 

Since the standard UVA CXL approach continues to apply dextran-containing Rf solution to the cornea during UVA irradiation, a layer of solution remains on the surface of the epithelium, potentially adding to the shielding effect of the Rf already contained within the epithelium itself. To assess the shielding effect of this layer, UVA transmittance alone was first measured through different thicknesses of Rf solution between 30 and 210 µm. This was accomplished by measuring the power of a UVA light transmitted through Rf solution trapped between two glass coverslips separated by gaskets of varying thickness. 

To test the effects of shielding on crosslinking, ex vivo rabbit eyes were treated with epithelial MCs followed by 30-minute application of 1% Rf in 1% methylcellulose in PBS. Eyes were then treated with UVA CXL for 30 minutes, with continued application of 1% Rf in 1% methylcellulose solution or rinsed with PBS and treated with 1% Rf in PBS alone. 

After treatment, corneas were removed and fixed overnight in 2% paraformaldehyde (PFA; Mallinckrodit Baker, Phillipsburg, NJ, USA), then sectioned and imaged for CXL-induced collagen autofluorescence (CAF), as previously reported.^14,20–23 Briefly, corneas were cut into 200-µm-thick sections using a vibratome (Campden Instruments, Loughborough, UK) and irradiated with a 760-nm FS laser light, collecting the 400- to 450-nm emission spectra within the CXL regions of corneal sections using either a Leica SP8 (Beckman Laser Institute; Leica, Wetzlar, Germany) or Leica Stellaris 8 multiphoton confocal microscope (Leica) and Chameleon FS laser (Coherent, Santa Clara, CA, USA). 

Live Rabbit Model

A comparison of MC NLO CXL and MC UVA CXL was performed in the right eyes in a total of 36 live New Zealand albino rabbits under 6 months of age. All animals were treated according to the Association for Research in Vision and Ophthalmology statement on the use of animals in vision research, and experiments were approved by the Institutional Animal Care and Use Committee (IACUC) of the University of California, Irvine (AUP 23-016). Eyes were followed sequentially and rabbits sacrificed at 2 weeks, 1 month, or 2 months (n = 3 male and 3 female/time point). Rabbits were followed postsurgery to monitor in vivo wound healing via Lissamine green staining, optical coherence tomography (OCT) imaging, and Cochet–Bonnet esthesiometry. After sacrifice, corneas were removed and examined for CAF to verify the presence of CXL, followed by immunohistochemistry (IHC) to analyze the cellular wound-healing response. 

Preoperative Evaluation

Baseline esthesiometry measurements were taken on both eyes of each rabbit using a Luneau Cochet–Bonnet Aesthesiometer (Western Ophthalmics, Lynnwood, WA, USA) on unanesthetized rabbits as described below. Rabbits were then sedated with a subcutaneous injection of 30 to 50 mg/kg ketamine hydrochloride and 5 to 10 mg/kg xylazine (MWI Animal Health, Boise, ID, USA), and OCT images were taken of each eye using a Topcon DRI OCT Triton Plus (Topcon, Tokyo, Japan) for baseline measurements of epithelial and stromal thickness and corneal haze as described below. 

Surgery: MC Formation

The creation of MCs was achieved as described above according to our previously reported protocol.¹⁴ Corneas were then imbibed with 1% Rf solution in 1% methylcellulose for 30 minutes. 

UVA CXL

UVA CXL was performed using the protocol previously described.²³ Rabbits were sedated as described above followed by topical anesthesia using 0.5% tetracaine hydrochloride (Alcon, Ft. Worth, TX, USA) to prevent pain during treatment. The central 8 mm of the cornea was exposed to 3 mW/cm² 370 nm UVA light for 30 minutes with continued application of 1% Rf in PBS every 5 minutes. After treatment, the rabbits received a subcutaneous 0.1-mL injection of buprenorphine hydrochloride (Reckitt Benckiser Healthcare Ltd., Slough, UK) and 0.3% gentamycin sulfate eye drops (Allergan, Inc., Irvine, CA, USA) in the treated eye. Gentamycin drops were repeated three times daily for 3 days to prevent infection. 

