The Pattern of Retinal Ganglion Cell Loss in OPA1-Related Autosomal Dominant Optic Atrophy Inferred From Temporal, Spatial, and Chromatic Sensitivity Losses

Purpose Progressive retinal ganglion cell (RGC) loss is the pathological hallmark of autosomal dominant optic atrophy (DOA) caused by pathogenic OPA1 mutations. The aim of this study was to conduct an in-depth psychophysical study of the visual losses in DOA and to infer any selective vulnerability of visual pathways subserved by different RGC subtypes. Methods We recruited 25 patients carrying pathogenic OPA1 mutations and age-matched healthy individuals. Spatial contrast sensitivity functions (SCSFs) and chromatic contrast sensitivity were quantified, the latter using the Cambridge Colour Test. In 11 patients, long (L) and short (S) wavelength–sensitive cone temporal acuities were measured as a function of target illuminance, and L-cone temporal contrast sensitivity (TCSF) as a function of temporal frequency. Results Spatial contrast sensitivity functions were abnormal, with the loss of sensitivity increasing with spatial frequency. Further, the highest L-cone temporal acuity fell on average by 10 Hz and the TCSFs by 0.66 log10 unit. Chromatic thresholds along the protan, deutan, and tritan axes were 8, 9, and 14 times higher than normal, respectively, with losses increasing with age and S-cone temporal acuity showing the most significant age-related decline. Conclusions Losses of midget parvocellular, parasol magnocellular, and bistratified koniocellular RGCs could account for the losses of high spatial frequency sensitivity and protan and deutan sensitivities, high temporal frequency sensitivity, and S-cone temporal and tritan sensitivities, respectively. The S-cone–related losses showed a significant deterioration with increasing patient age and could therefore prove useful biomarkers of disease progression in DOA.

transmit achromatic visual information of high spatial, but low temporal resolution, and also red-green chromatic information produced by opponent long (L) and middle (M) wavelengthsensitive cone signals. Parasol cells constitute a smaller proportion of RGCs, and their L-and M-cone inputs form the magnocellular pathway, which transmits achromatic visual information of high temporal but low spatial resolution. Parallel processing in the retina and early visual pathway has been the subject of several excellent reviews. [12][13][14][15][16][17][18][19] The relatively sparse small bistratified RGCs that form part of the koniocellular visual pathway are excited by short (S)-wavelength cones and inhibited by L-and M-cones, and they transmit a blue-yellow chromatic signal. The S-cones and their pathways have been the subject of several recent reviews. [20][21][22][23] Unlike conventional clinical visual tests that distinguish relatively poorly between different visual pathways, advanced psychophysical methods can be designed that selectively test the performance of one individual pathway. 24 However, it should be recognized that although concrete links are often made between these physiological pathways and psychophysical performance on various tasks, they are inevitably hypothetical and to some extent tenuous. In the context of DOA, psychophysical studies have focused mainly on color vision and have shown various deficits that tended to be most severe for discrimination along the tritan axis. 1,[25][26][27][28][29] There are two reports detailing deficits in the parvocellular and magnocellular pathways in DOA, with one showing abnormal achromatic contrast sensitivities and the other abnormal peripheral motion detection thresholds. 1,28 Although both parvocellular and magnocellular pathways seem to be involved in visually affected patients, the severity and progression of deficits with increasing age remain unclear.
In this study, we used temporal, spatial, and chromatic psychophysical tests to assess the magno-, parvo-, and koniocellular pathways in patients with DOA caused by pathogenic OPA1 mutations. The psychophysical temporal test protocols have previously been used in our laboratory to study various retinal pathologies. [30][31][32][33] Our aim in this work was to characterize the visual losses, and to infer any selective losses in visual pathways subserved by different RGC types. The deficits observed for these functional visual parameters were also related to structural measures, namely, the thicknesses of the macular RGC and the peripapillary retinal nerve fiber layers (RNFL), and to the patient's age.

