1. INTRODUCTION

Temporal contrast sensitivity measurements are valuable for basic research, for clinical research, and potentially also for clinical practice. For clinical applications, its special value lies in its direct relation to neural processing and the independence from changes in the optic media [1]. The value of isolating the different “channels” of temporal vision (i.e., color channel or luminance channel) enhances its value in basic research and shows great promise in clinical applications [2]. We have shown previously that isolation of L- and M-cone-driven temporal contrast sensitivities may be applicable in a clinical setting [3]. The isolation of S-cone- and rod-driven sensitivities is generally considered more difficult due to its larger interference with prereceptoral filters, such as the lens and the macular pigment.

The S-cones constitute only about 7%–10% of the cone population, and they are absent from the center of the fovea [4,5]. They feed into the koniocellular pathway, which conveys blue–yellow chromatic information with limited spatial resolution. The temporal resolution of S-cone-driven signals is also limited with a maximal critical flicker fusion (CFF) of 18–28 Hz [6–8], although there is some evidence that they may have a limited input to the luminance pathway, and thus the ability to resolve more rapid flicker [9]. S-cone-driven sensitivities have been known to be more vulnerable to damage and disorders than those driven by the L- and the M-cones [10], although the age-related threshold elevation in S-cones is more likely to be caused by changes in the optic media [11].

Rods are responsible for vision at low light levels. Signals are conveyed via two pathways [12,13]: the slower one via rod bipolars and AII Amacrine cells; the faster pathways via gap junctions with cones [14]. Rods are often differentially affected compared with cones in hereditary retinal disease, like retinitis pigmentosa, but testing rod function may also be useful in acquired conditions such as age-related macular degeneration [15,16]. Current methods to measure rod function require prolonged periods of dark adaptation. This limits application and may introduce errors into the measurements. Recently, it has been demonstrated that rod-driven electroretinograms (ERGs) may be measured without extensive dark adaptation under mesopic conditions using silent substitution stimuli [17]. This opens up the perspective that rod-driven sensitivities may also be measured in psychophysical experiments without the need for extended dark adaptation periods. Silent substitution stimuli could then be useful also for clinical studies where genotype-phenotype correlations are investigated [2] and for the monitoring of novel therapies (such as gene therapy).

For the isolation of single photoreceptor types, silent substitution may be superior to paradigms that so far have been used in clinical applications. The silent substitution paradigm allows creating stimuli that stimulate specific classes of photoreceptors without changing the mean state of adaptation [18], which is then an independent parameter that is under the control of the experimenter. Furthermore, with the silent substation paradigm, stimulus strength can be quantified in terms of Michelson contrast for all photoreceptor classes (L-, M-, and S-cones, rods) and allows quantifying detection thresholds and sensitivities.

However, a major drawback has been the need for observer calibration, which is time-consuming and simply not feasible for untrained observers [19,20]. Isolated responses of the S-cones and rods may be particularly difficult to obtain [21] because L- and M-cones are more sensitive than the other photoreceptor types under most light-adapted conditions.

The aim of this study was to evaluate the feasibility of a paradigm for measuring S-cone- and rod-driven temporal contrast sensitivities in normal subjects without observer calibration and without the need of extended dark adaptation periods to obtain rod-driven sensitivities. A second objective was to identify stimulus conditions in which rod and S-cone isolation can be confidently assumed, as well as conditions where intrusion may disturb measurements. To achieve this, contrast sensitivities for the photoreceptor-driven signals were measured over a broad range of mesopic conditions and temporal frequencies with silent substitution conditions to describe the temporal dynamics and adaptational behavior of photoreceptor-specific sensitivities. Furthermore, we present data from an S-cone monochromat. Finally, data from a control bleaching experiment with subsequent sensitivities to rod-isolating stimuli are presented.

2. METHODS

A. Apparatus

We used a two-channel, four-primary LED stimulator with a Maxwellian-view optical system that has been described in detail (including a construction plan and a photograph of the device) elsewhere [22].

