Abstract
purpose. To evaluate the effect of contrast and luminance attenuation on the multifocal electroretinogram (mfERG) responses of normal and glaucomatous eyes of cynomolgus monkeys.
methods. Nine adult male cynomolgus monkeys with unilateral experimentally induced glaucoma were used. Hypertension-induced damage was confirmed by tomography of the optic disc. mfERGs were recorded with five different stimulus contrasts and/or luminance settings. The first-order and the first slice of second-order responses were analyzed.
results. Waveforms of normal and glaucomatous eyes differed in shape and amplitude. Second-order responses contributed to first-order responses of the signals in the normal eyes, but made a negligible contribution to the signals in the glaucomatous eyes. Contrast and luminance attenuation affected both first- and second-order responses. The differences between signals in normal and glaucomatous eyes were sufficiently large for an unsupervised clustering algorithm to achieve accurate segregation.
conclusions. The observations in this study indicate that outer and inner retinal generators participate in first-order mfERG responses and that both inner and outer retinal contributors respond to contrast and luminance changes in stimulus. The hypertension-induced changes in the mfERG furthermore suggest damage to both inner and outer retina.
Successful management of glaucoma hinges, to a large extent, on early detection and on monitoring of retinal function once the disease has been diagnosed and treatment initiated. Functional glaucomatous damage can be assessed by perimetry. However, this method is not objective, is not useable in nonverbal patients, and is not sensitive enough to detect small focal defects.
1 2 3
The multifocal electroretinogram (mfERG) technique introduced by Sutter and Tran in 1992
4 allows simultaneous independent stimulation of small retinal areas. The mfERG produces a high-resolution map of the electrophysiological function of the retina. As such, it has a substantial advantage over the flash ERG (FERG) and the pattern ERG (PERG) that evoke a global response. This advantage is especially important in the diagnosis and assessment of diseases that affect the retina unevenly. The cellular sources of the mfERG response and its sensitivity to changes caused by disease processes have yet to be fully established. The potential of the mfERG for research and clinical applications in retinal diseases such as retinitis pigmentosa, diabetic retinopathy, and glaucoma, has been explored.
5 6 7 8 9 Some researchers suggest the presence of a component originating in the human optic nerve head and its attenuation in patients with glaucoma.
10 11 12 13 Other studies have described delayed peak latencies and decreased amplitudes in patients with glaucomatous eyes.
14 15 16
The purpose of this study was to test the effect of contrast and luminance attenuation on the mfERG response in cynomolgus monkeys and to evaluate the use of these parameters in the diagnosis of glaucoma in this animal model. Preferential damage to cells of the magnocellular pathway leads to a differential loss of contrast perception, as seen in PERG.
17 18 19 20 21 22 23 24 25 Patients who have ocular hypertension with no detectable field defects or optic nerve damage also demonstrate reduced contrast sensitivity
22 ; thus, reduced contrast perception may be a sensitive measure for early phases of the disease. Several researchers have evaluated the effect of attenuation of contrast and luminance on human mfERG response,
13 14 26 27 28 but no reports of similar studies in monkeys have been published, to the best of our knowledge. Assuming that these results are confirmed in monkeys, this could provide a tool for a better understanding of the pathogenesis of ocular hypertension.
Nine 8-year-old male cynomolgus monkeys (Macaca fascicularis) with uniocular experimental glaucoma were used in this study. Animals were permanently housed in an air-conditioned room with a temperature of 19°C to 25°C in a light–dark cycle (light from 7 AM to 7 PM). All animals were free from herpes virus (sanitary examinations are performed every 6 months). They were fed a normal pellet diet (Special Diet Service, Witham, Essex, UK), supplemented with fruit daily. Water was provided ad libitum. Experimental and animal care procedures adhered to the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research and local regulations and ethical considerations for the use of animals.
Ocular hypertension was induced by argon laser trabeculum photocoagulation of the midtrabecular meshwork between December 1994 and February 1995.
29 30 31 The aiming beam of the laser (75–100 μm in diameter, 1100 mW, 0.5-second duration) was focused on the midtrabecular meshwork. Between 50 and 65 laser beam spots were directed at the entire trabeculum (360°). One monkey was treated twice at a 1-month interval. The contralateral eye served as an untreated control.