NLO CXL

After Rf imbibition as described for UVA CXL, amplified NLO CXL was performed in the same manner as previously published.²⁰ Sedated animals received a drop of topical ophthalmic 0.5% tetracaine hydrochloride (Alcon). The central 4 mm of the right corneas then underwent amplified NLO CXL using 760 nm FS light, 2 µm line separation, 0.3 µJ/pulse pulse energy, 30 mW total power, and scan speed of 20 mm/s.²⁰ Rabbits received the same after treatment care as UVA CXL rabbits to prevent pain and infection. 

In Vivo Measurements

Lissamine Green Staining

For 3 consecutive days after CXL, rabbits in all groups were monitored for epithelial damage via Lissamine green staining (10 mg/mL in PBS; Sigma Aldrich). After sedation, the treated cornea of each rabbit received a drop of Lissamine green solution to stain any epithelial defects. The cornea was then rinsed with PBS and images taken using a Celestron Handheld Digital Microscope Pro (Celestron, Torrance, CA, USA). Images were imported into QuPath open-source software²⁴ for analysis of epithelial wound dimensions. The area that had taken up the stain was measured at 24, 48, and 72 hours, through to full epithelial healing. 

OCT

OCT imaging was also performed on both eyes of anesthetized rabbits (three per treatment group) before surgery and at each time point after surgery using a Topcon DRI OCT Triton Plus (Topcon) system. These images were used to measure epithelial and stromal thickness, as well as the depth of demarcation line throughout healing. 

Cochet–Bonnet

Cochet–Bonnet esthesiometry was performed on all rabbits prior to surgery and at each time point after surgery to measure the nerve damage in both treatments using a Luneau Cochet–Bonnet esthesiometer. For this, the central cornea of an awake rabbit was probed with the esthesiometer's fiber at the maximum length. If no response was observed, the length was shortened by half of a centimeter, thereby increasing the force, and the procedure repeated until a blink reflex was reached. This process was repeated four times, and the mean value was recorded for each rabbit at 7, 14, 21, and 28 days postoperatively. 

Sacrifice and Tissue Preparation

Animals were sacrificed at 2 weeks, 1 month, or 2 months posttreatment by an intravenous injection of Euthanasia III (Vedco, Inc., St. Joseph, MO, USA) into the marginal ear vein. Immediately after sacrifice, corneas were fixed in situ under 20 mm Hg pressure via perfusion of 2% PFA in PBS to maintain the in vivo collagen structure, as described in previous studies.²³ 

Corneas were then removed and allowed to fix overnight. Each cornea was cut through the center, perpendicular to NLO CXL scan lines when relevant. Half of the cornea was then vibratome sectioned and evaluated for CAF as discussed above. The remaining half was embedded in OCT and cryosectioned, and 10-µm-thick sections were prepared for IHC staining and imaging using DAPI (1:2000) and rhodamine phalloidin (1:100) to detect keratocytes and epithelial cells. 

Tissue Imaging

Two-photon detection of corneal CAF was done as described above.^14,20–23 Regions of CAF were used to confirm the presence of CXL within each sample. If the region of CXL could not be found in a sample, that sample was removed from all other portions of the study. Three samples from the 2-month NLO CXL group were removed from the study due to their lack of confirmed CXL. 

Fluorescent imaging was also performed on the stained cryosections by using 488-nm excitation and collecting at 500 to 550 nm to observe cellular differences between the two groups using a Leica DMI600B (Leica) with Metaview Image capture. Analysis of these images was performed in QuPath imaging software to measure the depth of any acellular zones, reported as the percentage of the stromal depth of the section and the keratocyte densities within sections. To measure the cellular density, cell counting was performed within a 6-mm-wide rectangle within the 10-µm-thick sections and reported as cells/mm³. 

Statistical Analysis

Statistical analysis was performed using the Tukey–Kramer method for a multiple-comparison analysis of variance in statistical software (SigmaStat, San Jose, CA, USA). Groups were considered to have a statistically significant difference with a P value of less than 0.05. 