METHODS Participants
We included 25 patients with DOA from 15 independent families carrying 13 different pathogenic sequence variants in the OPA1 gene. The genetic testing was performed in the Leeds Genetics Laboratory (St. James's University Hospital, Leeds, UK) and as part of two previously reported studies. 4,34 All patients underwent achromatic spatial contrast sensitivity and color discrimination tests and the following ophthalmologic investigations: best-corrected visual acuity (BCVA) using the Early Treatment Diabetic Retinopathy Study (ETDRS) charts, 35 static computer-automated white-on-white threshold perimetry (Swedish Interactive Testing Algorithm fast 30-2 Humphrey visual field [model 740; Humphrey Instruments, Dublin, CA, USA]), and optical coherence tomography (OCT) imaging of the macula and optic nerve head (see details below). The demographics, genotypes, and clinical characteristics of the DOA patients are given in Table 1. Eleven of the 25 DOA patients (P1-11) also performed an extended psychophysical test protocol that included L-and S-cone temporal vision tests. The genotypes and clinical data for all patients are given in Table 2. Normative spatial contrast sensitivity and chromatic sensitivity data were obtained from 15 individuals (aged 17-78 years old at the time of testing) with normal or corrected-tonormal visual acuities, and normal color vision as assessed by standard color vision tests (the Farnsworth-Munsell 100 hue test, The Hardy-Rand-Rittler [HRR] pseudoisochromatic test, and the Ishihara color vision test). These measurements were made as part of this study. Normative temporal acuity was also measured in 15 individuals (aged 17-67 years old at the time of testing). Three of them were studied as part of this study and 12 are taken from a previous study that used the same conditions as those used here. 31,33 Similarly, normative temporal contrast sensitivity data (n ¼ 8) were taken from a previous study. 31,33 L-and S-Cone Critical Flicker Fusion and L-Cone Modulation Sensitivity Measurements L-and S-cone temporal acuities (critical flicker fusion, CFF) and L-cone temporal modulation sensitivity were measured using a Maxwellian-view optical system described in detail elsewhere. 33,36 Two types of stimuli were used for the CFF measurements: The L-cone stimulus was a flickering, monochromatic, circular target with a diameter subtending 48 of visual angle and a wavelength of 650 nm. The target was presented in the center of a 98 diameter circular background field of 481 nm. The S-cone stimulus was also a flickering circular target of 48 diameter but of 440 nm in wavelength presented in the center of a 98 diameter background field of 620 nm in wavelength. For L-cone modulation sensitivity measurements, the time-averaged mean radiances of the 481nm background and 650-nm target were fixed at 8.3 and 10.6 log quanta s À1 deg À2 , respectively. The observers varied the modulation of the sinusoidally flickering 650-nm target to find the modulation at each target frequency at which the flicker was just visible. In the CFF measurements, observers varied the frequency of the 650-or 440-nm target, which was sinusoidally flickering with a contrast of 92%, to find the frequency at which the flicker was just visible for a range of target radiances. Before each run, the observers light adapted to the background and target for 2 minutes. The observers viewed the stimuli monocularly with the eye they preferred (as indicated by the asterisk in Table 2) and interacted with the computer by means of an eight-button keypad as previously described. 33 Each experiment was repeated two or three times on the same day. Details of these measurement techniques have been given elsewhere. 33 Achromatic Spatial Contrast Sensitivity Function (CSF) The achromatic spatial contrast sensitivity function, the SCSF, was measured as a function of spatial frequency. The target stimuli were generated on a gamma-corrected Sony Trinitron monitor (model GDM F520; Sony Electronics, Inc., Park Ridge, NJ, USA) with 85-Hz frame rate and a resolution of 1600 3 1200 pixels connected to a DataPixx video processor (VPixx Technologies, Inc., Saint-Bruno, QC, Canada). The full screen subtended a visual angle of 398 3 298 at the test distance of 0.57 m. The experiments were performed at a constant mean luminance of 44.57 cd.m À2 as measured using a ColorCal calibration device (Cambridge Research Systems Ltd., Rochester, UK). The stimuli were circular centrally presented, horizontally oriented Gabor patterns with spatial standard deviations of 68 and spatial frequencies ranging from 0.25 to 16 cycles per degree (cyc/deg). The target duration was 500 ms, preceded and followed by 100-ms cosine-windowed onsets and offsets. The order of presentation was from low to high spatial frequencies. Thresholds were measured using a Yes/No staircase procedure. By means of a two-button keypad, observers indicated whether they could detect the Gabor pattern. Stimulus contrast (Michelson contrast) was defined as (L max À L min )/(L max þ L min ), where L max and L min are the maximum and minimum luminances in the Gabor pattern, respectively. Contrast was adjusted following a one-up-one-down staircase procedure with a variable step size. For the first five reversals (changes in the direction of the staircase) the step size was 0.3 log units, after which the step size was reduced to 0.2 log units for two more reversals and then finally to 0.1 log unit for the last four reversals. A single run required a total of nine reversals, with the contrast ''threshold'' taken as the average of the last six reversals. Contrast sensitivity is the reciprocal of the contrast threshold. The spatial contrast sensitivity measurements lasted approximately 20 minutes for each observer. The

Modeling Contrast Sensitivity Functions
In order to characterize the temporal and spatial contrast sensitivities, and how they differ between patients and control subjects, we fitted a cubic function in log-log coordinates, log 10 (S) ¼ a log 10 (f) þ b log 10 (f) þ c log 10 (f) þ d; where S is temporal or spatial contrast sensitivity, f is the frequency (Hz for temporal, and cyc/deg for spatial), and a, b, c, and d are constants. The cubic function here has no theoretical basis, but was chosen for its simplicity and because it provided a good fit to both the temporal and the spatial contrast threshold data for both the normal controls and the patients and therefore could be used to parameterize the contrast sensitivity functions.