The stimulus geometry consists of a small circular field (2° diameter) and a larger surrounding annular field (2° inner diameter and 13° outer diameter), each being controlled by four LEDs with peak spectral outputs at 660 nm (red), 558 nm (green), 516 nm (cyan), and 460 nm (blue). Interference filters are used to narrow the bandwidths at half-height to 8–10 nm.

The eight channels of a sound card (Asus Xonar D2-PM, ASUSTek, Taipeh, Taiwan) were used to modulate the luminance of each LED independently. The method of how the sound card controlled the LED output was described previously [23]. Briefly, the sound card produced an amplitude-modulated signal with a carrier frequency of 20 kHz, which was subsequently converted into a frequency-modulated signal (up to 250 kHz) using a voltage frequency converter (Texas Instruments VFC320CP).

To calibrate the retinal illuminance produced by the LEDs, we used the method of Nygaard and Frumkes [24]. The spectral outputs were measured at 2-nm intervals with a CAS 1401 spectroradiometer (Instrument Systems, Munich, Germany). These measurements were shown previously [3], where a detailed description of the calibration procedure can also be found. Fourth-order polynomials were used to linearize the LED outputs.

B. Stimuli

To minimize influence of the macular pigment on the fundamentals, we used the outer annular field as a test field, in which we presented photoreceptor-specific modulations. The mean chromaticities of center and surround fields were white with equal CIE coordinates (Commission Internationale de l’Eclairage; $x=0.38$, $y=0.28$).

The maximal time-averaged retinal illuminance produced by the surround field was 587 photopic troland (phot Td). To control this mean retinal illuminance neutral density filters in steps of 0.3 up to 3.3, log units were inserted into the optical pathway. The central field served as fixation target that was not modulated and set to half the time-average luminance of the surround field. Subjects were instructed to shift fixation within the 2° field to avoid Troxler’s fading. The presence of a dim center stimulus also helps to minimize the sensitivity to stray light that may otherwise stimulate a very sensitive, dark-adapted central retina. No subject reported a counterphase flicker of the central field that has been observed under certain conditions in the past [25]. To create silent substitution stimuli, LED luminances were modulated with fixed ratios of modulation depth at equal temporal frequencies either in phase or in counterphase. During the psychophysical procedure, the contrasts in the four LEDs were altered in concert without changing the modulation ratio, ensuring that contrast was changed only in the isolated photoreceptor type and remained 0% in the other three types. We used the Stockman and Sharpe 10° cone fundamentals [26] for cones and the scotopic luminous efficiency function for rods (available at [27]). The maximally achievable contrast (gamut) of the stimulator was 24.95% L-cone contrast, 22.33% M-cone contrast, 82.75% S-cone contrast, and 27.30% rod contrast. The exact conditions and the gamuts are shown in Table 1.

Table 1. LED Contrasts at the Instrument Gamut for the Four Different Isolating Stimuli^a

View Table | View all tables in this article

C. Procedure

Subjects viewed the stimulus through an artificial pupil of 3 mm diameter. The artificial pupil was always smaller than the natural pupil. Temporal contrast thresholds for the different photoreceptor classes were determined using a modified PEST procedure (parameter estimation by sequential testing) [28]. The subjects reported whether they perceived flicker (either color or luminance) by pressing a button on a game pad (yes/no paradigm). A forced-choice paradigm would have further reduced response bias, but was impractical to implement with the stimulator. Thus, it cannot be completely ruled out that subjects adopted different response criteria at different temporal frequencies [29]. We used a staircase algorithm with two randomly interleaved staircases. One staircase started at maximal and the other at 0% contrast. Thus, it was nearly impossible even for a trained observer to guess which contrast change could be expected next. If the subject did not perceive flicker, then the contrast was increased; otherwise it was decreased. The initial contrast change was 20% of the maximal modulation, and this step size was halved after each change from detection to no detection and vice versa. Stop criterion was a step size of less than one-seventh of the current photoreceptor modulation depth after at least two changes in direction. If the subject indicated no flicker at maximal contrast three times, it was assumed that a threshold could not be reached within the gamut of the stimulator. Consequently, the procedure was interrupted. Also, the procedure was terminated if the subject reported flicker at zero contrast three times. Temporal contrast sensitivity was defined as the inverse of the contrast at threshold.

