Correlation Between Multimodal Microscopy, Tissue Morphology, and Enzymatic Resistance in Riboflavin-UVA Cross-Linked Human Corneas

Abstract

Purpose.: To explore the utility of multimodal microscopy as a noninvasive tool to assess corneal collagen cross-linking (CXL) efficacy, we investigated the correlation between riboflavin (RF) axial profile, second harmonic generation (SHG) imaging, and histological/biochemical changes of human corneas after RF-ultraviolet A (UVA)–catalyzed CXL.

Methods.: De-epithelialized human corneoscleral tissues were imaged by confocal and multiphoton microscopy to study RF tissue diffusion profile and SHG-based roughness index (Rq) after CXL. We installed 0.1% RF for 5, 10, and 20 minutes, respectively, followed by UVA irradiation, while dextran drug vehicle–treated corneas served as controls. Masson's trichrome staining and collagenase digestion assay were employed to assess ultrastructural modifications of collagen lamellae and bioenzymatic strength following RF-UVA CXL.

Results.: Stromal absorption of RF was significantly higher in 20 minutes compared with 5- and 10-minute drug instillations. The roughness index of SHG images was reduced after RF-UVA CXL at all RF instillation time points compared with dextran controls. Interestingly, correlation between axial profiles of RF dosage and Rq index was only observed in 10- and 20-minute RF instillations (R² = 0.13 and 0.28, respectively, all P < 0.05), but not in the 5-minute group. Masson's trichrome staining revealed collagen fibril compaction in cross-linked corneas in an RF dose-dependent manner. Collagenase digestion assay showed significantly increased biochemical strength by higher RF doses in cross-linked corneas.

Conclusions.: Intrastromal RF distribution profiles correlated with histological and functional property changes in RF-UVA cross-linked corneas. A riboflavin-defined threshold further determined the sensitivity of SHG imaging as a noninvasive imaging modality to assess the efficacy of RF-UVA CXL.

Corneal collagen cross-linking (CXL) is a photo-polymerizing therapy used to treat corneal ectatic disorders, such as keratoconus, pellucid marginal degeneration and post-LASIK corneal ectasia.^1–6 The treatment method is based on ultraviolet-A (UVA) irradiation of riboflavin (RF)–sensitized corneal stroma to catalyze generation of de novo, covalent, inter- and intrafibrillar collagen bonds by singlet oxygen species.^1,3–5 This photochemical reaction confers physical strength and biodegradative resistance to the ectatic cornea. In spite of accumulating clinical evidence demonstrating the efficacy of CXL in halting the progression of keratoconus,^2,7–11 there is no unique, reliable method capable of predicting the clinical outcome in individual cases to date. 

Multiphoton microscopy—a nonlinear optic imaging technique with superior resolution, low excitation volumes, reduced phototoxicity, and deep tissue penetration by virtue of long-wavelength lasers—recently has been introduced to study ultrastructural changes of corneal collagen organization after RF-UVA CXL in lapine, bovine, and porcine eyes.^12–16 Second-harmonic generation (SHG) is a unique imaging module of multiphoton microscopy that probes hierarchical organization from molecular scale up to tissue architectural levels by detecting anisotropic biological structures such as collagen.¹⁴ Recent studies have shown the feasibility of employing SHG imaging in qualitatively and quantitatively assessing the degree and depth of collagen cross-linking following RF-UVA CXL procedures in bovine and porcine eyes.^16,17 Microscopy with SHG is a noninvasive imaging technique capable of penetrating the entire corneal tissue, thereby potentially providing thorough and insightful information on CXL-induced changes of collagen lamellae at ultrastructural levels. Hence, it represents an attractive methodology to determine CXL efficacy. However, the fidelity and sensitivity of SHG imaging measurements to histological changes in collagen composition and, even more importantly, to the functional property of corneal stroma following RF-UVA CXL in clinical settings still remains undetermined. To this end, we evaluated the applicability of SHG imaging in the assessment of morphological and functional changes after RF-UVA CXL in human corneas. 