Chromatic Discrimination
Chromatic discrimination was tested using the Trivector test procedure implemented as part of the  The test conditions were modified to be appropriate for observers with markedly reduced visual acuity. The viewing distance was set so that the gap in the letter C opening subtended 58 instead of 18 of visual angle. (For our system, this corresponded to a viewing distance of 62.6 cm.) Thresholds were measured along the three dichromat confusion lines, protan, deutan, and tritan, but the highest target saturation was increased to 1600 3 10 À4 u 0 v 0 units and the number of steps from 6 to 10 in order to maintain the standard unit difference between the steps. The time allowed for the subject to respond was increased from the standard 8 to 20 seconds. The test was then run using a standard one-up, one-down psychophysical staircase procedure with six reversals. 37 The observers were instructed to respond to the gap position (four-alternative forced choice) by means of a four-button keypad. The test was performed binocularly with appropriate near corrections if needed.
The CCT depends on commercial software and hardware that uses a derivation of the 28 CIE standard observer data to generate the test stimuli and interpret the results. Although the 28 observer is plausible for normal control measurements, since normal observers used central vision and stimuli with gaps of 18, it is not ideal for DOA patients who use larger targets and who can suffer from central scotomas. Unfortunately, the CCT cannot be modified for a larger field. Failures to isolate an aberrant cone response, however, should decrease thresholds, not increase them-as we find.

Structural Studies
The Spectralis platform (Heidelberg Engineering Ltd., Heidelberg, Germany) was used for macular SD-OCT imaging. The pink symbols denote the mean 6 1 SEM across all DOA patients (except for the lowest two radiances that could be detected only by six and nine patients, respectively), and gray symbols again denote the mean of 15 normal observers (61 SEM across observers). The Ferry-Porter slopes of mean CFF for DOA and normal observers are indicated by the blue lines. (C) The L-cone CFF frequencies at 8.5 log 10 quanta s À1 deg À2 plotted as a function of observer age for normal (gray symbols) and DOA patients (pink symbols). The solid gray and pink lines show the best fits for linear regression for normal and DOA patients, respectively-the dashed lines, the 95% confidence interval for the best-fitting linear regressions. Automated segmentation and thickness analyses were performed for perifoveal volumetric retinal B-scans using the Heidelberg Engineering segmentation tool, included in the Spectralis Glaucoma Module software (version 6.0). The combined thickness of the ganglion cell layer (GCL) and inner plexiform layer (IPL) was recorded as mean of the four sectors of the inner ring (between 1 and 3 mm in diameter) of the nine macular ETDRS subfields as described elsewhere. 38 Normative data for the GCL-IPL thickness was generated from SD-OCT images of 48 healthy eyes of 48 subjects. 38 Peripapillary RNFL thickness measurements were performed using the Cirrus HD-OCT 4000 (Carl Zeiss Meditec, Inc., Dublin, CA, USA) platform. The mean RNFL thickness was recorded.

Statistical Analyses
Mann-Whitney U independent samples test was used for comparison of distribution of continuous variables in DOA and normal observers, and in males and females, as indicated. Pearson product-moment correlation coefficient was used for the analysis of statistical dependence between various variables indicated in the Results section. Temporal functions were analyzed in DOA P1 to P11 within the data for the eyes used for the monocular temporal tests. All other variables were studied in the entire DOA study group. Binocular test data (CSF and CCT) were compared with the monocular data of the eye with the better BCVA. Monocular tests were analyzed for each eye separately.
The study was approved by local ethics committees at Moorfields Eye Hospital and at University College London and conformed to the standards set by the Declaration of Helsinki. All subjects or their parents signed informed consent forms. Figure 1A shows L-cone CFF (temporal acuity) data for the 11 DOA (P1-P11) and 15 normal observers plotted as a function of log 10 target radiance. The mean data for 15 normal observers are shown by gray symbols, and the individual data for each DOA patient are shown by the colored symbols, as means 6 1 SEM. In the normal observers, the mean lowest target radiance at which flicker was seen was 6.6 log 10 quanta s À1 deg À2 , after which the CFF rises until it approaches a plateau near 40 Hz. Most DOA patients are unable to see the flicker at such a low radiance and require, on average, a radiance of 7.4 log 10 quanta s À1 deg À2 before flicker can be observed. For them the asymptotic upper bound on flicker rate is highly variable, ranging from 24 to 35 Hz.

