We describe a dual, second harmonic generation (SHG) and third harmonic generation (THG) microscope, with the aim to obtain large-scale images of the cornea that can simultaneously resolve the micron-thick thin layers. We use an Ytterbium femtosecond laser as the laser source, the longer wavelength of which reduces scattering and allows simultaneous SHG and THG imaging. We measure one-dimensional SHG and THG profiles across the entire thickness of pig cornea, detected in both the forward and backward directions. These profiles allow us to clearly distinguish all the porcine corneal layers (epithelium, stroma, Descemet’s membrane and endothelium). From these profiles, longitudinal cross sectional images of the corneal layers are generated, providing large scale topographic information with high-spatial resolution. The ability to obtain both SHG and THG signals in epi-detection on fresh eyes gives promising hopes for in vivo applications.
©2008 Optical Society of America
Imaging of the whole thickness of the cornea at a macroscopic scale provides key information about the characteristics of this tissue (thickness, topography and curvature) , such information being used by the ophthalmologist to refine the diagnosis of corneal pathologies and initiate the treatment. This macro-scale issue also involves the precise determination of each corneal layer, including the two thin posterior layers of the cornea, the endothelium and Descemet’s membrane. The endothelium regulates the level of corneal hydration essential for corneal transparency and corneal endothelial malfunction may be accompanied by changes in Descemet’s membrane ultrastructure, as seen for instance in Fuchs’ dystrophy .
In conventional (linear) microscopy, two techniques are commonly used for noninvasive in vivo imaging of the cornea: confocal microscopy and optical coherent tomography (OCT). While confocal microscopy can achieve cellular imaging in the cornea , it cannot detect the main component of the stroma, i.e. type I collagen. In addition, this technique uses a visible light source, which causes more scattering through the media. OCT is based on the detection of backscattering photons from the ocular structures. The actual resolution of the commercially available OCT systems is typically limited to 10-15 µm, which is insufficient to resolve the thin posterior layers of the cornea.
Second harmonic generation imaging (SHGI), Third harmonic generation imaging (THGI) and two-photon fluorescence imaging (TPFI) are three nonlinear optical phenomena of interest for imaging biological tissues. First demonstrated in 1971 , SHGI is now well known for second-order nonlinear optical imaging. It is well suited for corneal imaging because of the significant amount of type I collagen fibers present in the stroma. SHGI has also allowed to identify Bowman’s layer in the human cornea, owing to the organization of the collagen fibers in this area .
THGI, a third-order nonlinear optical imaging method, was first used in microscopy in 1997 by Barad et al. . THGI is newer than SHGI to the biomedical community. It was shown that under tight focusing conditions, an interface separating two media with a different refractive index, or third-order nonlinear susceptibility, generates third harmonic signals . Quickly following these first results obtained with optical fibers, the method was applied to biological samples providing high resolution cellular imaging .
One of the strengths of nonlinear microscopy is multimodality , based on the complementarity of the signals generated by the different sources. For imaging of the eye, the method currently used to complement SHGI is TPFI [9,10]. However, TPFI is limited by the fact that fluorescence is linked with the intrinsic fluorophores and the signal is limited to specific molecules. Combination of SHGI and TPFI nonlinear microscopy does not allow clear differentiation of the posterior layers of the cornea, Descemet’s membrane and the endothelium. While the stroma can easily be identified by SHG, the collagen present in Descemet’s membrane does not generate a second harmonic signal and posterior to the stroma, the emitted TPF signal is homogeneous.
THGI is less restrictive than fluorescence. Emission occurs at the optical interfaces and it covers a wider field than TPFI . Furthermore, contrary to fluorescence, THGI is not limited spectrally since it is not a resonant process. Longer excitation wavelengths can be used, which also reduces scattering in the tissue.