Monkeys were premedicated with an intramuscular injection of ketamine (10 mg/kg). The intraocular pressure (IOP) in both eyes was measured using a calibrated tonometer (Tono-Pen XL; Mentor Ophthalmics, Inc., Norwell, MA). To avoid manual pressure on the globe, the lids were opened with a Barraquer-Colibri speculum. Anesthesia was induced with intravenous (IV) propofol, and the monkeys were intubated and maintained on an inhalation mixture of N2O and O2, supplemented with a continuous IV drip of propofol at a rate of 5 mg/kg per hour throughout the recording session. Eye movement was prevented by periodical IV boluses of a nondepolarizing muscle relaxant (0.1 mg/kg vecuronium or 0.45 mg/kg rocuronium). Expired CO2, heart rate, respiratory rate, core temperature, blood oxygen level, and blood pressure were monitored continuously and maintained within normal range. Pupils were dilated with topical tropicamide and atropine. Hypertension-induced damage was assessed by retinal tomography (Heidelberg Engineering, Heidelberg, Germany). Optical correction of the eyes was evaluated with the tomograph and manually with a retinoscope, and the eyes were optically corrected accordingly.
Jet contact lens electrodes (Metrovision, Perenchies, France) were placed on the cornea of each eye and served as an active or a reference electrode interchangeably. Hydroxyethylcellulose gel was applied for corneal protection and for good adhesion of the Jet lenses. A gold cup electrode attached to the shaved skin rostral to the ear served as a ground electrode. The eyes were recorded in random order.
A sequence of brief positioning recordings was conducted for each eye before commencing the experimental recordings. The eye’s position was adjusted so that the macula was in the center of the stimulated field, and the optic disc was included in the retinal area stimulated.
The monitor (Landscape UHR 2.1L; Nortech Imaging Technologies, Plymouth, MN) was viewed by the monkey at a distance of 33 cm, which provided a visual field with an approximate 23° radius. The stimulus consisted of 103 densely packed unscaled hexagons alternating between black and white. Recordings were performed under five settings, which varied in contrast or in luminance
(Table 1) . The background was set on average luminance. The hexagons alternated according to a pseudorandom binary sequence of 2
15 − 1 steps. The frame rate was 75 Hz, yielding a total recording run of approximately 7 minutes, recorded as one segment. An identical stimulating sequence was used for all the hexagons. However, the hexagons were temporally modulated with a relative lag among them, enabling the extraction of local responses from the overall response derived at the cornea. A detailed description of the mfERG system is given elsewhere.
4 Signals were recorded with a band pass of 1 to 300 Hz; a notch filter was not used. Signals were amplified by a gain of 100,000 (Model 12; Grass Instrument Co., West Warwick, RI). In response to an occluded stimulus, mfERGs were recorded to estimate the residual noise. Signals that were not satisfactorily different from the noise in wave shape and amplitude were discarded.
The first-order kernel (K1) and the first slice of the second-order kernel (K2) were evaluated to assess glaucoma-related changes to retinal function. To study the impact of K2 responses on the K1 responses we used an averaged response of the central seven hexagons
(Fig 1A) , because K2 responses were largest in this region in our study, in agreement with previous findings.
32 However, a similar relationship between K2 and K1 components exists in all retinal areas. The amplitude and implicit time values of distinct peaks in the responses were analyzed statistically (Excel; Microsoft, Redmond, WA). The root mean square (RMS) was calculated according to
\[\sqrt{\frac{{{\sum}_{n{=}1}^{N}}\ n^{2}}{N}}\]
where
n is the amplitude in nanovolts per degree squared measured every 0.83 ms from 0 to 80 ms, and
N is the number of
n values. Statistical significance was determined by the nonparametric directional Wilcoxon signed rank test (unless specified otherwise).
The combination feature of the visual evoked response imaging system software (VERIS; Electro Diagnostic Imaging [EDI], San Mateo, CA), which combines a number of raw data sets into one single record, has been used previously to represent a group of similar mfERG recordings.
33 This tool holds the advantage of providing both a representative waveform and an increased signal-to-noise ratio. However, the useful application of this method requires good alignment of the combined trace arrays. Therefore, the data of three eyes in each group in which the trace array was not exactly centered had to be excluded from this analysis.