L-Cone Temporal Functions
The mean data (61 SEM across observers) for normal and DOA observers are shown in Figure 1B with gray and pink symbols, respectively. Over two and half log 10 units the normal CFF increases as an approximately linear function of log 10 radiance with a slope of 9.2 Hz per decade as shown by the blue line fitted to the mean normal data. This behavior is known as obedience to the Ferry-Porter law. 39,40 For the DOA patients (pink triangles in Fig. 1B), the Ferry-Porter slope, also shown by a blue line, is 7.3 Hz per decade. Slope fits were also made to the individual DOA data over the range of CFFs consistent with the Ferry-Porter law and are tabulated in Table 3. Figure 1C shows the L-cone CFF obtained at 8.5 log 10 quanta s À1 deg À2 plotted as a function of observer age for normal (gray symbols) and DOA observers (pink symbols). The solid gray and pink lines show the best fits for linear regression for normal and DOA observers, respectively, and the dashed lines indicate the 695% confidence interval for the fits. Over the range of ages observed, L-cone CFF did not show any significant age-related variation either for the DOA or for the normal observers (Pearson's r ¼ À0.474, P ¼ 0.140 for DOA observers, and r ¼À0.018, P ¼ 0.950 for normal observers). For the DOA observers (see Table 3), the L-cone plateau showed weakly significant age-related decreases (Pearson's r ¼ À0.596, P ¼ 0.053), and the Ferry-Porter slope weakly significant increases (r ¼ 0.613, P ¼ 0.045 for the Ferry-Porter slope). Table 3 summarizes the CFF data and the Ferry-Porter fits: Column 2 shows the lowest detected target radiance and the mean value (6SEM) for 11 DOA patients, and the mean value (6SEM) for the normal observers. Columns 3 and 4 show the best-fitting Ferry-Porter slope together with the R 2 value for the fit; columns 5 and 6 show the CFF obtained at 8.5 log 10 quanta s À1 deg À2 interpolated from the Ferry-Porter slope together with the plateau frequency. 33 Dominant optic atrophy patients first detected flicker at the mean radiance eight times (0.91 log unit) more intense than that for normal observers. In all DOA patients, CFF followed the Ferry-Porter law over a radiance range similar to that of normal observers, but in all patients the Ferry-Porter slope was shallower. This and the overall loss of sensitivity gives a mean reduction in CFF of 8.89 6 0.93 Hz (mean 6 SE) at 8.5 log 10 quanta s À1 deg À2 . The plateau frequency was considerably lower in patients compared to normal observers (means of 29.88 and 40.18 Hz, respectively). Figure 2 shows the logarithm of L-cone temporal modulation sensitivity as a function of temporal frequency measured with 48 targets at a time-averaged 650-nm target radiance of 10.6 log 10 quanta s À1 deg À2 for all observers. The mean data for 8 normal observers are shown in each image as gray triangles, while those for each of the 11 DOA patients are shown in the first 11 images as colored triangles. The mean data for the DOA patients are shown in the bottom right-hand image as filled triangles. The sensitivity differences between the DOA and mean normal data are shown in each image by the circles. The fits of the cubic polynomial model (described in Methods) are shown as black curves. The fitting parameters are given in Table 4.
The mean L-cone modulation sensitivity function for normal observers peaks in sensitivity at 6.48 6 0.30 Hz and falls off in sensitivity at both low and high temporal frequencies. [31][32][33] This shape of sensitivity function is known as a band-pass function; the decline in sensitivity at low temporal frequencies is often attributed to surround antagonism. 41,42 The temporal contrast FIGURE 2. L-cone modulation sensitivity. Each image shows, for a single DOA patient, the log 10 L-cone modulation sensitivities measured using a sinusoidally modulated, 650-nm 48 target with a mean radiance of 10.3 log 10 quanta s À1 deg À2 superimposed on a 98, 481-nm background of 8.3 log 10 quanta s À1 deg À2 and plotted as a function temporal frequency colored triangles. (The gray symbols in each image show the mean normal data set for comparison.) The 12th (gray) image shows the mean DOA data (black triangles). The error bars are 61 SEM either between runs for the individual patients, or between observers for the mean data. In the lower part of the first 11 images, the differences in log 10 sensitivity between each DOA patient and the normal mean data (n ¼ 8) are shown as circles. The 12th image shows the differences between the mean data (filled circles). The black curves are the fits of the cubic polynomial model described in Methods. sensitivity functions for all but one of the DOA patients (P1) were also band-pass in shape with a lower peak to the normal function (mean 6 SEM: 5.53 6 0.59 Hz), but they showed a mean peak sensitivity loss of 0.66 log 10 units (mean 6 SEM) (see image 12). The mean loss, shown in image 12, is roughly independent of temporal frequency. However, there are individual differences. The band-pass peaks for P9 and P10 are shifted to lower temporal frequencies and therefore showed bigger losses at high temporal frequencies. And the cone temporal contrast sensitivities (TCSFs) for P1 and P3 were fairly flat, so that the sensitivity relative to normal observers increased at higher temporal frequencies.