In this work, we describe a multimodal microscope that combines SHGI and THGI. Multimodality and high spatial resolution have allowed us to characterize even the deepest layers of the cornea. This paper presents experimentally measured one-dimensional (1D) and two-dimensional (2D) SHG and THG profiles across the entire thickness of the pig cornea, detected in both forward and backward directions. To our knowledge, this is the first example of backward detection THGI for such a transparent tissue. These profiles allow to clearly distinguish the four layers (epithelium, stroma, Descemet’s membrane and endothelium) of the porcine cornea (there is no Bowman membrane in the pig cornea ). From these profiles, longitudinal cross sectional images of the corneal layers are generated, providing large scale topographic information with high-spatial resolution of 2 µm in the axial direction.
2. Material and method
2.1. Experimental set-up
Our experimental setup has been described previously  and is illustrated in Fig. 1. The custom-made upright microscope was built with opto-mechanical parts (AFOptical Inc., Fremont, CA). The laser source is a femtosecond Yb:KGW oscillator (model t-Pulse, Amplitude Systemes, Pessac, France) that delivers 200 fs pulses with 1.03 µm wavelength, at a repetition rate of 50 MHz and an average power of 1.2 W. The 3D scanning of the sample is performed by motorized stages in the XY direction (perpendicular to the laser beam) and a micrometer along the Z-axis (Micos GmbH, Eschbach, Germany). For focusing the excitation beam onto the tissue and collecting backward harmonic signals, two types of objectives are used, depending on the experiment: a 63x dry objective (Zeiss Plan NeoFluar 63x/numerical aperture 0.75/Working distance 2mm, Carl Zeiss MicroImaging, Jena, Germany) and a 60x water immersion objective (LUMplanFL 60x/numerical aperture 0.9/Working distance 2mm, Olympus, Tokyo, Japan). After filtering out the excitation laser beam, harmonic signals are recorded in the backward and forward directions using two photomultiplier tubes (PMT H9305-04, Hamamatsu, Hamamatsu City, Japan). For these ex vivo experiments in which there is a constraint on the working distance (because of the use of half eyes), the forward detection is done with the photomultiplier tube put directly under the sample. Its large window allows collecting the harmonic signals generated by the sample. To complement the analysis of the posterior corneal layers, the Descemet’s membrane and endothelium, additional experimentation was performed on isolated corneas using the Zeiss Plan Neofluar 63x objective (Carl Zeiss MicroImaging) to collect in the forward direction the harmonic signals which are split by a dichroic mirror before reaching the PMT. Photocurrents are amplified and collected by a data acquisition board (PCI-6122S, National Instruments, Austin, TX). The computer synchronizes both the scanning process and the data acquisition. Theorical calculations  of the axial resolution of the focused excitation beam onto the porcine cornea give values around 5 µm for the Zeiss Plan Neofluar dry objective (Carl Zeiss MicroImaging) and around 3.5 µm for the LUMplanFL water immersion objective (Olympus). Acquisition along longitudinal sections was achieved with increments ranging from 400 nm to 1 µm, allowing optimal scanning resolution along the longitudinal axis.
2.2 Tissue preparation
Fresh pig eyes were obtained from a slaughterhouse within 6 hours from death and kept in a humid chamber at 4°C until the experiment. Except for some of the experiments on endothelium and Descemet’s membrane imaging for which isolated corneas were used, all experiments were done on whole half eyes with an intact anterior segment (cornea, aqueous humour, iris, lens), in order to preserve the natural curvature and anatomical structure of the cornea (Fig. 1(b)). The half eye is placed directly on a microscope glass slide that transmits the UV third harmonic, and maintained with medical gauzes. For the immersion objective, an ophthalmic gel (Tear-gel, Novartis, Basel, Switzerland) is used. The refractive index (n=1.339 at 20°C and 589 nm wavelength) of this gel is closer to the average index of the cornea (1.377) than is air. It keeps the cornea moist and maintains an adequate meniscus between the focusing objective and the curved cornea.