Materials and Methods

Tissue Preparation and RF-UVA CXL Protocol

Human corneoscleral tissues from the cornea bank at the Medical University of Vienna (MUW; Vienna, Austria) not suitable for corneal transplantation were used with the approval of the internal review board of the MUW Ethical Committee (MUW 1578/2013) and stored in corneal preservation medium (Carry-C medium; Alchimia, Ponte San Nicolò, Italy). Corneal epithelium was surgically debrided with a surgical knife (Parker 15; BD Biosciences, San Diego, CA, USA) after soaking with 20% (vol/vol) ethanol for 30 seconds. For imaging purposes, debrided corneosclera were fixed with corneal apex facing up onto the membrane of a cell strainer (40 μm pore size; BD Biosciences). Three loose stitches at the peripheral sclera with 10-0 nylon (Alcon Laboratories, Inc., Fort Worth, TX, USA) were placed without changing corneal contour (Supplementary Fig. S1). A drop of 0.1% (wt/vol) RF solution in 20% (wt/vol) dextran vehicle (MedioCROSS D; Medio-Haus Medizinprodukte GmbH, Kiel, Germany) was applied to the corneal apex every minute for 5, 10, and 20 minutes of instillation time (n = 4 corneas/group). In parallel, the drug vehicle (Dextran T500, 20% [wt/vol]; Sigma-Aldrich Corp., St. Louis, MO, USA) was administered to control corneas for 20 minutes. Riboflavin tissue profiles were immediately imaged by confocal microscopy. For corneal CXL, RF-sensitized corneas were exposed to UVA light (370 nm) at a tube-to-corneal apex distance of 15 cm using a UVA-1 irradiation device (Sellamed 3000; Sellas, Ennepetal, Germany). We employed an irradiance of 3 J^.cm⁻², compatible with the conventional clinical CXL protocol. Energy output of a UVA lamp was monitored with a UV meter (Variocontrol; Herbert Waldmann GmbH & Co. KG, Villingen-Schwenningen, Germany). Following UVA irradiation, corneas were submerged in isotonic Dulbecco's PBS (DPBS; Gibco, Grand Island, NY, USA) for 90 minutes to allow cross-linking and reverse dextran vehicle–induced corneal dehydration. After cross-linking, transverse optical sections of corneoscleral buttons were imaged from corneal apex to Descemet's membrane by SHG imaging. To ensure corneal deturgescence, corneas were then incubated in corneal preservation medium (Alchimia) for 22.5 hours before histology and functional assay. A schematic presentation of experimental design is illustrated in Figure 1A. 

Figure 1

 

Experimental design and corneal hydration kinetics. (A) Schematic illustration of experimental design: (1) alcohol-assisted epithelial de-epithelialization and corneal suture fixation; (2) RF instillation; (3) confocal and multiphoton imaging; (4) RF-catalyzed UVA irradiation; (5) crosslinking of irradiated corneas in DPBS for 90 minutes and rehydration; (6) multiphoton imaging of cross-linked corneas; (7) immersion of corneas in medium (Alchimia) for restoration of corneal deturgescence; (8) dissection of corneas into two hemispheres; (9) OCT embedment for cryosectioning; (10) MTS histology; (11) collagenase digestion assay of central cornea; (12) data integration and correlation analysis. (B) Time course of corneal hydration kinetics in distilled water, DPBS, medium (Alchimia), and 20% dextran vehicle (n = 4/group). *P < 0.05 compared with medium (Alchimia) at given time points. (C) Representative graphs of corneal turgescent/deturgescent state and corneal thickness after 24 hours immersion in distilled water, DPBS, medium (Alchimia), and 20% dextran vehicle. Scale bar: 2 mm.

Hydration Control Experiment

To evaluate the hydration dynamics of corneal tissue during experimental procedure, a swelling/deswelling kinetics experiment was performed on naïve corneas. Corneas were submerged in two hypo-, one iso-, and one hypertonic solution: (1) double-distilled water; (2) DPBS; (3) cornea preservation medium (Alchimia); and (4) 20% (wt/vol) dextran drug vehicle (Sigma-Aldrich Corp.). A time course up to 24 hours was determined. To detect swelling/deswelling effects, weight of corneal tissues in all groups at different time points was measured by an analytical balance (sensitivity 210 ± 0.0001 g; BP211D; Sartorius, Göttingen, Germany) and normalized to the initial weight. 