In conclusion, L-cone TCSFs were characterized by reduced temporal acuity or CFF, a mild but significant reduction in the Ferry-Porter CFF slope, and a general loss of sensitivity. L-cone temporal acuity showed weak age-related effects in terms of an increase in Ferry-Porter slope and a decrease in the plateau frequency, but CFF was approximately constant at middle intensities (see Fig. 1C).

S-Cone Temporal Function-Critical Flicker Fusion
Figures 3A and 3B show S-cone CFF data plotted as a function target radiance, and Table 5 tabulates CFF characteristics: the lowest detected target radiance, the Ferry-Porter slope, and the CFF frequency at 8.5 log 10 quanta s À1 deg À2 target radiance, for DOA and normal observers as in Table 3 but for S-cones. The data for 15 normal observers are again shown with gray symbols as mean 6 1 SEM. In the normal observers, S-cone CFF starts to rise above a radiance of 6.5 log 10 quanta s À1 deg À2 until reaching a plateau of approximately 22 Hz at 9.0 log 10 quanta s À1 deg À2 . The linear relation between CFF and radiance, consistent with the Ferry-Porter law, is 7.3 Hz per log 10 quanta s À1 deg À2 for normal observers at radiances below the plateau. (The rise after 9.9 log 10 quanta s À1 deg À2 is due to M-cones becoming more sensitive than S-cones at high target radiances. 33,43 ) The data for each DOA patient are presented with individually colored symbols in Figure 3A. The data for S-cone CFF was severely abnormal for the oldest DOA patients (P9 and P10, 52 and 54 years old, respectively), so that P9 could not detect any flicker with a 440-nm target and P10 detected flicker only at the highest radiance (a radiance for which, in normal observers, M-cones are more sensitive than S-cones).
The youngest DOA patients, in contrast, had normal Ferry-Porter slopes and near normal CFF frequencies at 8.5 log 10 quanta s À1 deg À2 as shown in Figure 3C where the S-cone CFF frequencies at 8.5 log 10 quanta s À1 deg À2 are plotted as a  The S-cone CFF frequencies at 8.5 log 10 quanta s À1 deg À2 plotted as a function of observer age, and the best fits for linear regression (solid lines) and 95% confidence intervals (dashed lines), for DOA patients (cyan symbols and lines) and for normal observers (gray symbols and lines).
function of observer age for normal (gray symbols) and DOA (cyan symbols) observers. S-cone CFF at 8.5 log 10 quanta s À1 deg À2 showed a significant age-related variation for the DOA patients (Pearson's r ¼ À0.920, P < 0.001) but not for the normal observers (Pearson's r ¼À0.326, P ¼ 0.236). Also S-cone plateau frequency at 9.2 to 9.4 log 10 quanta s À1 deg À2 target radiance and the lowest detected radiance showed significant age correlation (Pearson's r ¼À0.838, P ¼ 0.005 for plateau and Pearson's r ¼ 0.667, P ¼ 0.050 for lowest radiance). (Severely impaired S-cone temporal acuity associated with lowered Sand M-cone-related chromatic sensitivities as will be presented in detail later on). The mean CFF data for the DOA patients are shown in Figure 3B (cyan triangles). Figure 4 and Figure 5A show achromatic SCSFs, that is, sensitivity plotted in log 10 units as a function of spatial frequency (cyc/deg) plotted on a logarithmic scale. The mean data for 15 normal observers are shown with gray symbols and have a peak mean contrast sensitivity of 2.19 6 0.03 log 10 (mean 6 SEM) at a mean spatial frequency of 2.18 6 0.15 cyc/ deg (mean 6 SEM). Like the TCSF, the SCSF also has a bandpass shape with losses at lower spatial frequencies probably resulting from surround inhibition. 41,44,45 The response at the highest spatial frequencies is limited jointly by the modulation transfer function of the optics and the size of the receptive field center, so that only neurons with the smallest receptive fields are able to respond to the highest frequencies. [46][47][48] The data for DOA P1 to P25 are shown with individually colored symbols as mean 6 1 SEM in the upper part of each image (colored triangles), together with normal data (inverted gray triangles). Differences in sensitivity between individual DOA patients and normal are shown as colored circles in the lower part of each image. There is little loss in sensitivity at low frequencies for the DOA patients, with the exception of P8, P9, P18, P19, and P24, but all DOA patients show substantial losses relative to normal observers at high spatial frequencies.