3. Results and discussion
3.1. SHG/THG full thickness corneal 1D profile
We first performed 1D acquisitions of the SHG and THG signals along the full thickness of the cornea, with detection in the forward and backward direction. All of the data shown in this section were obtained at the center of the cornea. Central corneal thickness was also measured using an ultrasound pachymeter (SP-3000, CBD Tomey, Phoenix, AZ), since ultrasound pachymetry is still considered to be the gold standard for corneal thickness measurements. The Zeiss Plan Neofluar dry objective was used for this experiment. SHG was measured in the backward direction and THG in the forward direction. The profile shown in Fig. 2 corresponds to the first 300 µm of the porcine cornea, with 0.5 µm increments in the z direction. As we can see, the forward THG (F-THG) reveals a strong signal at the air-epithelium interface, but no signal in the deeper positions (which is the reason why the full thickness is not displayed). On the other hand, the majority of the backward SHG (B-SHG) signal begins at the entrance of the stroma, and continues to be detected deeper in the stroma, before it starts to decrease progressively.
We then repeated the same experiment through the full thickness of the pig cornea (B-SHG/F-THG profile) using the LUMplanFL 60×/NA 0.9 water immersion objective and the ophthalmic gel. Here, 1 µm increments in the z-direction were used. Figure 3(a) shows that the B-SHG signal is detected over the entire thickness of the stroma, showing both the entrance and the end of the stromal layer. Furthermore, the F-THG signal reveals all the important interfaces across the entire corneal thickness. With the immersion objective, the strongest THG signal is obtained at the surface of the cornea, as is the case for the dry objective. However, the THG intensity at the outmost surface of the epithelium measured by the immersion objective is much weaker than that measured by the dry objective. In the latter case, the large difference in refractive index between air and epithelium generates a strong signal, while scattering effects in the cornea  result in a rapid decrease of the signal as we increase the depth of focus in the cornea. With the immersion objective, this index difference is less, and so is the signal at the surface. However, one can detect THG signal within the cornea, although there is still a decreasing trend due to scattering. Therefore, by using the immersion objective with the ophthalmic gel, the combination of the SHG and THG signals makes it possible to identify the different layers of the pig cornea.
On the trace shown in Fig. 3(a), the distance between the first peak (gel/cornea interface) and the last peak (endothelium/aqueous interface) is 953 ± 4 µm. For the same sample, the ultrasound pachymetry value was 922 ± 5 µm. The difference between the two measurements can be due in part to the positioning of the hand held pachymetry probe or to an increase in corneal thickness during the experiment due to the loss of intra ocular pressure.
As shown in Fig. 3(a), the main layers of the cornea can be easily identified. The SHG signal is emitted only in the stroma, because of the non-centrosymetric characteristic of this tissue composed of collagen fibers. On the other hand, THG signal is not linked to the composition of the material, and is theoretically emitted at each optical interface. Six sources of emission can be identified for the THG signal: (1) the gel-epithelium interface; (2) the epithelium-stroma interface; (3) the multiple interfaces within the stroma; (4) the stroma-Descemet’s membrane interface; (5) the Descemet’s membrane-endothelium interface; and finally (6) the endothelium-aqueous interface. Figure 3(b) shows a magnified trace of the region between (4) and (6). We notice a region in which the SHG is decreasing (starting at z=1090 µm), followed by a last THG peak (at z=1110 µm). After this THG peak, no more signal is observed, which is in agreement with the knowledge that the aqueous humour is a homogeneous medium (e.g. without interfaces). SHG is not detected outside stroma. The region before the last THG signal and after the last SHG peak signal corresponds to the posterior layers of the cornea, Descemet’s membrane and the endothelium. The 2D profiles presented in chapter 3.4 provide a more evident demonstration of this identification since the typical shape of these structures can be recognized.