Confocal and Multiphoton Microscopy

A multiphoton microscope (TSC SP5; Leica Microsystems GmbH, Solms, Germany) was used to acquire a z-stack of en face images throughout the corneal tissues with a 15-μm interslice interval at a maximal depth of 490 μm. A ×10 air objective (HCX PL APO ×10/0.40 CS; Leica Microsystems GmbH) was used to image the corneal apex from anterior to posterior stroma. Confocal microscopy was performed to profile RF tissue diffusion. An argon ion laser with a wavelength of 488 nm was used to excite RF with a power of 5%. Fluorescent signals were detected from 500 to 530 nm by a hybrid detection technology (HyD; Leica Microsystems, Inc., Buffalo Grove, IL, USA). For multiphoton microscopic SHG imaging, a Mai Tai Ti:sapphire laser (Spectra Physics, Irvine, CA, USA) providing a wavelength of 800 nm was employed with a fully opened pinhole (600 μm). Backward-scattered SHG signals were collected by a photomultiplier tube (PMT) set to detect a range of 400 ± 20 nm. 

Morphological Assessment by Masson's Trichrome Staining

Following CXL treatment and SHG imaging, corneas were restored in medium (Alchimia) for 22.5 hours and dissected into two hemispheres. One hemisphere without fixative was snap-frozen and embedded in OCT compound (Tissue-Tek containing medium; Sakura, Alphen aan den Rijn, Netherlands). Tissues were then cryosectioned at the sagittal plane of the corneoscleral buttons with 4-μm thickness by using a cryotome (CM3050 S; Leica Microsystems, Inc.). Masson's trichrome staining was performed using a stain kit (Artisan; Dako Denmark A/S, Glostrup, Denmark) according to the manufacturer's instructions. Bright images of Masson's trichrome-stained tissues were acquired by an inverted microscope (Axiovert 40; Leica Microsystems, Inc.) at ×20 magnification. For statistical analysis, at least eight images were acquired from each sample. 

Collagenase Digestion Assay

Central corneal tissue from the other corneoscleral hemisphere was punched by a dispensable 6-mm trephine on RF–UVA CXL and control corneas for collagenase digestion assay. Collagenase (10 U/mL; collagenase A from Clostridium histolyticum, Roche, Basel, Switzerland) in DPBS at 37°C on a shaker with 70 rpm was used for up to 160 minutes as described previously.^18,19 Corneal weight was measured before and after enzymatic digestion by an analytical balance. Final weight was normalized to the initial weight to obtain relative collagen digestion (n = 6 corneas/group). 

Image Processing and Data Analysis

Multiphoton microscopy z-stack images were analyzed using image processing software (Just ImageJ; Fiji; Laboratory for Optical and Computational Instrumentation, Madison, WI, USA).²⁰ To quantitatively characterize corneal stromal texture, a plug-in (SurfCharJ; Fiji) was employed to process SHG images by which SHG signals were transformed to roughness indices, indicated by the root mean square deviation (i.e., Rq) values.^21,22 To determine the collagen compaction effect caused by CXL, micrographs of Masson's trichrome staining were analyzed by a graphics editing program (Adobe Photoshop CS5 v12.1; Adobe Systems, San Francisco, CA, USA) as follows: Blue Masson's trichrome histology images were transformed to grayscale mode followed by subtraction of the tissue-free background. Then, the threshold for binary transformation was set at 128, at a scale of 0 through 255 for all images. The relative collagen density (%) was determined by an area of collagen bundle divided by total corneal tissue area. 

Statistics

Statistical analysis was performed by employing one-way ANOVA and Tukey's multiple comparisons in a depth-wise manner (GraphPad Prism 6.05; GraphPad Software, Inc., La Jolla, CA, USA). Data are shown as mean ± SEM with P < 0.05 considered statistically significant. Linear regression analysis was used to determine the relation between RF gradient and axial roughness profile. Correlation coefficient (R²) and P values were determined by Pearson's correlation test with graphing and statistical software (GraphPad Software, Inc.). 