Results

Riboflavin Delivery Optimization

Effects of Osmolarity

Though not significant, the highest calculated stromal Rf concentration was achieved with the 300-mOsm solution (Fig. 1). These results indicate that the optimal osmolarity is likely equivalent to that of the natural tear film. 

Figure 1.

 

Riboflavin osmolarity. Stromal Rf concentration values, reported in µg/mL, after being treated with MC machining and 30 minutes of dripping with Rf solution while varying osmolarity between 200 and 400 mOsm. No values are significantly different.

UVA Transmission

As shown in Figure 2A, a 30-µm-thick solution of 1% Rf—nearly equivalent to the thickness of the corneal epithelium—transmitted only 20% of the incident UVA light, indicating that Rf solution both covering and within the epithelium will shield a significant amount of irradiation, preventing photoactivation of Rf and CXL of the stroma. 

Figure 2.

 

UVA transmittance through riboflavin barrier. (A) The dropoff in UVA power detectable through increasing thicknesses of Rf solution. (B, D) CAF and cellular damage detected after 24 hours in corneas after MC UVA CXL with continued Rf dripping. (C, E) Corneas were rinsed before UVA CXL began and did not receive continued dripping.

To test this hypothesis, we next performed MC UVA CXL on ex vivo rabbit eyes using a 30-minute UVA exposure as described above. Eyes that were continually treated with topical Rf in methylcellulose every 5 minutes during UVA exposure showed no CAF or stromal CXL (Fig. 2B). However, eyes that were rinsed with 50 mL PBS to remove surface Rf prior to UVA exposure for 30 minutes and continued dripping with Rf in PBS showed strong CAF within the stroma (Fig. 2C). Furthermore, when eyes were incubated overnight in organ culture and then removed, fixed, cryosectioned, and stained for live/dead biomarkers, eyes treated with the standard Rf application showed no damage to either the epithelium or stroma (Fig. 2D). By contrast, eyes that were rinsed of Rf before UVA CXL and then organ cultured showed marked thinning and damage to the corneal epithelium in addition to loss of anterior stromal keratocytes (Fig. 2E). 

UVA CXL Versus NLO CXL

Epithelial Damage: Lissamine Green Staining, Cochet–Bonnet, and OCT

While rabbits receiving MC UVA CXL all showed epithelial defects after 24 hours, averaging 23.89 ± 5.6 mm² (Figs. 3A, 3C), those treated with MC NLO CXL showed no epithelial damage (Fig. 3B). Eyes showing epithelial defects also healed by day 3 after UVA CXL (Fig. 3C). Rabbits receiving MC UVA CXL also showed a complete loss of corneal sensitivity after 24 hours (Fig. 4), averaging 0.83 ± 0.24 cm needed to elicit a blink reflex compared to the control value of 4.25 ± 0.35 cm. The sensitivity of the UVA CXL–treated eyes remained significantly lower through the first month postoperatively. Rabbits receiving MC NLO CXL showed no significant loss of corneal sensitivity. Furthermore, OCT imaging measurements, as seen in the graphs in Figure 5, revealed that MC UVA CXL–treated eyes exhibited a significant increase in stromal thickness of 232.55 µm and a complete loss of epithelium in the central cornea, both of which returned to baseline values by day 7. MC UVA CXL also resulted in a clear line of demarcation, peaking in depth at day 7 and averaging 236.78 ± 11.88 µm, significantly deeper than all further days. Representative OCT images of MC UVA CXL eyes at each time point can be seen in Figures 5A–E, with arrows representing the depth of the demarcation line (shown at days 1, 7, 14, 28, and 56, respectively). NLO CXL–treated eyes showed no significant change in any thickness value, and no demarcation lines could be seen in any animals at any time point. 

Figure 3.

 

Lissamine green epithelial staining. Corneas were stained with Lissamine green for 3 consecutive days following each CXL procedure. (A) A representative epithelial wound 24 hours after MC UVA CXL and (B) lack of wound 24 hours after MC NLO CXL. The graph shows the wound area measured each day in MC UVA CXL eyes, which fully heals by day 3. All days showed a significant decrease in wound area, indicated with asterisks.