Achromatic Spatial Contrast Sensitivity Function
The fits of the cubic polynomial functions are shown as black curves in Figures 4 and 5. The maximum visible frequency is taken as the zero crossing of the cubic function, and the maximum sensitivity and peak frequency can be found as the local maximum in the cubic function, if it exists. Table 6 tabulates these values. Note that for nine patients (denoted by asterisk) the TSCF appears low-pass with no discernible peak in sensitivity, and in these cases, for analysis purposes, the peak frequency was defined as the lowest frequency tested (0.5 cyc/ deg) and the peak sensitivity was simply the sensitivity at that frequency. Figure 5A presents the mean achromatic SCSF for the entire 25 DOA patient group. The patients have a significantly lower achromatic spatial contrast sensitivity than normal with a mean peak sensitivity of 1.42 6 0.07 log 10 (mean 6 SEM) shifted to the spatial frequency of 0.72 6 0.06 cyc/deg (mean 6 SEM). The sensitivity loss increased nearly linearly toward high spatial frequencies, and mean CSF function was low-pass or band-pass for the DOA patients instead of being consistently band-pass in the normal observers. The difference in peak sensitivity for patients compared to normal observers was not correlated with the sensitivity losses found in the TCSF data (Pearson's r ¼ 0.284, P ¼ 0.426).
The spatial acuity at maximum contrast derived from the model fits for each is plotted in Figure 5B. The highest observed spatial frequency under maximal contrast (corresponding to zero sensitivity) relates to spatial visual acuity expressed as logarithm of the minimum angle of resolution (logMAR), so that 0 logMAR corresponds to 30 cyc/deg. 49 This value for spatial frequency corresponds to the extrapolated intercept of the contrast sensitivity function at zero sensitivity. Figure 5A shows the mean achromatic SCSF for DOA and normal observers and extrapolation of SCSF to zero sensitivity at the intersection with the horizontal gray dashed line. Figure  5B shows the spatial frequencies obtained by extrapolation of SCSF of each observer at zero sensitivity plotted as a function of BCVA expressed in logMAR units. In normal observers with normal visual acuities (0 logMAR), the extrapolated SCSF at zero sensitivity had median intercept at 23.8 6 1.09 cyc/deg (median 6 SEM). In the DOA patients with BCVA better than 1.0 logMAR, the SCSF intercepts correspond roughly with the BCVA values. The logMAR is a reasonable predictor of spatial acuity below 0.75, but not above it. There was no sex-related variation in the BCVA (Mann-Whitney U test, P ¼ 0.210).
The achromatic SCSF has a spatial frequency-dependent contribution from several RGC subpopulations: (1) Midget cells with the smallest receptive field probably mediate the highest spatial frequencies 44,48 ; (2) OFF midget cells may have more significant contribution at low spatial frequencies than ON midget cells 45 ; (3) parasol RGCs are also sensitive to low spatial frequencies. 50 We analyzed the statistical dependence between the nonspatial variables (age, the structural findings, visual field defect, temporal data, chromatic data) and the spatial losses at low (0.75 cyc/deg) and higher (4 cyc/deg) spatial frequencies separately. Losses at both frequencies, however, showed similar associations with the field and color vision defects and the structural variables, and no correlation with the temporal data (studied within the 11-patient subgroup) or dependence on age. Figure 6 shows the vector lengths of the three confusion lines (protan, deutan, and tritan) of the CCT in 10 À4 u 0 v 0 units (the CIE 1976 u 0 v 0 color space) for 21 DOA and 15 normal observers presented as a function of observer's age. The longer the vector length, the more saturated color was required for color discrimination, with the maximum saturation corresponding to the vector length of 1600 10 À4 u 0 v 0 units (the CIE 1976 u 0 v 0 color space). In the normal observers (smaller partially transparent symbols), the vector lengths were similar to those previously reported using the  Figure 6. Two of them, aged 34 and 65 years, carried OPA1 c.733A>T mutation, and the third patient, aged 39 years old, OPA1 c.112C>A mutation. In addition, OPA1 c. 1296delCAT mutation associated with early loss of red-green color vision, unrelated to the visual acuity or field losses, and with an equal loss of color discrimination in all three confusion lines, whereas in the other genotypes the tritan axis was always more severely affected. There was thus evident genotyperelated variation in the chromatic discrimination. However, in general, color vision deteriorated with increasing patient age (Pearson's r ¼ 0.545, P ¼ 0.011 for protan, r ¼ 0.581, P ¼ 0.006 for deutan, and r ¼ 0.710, P < 0.001 for tritan axes). There was no sex-related variation in the chromatic discrimination (Mann-Whitney U test, P ¼ 0.891 for protan, P ¼ 0.891 for deutan, and P ¼ 0.945 for tritan axes).

