Nonlinear optical collagen cross-linking and mechanical stiffening: a possible photodynamic therapeutic approach to treating corneal ectasia

Abstract. In this study we test the hypothesis that nonlinear optical (NLO) multiphoton photoactivation of riboflavin using a focused femtosecond (FS) laser light can be used to induce cross-linking (CXL) and mechanically stiffen collagen as a potential clinical therapy for the treatment of keratoconus and corneal ectasia. Riboflavin-soaked, compressed collagen hydrogels are cross-linked using a FS laser tuned to 760 nm and set to either 100 mW (NLO CXL I) or 150 mW (NLO CXL II) of laser power. FS pulses are focused into the hydrogel using a 0.75 NA objective lens, and the hydrogel is three-dimensionally scanned. Measurement of hydrogel stiffness by indentation testing show that the calculated elastic modulus (E) values are significantly increased over twofold following NLO CXL I and II compared with baseline values (P<0.05). Additionally, no significant differences are detected between NLO CXL and single photon, UVA CXL (P>0.05). This data suggests that NLO CXL has a comparable effect to conventional UVA CXL in mechanically stiffening collagen and may provide a safe and effective approach to localize CXL at different regions and depths within the cornea.


Introduction
The cornea is the outer, transparent part of the eye that provides two thirds of the refractive power for focusing light back to the retina. The cornea is composed of five tissue layers: the epithelium, Bowman's layer, stroma, Descemet's membrane, and endothelium. The corneal stroma represents 90% of the corneal thickness (∼500 μm) and is comprised predominantly of a fibrous collagenous extracellular matrix that provides strength and shape to the cornea.
Corneal shape and refractive power are thought to be controlled by the mechanical properties of the stromal collagen and that regional differences in collagen organization influence local tissue mechanics. Past studies by Kokott suggested the presence of both a circumferential band of collagen that constrains the corneal curvature at the limbus, and a preferred nasal-temporal/superior-inferior orientation of deeper collagen layers that aligns with the extra ocular muscles. 1 Studies using x-ray diffraction have confirmed these earlier findings and shown a preferential alignment of collagen fibers in the mid-to posterior central cornea and an apparent tangential orientation in the corneal periphery. 2 More recent studies using nonlinear optical microscopy to image second-harmonic-generated (SHG) signals from collagen have also identified sutural or bow-string collagen fibers in the anterior stroma that insert into Bowman's layer and extensively intertwine with orthogonally arranged collagen fibers. 3 The degree of collagen intertwining has also been associated with increased mechanical stiffness of the anterior cornea, and that the mechanical weakening of the cornea as occurs in Keratoconus is associated with loss of sutural collagen fibers and decreased collagen intertwining. 3,4 Keratoconus affects one in 2000 individuals in the United States and is the most common cause for corneal transplantation surgery. 5,6 A strategy for treating keratoconus has been introduced by Spoerl et al. that involves photoactivation of riboflavin by ultraviolet A light (UVA) inducing collagen cross-linking (CXL) leading to mechanical stiffening of the cornea. 7,8 Clinical results suggest that UVA CXL is effective in treating ectasia and that treatment substantially, if not permanently, halts the progression and delays the need for corneal transplantation. [9][10][11] Currently, UVA CXL is suggested for children and adolescents at the time of keratoconus diagnosis to prevent early progression of the disease. 12 In addition to treating other ectatic disorders such as post-LASIK ectasia, recent studies have shown that UVA CXL may have a wider range of therapeutic effects including enzymatic inhibition, antimicrobial activity for the treatment of corneal ulceration 13 and infectious keratitis. 14 Overall, UVA CXL therapy may substantially reduce the need for annually nearly 50,000 corneal transplant procedures performed in the United States. 9,12 UVA CXL therapy involves an approach where the cornea is presoaked with riboflavin. 15 This currently requires the removal of the overlying corneal epithelium to allow penetration of riboflavin into the stroma, albeit alternative approaches not requiring epithelial removal are being intensively investigated. 16,17 Corneas are then exposed to 5.4 J∕cm 2 of 370 nm UVA light for two to 30 min depending on the illumination intensity (3.0 to 45 mW∕cm 2 ). 18,19 Photoactivation of riboflavin induces the formation of free radicals that interact with corneal proteins leading to covalent bonding within collagen fibrils. 20 Since generation of free radicals also causes cell death, riboflavin concentration, UVA irradiance, and treatment duration have all been selected to minimize cellular damage and protect the corneal endothelium. 10 These parameters effectively limit CXL to the anterior 300 μm of the corneal stroma 21 and also restrict UVA CXL therapy to corneas thicker than 400 microns. Alternative UVA CXL protocols that either preswell the cornea or reduce radiance exposure have been proposed for treating thin corneas, 13 however, endothelial damage even in thicker corneas has been reported and remains a concern. 22 Because of these concerns, it would be beneficial to develop methods that could improve both the lateral and axial control of corneal CXL to provide better safety and improve efficacy and expand therapy to treat other refractive disorders and corneal diseases.
Recently, femtosecond (FS) lasers have been used to photoactivate photosensitizers through the nonlinear optical (NLO), multiphoton processes within small focal volumes to provide high lateral and axial control of cross-linking for nano-fabrication (NLO CXL). 23 Studies show that NLO CXL can be used to create detailed three-dimensional (3-D) microstructures that integrate collagen and other proteins, 24,25 and that the material stiffness of these nanostructures can be modulated by laser dwell time. 26 Recently, NLO CXL has been used to stiffen bioengineered cardiac tissue, although the degree of mechanical stiffening was limited to less than 35% and not comparable to the therapeutically utilized UVA CXL of the cornea, which induces the mechanical stiffening of more than 200%. 27 In this study, we have evaluated the effectiveness of NLO CXL to mechanically stiffen compressed type I collagen hydrogels and compared the stiffening effect with that obtained using conventional UVA CXL.