Figure 4.

 

Cochet–Bonnet esthesiometry. The corneal sensitivity of rabbits was tested over time in both treated and untreated eyes via Cochet–Bonnet esthesiometry. MC UVA CXL (red) resulted in a significant decrease (P < 0.05) in sensitivity, as indicated by asterisks. The sensitivity of MC UVA CXL–treated eyes returned to significantly normal values after day 28. MC NLO CXL (blue) never resulted in any drop in corneal sensitivity.

Figure 5.

 

OCT imaging. Representative OCT images of MC UVA CXL treatment at the time points of 1, 7, 14, 28, and 56 days are shown in A to E, respectively, with arrows denoting the line of demarcation. Graphs show the depth of the demarcation line, stromal thickness, and epithelial thickness over the same time points for both UVA and NLO treatments (red and blue, respectively). MC UVA CXL was shown to result in an increased stromal thickness at day 1 postprocedure accompanied by a total loss of epithelial thickness. MC NLO CXL resulted in no change in either value at any time point. A clear line of demarcation was only observed in MC UVA CXL eyes. Asterisks denote a significant difference of P < 0.05.

Image Analysis

CAF was used to confirm that all corneas included in the study contained CXL. We were unable to confirm the region of CXL via CAF in three rabbits within the 2-month NLO CXL group and therefore excluded the results from those animals. 

IHC of NLO CXL detected very rare acellular patches within 2-week samples, (Fig. 6A, double arrowhead) and no acellular regions at all at any further time points, indicating cells had already begun migrating back to the area (Fig. 6B). This is consistent with previous data.²⁰ UVA CXL eyes, however, showed regions of cellular death at all time points (Figs. 7B, 7D, 7F), consistent with previous studies.²³ Acellular regions from both procedures corresponded directly with regions of CAF (Figs. 7B, 7D, 7F). Analysis of IHC resulted in decreasing depths of acellular zones at each time point, listed in Table 1 as a percentage of the total thickness of the depth measurement of each sample. The depth of this region was also significantly shallower than the depths of the demarcation lines measured from the OCT images until the 2-month time point, when the demarcation line became hard to distinguish at an exact location in any OCT images (Fig. 7). Keratocyte density measurements from the IHC images identified lower densities of 636.7 ± 111.01, 686.9 ± 104.68, and 773.0 ± 147.37 keratocytes/mm² (K/mm²) for UVA-treated eyes at 2 weeks, 1 month, and 2 months, respectively, while NLO-treated eyes showed cell densities of 882.7 ± 113.09, 853.8 ± 117.06, and 957.5 ± 120.57 K/mm² (Table 2). While cellular densities were higher in NLO CXL eyes at all time points, they were only significantly different at 2 weeks. 

Figure 6.

 

Collagen autofluorescence and immunohistochemistry of MC NLO CXL. After sacrifice, corneas were removed and processed for either IHC or CAF imaging. For IHC, sections were treated with DAPI for staining cell nuclei (cyan) and rhodamine phalloidin for staining actin (red). Cellular staining of MC NLO CXL showed only small, sparse regions of cellular death at 2 weeks (A), while cells had fully migrated back into the region of CAF (green) by 1 month (B).

Figure 7.

 

OCT compared to immunohistochemistry after MC UVA CXL. In vivo OCT imaging showing representative demarcation lines at the time points of 14, 28, and 56 days (A, C, and E, respectively) with arrows to represent the demarcation line compared to their corresponding IHC within the same animals with arrows to represent depth of the acellular zone. Sections were treated with DAPI for staining cell nuclei (cyan) and rhodamine phalloidin for staining actin (red), with CAF shown in green.

Table 1.

 

Demarcation Line Versus Acellular Zone

 

Table 2.