Evoked potential waveforms are typically characterized by latency and amplitude measures of peaks and troughs along the curve. Features that may be overlooked include latencies and amplitudes of interposed data points that comprise the waveform of the recording. Consequently, valuable information may be lost. In our case, an average of the seven macular K1 responses was extracted from each recording and grouped into data sets according to the stimulus setting. Each ensemble of ERGs was decomposed into a set of principle components (PCs) and the unsupervised fuzzy clustering (UFC) algorithm (in MatLab; The MathWorks, Natick, MA) was applied to the corresponding coefficient of these principle components, to classify the ERGs into the optimal number of clusters.
34 35
The principal component analysis procedure was performed by the discrete Karhunan-Loeve transform using the eigenvectors of the signal’s covariance matrix.
34 35 The UFC algorithm partitions the data by a combination of the modified fuzzy k-means (FKM) algorithm
34 35 and the fuzzy maximum-likelihood estimation (FMLE) algorithm,
34 35 for a mixture of Gaussian distributed clusters. The advantage of the UFC algorithm is the unsupervised initialization of cluster prototypes and the criteria for cluster validity using fuzzy hypervolume and density functions.
34 35 It performs well in a situation of large variability of cluster shapes, densities, and number of data points in each cluster.
IOPs in the eyes over a period of 30 months preceding the mfERG recordings are documented in
Table 2 . On the day of mfERG recordings, the median IOP and 5 to 95 percentile range was 15 (13–18) mm Hg in the control group and 42 (33–69) mm Hg in the treated group. Median IOP in the treated eyes was significantly higher than in the normal eyes (
P < 0.005). It has been reported that the Tono-Pen (EDI) underestimates pressure in the cynomolgus monkey, especially at high IOPs.
36 Therefore, it is conceivable that the true IOPs were even higher in the treated eyes.
Median and 5 to 95 percentile range of cup-to-disc (C/D) ratios on the day of mfERG measurements were 0.14 (0.09–0.40) in the control group and 0.80 (0.66–0.90) in the treated group. The C/D ratios in the treated eyes were significantly higher than in the normal eyes (P = 0.005). The mean retinal nerve fiber layer thickness on the day of the experiment was also significantly different (P = 0.025) between the control and treated eyes (median and percentile values were 0.29 [0.23–0.36] mm in the normal eyes and 0.16 [0.09–0.20] mm in the treated eyes). Based on the remarkable changes in the optic nerve head, we felt confident in the determination that the treated eyes had sustained glaucomatous damage. No correlation was found between tomography findings and IOP or between tomography findings and glaucoma-induced changes in the mfERG responses.
Trace arrays of the normal eyes showed variations across the retinal field in waveform and amplitude. Trace arrays of the glaucomatous eyes differed from the normal by displaying uniform wave shapes, varying across the retinal field only in amplitude
(Fig. 1) . The insets in
Figure 1 portray typical single traces. The mfERG response of the normal eyes was characterized by a negative trough designated N1, followed by a double positive potential (P1, P2) and smaller oscillations. The most prominent feature distinguishing the waveforms of the glaucomatous eyes from the normal eyes was that the positive potential (15–45 ms) appeared as a single peak. Furthermore, in glaucomatous eyes the signal after the double peak showed less prominent oscillations. These observations suggest that normal and glaucomatous eyes differ in the characteristics of the components contributing to the K1 response or in the superposition of these components. The RMSs of the waveforms, a general amplitude measure of the responses, were significantly different between normal and glaucomatous eyes under the baseline setting and under all stimulus conditions
(Table 3) . Although RMS does not directly represent any retinal process, this observation is a quantitative measure, suggesting that the underlying components are different between normal and glaucomatous eyes.
Attenuation of contrast induced changes in amplitude and wave-shape in the K1 responses of the normal eyes. Generally, the amplitude of the responses correlated directly with the contrast
(Fig. 2A) . The substantial reduction of P1 caused an increase in the P2-to-P1 amplitude ratio with diminishing contrast
(Table 4) . Reduction in mean luminance induced a delay in wave components
(Fig. 2B) . The median and 5 to 95 percentile implicit times of N1 were 13.3 (12.0–13.8), 13.3 (13.3–14.2), and 15 (14.2–16.7) ms under mean luminance of 100, 50, and 20 cd/m
2, respectively. The delay in N1 was statistically significant under medium and low luminance levels in comparison with baseline (
P ≤ 0.05). In addition, P1 decreased substantially with luminance reduction, whereas the amplitude of P2 increased
(Table 4) .