D. Ethics Statement

The study was approved by the local ethics committee and adhered to tenets of the Declaration of Helsinki. All subjects gave written informed consent.

E. Bleaching Adaptation

We verified that perception of the rod stimuli was indeed mediated by rods by demonstrating differences in the dynamics of the recovery of contrast sensitivity for rod and cone stimuli after adaptation to bleaching light [30,31]. The two authors participated as normal subjects in these experiments (CH, male, 35 years old and JK, male, 54 years old). Both subjects had normal color vision as established with Ishihara plates and the anomaloscope, normal best-corrected acuities, inconspicuous visual fields on static perimetry (Octopus G1, Haag-Streit, Switzerland), clear optic media, and normal findings on funduscopy. In both subjects, only the right eye was examined. Bleaching was carried out using a Ganzfeld bowl (Q450SC in a Retiport system, Roland Consult, Germany). After adaptation to bleaching condition for either 2 or 4 min (see below), the subjects returned to the LED stimulator within approximately 30 s, and flicker thresholds were continuously measured back-to-back over a time span of 15 to 20 min. These thresholds indicate the modulation contrast around a mean retinal illuminance at threshold, not absolute detection threshold. This was repeated for different mean retinal illuminances, for different frequencies, and for different isolating stimuli (L-cones, M-cones, and rods).

For subject JK, baseline measurements were performed three times for each stimulus type (L-/M-cones and rods) before he was light-adapted to bleaching conditions using a 50,000 phot Td light for 4 min (pupils dilated with phenylephrine and 0.5% tropicamide). Measurements were only performed at 4 Hz and 0.5 phot Td.

For subject CH, measurements were performed at two different frequencies and three different light levels. During the bleaching condition, he was adapted to a 10,000 Td light for 2 min (pupils were not dilated). Dark adaptation was measured at two different frequencies (4 and 10 Hz), and three different retinal illuminance levels (0.3, 9.3, and 294 phot Td) for rod stimuli, and at 4 Hz and 9.3 Td for L- and M-cone stimuli.

F. Temporal Contrast Sensitivities at Different Frequencies and Retinal Illuminances

The two authors and one additional normal observer (LH, female, 23 years old), who was naïve concerning the goals of the measurements, participated in the main experiments. The third participant underwent the same examinations as described above to assure normal observer status as the other two. The number of subjects was relatively small because of the extensive psychophysical testing protocol. Measurements were performed at three to four different light levels per session, starting at a maximal illuminance of 587 phot Td. In subsequent measurements, retinal illuminance was reduced in 0.3 log steps to the minimal level of 0.07 phot Td using neutral-density filters. The subjects were allowed to insert breaks as wished. At the beginning of each session and after each break, the subjects adapted in the dark room for at least 15 min prior to performing the experiments. After each change in retinal illuminance, they adapted for only approximately 2 min to the new condition because the change in retinal illuminance was small. For each retinal illuminance, frequency (1, 2, 4, 6, 8, 10, 12, 16, 20, and 28 Hz) and isolated photoreceptor type (L-, M-, S-cones, and rods) were varied independently in a randomized fashion. For retinal illuminance, the systematic change was chosen instead of a random paradigm to reduce adaptation time.