Results

Conventional confocal microscopy was performed to obtain the issue gradient profiles of different RF instillation time points. Multiphoton SHG images were acquired before and after RF-UVA CXL to determine tissue architectural changes. We further determined CXL-induced alterations by histological and enzymatic analyses in order to evaluate the fidelity of SHG imaging of assessing corneal structural changes caused by CXL treatment. 

Hydration Kinetics of Control Corneas

In order to evaluate hydrodynamics of corneas during experiments, we conducted a hydration kinetic experiment on naïve corneas. Corneal swelling indexed by relative hydration weight was observed in double distilled water and DPBS groups while corneas in cornea preserving medium (Alchimia) stayed at the baseline level (Fig. 1B). Intriguingly, dextran vehicle–immersed corneas initially showed deswelling up to 1.5 hours; thereafter, a gradual swelling was observed. By 24 hours, corneal swelling in distilled water, DPBS, and dextran showed an average of 58.9%, 39.2%, and 43.1% increase in tissue weight, respectively (all P < 0.05 versus cornea preserving medium [Alchimia]; Supplementary Table S1). The appearance of corneas in distilled water turned severely cloudy and wrinkled (turgescent). Corneas in DPBS and dextran were mildly cloudy. By comparison, corneas in medium (Alchimia) retained good transparency (deturgescence) with the least thickness among all groups (Fig. 1C). These results excluded the possibility of hydration-related artifacts in the following imaging and histological measurements. 

Depth-Resolved Imaging of Intrastromal RF Concentrations Revealed Differential Tissue Distribution Profiles in an RF Dose–Dependent Manner

Fluorescent signals of 5-, 10-, and 20-minute RF instillation were obtained throughout the corneal stroma. The mean fluorescence intensity (MFI) of background signals from dextran controls was subtracted from each optical slice of RF images and plotted against corneal depth. A characteristic RF diffusion pattern as shown in Figure 2A was initially increasing in the anterior stroma, peaking at ∼200 μm in depth, followed by gradual attenuation in the posterior stroma toward the Descemet's membrane. Quantitatively, the 20-minute RF group revealed higher RF fluorescence intensities from 60 to 330, and 225 to 510 μm of stromal depth, compared with the 5- and 10-minute RF groups, respectively (P < 0.05; Fig. 2B, Supplementary Table S2). The axial fluorescence profiles of RF signals were reconstructed from z-stack images (Fig. 2C). To determine total RF tissue dosages, cumulative MFI values were averaged within every group. While a 20-minute RF instillation revealed a higher RF tissue dose than 5 and 10 minutes of drug administration, no difference was observed between the 5- and 10-minute RF groups (Fig. 2D). Taken together, these data indicated that depth-dependent distribution of RF was related to the topical application time. 

Figure 2

 

Riboflavin diffusion profile in human corneas determined by depth-resolved multiphoton imaging. (A) Representative en face images of 20-minute RF instillations at various stromal depths from 100 to 500 μm. Scale bar: 200 μm. (B) Tissue diffusion profiles of 5-, 10-, and 20-minute RF instillations were plotted as the MFI in respect to corneal depths. Fluorescent intensities of RF were dextran control–derived autofluorescence background corrected. Boxed area under * indicates the zone of corneal stroma with significantly higher MFI in the 20-minute RF group than in the 5-minute RF group. Boxed area under † denotes the stromal segment where the MFI of the 20-minute RF group was significantly higher compared with the 10-minute RF group (P < 0.05). Statistical details are outlined in Supplementary Table S1. Data were shown as mean ± SEM of four corneas. (C) Spatial profiles of representative RF penetration gradients from corneal apex to the endothelial side. (D) Corneal stroma RF doses shown as cumulative MFI were plotted as a function of RF instillation time. *Groups with higher MFI compared with dextran control. Δ Denotes higher MFI compared with 5-minute RF group. †Higher MFI than the 10-minute RF group. Data were shown as mean ± SEM of four corneas.

SHG Microscopy Revealed Altered Collagen Organization Post-RF–UVA CXL

En face SHG imaging revealed a distinct lamellar arrangement of discernible collagen bundles throughout the entire stroma of native corneas. In contrast, RF-UVA cross-linked corneas displayed a smoothened, homogenous lattice of short, fine, and reticular collagen fibrils. The anterior and middle stroma of 5-, and 10-minute RF–UVA cross-linked corneas exhibited a reduced undulation of collagen lamellae compared with controls. In contrast, the counterpart of 20-minute RF-UVA CXL showed increased homogeneity of tissue patterns without distinguishable collagen fibers. In the posterior stroma, a uniform, smooth tissue texture with low SHG signal intensities was observed in all RF groups (Fig. 3A). 