The responses of the glaucomatous eyes to contrast and to luminance attenuation showed features in common with the analogous responses of the normal eyes
(Figs. 2C 2D) . Specifically, contrast reduction elicited amplitude reductions (
Fig. 2C ,
Table 4 ). P2 was diminished in the glaucomatous eyes and was therefore masked by P1 in the response to a stimulus of 100% contrast, in which the amplitude of P1 was high. A double peak was revealed only under stimulation at lower contrasts.
Reduction in mean luminance induced a significant delay (
P ≤ 0.05) in the latency of the response of the glaucomatous eyes. The median and 5 to 95 percentile range of the implicit time of N1 was 14.2 (12.3–15.5), 15 (12.7–17.2), and 17.5 (14.7–18.8) ms in the signals evoked by luminance of 100, 50, and 20 cd/m
2, respectively. Under all three luminance levels, it was not possible to distinguish clearly between P1 and P2
(Fig. 2D) . Nevertheless, it is conceivable that P1 was largely reduced in amplitude (as was observed in the responses of the normal eyes), whereas P2 was unaffected. Consequently, the predominance of P2 was manifested as a latency increase of the entire positive component. A conspicuous effect on the signal, reflected both by the reduction in amplitude and by the delay of N1, was observed when the luminance was reduced from 50 to 20 cd/m
2, whereas a reduction from 100 to 50 cd/m
2 elicited subtle changes in the waveform. This shows that a stimulus of 100 cd/m
2 did not evoke a response considerably different from that evoked by the stimulus of 50 cd/m
2 in the glaucomatous eyes.
Second-order responses of normal eyes were characterized by two peaks at approximately 25 and 45 ms, separated by a trough (p1, n, p2;
Figs. 3A 3B ). The K2 responses reflected the effects of contrast and luminance attenuation observed in the K1 responses
(Table 5) . Contrast reduction elicited a gradual amplitude reduction in the K2 responses. Luminance reduction evoked delayed responses. The median and 5 to 95 percentiles of the implicit time of p1 were 23.3 (22.0–26), 24.2 (22.8–26.3), and 25.8 (23.5–28.5) ms at high, medium, and low luminance. The delay in p1 induced by low luminance was statistically significant when compared with the implicit time of the baseline setting (
P < 0.01). The glaucomatous eyes had negligible K2 responses, with only minor changes in response to the contrast and luminance attenuation of the stimulus
(Figs. 3C 3D) .
To study the contribution of K2 to K1 responses, K2 responses were subtracted from K1 responses (the K1-K2 component). In the normal eyes, this rendered a component characterized by Sn1, which was identical with N1 of the K1 responses in amplitude and latency, a positive double peak (Sp1, Sp2), and a second negativity Sn2
(Figs. 4A 4B) . Contrast reduction elicited a notable amplitude reduction of all signal elements except Sp1
(Fig. 4A) , which implies that P1 of the K1 responses emanates fundamentally from the K2 responses. Luminance reduction was manifested in a delay of the K1-K2 component, in correlation with the K1 responses. The implicit time of Sn1 was 12.5 (11.5–13.3), 13.3 (12.5–14.3), and 14.9 (12.6–15.8) ms for 100, 50, and 20 cd/m
2 respectively, with a statistically significant delay under low luminance (
P < 0.01). In the glaucomatous eyes the K1-K2 component was almost identical with the K1 response, substantiating the minor contribution of K2 responses to the K1 responses. Another conspicuous difference was that the signals of normal eyes showed a prominent Sn2, whereas in the signals of the glaucomatous eyes, Sn2 was a smaller constituent of the waveform.
It is evident that N1 of the K1 response, as well as the oscillations subsequent to the double peak, emanate from the K1-K2 response. The K2 response contributed mainly to the positive potential of the K1 response and accentuated the double positive peak. Both K2 and K1-K2 responses changed with contrast and luminance attenuation. Therefore, the effect of variations in stimulus on the K1 response arises from a combined effect on these components.