We also show THG images acquired in the epi-detection across the whole thickness of the cornea. In Fig. 4(a), SHG and THG profiles are recorded across the cornea, but this time with THG detected in the backward direction (B-THG) and SHG detected in forward direction (F-SHG). We can see that enough B-THG signals are collected to reconstruct a profile similar to that previously observed with F-THG. The B-THG signal is typically 40 times weaker than the F-THG signal. The F-THG signal reflected from the microscope slide and contributing to the B-THG signal is estimated to be negligible. This is because the Fresnel reflection of the F-THG beam from the microscope glass slide will be about 3% (for an angle of incidence AOI=0°), and only 1x10-4 of this reflected beam reaches the PMT (due to the reflected beam being strongly diverging). If we take into account the transmissivity of the various optics (50%), the back reflected F-THG signal is 1x10-6 times that of the F-THG. The source of the B-THG signal in our experiment needs to be clarified, which is out of the scope of the present paper. However, our calculations estimate that our experimentally observed B-THG signal is much stronger than the back-reflected F-THG signal. As observed for the case of F-THG/B-SHG in Fig. 3(b), we can distinguish the same layers. Clearly, the capability of performing backward detection of the endothelial layer would be of great interest for clinical applications in corneal pathologies, such as in Fuchs’ dystrophy, a pathology for which the surgical treatment (posterior lamellar transplantation) consists in the precise removal of this part of the cornea.
3.2. 2D Topography of the full thickness corneal profiles
In this section we investigate the use of combined SHG/THG microscopy to obtain topographic images of the cornea on a large field of view, like those obtained by OCT, but with a better delineation of the different corneal layers. The following images were acquired with the 60×/NA 0.9 LUMplanFL water immersion objective and gel between the objective and the cornea.
As demonstrated in the previous section, SHG and THG profiles along the full corneal thickness reveal the various layers of the cornea. Therefore, by scanning along the x or y axes, it is possible to reconstruct a topographic image of the epithelial, stromal and endothelial corneal layers. Figure 5 shows a 2D B-SHG/F-THG image in the (x, z) plane of the cornea. The focused excitation beam scans the sample along the z-axis in its whole thickness with increments of 1 µm. The corneal sample is moved along the x-axis over a distance of 1740 µm from the corneal apex with increments of 60 µm.
This large x-axis increment was chosen to reduce the image acquisition time to a reasonable limit (the present picture was acquired in about 15 minutes). Naturally, images with higher resolution in the x-direction could be obtained, with the consequence of either a longer acquisition time or a smaller field of view.
For comparison purposes, Fig. 5(b) shows a histology section of a pig cornea. In both figures (Fig.5 (a and b)), the epithelium, stroma, Descemet’s membrane and endothelium can be clearly localized. Bowman membrane is typically absent in the pig eye .
3.3. 2D profiles of the corneal epithelium
The corneal epithelium is composed of around 5 stratified cellular layers in which superficial, wing and basal cells are lined up from surface to bottom. Figure 6(a) shows a histological cross-section of an anterior cornea stained with Masson’s trichrome in which the three cell types are clearly identified. In Fig. 6(c), which gives a magnified view of the epithelial layer shown in Fig. 6(b), B-THG/F-SHG imaging allows distinction of the three types of epithelial cells. Some of the individual wing and basal cells can even be distinguished.
We believe that this high cellular contrast is due to the shape of the cells themselves, and more specifically to the architectural distribution of their cell membranes. The superficial cells are flatter, with high contrast inhomogeneities due to the dense stacking of their cellular membranes. The intermediate wing cells are rounder and the deep basal cells clearly larger and elongated, with their longer axis oriented perpendicular to the stromal surface. This structure could explain why the THG signal is stronger in the superficial layers of the epithelium. The increase scattering deeper in the tissue may also explain the decreased THG signal in the deeper layers of the epithelium.
It can be noticed that the thickness of the epithelium is not constant along the transverse direction. Similar to histology, SHG and THG microscopy reveal local variations in the order of 25 µm (Fig. 6(a and b)). Thickness was also found to change as a function of animal age.
3.4. 2D profiles of the corneal endothelium
Determination of the deepest layers of the cornea is an important issue. As shown by histology (Fig. 7(a)), the most remarkable characteristic of Descemet’s membrane appears to be its homogeneity compared with the endothelium (which is a monocellular layer) or the stroma (which is composed of collagen bundles and keratocytes). The picture shown in Fig. 7(b) contains a peculiar void in the signal (both in SHG and THG) limited by two strong borders of THG signal. This indicates that this area corresponds to Descemet’s membrane.