 

Cellular Density

 

Discussion

Traditional UVA CXL requires removal of the corneal epithelium to facilitate penetration of sufficient Rf into the corneal stroma to produce an effective CXL reaction. Removal of the corneal epithelium is painful, leads to delayed visual recovery, and increases the risk of corneal infection.^3–9 While the search for the perfect transepithelial approach to UVA CXL is a highly active field of research, no current attempt at transepithelial UVA CXL has been as successful as the traditional method.^9–13 Most of these poor results can be attributed to the lack of adequate Rf within the corneal stroma to induce a successful reaction, but this ignores a second hurdle. Even if a sufficient stromal Rf concentration is achieved, the subsequent UVA irradiation would still need to pass through the Rf-soaked epithelium, partially blocking the stroma and likely causing damage to the epithelium. This epithelial MC technique has already been shown to achieve stromal Rf concentration sufficient for CXL with only 1% epithelial damage.¹⁴ Additionally, in contrast to UVA CXL, NLO CXL has the ability to bypass the epithelium entirely, preventing the second hurdle of transepithelial CXL. The purpose of this study was to test and compare the safety of both CXL modalities combined with epithelial MCs. 

Our first step was to evaluate the most efficient manner of Rf solution application by testing the osmolarity of the solution and the transmittance of UVA light through layers of methylcellulose-thickened Rf solution. It was found that altering the osmolarity of Rf solution had no significant impact on the concentration of Rf that diffused into the stroma within the range tested, and therefore the original 300-mOsm solution was used in subsequent experiments. Additionally, it was verified that continued application of thickened Rf solution severely dampened the CXL effect during UVA CXL. The unrinsed cornea exhibited both very little CAF and cellular damage, indicating the UVA irradiation did not make it beyond the layer of Rf on the surface of the eye. In contrast, the opposite was observed in rinsed corneas, with CAF and an overlapping region of cellular damage. Furthermore, when UVA power was measured going through increasing thicknesses of Rf solution, it was found that a thickness of only 30 µm was sufficient to block 80% of UVA irradiance. For this reason, we determined that not only the Rf-soaked epithelium but also the layer of methylcellulose-thickened Rf remaining on the surface of the epithelium blocked the necessary UVA irradiation from reaching the corneal stroma, producing a reduced CXL effect. Because of this, all in vivo studies incorporated a corneal rinse with PBS and continued application of Rf solution in PBS, as opposed to thickened methylcellulose, during the CXL portion of the procedure. These results are somewhat contradictory to what others have published.²⁵ Morgan et al.²⁵ found both decreased Rf concentration and CXL efficacy in corneas rinsed prior to UVA exposure, but this can be explained by the fact that that study did not continue Rf application for the duration of CXL. 

The in vivo portion of this study was designed to evaluate the second hurdle of transepithelial CXL: getting irradiation past an intact epithelium once sufficient Rf levels have been reached without further damaging the epithelium with that irradiation. For this, we tested the safety and efficacy of combined MC NLO CXL therapy compared with MC UVA CXL therapy in both male and female rabbits. At no point was there a significant difference in any measurement between the sexes. For UVA CXL, Lissamine green epithelial staining showed a region of complete epithelial death, which took 3 days to heal, as expected.¹⁷ Cochet–Bonnet also showed a significant decrease in corneal sensitivity that lasted through day 28, indicating nerve damage following UVA CXL. These findings are consistent with those reported by Xia et al.,²⁶ who provided data suggesting that nerve regeneration was not fully complete until day 90 after traditional UVA CXL, and Chen et al.,¹⁶ who noted significant changes in nerve density after BAK transepithelial UVA CXL. Additionally, OCT imaging of UVA CXL–treated rabbits showed a clear demarcation line, significantly increased stromal thickness, and significantly decreased epithelial thickness. Notably, the depth of the demarcation line decreased over time until 1 month postoperatively, when it then became extremely rare or difficult to identify. Similarly, the same trend was observed for the depth of the acellular zone measured via IHC imaging. All of these results demonstrate a significant and continued pattern of damage following MC UVA CXL. 

In contrast, MC NLO CXL produced very different in vivo results. No sample showed a measurable epithelial wound, proving that NLO CXL is capable of leaving the epithelium entirely intact throughout the procedure. Cochet–Bonnet testing showed no significant difference in corneal sensitivity at any point, showing that CXL can be performed without damaging superficial nerves. OCT imaging showed no region of increased intensity or demarcation line, and there were no changes in any thickness measurements. In fact, it was difficult to tell if CXL had actually occurred within the NLO CXL group before sacrifice. 