We tested data sets assembled according to various stimulus settings. Maximal segregation between signals of normal and glaucomatous eyes was achieved when clustering the signals evoked by contrast attenuation. When signals under high, medium, and low contrast levels were grouped together as one data set, the signals of the glaucomatous eyes were generally set apart from the other signals, based on the primary five PCs
(Figs. 5A 5B) . The PCs are mathematical components of the waveforms and do not represent physiological functions of the retina. However, it is interesting to note that the main elements of the second PC corresponded to the p1, n, and p2 of the K2 responses.
At each separate contrast level, the algorithm yielded a fundamentally good segregation between signals of normal and glaucomatous eyes, but several outliers were registered. For example,
Figures 5C and 5D show the clustering of the signals under low-contrast stimulation. The algorithm clustered the signals of the normal eyes (signals 1 to 8) together as cluster 1 with two “outliers” (signals 2 and 7). The signals of the glaucomatous eyes were grouped as cluster 2 with one outlier (signal 11). However, assembling the clustering results of the data sets under high, medium, and low contrast levels enabled us to overcome the outliers and yielded the optimal segregation. By comparing the clustering results to the signal numbers that were entered, as was now shown for 20% contrast, we found that when a signal was classified as glaucomatous under each contrast level, it was in fact glaucomatous. In all data sets, clustering into three centers yielded an additional group consisting mainly of the normal and glaucomatous signals that were incorrectly classified when two cluster centers were forced, indicating that the algorithm was able to isolate the deviant signals.
In this study K1, K2, and K1-K2 responses were characterized for mfERG responses of normal and glaucomatous eyes in a monkey model of glaucoma. First-order responses (K1) are considered the mean response to a focal flash. However, the effect of a flash extends beyond the duration of one frame and is inevitably integrated in the consecutive flashes. By chemically blocking the activity of specific retinal cells, it was estimated that K1 responses are dominated by bipolar cells and photoreceptor activity, with contributions from the inner retina.
37 The first slice of the second-order kernel (K2) measures how the mfERG response is influenced by the adaptation to the immediately preceding flash, a process associated with the inner retina.
6 37 38 Removing the contribution of the inner retina (K1-K2) could be considered a refined approximation of the response to the stimulating flash only. Breaking down the signals into K1-K2 and K2 responses is a simplification, because additional higher order components also contribute to the K1 response. Nevertheless, this analysis approximates the contributions of outer (K1-K2) and inner (K2) retinal sources to the K1 mfERG response.
The superposition of the inner retinal contribution (K2) on the outer retinal responses (K1-K2) modified the waveforms. First, it scaled down the amplitudes of the positive potential. Second, it shaped the positive potential into a prominent double positive peak. Third, it masked the negative component Sn2, which may be comparable to the photopic negative response (PhNR) of the FERG in monkeys.
39 Although the full-field and the mfERG paradigms and signals differ, Sn2 bears similarity to the PhNR. Both appear immediately after the positive potential of the response, and both are diminished in glaucoma. However, the PhNR was attributed to inner retinal cells, whereas in the present study, Sn2 appeared to originate in outer retinal layers. The precise origin of Sn2, as of the other mfERG signal elements, remains to be determined.
A clear difference between signals of normal and glaucomatous eyes was seen in the K1 responses evoked by the commonly used mfERG stimulus setting of high contrast and high luminance (the baseline setting). The signal of the glaucomatous eyes did not show the prominent oscillatory elements evident in the signal of the normal eyes. This suggests that the glaucomatous eyes are missing an oscillatory component, which is typical of neuronal spiking activity. Although the origin of the oscillatory component in the mfERG response is not certain, it may reflect oscillatory potentials (OPs) generated by cells in the inner plexiform or inner nuclear layers of the retina.
40 41 42 Amacrine, interplexiform and ganglion cells were suggested as the possible generators of OPs in FERG responses.
13 Oscillatory potentials were shown to be affected by glaucoma as well as by other inner retinal disorders.
40 43 44 45 Oscillatory potential losses were also significant in eyes of patients with glaucoma with no obvious clinical manifestations, implying that OPs may be a leading indicator of glaucomatous damage.
36 39 The results of this study suggest that OPs contribute to the first-order response of the mfERG and that they are diminished in responses of glaucomatous eyes.
This study shows that K2 responses are greatly diminished in signals of glaucomatous eyes. This implies an abnormal process of flash adaptation in the glaucomatous eyes. Previous studies have attributed the second-order kernel responses to inner retinal layers.