We can recognize on this same image some cells in the endothelium (ec). The endothelium is a thin monolayer of cells lying flat on Descemet’s membrane. The dark elongated structures seen in this layer are believed to correspond to the nuclei of the endothelial cells. Precision of these structures is at least equal or better than that obtained in Fig. 7(a) with standard histology. As for the thin anterior corneal epithelial layers, combined SHG/THG imaging of the thin posterior layers of the cornea reaches a degree of precision in the detail beyond that obtained with any commercially available OCT system. Finally, some cells in the stroma, namely the keratocytes (k), could also be recognized by their typical elongated shape, squeezed between the collagen bundles.
We have demonstrated that the combination of simultaneous SHG and THG allows determining all the layers of a porcine cornea in large scale imaging in forward and in backward direction. Large-scale acquisitions using focusing immersion objective are important, since they provide images of the whole corneal thickness. Longitudinal cross-section corneal images similar to those obtain with OCT are generated, with the difference that SHGI and THGI provide the high spatial resolution needed to clearly distinguish the structure of the main corneal layers. Combination of both B-SHG and B-THG will offer unique opportunities for the development of new non-invasive tools for in vivo corneal imaging.
We thank Thouria Bensaoula, MD, François Blanchard, MSc, Marie-Eve Choronzey, BA, Louis Hoffart, MD, Ossama Nada, MD and Gilles Olivié, PhD, for their technical support. This work was supported by the Natural Sciences and Engineering Research Council (NSERC, ON, Canada), the Canadian Institutes of Health Research (CIHR, ON, Canada) and the FRSQ Reasearch in vision network (QC, Canada).
References and links
1. A. C. M. Wong, C. C. Wong, N. S. Y. Yuen, and S. P. Hui, “Correlational study of central corneal thickness measurements on Hong Kong Chinese using optical coherence tomography, Orbscan and ultrasound pachymetry,” Eye 16, 715 (2002). [CrossRef] [PubMed]
5. N. Morishige, W. M. Petroll, T. Nishida, M. C. Kenney, and J. V. Jester, “Noninvasive corneal stromal collagen imaging using two-photon-generated second-harmonic signals,” J. Cataract Refract. Surg. 32, 1784 (2006). [CrossRef] [PubMed]
6. Y. Barad, H. Eisenberg, M. Horowitz, and Y. Silberberg, “Nonlinear scanning laser microscopy by third harmonic generation,” Appl. Phys. Lett. 70, 922 (1997). [CrossRef]
9. B. G. Wang, A. Eitner, J. Lindenau, and K. J. Halbhuber, “High-Resolution Two-Photon Excitation Microscopy of Ocular Tissues in Porcine Eye,” Lasers Surg. Med. 40, 247–256 (2008). [CrossRef] [PubMed]
10. S. W. Teng, H. Y. Tan, J. L. Peng, H. H. Lin, K. H. Kim, W. Lo, Y. Sun, W. C. Lin, S. J. Lin, S. H. Jee, P. T. C. So, and C. Y. Dong, “Multiphoton autofluorescence and second-harmonic generation imaging of the ex vivo porcine eye,” Invest. Ophthalmol. Vis. Sci. 47, 1216–1224 (2006). [CrossRef] [PubMed]
11. E. J. Gualda, G. Filippidis, G. Voglis, M. Mari, C. Fotakis, and N. Tavernarakis, “In vivo imaging of cellular structures in Caenorhabditis elegans by combined TPEF, SHG and THG microscopy,” J. Microsc. 229, 141–150 (2008). [CrossRef] [PubMed]
12. E. Svaldeniene, V. Babrauskiene, and M. Paunksniene, “Structural features of the cornea: light and electron microscopy,” Veterinarija ir Zootechnika 46, 50–55 (2003).
13. A. Brocas, L. Jay, E. Mottay, I. Brunette, and T. Ozaki, “Corneal imaging by second and third harmonic generation microscopy,” Proc. SPIE 6860, 68600C (2008). [CrossRef]
14. J. Squier and M. Müller, “High resolution nonlinear microscopy: A review of sources and methods for achieving optimal imaging,” Rev. Sci. Instrum. 72, 2855–2867 (2001). [CrossRef]