Figure 3

 

Second harmonic generation–derived roughness profiles of UVA-crosslinked corneas at different RF doses. (A) Representative en face SHG images of native corneas (controls) and UVA-crosslinked corneas with 5-, 10-, and 20-minute topical RF applications at anterior (120 μm), middle (240 μm), and posterior (360 μm) stroma. Closed arrowheads reveal coarse and long collagen bundles in an organized distribution. Open arrowheads show short and intercalated collagen fibrils. Open arrows indicate homogenous tissue texture without discernible collagen fibers. *Uniform tissue texture with low SHG signal intensities. Scale bar: 200 μm. (B) Second harmonic generation signals at different depths were shown as roughness index (i.e., Rq). Boxed area marks the stromal zone with lower roughness of RF-UVA crosslinked corneas than that of native control corneas. Data were shown as mean of Rq ± SEM. Detailed statistics are outlined in Supplementary Table S3. (C) Bar graph shows averaged Rq values of the whole corneal stroma. Data were shown as mean ± SEM. *P < 0.05 versus control.

In order to quantitatively assess the alterations of tissue texture caused by RF-UVA–assisted CXL, the SHG signal was transformed to a root mean square deviation (Rq) value indicated as roughness index. All groups of RF-UVA–cross-linked corneas displayed decreased Rq values compared with those of native control corneas. A significant decline of roughness was observed in the middle stromal segment (approximately from 165 to 270 μm) of all RF-UVA CXL groups compared with controls (boxed area in Fig. 3B, Supplementary Table S3). In addition, the averaged Rq values per cornea showed a consistent decrease in 5-, 10-, and 20-minute RF-UVA–cross-linked corneas versus controls (Fig. 3C). Intriguingly, no difference in tissue roughness was observed between the three RF-UVA treated groups. Collectively, these data suggested SHG imaging as a sensitive tool to detect RF-UVA CXL–induced structural changes. 

Density of Corneal Collagen Lamellae Was Increased Following RF-UVA CXL in an RF Concentration Gradient-Dependent Manner

After multiphoton microscopy studies, we sought to evaluate the differences between imaging- and histology-based approaches to determine CXL-induced structural changes. In general, Masson's trichrome staining revealed CXL-induced reductions of interfibrillar space in all RF-UVA CXL groups compared with controls along with depth-related morphological differences within treatment groups (Fig. 4). The spatial pattern of collagen compaction was dependent on RF dosages. The anterior stromal segment displayed intense lamellar compaction in all groups of RF-UVA CXL, while the middle stroma exhibited such morphological alterations only following 10- and 20-minute RF-UVA CXL procedures. A condensed collagen organization in the posterior segment was exclusively observed in the 20-minute RF-UVA CXL group (Fig. 4A). For quantitative assessment of differences in collagen density between groups, micrographs of Masson's trichrome staining were transformed to binary photos. Compared with large and heterogeneous interfibrillar areas in controls, RF-UVA cross-linked corneas exhibited narrow and uniform interstitial space (Fig. 4B). Relative collagen densities significantly increased from 67.91% in controls to 86.43% and 98.01% in the 10- and 20-minute RF-UVA CXL groups (P < 0.05 and P < 0.01, respectively). Moreover, a significant increase of corneal density from 78.62% in 5- to 98.01% in 20-minute RF-UVA CXL was observed (P < 0.05; Fig. 4C). Morphological study conducted by Masson's trichrome staining demonstrated compelling collagen compaction at high RF dosage–induced CXL (i.e., 10 and 20 minutes of RF instillation). 