G. S-Cone Monochromat

We measured sensitivities in one subject with S-cone monochromacy (male, 22 years old). This subject had nystagmus and a reduced visual acuity (Snellen visual acuity of 0.15, corresponding to 20/135). According to the patient, the diagnosis of S-cone monochromacy was ascertained genetically at another hospital. We, however, did not have access to the results of the genetic test. We confirmed the plausibility of this diagnosis with the anomaloscope using the Rayleigh equation and with a full field ERG following the standards of the International Society for Clinical Electrophysiology of Vision (ISCEV). The subject was dark-adapted for 15 min, and then we measured temporal contrast sensitivities for different isolating stimuli in a randomized fashion at 10 different frequencies (1, 2, 4, 6, 8, 10, 12, 16, 20, and 28 Hz), as in the normal subjects. The time-averaged retinal illuminances were 0.59, 9.3, and 294 phot Td. At the highest retinal illuminance, he was affected strongly by glare.

H. Data Analysis

Data analysis was performed using R (R: a language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria). Figures were created using the package ggplot2 [32].

1. Temporal Contrast Sensitivity Functions (DeLange Curves)

The temporal contrast sensitivity functions (tCSFs) show the contrast sensitivity as a function of temporal frequency [33]. These curves were analyzed qualitatively by comparing the shapes of the functions at different retinal illuminances and by classifying them as low-pass or bandpass.

2. Sensitivity-versus-Intensity Curves

Sensitivity-versus-intensity (SVI) curves show the temporal contrast sensitivity as a function of retinal illuminance for a given frequency. Contrast modulation is used instead of absolute amplitude, as was the case in threshold-versus-intensity (TVI) curves shown in earlier studies [33]. In TVI curves, normally three different adaptational behaviors can be observed: (1) a linear range, where absolute amplitude of modulation at threshold is constant and the TVI functions have slopes of 0, (2) the DeVries-Rose range with slopes of 0.5, and (3) the Weber range with slopes of $+1$ (here, amplitude at threshold is proportional to retinal illuminance) [33]. SVI curves have slopes of zero when Weber’s law holds and a slope of $-1$ in the linear range.

I. Estimation of the Quality of Isolation

Calculations were performed to estimate a possible residual modulation contrast in the silenced photoreceptor types when the subjects’ individual fundamentals or prereceptoral filters deviated from the normal observer’s. First, LED contrasts were calculated using the standard cone fundamentals. Using these LED contrasts, rod and cone contrasts were calculated using modified fundamentals in which we simulated the adjustments generated by certain types of deviation. Cone and rod fundamentals were modified in several ways: (1) by shifting them 2, 4, or 6 nm toward shorter or longer wavelength, (2) by multiplying them with macular pigment spectral absorption curves with an macluar pigment optical density (MPOD) at 460 nm of $-0.3$, $-0.2$, $-0.1$, 0.1, 0.2, or 0.3 [34], or (3) by correcting for the aging of the ocular media as described below [35]. We have not performed calculations for shifted-peak absorption of the rods because there are no known polymorphisms of the rhodopsin [36]. For age-dependent changes of the ocular media, we calculated “aphakic” cone fundamentals by eliminating lens absorption from the Stockman and Sharpe 10° cone fundamentals and the rod fundamentals [26,37,38]. This was achieved by dividing the latter by the lens transmission spectrum $(={10}^{-\text{lens density}})$ [39]. These data were obtained from [27]. Then, the age-related optical densities of the optical media according to Eq. (1) for large field sizes from Van de Kraats and Van Norren were calculated, converted to transmission and multiplied with the “aphakic” cone fundamentals to estimate the age-related cone fundamentals [35]:

The results of these calculations are shown in Table 2. The L- and M-cone-isolating stimuli are relatively robust against variability in prereceptoral filtering. However, deviations in the spectral sensitivity of the L-cone opsin may cause intrusion of L-cone signals in some experimental conditions. We have shown earlier that it may result in flicker detection of M-cone-isolating stimuli in deuteranopes, which, however, is not relevant in trichromats [3]. S-cone- and rod-isolating stimuli are more vulnerable to variability in optical density of the macular pigment and the lens. Although large deviations from the assumed fundamentals are necessary to cause intrusion by L- or M-cones in S-cone-isolating stimuli, residual rod modulation may occur. In contrast, optical density changes may induce residual modulation in M-cones in rod-isolating stimuli.