Determination of an Optimal Wavelength for Two-Photon Excitation of Riboflavin
To determine the optimal wavelength for NLO riboflavin photoactivation, riboflavin fluorescent signal intensity as a function of femtosecond (FS) laser wavelength was measured. , and the fluorescent signal was collected using the Meta Detector (Carl Zeiss) over the spectral range of 506 to 559 nm to include the peak signal (520 to 530 nm). 28 To appropriately set detector gain and offset and to provide maximum intensity of signal at peak excitation, the edge of the riboflavin solution in the Petri dish well was imaged providing the simultaneous detection of riboflavin fluorescence and negative background. Initial experiments indicated that peak excitation of riboflavin was obtained with 760 nm excitation at 100 mW of laser power. Gain and offset were then adjusted to achieve a full range of pixel intensity. Images were then taken at different FS laser wavelengths from 740 to 960 nm with the laser power adjusted to 100 mW at each wavelength maintaining the same gain and offset. The intensity of signal was then averaged over a 0.031 mm 2 rectangular region from the region of riboflavin fluorescence and negative background using Metamorph image processing software (Molecular Devices, Sunnyvale, California).

Formation of Compressed Collagen Hydrogels
Compressed collagen hydrogels were produced using methods previously described. 29 Briefly, acid soluble, rat-tail type I collagen (3 mg∕ml) was neutralized with 1N NaOH (Acros, New Jersey), poured into a 24 well tissue culture plate (3 ml∕well) and allowed to polymerize at 37°C in a 5% CO 2 humidified incubator for 30 min. Collagen hydrogels were then placed on top of a compression platform comprised of a #54 Whatman filter paper (GE Healthcare, Piscataway, New Jersey) having a 7.6-mm-diameter central window, which was used to transport the gel following compression. Beneath the transporting paper was a 50-μm nylon mesh (Sefar Filtration Inc., Depew, New York), an absorbing layer (two #1 Whatman filter papers, GE Healthcare) and a plastic plate as shown in Fig. 1(a). The hydrogel was then covered by a second nylon mesh, two #1 Whatman filter papers and a plastic plate, and then compressed by a 90-g mass for 5 min to squeeze out excessive water and form a thin collagen hydrogel with high collagen fiber density [ Fig. 1 Compressed gel thickness was controlled by placing a 250-μmthick spacer placed around the gel. Compressed gels were then moved for mechanical testing and collagen cross-linking by using the transporting paper to avoid possible damage [ Fig. 1(c)] during transportation. No visible damage to the hydrogel during these movements was detected.