6 37 38 Therefore, we propose that glaucomatous eyes sustain damage in the inner retinal layers, and that this is evident in the mfERG responses. Similar findings were recently reported.
32 46 However, the results of our experiment indicated more extensive retinal damage, involving outer retinal layers as well. Because K2 responses were negligible in the signals of glaucomatous eyes, we expected to find a close resemblance between the K1 responses of the glaucomatous eyes and the K1-K2 responses of the normal eyes and were surprised that this was not the case (compare
Figs. 2C 2D with
Figs. 4A 4B ). The differences between these responses suggest damage to the outer retina in the glaucomatous eyes. It is probable that the long-standing ocular hypertension in our monkeys caused more severe damage than in the animals used by other researchers, encompassing outer retinal layers. Alternatively, damage to outer retinal cells may have occurred independent of inner retinal damage. It has been shown that, in this monkey model of glaucoma, swelling of the red and green cones occurs as early as 7 weeks after induction of glaucoma, possibly due to decreased choroidal circulation and ischemia.
47
To enhance differences between normal and glaucomatous eyes, stimuli were reduced in contrast or luminance. The inner retina processes contrast and luminance perception, and a loss of sensitivity to these parameters is a known functional manifestation of glaucomatous damage.
20 21 24 26 Bearse and Sutter
27 suggested that decreased contrast increases the relative contribution of the inner retina to the mfERG signal, and stimuli of reduced contrast were used in an attempt to detect glaucoma.
27 48 In the signals of normal and glaucomatous eyes in this experiment, the amplitude of the responses was directly correlated with the contrast of the stimulus. A lower contrast elicited weaker responses. Similar findings in PERG and in mfERG were previously reported in normal human eyes.
26 49 In this study, both the K2 and the K1-K2 responses decreased with contrast reduction
(Figs. 3A 3C 4A 4C) . If indeed these components represent activity of different retinal layers, this suggests that both the outer and the inner retina responded to contrast attenuation. Furthermore, in this study, the inner retinal component of the monkey did not saturate under medium contrast as in humans,
27 but continued a gradual, yet constant, increase with contrast
(Fig. 3A) .
Contrast and luminance attenuation helped elucidate the differences in K1 responses between normal and glaucomatous eyes. It is evident that the single positive potential in glaucomatous eyes correlated to P1 of the responses of normal eyes, whereas P2 was diminished in the glaucomatous eyes. Changes in waveform of the positive potential suggest ocular hypertension induced damage in the inner plexiform layer.
37 P1 responded more vigorously to contrast and luminance changes than did P2
(Fig. 2) —thus, the common effects in both eye groups. A recent study in human patients in which responses to 100% and 50% contrast stimulation were compared showed only subtle differences between normal and glaucomatous eyes.
14 The difference in results between the studies could be due to species difference or to a different degree of damage.
Luminance reduction caused a delay of the signals in normal and glaucomatous eyes, particularly when the luminance was reduced from 50 to 20 cd/m
2 (Figs. 2C 2D) . Implicit time changes and the effect of luminance attenuation on the implicit time of the signals has not been established, and there are reports of constant implicit time over a range of luminance levels, as well as reports on decreases in peak implicit time with increasing luminance intensity.
8 37 50
The differences in waveforms between normal and glaucomatous eyes were not only evident in the signals, but were also sufficient to set the groups apart objectively. The UFC algorithm distinguished between signals of normal and glaucomatous eyes. Applying the algorithm to the signals of each contrast level separately was the most proficient method.
In summary, in this study advanced glaucomatous damage in monkeys eliminated K2 responses. In addition, a closer approximation of the responses to the stimulating flash (K1-K2 responses) illustrated differences between normal and glaucomatous eyes. Together, these findings indicate damage to outer and inner retinal layers. Contrast and luminance attenuation have a complex effect on the K1 responses by modifying outer and inner retinal contributions. Future studies should investigate the mfERG changes at early stages of ocular hypertension and further examine the origin of the signal components.
The authors thank Erich Sutter (Electro Diagnostic Imaging, San Mateo, CA) for valuable advice, Hillary Voet (Center for Agricultural Economic Research, Faculty of Agriculture, Hebrew University of Jerusalem) for help with the statistical analyses, and Isabelle Questel and Emmanuel Faure (Novartis Ophthalmics) for their dedicated work.