Figure 4

 

Histological assessment of intrastromal structure by Masson's trichrome staining. (A) Representative cross-sectioned images of the anterior, middle, and posterior stroma from native and RF-UVA crosslinked corneas. Interfibrillar and interlamellar space indicated by closed arrowheads was reduced in an RF-tissue dose-dependent manner. Dotted line indicates Descement's membrane. Scale bar: 20 μm. (B) Micrographs of Masson's trichrome stained corneas were transformed to binary images for qualitative demonstration of collagen organization and for quantitative assessment of collagen density. Scale bar: 20 μm. (C) Relative collagen density of four corneas per group was shown in a box whiskers plot. *P < 0.05. **P < 0.01.

Increase in Biochemical Stability of Cross-Linked Corneas Was Related to RF Dosage

Next to the histomorphological study, we questioned whether the functional properties of corneas were altered by CXL and whether these functional changes were correlated to structural alterations assessed by SHG imaging. After 120 minutes of collagenase digestion, an average of 50.5% ± 14.6% and 80.0% ± 20.2% of cross-linked stroma remained undigested in the 10- and 20-minute RF-UVA CXL groups, respectively (n = 6, P < 0.05, Figs. 5A, 5B). In contrast, the corneal stromas of the control and 5-minute RF-UVA CXL groups were completely digested after approximately 100 minutes (data not shown). By 160 minutes of digestion, residual stroma was only observed in the 20-minute RF-UVA CXL group (Fig. 5C). These data suggested that the higher RF dosage in the UVA CXL conferred enzymatic resistibility to corneal stroma. 

Figure 5

 

Collagenase digestion assay of RF-UVA crosslinked corneas. (A) Collagenase A (10 U/mL) was used to determine the enzymatic resistance of UVA-crosslinked corneas with different RF doses. The weight of undigested corneas after 120 minutes of collagenase digestion was shown as relative mean ± SEM from independent experiments. *P < 0.05. **P < 0.01. (B) Representative micrographs of remaining corneas after 120 minutes of collagenase digestion. (C) By 160 minutes of collagenase digestion, undigested corneas were only found in the 20-minute RF-UVA group. Scale bar: 200 μm.

Tissue Roughness Profile Predicted by SHG Signals Was Correlated With RF Distribution When RF Exceeded a Dose-Determined Threshold

To probe the relation between RF gradients and SHG axial profiles, we conducted regression analysis for 5-, 10-, and 20-minute RF-UVA CXL groups. Interestingly, at a low RF tissue dose (5-minute RF group), no correlation was observed. A low degree but significant correlation was noticed in the 10-minute RF group, (R² = 0.126, P < 0.05), while the 20-minute RF group displayed a higher degree of correlation (R² = 0.282, P < 0.01; Fig. 6). The global maxima of the RF diffusion profile–derived MFI values were in proportion to topical RF application times. Namely, the maximal MFI values were 30.3, 38.3, and 53.8 arbitrary units (AU) in the 5-, 10-, and 20-minute RF installation groups, respectively. Moreover, the corneal depth at which RF diffusion peaked was correlated to the RF instillation time (Fig. 6). Taken together, these data suggested that the correlation between axial profiles of RF diffusion and roughness in CXL was determined by an RF dose–defined threshold and strongly relied on intrastromal RF diffusion pattern. 

Figure 6

 

Depth-resolved regression analysis of RF intensity and collagen roughness. We plotted the MFI of RF against the Rq generated by SHG in a depth-wise manner. Correlation was only observed in high doses of RF (10-, and 20-minute groups) but not in low dose of RF (5-minute group). Arrows indicate the middle plane of corneal stroma.