Nonlinear Optical and UVA Collagen Cross-Linking
To perform NLO and UVA CXL, gels were initially placed into a modified POC (Perfusion, Open and Closed)-R environmental chamber (Catalog #000000-1116-079, Carl Zeiss), as shown in  For NLO CXL the chamber was mounted onto the Zeiss LSM 510 Meta and a 760-nm FS laser beam was focused into gel through a 20 × ∕0.75 NA objective lens (Zeiss). As shown in Fig. 2(b), the hydrogel was tile scanned over a 5.2 × 5.2 mm area through the thickness of the gel at 2 μm axial steps using the MultiTime function in the LSM software. As described in Fig. 2(c), each tile was raster scanned (512 lines) over a 455 × 455 μm area at 28 cm∕s scan speed.
Hydrogels in environmental chambers used for control or UVA cross-linking were stored over night in riboflavin during NLO CXL of hydrogels to provide the same condition as NLO CXL gels. Hydrogels were then transferred to a 60-mmdiameter tissue culture dish (Becton Dickinson Labware, Franklin Lakes, New Jersey) containing fresh riboflavin solution and either exposed to 370-nm UVA light at an irradiance of 3.0 mW∕cm 2 for 30 min or were mock treated for 30 min. The area of cross-linking was limited to 5.2 × 5.2 mm by placing a mask over the hydrogel.

Measurement of Mechanical Stiffness
To measure the mechanical properties of the compressed collagen gels before and after CXL, gels were clamped between an O-ring and gasket using two metal plates with a 7.6-mmdiameter central window ( Fig. 3 and inset). The bottom plate was supported by a transparent cylinder and the device attached to a 3-D position control assembly as previously described. 21 A small amount of low viscosity silicon oil (Pure Silicone Fluid 1.5cSt, Clearco Products Co, Bensalem, Pennsylvania) was then applied to the surface of each gel to prevent dehydration and adhesion of the indenting probe tip to hydrogel surface prior to measurement of mechanical stiffness. Gels were indented at the center using a 250-μm-diameter round tip probe attached to a 5-μN resolution force transducer (F10, Harvard Apparatus, Holliston, Massachusetts). The axial position of the probe tip was controlled by a computer using LabVIEW 8.2 (National Instrument, Austin, Texas) through an electronic controller (ESP300, New Port Corporation, Irvine, California) so that the gel was indented at 20 μm∕s through 1 mm depth  Fig. 3 Schematic of the indenting force-measurement system. Hydrogels were clamped between metal plates on cylinder on a XY positioning system using translation stages. 250-μm-diameter indenting probe was screwed into the tip of force transducer attached to Z positioning translation stage driven by a step motor. Hydrogel was clamped using O-ring and gasket and indented at the center of window (inlet). over 10 cycles. The position of the tip and the signal from the force transducer were recorded every 0.05 s as previously described. 3 Following indentation measurement, the collagen hydrogel thickness was microscopically measured by 3-D imaging the gel using NLO second harmonic generation (SHG) to detect collagen. Gels were placed on the LSM 510 Meta, and SHG signals were generated by a 800-nm FS laser beam using a 20×∕0.75 NA objective (Zeiss). Forward scattered SHG signals were collected using a 400 AE 40 nm bandpass filter, and the hydrogels were axially scanned at 5 μm steps. Using LSM software, 3-D data sets were collected, reconstructed, and the thickness of the gels determined (Fig. 4). To estimate thickness, data sets were collected in five locations per gel, the center and 2 mm from the center at 0 deg, 90 deg, 180 deg, and 270 deg, and the average was calculated.
Elastic modulus was calculated using the modified Schwerin point-load solution: 30 where E ¼ elastic modulus, δ ¼ indentation displacement, v ¼ Poisson's ratio, a ¼ radius of window holding gel, h ¼ gel thickness, and P ¼ indenting force at the 10th cycle. Poisson's ratio was assumed to be 0.5 as in the past study. 31 Elastic moduli before and after each treatment, E Baseline and E post , were calculated by inputting baseline and post-treatment measurements and thickness into the equation.

Experimental Design
A total of 12 compressed collagen hydrogels were manufactured for the study. Hydrogels gels were then divided into four groups: (1) not cross-linked, (2) conventional UVA CXL, (3) 760-nm FS light at 100 mW power (NLO CXL I), and 4) 760-nm FS light at 150 mW power (NLO CXL II). All hydrogels were soaked in 0.1% riboflavin in PBS for 30 min in a Petri dish. All hydrogels were also stored in the environmental chamber overnight to simulate conditions of NLO CXL and then transferred to a Petri dish to simulate conditions of UVA CXL. Mechanical stiffness and gel thickness was measured before and after treatment for all hydrogels. Each experimental run required approximately 12 h to complete, started and finished at the same time, and compared a NLO CXL to either a control or UVA CXL hydrogel. As a result, each hydrogel was treated exactly the same except for UVA, NLO, or mock light exposure.