Chromatic Discrimination
For comparison of regression rates between color vision along the tritan confusion line and S-cone temporal acuity, data for the 11 DOA patients undergoing both tests were converted to percentage of the normal values and referred to the observer age. Both short wavelength-related visual functions showed similar regression (Pearson r ¼ À0.734, P ¼ 0.010 for tritan vector length, and r ¼ À0.856, P ¼ 0.001 for S-cone temporal acuity) corresponding to annual deterioration of 1.9% for both functions.

OCT Structural Data
The mean macular GCL-IPL thickness in 50 eyes of 25 DOA patients was 39.7 6 1.10 lm (mean 6 SEM), that is, 42% of the normal, 93.5 6 1.4 lm (mean 6 SEM). The mean peripapillary RNFL thickness was 61.9 6 1.2 lm (mean 6 SEM), corresponding to 63% of the normal and to that previously reported in another cohort of nonsyndromic DOA. 52 Figure 7 shows the thickness of the GCL-IPL complex measured from the macular SD-OCT images of 50 eyes of 25 DOA patients (pink triangles) and of 48 normal eyes (gray inverted triangles) plotted as a function of age. Dominant optic atrophy patients had significantly thinner GCL-IPL complex than normal at all ages, but the best fit for linear regression of the GCL-IPL thickness both for normal subjects (Pearson's r ¼ À0.470, P < 0.001) and for DOA patients (r ¼ À0.685, P < 0.001) corresponded to 0.2 lm loss and 0.2% per year. Dominant optic atrophy patients did not show any sex-related variation in the GCL-IPL thickness (Mann-Whitney U test, P ¼ 0.240).

DISCUSSION
Visual impairment in DOA was characterized by losses that can be associated with all three visual pathways, but with inhomogeneous severity and age-related variation. The loss of midget RGCs could account for the loss of high spatial frequency sensitivity that was markedly subnormal in most DOA patients at all ages. It could also account for the loss of chromatic resolution along the protan and deutan color confusion lines, which showed clear deterioration with increasing patient age. By contrast, the loss of parasol RGCs, which could account for the loss of high temporal frequency acuity, was mild and without significant age-related variation. The most striking age-related visual losses, however, were observed for the S-cone-mediated chromatic and temporal sensitivities, which could result from the loss of small bistratified RGCs.
Visual perception results from the sequential and parallel transmission and processing of signals from the photoreceptors through the visual system. In DOA, there is no evidence of functional or structural abnormalities in the outer retina 7,53-55 ; and although a subgroup of patients carrying pathogenic OPA1 mutations can develop a more severe neurologic phenotype and nonspecific white matter cortical abnormalities, the neuropathology in the majority of patients is limited to the optic nerve. 56 The psychophysical abnormalities identified in our patient cohort could therefore be attributed to RGC loss or dysfunction. At this level of the visual pathway, perception depends on the parallel transmission of signals by different subpopulations of RGC types that encode different characteristics of visual target. 44,57 In DOA patients, the parasol RGCrelated visual functions showed a lowered temporal acuity, a slower increase in acuity with increasing radiance than normal, and a roughly constant sensitivity loss across temporal frequencies. Could these findings be attributed to loss of a particular population of RGCs (e.g., the parasol cells) or to functional abnormalities within surviving members of that population? Psychophysical studies following a magnocellular lateral geniculate lesion in macaques revealed TCSF losses that increase with temporal frequency, but with some preservation of low frequency sensitivity, thus changing the shape of the TCSF from band-pass to low-pass. 58,59 The pattern of magnocellular visual function in DOA differs from that reported for the anatomic loss of the magnocellular pathway. In contrast, nearly identical temporal losses to that observed in DOA have been described for single parasol RGC firing pattern in experimental primate glaucoma. 60 In this model there was loss of the temporal acuity by 10 Hz, attenuation of the temporal sensitivity in the most dysfunctional RGCs also at low frequencies, and preservation of the band-pass sensitivity function. The extent of the functional abnormalities correlated with the loss of normal RGC dendritic arborization that is thought to represent the first sign of RGC degeneration in glaucoma. 61 A similar structural feature has also been found in the B6;C3-Opa1 Q285STOP mouse model as a precursor of RGC degeneration. 62 The architecture of dendritic arborization in patients with OPA1 mutations is not known, but our data support the idea that loss of dendritic arborization may be involved in the pathogenesis of DOA rather than just population loss.