Image analysis, both CAF and IHC, was used as a means of confirming CXL within all samples. While every UVA CXL and all other NLO CXL samples showed clear CAF overlapped with regions of cellular death, three rabbits within the NLO CXL 2-month time point were removed from the study due to lack of CAF. This could be due to either an off-center CXL treatment, making the spot hard to find during postsacrifice processing, or the depth being set too deep on the day of treatment. All three rabbits in question had their treatments performed on the same day, and since the depth of CXL is set once each day, we believe the second scenario is most likely. Postsacrifice imaging analysis in the remaining samples revealed depths of acellular zones in MC UVA CXL that were significantly shallower compared to the depth of the demarcation line measured in vivo. Both measurements decreased throughout the first month of healing. Others have used the depth of the demarcation line as a tool for measuring CXL efficacy by equating it to simply a measure of the region where stromal CXL exists,²⁷ but both the decrease in the depth of the demarcation line and the decrease in the acellular zone over time suggest that cell activation and migration correlate more closely with the appearance of the demarcation line. If the demarcation line exists as only a transition between the CXL and non-CXL areas of stroma, that could imply that any movement or loss of the demarcation line would also result in loss of CXL effect, but many previous studies have shown that the effects of CXL such as CAF, increased stiffening, or corneal flattening can last months to years.^9,23,28–35 The correlation shown in this study between the depths of the demarcation line and the depths of acellular zones hints at a correlation between cellular healing and the origins of the increased refractive area of cornea marking the line of demarcation after the procedure. Alternatively, cell migration back into the wounded area is connected to the migrating demarcation line. It should be noted that this trend did not continue through the full 2 months. A noticeable demarcation line at this time point, however, was extremely rare and did not appear in all images of samples where one was spotted. Additionally, in the instances where it was seen, the line was more diffuse compared to previous time points. It is possible that there were too few measurable images to make a statistically sound conclusion (i.e., the demarcation line had mostly disappeared) or that less activated keratocytes produced a more dispersed demarcation line. 

More evidence for a connection between cellular migration and the demarcation line is the notable lack of a demarcation line found in any NLO CXL sample in this study. As only sparce, patchy acellular regions were visible following NLO CXL, it is logical to say that this lack of an acellular zone could be connected to the lack of a demarcation line. This idea could also be supported in part by the appearance of a demarcation line in other corneal injuries that cause cellular damage and keratocyte activation without any CXL effect such as alkali burn.³⁶ 

In summary, FS epithelial machining, when paired with NLO CXL, completely avoids epithelial damage and loss of corneal sensitivity, and it shows dramatically reduced anterior stromal damage compared to MC UVA CXL. Cellular staining also showed NLO CXL was capable of faster cellular recovery after crosslinking than UVA CXL. This suggests that MC NLO CXL—unlike MC UVA CXL—can achieve a faster visual recovery without postoperative pain or risk of infection, producing a truly safe and effective transepithelial CXL procedure. 

Acknowledgments

The authors acknowledge support to the Gavin Herbert Eye Institute at the University of California, Irvine from an unrestricted grant from Research to Prevent Blindness and from NIH core grant P30 EY034070. 

Supported in part by an R01 research from the NIH NEI EY024600, an NEI Core p30 grant EY034070, and funding from the Discovery Eye Foundation, an Unrestricted Grant from Research to Prevent Blindness (RPB-203478), and the Skirball Program in Molecular Ophthalmology and Basic Science. 

This study was made possible in part by access to the Nonlinear Optical Microscopy Lab, part of the Optical Biology Core Facility at UC Irvine, a shared resource supported by the Chao Family Comprehensive Cancer Center (P30CA062203) and the Office of the Director, National Institutes of Health award S10OD028698. 

The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. 

Disclosure: S. Bradford, None; R. Joshi, None; S. Luo, None; E. Farrah, None; Y. Xie, None; D.J. Brown, M2Cor (I); T. Juhasz, M2Cor (I); J.V. Jester, M2Cor (I)