Discussion

In the present study, we investigated the feasibility of employing multiphoton microscopy as a noncontact in situ imaging tool for determination of RF diffusion profile and evaluation of CXL efficacy. We further provided evidence of morphofunctional changes of cross-linked corneas as key references to ascertain the sensitivity and fidelity of SHG imaging for detecting CXL-induced alterations. Since corneal collagen organization exhibits significant interspecies differences,¹⁷ we used human corneas in order to obtain the most clinically relevant data. Despite the fact that CXL was proposed as a promising procedure for treating ectatic corneal disorders in the early and intermediate disease stages more than a decade ago,^1,2 a clinical tool providing reliable assessment of early CXL therapeutic efficacy and capable of predicting long-term treatment outcomes is unavailable so far. Current clinical parameters to determine CXL efficacy mainly rely on measuring corneal optical properties such as refractive power by autorefractor and wavefront aberrometry, uncorrected and best-corrected visual acuity, as well as corneal physical properties such as topography by photokeratoscopy and tomography by anterior segment optical coherence tomography (AS-OCT).^17,23,24 Nevertheless, none of these conventional methods is sensitive enough to reveal subtle changes of lamellar collagen organization caused by CXL at an ultrastructural level, neither immediately nor during long-term follow-ups after RF-UVA–based phototherapy. A new technique combining corneal pachymetry and topography, known as Scheimpflug imaging, was recently introduced to evaluate the clinical efficacy of CXL.²⁵ Although this imaging technique offers information on both refractive and physical changes in cross-linked corneas, depth-dependent lamellar structural changes of the cornea are not resolved and the functional significance of such measurements remains to be validated by long-term clinical outcomes. Although conventional OCT has been extended to study anterior corneal structures (i.e., AS-OCT), this method is not sensitive enough to reveal collagen architectural alterations following CXL. In contrast to these currently applied clinical measurements, the noninvasive in situ multiphoton microscopy, offering insightful structural information at a supramolecular level, has recently been used to study CXL effects in several animal models.^12–16,26 Second harmonic generation, a key imaging module of multiphoton microscopy, has been shown to depict detailed fibrillar collagen ultrastructure of corneas.¹⁴ In agreement with previously reported data from Gupta et al.¹⁶ in porcine eyes, we observed an increased homogeneity in stromal collagen texture and a reduced roughness indexed by declined Rq values of cross-linked human corneal stroma (Fig. 3). Decreased Rq values of SHG images reflect the en face morphological changes from densely packed, parallel collagen bundles to interwoven, homogenous, and short bands of collagen fibrils, indicating an increased number of collagen cross-links in RF-UVA–treated stroma. Interestingly, the roughness profiles of cross-linked corneas showed the maximal reduction of Rq values mainly in the middle segment of corneal stroma in all RF doses studied (Fig. 3, Supplementary Table S3). Our findings are in line with quantitative studies employing different imaging modalities. Using two-photon collagen autofluorescence microscopy, Chai et al.¹² stated that CXL-induced alterations of collagen autofluorescence intensities were confined to the anterior 220 to 280 μm of lapine corneal stroma, independent of administered RF doses. By Brillouin optical microscopy, Scarcelli et al.¹³ detected an anterior segment–defined Brillouin frequency shift in cross-linked bovine corneas. Compared with the aforementioned imaging techniques, SHG microscopy has the advantage of superior resolution by delineating the lamellar collagen ultrastructure as a noninvasive therapeutic parameter. Despite the compelling difference in average tissue roughness between native and RF-UVA cross-linked corneas, we did not detect any differences within treatment groups (Fig. 3C). Taken together, these data indicate a high sensitivity of SHG multiphoton microscopy to CXL-induced alterations in human corneas, while the specificity of SHG in determining the magnitude and dimension of structural modifications caused by different doses of RF remains suboptimal. A plausible explanation for this observation is that the laser intensity of multiphoton microscope is intrinsically attenuated in deeper stroma,²⁷ thereby impeding the power of SHG microscopy to discriminate subtle CXL changes within treatment groups. 

Concomitant to multiphoton microscopy, we evaluated CXL efficacy by histological examination and collagenase digestion. Masson's trichrome staining revealed a CXL-induced collagen lamellae compaction uniformly throughout the entire stroma at higher RF dosage (20-minute RF), in contrast to the anterior stroma–confined structural changes at lower RF dosage (Fig. 4A). Furthermore, the average collagen density of cross-linked corneas was significantly increased by higher RF doses compared with controls, suggesting a dosage effect of RF in CXL (Fig. 4C). Additionally, collagenase digestion assay was employed to assess differential functional properties of corneas cross-linked by various RF doses. The tissue biostability of cross-linked corneas was positively related to RF dosage within the treatment groups (Fig. 5). Although morphofunctional analyses can reflect the RF dose-dependent CXL efficacy more accurately than SHG microscopy, their applicability is limited due to their invasiveness. 