Statistics
All measurements are reported as means AE standard deviations. The stiffening effect of each treatment was assessed by evaluating the statistical difference between E Baseline and E post values at 1-mm indenting depth using a repeated measures two-way ANOVA with a critical level of 0.05 (SigmaStat ver.3.1; Systat Software Inc., Point Richmond, California). The amount of stiffening was also assessed by comparing the ratio of elastic modulus, E post ∕E Baseline , between groups using a one-way ANOVA (SingmaStat ver 3.1) with the same critical level. The Holm-Sidak Method was used for both analyses. Figure 5(a) shows the riboflavin background and negative background with a rectangle on each side for analysis of fluorescence signal. Figure 5(b) shows the NLO-induced riboflavin fluorescence intensity as a function of FS wavelength normalized to the maximum value at 760 nm. It can be seen that riboflavin fluorescence was higher with FS excitation wavelengths of 740 and 760 nm with a peak at 760 nm. There was a marked reduction in intensity at FS wavelengths higher than 780 nm. These data indicated that 760 nm was the optimal wavelength to induce maximum NLO excitation of riboflavin in this experimental system, and the reduction of fluorescence from 780 nm wavelength is in agreement with the absorption over wavelength in the past study. 28 Although riboflavin is known to have a second absorption peak at 450 nm, a nonlinear absorption spectra peak above 760 was not detected. Figure 6(a) is a representative indenting force curve for 10 cycles showing the viscoelastic properties of collagen hydrogels with hysteresis defined by the difference between loading and unloading curves. Comparing the indenting force at a 1 mm depth in each cycle, it can be seen that there was a noticeable difference between the first and second cycle that diminished with increasing cycle number and became negligible (less than 1%) by the 10th cycle. This suggests that a reasonably steadystate behavior was achieved, and the indenting force values from the 10th cycle could be used to calculate elastic modulus with minimum viscoelastic contribution. Figure 6(b)-6(d) shows representative indenting force curves from the 10th cycle of baseline and post-treatment measurements in UVA CXL, NLO CXL (I) and (II) groups, respectively. Figure 7 shows the comparison in the average E values obtained before and after treatment for each group. First, there was no significant difference in the baseline measured E values between the different groups, although there was considerable variation between groups with average E values ranging from 0.92 AE 0.23 MPa to 1.53 AE 0.39 MPa. Second, photoactivation of riboflavin using either UVA or NLO at the two different powers showed significantly increased post-treatment E values compared to baseline (P < 0.05). Control hydrogels showed a slight increase in post-treatment E values, but this difference was not significant (P > 0.05). Because of the variation in the baseline E values between groups, a ratio between E post ∕ E Baseline , was calculated and the differences in stiffening between groups compared. As shown in Fig. 8, the E value ratios for the UVA CXL and NLO CXL groups were significantly higher compared to the control group (P < 0.05), while no significant difference was detected between UVA CXL and NLO CXL (P > 0.05) groups. Previous studies have shown that multiphoton processes can be used to activate photosensitizers and create detailed 3-D nanostructures, 24,25 with modified material stiffness characteristics based on FS laser dwell time. 26 It has also been demonstrated 27 that mechanical stiffening of bioengineered cardiac tissue can be achieved using a 5-KHz FS laser light focused using long focal distance lenses, although the degree of stiffening was less than 35%. The current report demonstrates for the first time that NLO photoactivation of riboflavin using 75-MHz FS laser focused using a high NA lens within a collagen hydrogel induces similar amount of collagen stiffening to that achieved by UVA CXL. Considering that NLO photoactivation of riboflavin is theoretically limited to the two photon focal volume, which for a  0.75 NA lens is 0.5 μm lateral and 2.5 μm axial, 32 the current NLO CXL approach has the potential to address some of the limitations of UVA CXL. Specifically, NLO CXL can provide improved lateral and axial control of the volume of cornea that is cross-linked, thereby avoiding damage to underlying structures outside the focal volume of FS laser beam.