Previous psychophysical studies in nonhuman primates on the effect of the parvocellular lateral geniculate lesion have shown losses of the high and low spatial frequency contrast sensitivity and a change from a band-pass to a low-pass SCSF consistent with detection by the magnocellular pathway. 63 Achromatic spatial contrast sensitivity in our DOA patient cohort had similar characteristics. However, experimental models of parvocellular pathway lesion also resulted in complete loss of color vision, 58 which was observed in only a few DOA patients carrying a particular genotype (genotype 10, c.733A>T), or with the most advanced disease (genotype 11, c.112C>A). This suggests that the parvocellular lesion in DOA preserves sufficient midget RGCs to support chromatic resolution along the protan and deutan axes, 64 but because of progressive loss there is also a steep age-related regression. Our results are in line with the report by Reis and colleagues 28 that showed increasing contrast thresholds at higher spatial frequencies for their DOA patients, although less than half of their cohort carried confirmed pathogenic OPA1 mutations. Our study also supports the structural and functional evidence pointing toward the midget cells as being the mainly affected RGC type in DOA, such as the predominance of the midget cells in the macular RGC population subserving central vision, and the preferential loss of the small-diameter axons in postmortem histopathologic studies 65,66 and in a recent OPA1 mouse model. 67 Toxic and inflammatory insults to midget RGCs do not result in loss FIGURE 7. Macular ganglion cell-inner plexiform layer (GCL-IPL) thickness (lm) for 50 eyes of 25 DOA patients (pink symbols) and for 48 healthy eyes (gray symbols) plotted as a function of subject age and including lines for the best fits of linear regression and 95% confidence intervals. * SCSF is not band-pass so ''peak frequency'' set to lowest measured and ''peak sensitivity'' set to the measured sensitivity at this frequency.
† Fitted function does not have a plausible value of the zero crossing, so maximum visible spatial frequency is undefined. of mainly the high spatial frequencies, [68][69][70] which implies that the preferential loss of the small receptive field midget cells may be specific to DOA.
The most extensive visual impairment was loss of S-cone chromatic and temporal sensitivities in the oldest DOA patients. This observation could be related to dysfunction of the small bistratified RGCs, which support S-cone vision (see above). Unlike normal age-matched observers, this group of DOA patients had unmeasurable chromatic resolution along the tritan confusion line and they were unable to detect flickering short-wavelength light. Loss of perception of the blue colors in DOA was also among the early observations by Kjer 7 and the most frequently reported color vision defect in this disorder. 1,4,[25][26][27][28][29] Reports showing clear age-related variations are, however, scarce; and this could be explained by the differences in the testing conditions used. The Trivector test procedure of the CCT enables better discrimination and quantitative assessment of the individual color axes than the elliptical test procedure previously used in another study. 27 Furthermore, our modifications of the viewing distance and response time, in turn, enabled use of the test for patients with varying impairment of the spatial resolution. Temporal sensitivity test of the koniocellular pathway (i.e., S-cone CFF) introduced an independent method for the assessment of the small bistratified RGC function. Importantly, this modality showed loss with increasing patient age identical to that observed for chromatic sensitivity.
A cross-sectional study does not allow disease progression to be followed at the individual level, but a prospective longterm longitudinal study in DOA is challenging. Despite these caveats, we carefully designed our study to include DOA patients covering a broad range of disease duration, and this approach enables us to speculate on a particular pattern and chronology of RGC loss in this disorder. Firstly, the macular GCL-IPL thickness in DOA patients was less than 50% of normative values and with a similar rate of decline with increasing age compared with healthy controls. Similar findings have been reported in other DOA cohorts, [9][10][11] and taken together, the evidence indicates that a major proportion of the RGC loss has already occurred in early childhood, and probably starting in utero. If the remaining 50% of the GCL-IPL thickness is liberally converted to RGC density, that is, 50% of the foveolar maximum, this corresponds approximately to the RGC density found at a foveolar eccentricity of 78 in normal retina. 48 Such an RGC density enables spatial resolution of 9 cyc/deg, 48 which could be used as an estimate for the theoretical upper limit of spatial visual acuity in DOA. Interestingly, in the majority of our DOA patients, the spatial acuities were of that level, or below, leading to an interpretation that the high-density small receptive field midget RGCs could constitute the majority of the 50% RGC population that has already been lost.
Parasol and small bistratified cells form only a small proportion of the total RGC population, and detecting their loss may not be possible with current in vivo imaging techniques, requiring instead more sensitive functional psychophysical testing methods. Our results indicate that parasol RGCs are functionally compromised in DOA, but there is no clear evidence of continuing cell loss. In contrast, the functioning of the koniocellular S-cone pathway seems to rapidly deteriorate with increasing patient age, an observation that is consistent with the ongoing loss of small bistratified RGCs. The majority of patients with DOA will experience gradual loss of visual function, and although this provides an extended window of opportunity for therapeutic intervention, designing cost-effective clinical trials to detect a favorable outcome measure within a reasonable time frame is challenging. Although a prospective longitudinal study is needed to confirm our findings for individual OPA1 mutation carriers, monitoring koniocellular pathway function with S-conemediated chromatic and temporal sensitivities could potentially provide useful biomarkers of disease progression in DOA.