The efficacy of photo-oxidative CXL is known to rely on several key factors (i.e., photosensitizer distribution, UVA penetration, tissue oxygen gradient, and collagen organization intrinsic to corneal ectatic pathology).^5,28 Among these causative factors, lately there has been increasing research interest in investigating the correlation of RF tissue dose or distribution with CXL therapeutic effects.^12,15,29,30 To our knowledge, the current study is the first report correlating RF diffusion profile to imaging, histological and functional data for evaluating CXL efficacy in human corneas. Riboflavin doses were inversely correlated to roughness (Rq values) in a stromal depth-dependent manner (Figs. 2B, 3B). The peak of RF concentrations (10 and 20 minutes, Fig. 2) overlapped with the maximal reduction of Rq values from approximately 165 to 270 μm (Fig. 3, Supplementary Table S3). Moreover, RF doses reached higher concentrations in the middle and posterior segments by prolonged RF instillation (Supplementary Table S2). Such a deeper RF diffusion pattern might explain the stark collagen fibrillar compaction observed in the middle and posterior stroma of 10- and 20-minute RF-cross-linked corneas (Fig. 4) and increased biochemical stability of 10- and 20-minute RF groups against enzymatic digestion (Fig. 5). Furthermore, we found a low but significant correlation between the axial profiles of RF and SHG signal–derived Rq values of cross-linked corneas only at high stromal RF doses, namely 10 and 20 minutes, but not in the 5-minute RF group (Fig. 6). This line of data suggested an RF dosage-defined threshold determining the fidelity of SHG imaging in assessing CXL efficacy that might be attributed to the amount of photosensitizer available in the UVA-irradiated tissues. A recent work demonstrated that effective intrastromal RF concentrations (>0.06% [wt/vol]) reached a stromal depth of 450, 300, and 200 μm by 20, 10 and 5 minutes of 0.1% RF instillation, respectively.¹⁵ The reported RF spatial profile might explain our observation that RF distribution and SHG signals correlated only when the corneal stroma attained a therapeutic dose of RF required for effective photo-oxidative CXL. Despite being significant, the “low” correlation might be due to the multiple cofactors other than photosensitizer concentration concomitantly influencing the CXL process. 

Although our data provided a link between RF tissue distributions and dosage, SHG imaging, histology and bioenzymatic properties of RF–UVA-treated corneas, there are some limitations in the current study. First, the correlation between RF concentrations and SHG-derived roughness index was only determined in normal corneas. Whether or not this correlation can be extrapolated to the disease-perturbed corneas (e.g., keratoconus)^31,32 requires further studies. Secondly, the clinical relevance of the imaging-architecture association uncovered in the present work to the corneal topography and patient's visual function after CXL procedure remains to be determined. Lastly, for the incompatibility of human eye anatomy and forward-scattered SHG imaging physics, we could only detect the backward-scattered SHG emission, which is known to exhibit suboptimal sensitivity in detecting corneal architecture.^14,33 A breakthrough in in vivo forward-scattered SHG imaging would expand our understanding of corneal CXL. 

In summary, we demonstrated the utility of multiphoton microscopy in measuring RF tissue gradient and assessing CXL efficacy in human corneas in situ. We further validated the imaging measurements with histological analysis and enzymatic functional assay. Collectively, our data suggested SHG microscopy as an attractive and noninvasive imaging diagnostic candidate for evaluating and predicting CXL efficacy with clinical therapeutic dosage of photosensitizer. 

Acknowledgments

We thank the Cornea Bank at the Medical University of Vienna (MUW) for assistance with procurement of donor corneal tissues; the Department of Rheumatology, MUW, for providing multiphoton microscopy equipment; Yi-Hsun Huang from the Department of Ophthalmology, National Cheng Kung University Hospital, for his technical support in multiphoton imaging; and Agnes Boltz from the Department of Ophthalmology and Optometry, MUW, for her review of this manuscript. 

Supported by funding for Translational Ophthalmic Research, University Clinic of Ophthalmology and Optometry, Medical University of Vienna, Austria. The authors alone are responsible for the content and writing of the paper. 

Disclosure: M. Laggner, None; A. Pollreisz, None; G. Schmidinger, None; R.A. Byrne, None; C. Scheinecker, None; U. Schmidt-Erfurth, None; Y.-T. Chen, None