Results
Improved control of CXL volume may become increasingly important as UVA CXL therapy gains wider acceptance. While UVA exposure levels during CXL was originally designed to avoid damage to the underlying corneal endothelium and the intraocular lens, recent cases of corneal endothelial failure following CXL in corneas thicker than 500 μm have been reported. 22 Additionally, a retrospective study of 350 patients identified 2.9% developing visually significant corneal edema presumed to be related to endothelial damage after receiving UVA CXL with stromal thicknesses greater than 400 μm. 33 Furthermore, in a prospective study of 12 patients an average 10% endothelial cell loss was identified after receiving UVA CXL with stromal thicknesses less than 400 μm. 34 While application of hypo-osmotic solutions to swell the cornea are recommended in such cases, 13 the effect of this treatment on endothelial cell density has not been evaluated.
The ability to control the CXL volume may also help to more safely expand the therapeutic domain to other ectatic disorders including post-LASIK ectasia. While the incidence of post-LASIK ectasia is low (<1%), it is a severe complication that can occur when the residual stromal bed thickness is not sufficient to maintain normal mechanical strength. 35 While UVA CXL at the time of surgery has been reported, 18 cross-linking of the residual bed might result in significant corneal endothelial cell damage and potentially damage the lens epithelium when residual bed thickness becomes less than 250 μm. By comparison, NLO CXL would be limited to the two-photon volume of the focusing lens, which can be precisely controlled by moving the lens respective to the cornea. In this manner the entire residual bed could undergo collagen cross-linking without exposing the corneal endothelium or underlying structures to the damaging effects of free radicals.
Another potential disadvantage of UVA CXL is that crosslinking is limited to the anterior corneal stroma. 21,36,37 Recent studies have shown that the collagen organization and mechanical properties of the corneal stroma show regional variations including a stiffer anterior stroma with collagen fiber intertwining and branching, a more compliant posterior stroma with collagen fibrils aligned along the major nasal/temporal and superior/inferior meridians, and a tangential collagen fibril orientation in the periphery. 2,3 How these different collagen fiber and fibril alignments control corneal mechanical properties and shape remain unknown. Although previous studies have shown that keratoconus corneas show structural changes in the anterior cornea, 4,38,39 whether cross-linking limited to the anterior 300 μm of stroma provides the most effective treatment is also not known. Since NLO CXL can be performed regionally and is not limited solely to the anterior cornea, the effects of different treatment strategies involving different depths, thicknesses, and patterns could be easily tested and may provide improved treatment outcomes.
In the current study, NLO CXL was performed using a Zeiss LSM 510 Meta, which limited the area available for photoactivation to a small 455 × 455 μm field of view using the Zeiss 20×∕0.75 NA objective. A large field was successfully crosslinked by tiling over the collagen hydrogel using an automated XY translational stage and the built-in LSM software MultiTime macro function. Because of the slow y-and z-scanning rate and XY translational stage movement, cross-linking using this  . Increase ratios in UVA and NLO CXL treated groups were significantly different from that of control (P < 0.05). Also note that the increase ratios of NLO CXL groups are not significantly different from that of UVA CXL (P > 0.05). system generally required 3 h to complete, compared to 2 to 30 min for UVA CXL, depending on the intensity of UVA light. Optimization of the scanning approach by adjusting scanning parameters-including two-photon focal volume, line speed, plane separation and intensity of laser beam-could markedly shorten the time for CXL, making this approach clinically feasible. However, increased scan speed will decrease FS laser dwell time leading to reduced generation of free radicals needed for cross-linking. However, it should be noted that NLO excitation of riboflavin can be achieved using concentrations of riboflavin above the clinically used 0.1% and at deeper depths in swollen corneas (∼1 mm). 40 These findings suggest that decreased dwell time may be offset by increasing riboflavin concentration to generate free radicals.
In the current study, changes in material stiffness related to NLO and UVA CXL were assessed by tissue indentation as used in previous studies. 30,41 In addition, type I collagen hydrogels were used since type I collagen is the principal fibrous protein of the cornea and through compression can be reproducibly fabricated to generate a thin, homogeneous, dense collagen fiber structure similar to the cornea that was thin enough to facilitate NLO CXL throughout the corneal phantom. Since collagen organization in collagen hydrogels is random, tissue-indentation testing was used to measure the stiffening of collagen regardless of orientation. As shown in a past study, 42 extensiometry with unidirectional tensile measurements of materials with random collagen fibers orientation only assess the collagen fibers parallel to the stretching direction. On the other hand, indentation testing is more convenient when using thin collagen hydrogels as used in the current experiment. Considering the ratio of hydrogel thickness and indenting depth, 0.08 to 0.1, it can also be assumed that the force measured during indentation is from stretching of the hydrogel suggesting that indentation testing can be used as an alternative for measuring tensile modulus for small and thin materials.

Conclusions
This study shows that NLO photoactivation of riboflavin within collagen hydrogels can lead to mechanical stiffening of equal magnitude to that achieved using conventional UVA CXL. We propose that NLO CXL has the promise to provide higher lateral and axial control of corneal collagen cross-linking and a safe and effective treatment for corneal ectasia. Future work to develop an instrumental capable of rapidly inducing NLO CXL within the corneal stroma is needed to test this hypothesis.