Table 2. Calculation of Residual Modulation in the Targeted and the “Silenced” Photoreceptor Types When Assumptions That Were Made for the Calculation of the Stimuli Do Not Hold^a

View Table | View all tables in this article

3. RESULTS

The tCSFs of the different photoreceptor classes at different light levels are presented in Fig. 1. We have described the tCSFs of the L- and M-cone-isolating stimuli before [3] and show these only for reference. All three subjects had similar tCSFs, which indicates that interindividual differences in preretinal absorption had only a minor influence on the results [despite the wide age range of the subjects that may have caused substantial differences in lens optical density (OD)].

 

Fig. 1. tCSFs of three normal, trichromatic subjects for S-cone- (blue triangles, dotted lines) and rod-isolating stimuli (black diamonds, broken lines) at the different retinal illuminances. As a reference, the tCSFs for L-cones (red squares) and M-cones (green circles) are shown in the background [3].

Download Full Size | PDF

A. S-Cones

Compared with the L- and M-cone stimuli, S-cone stimuli were detected only at substantially higher contrasts, and the sensitivities were accordingly smaller, particularly at low temporal frequencies. When the central field was set to a similar retinal illuminance as the annular test field, the border seemed much blurrier than for L- and M-cone-isolating stimuli. This observation is also compatible with S-cone mediated detection [40]. Furthermore, sensitivities to S-cone-isolating stimuli were also lower compared with those of rod-isolating stimuli. Only at low frequencies and retinal illuminances above 37.02 phot Td, were they slightly higher than for rods. In contrast to the more complex shape of the L- and M-cone tCSFs, the S-cone-driven tCSFs were low-pass.

The SVI curves are shown in Fig. 2. Although the absolute sensitivity was lower, the curve forms were quite similar to those of the L- and M-cones with an ascending portion at low and intermediate retinal illuminances and a horizontal portion (indicative for Weberian properties) at high retinal illuminances. S-cones were in the Weber range at retinal illuminances between 37.02 and 589.69 phot Td at most frequencies. The slopes of the SVI curves were much steeper than those of the rods below the Weber range (see below). Therefore, detection on the basis of residual rod responses is unlikely.

 

Fig. 2. SVI curves for the S-cone-isolating stimuli at different temporal frequencies (blue triangles, dotted lines). This plot shows the averaged data of the three color normal observers (error bars: standard deviation). The SVI curves for L- and M-cones are shown for reference (L-cones, red squares; M-cones, green circles) [3].

Download Full Size | PDF

1. Combined Linear Model

We used a combined linear model to characterize the different patterns of the SVI curves of the different photoreceptor types (fits are shown with the corresponding SVI curves). Especially for cone-isolating stimuli, the SVI curves consisted of two parts: at low and intermediate retinal illuminances, there was a linear increase in sensitivity $S$ with increasing illuminance $I$ in the double logarithmic plot, whereas sensitivities were constant at retinal illuminances above a certain threshold $I_{\text{weber}}$ (consistent with Weber’s Law). We formalized this as follows:

The parameter $a$ is the slope of the ascending portion, and $b$ is the offset. $I_{\text{weber}}$ is the retinal illuminance above which Weber’s law holds. The results of the fits are plotted in Fig. 3. $I_{\text{weber}}$ was generally larger for S-cone-isolating stimuli compared to L- and M-cone-isolating stimuli at low to intermediate frequencies (Fig. 3, upper plot). Sensitivities, $I_{\text{weber}}$ for S-cone-isolating stimuli increased less at high frequencies than for L- and M-cones, but the fits were much less reliable due to the small number of retinal illuminances where a sensitivity could be measured. The slopes of the S-cone-driven SVIs were relatively constant at low and intermediate frequencies, somewhat smaller than those of L- and M-cones between 1 and 6 Hz and comparable between 10 and 16 Hz.

 

Fig. 3. Parameters obtained from fits of a combined linear model to the SVI data. The fits are shown together with the SVI curves in Figs. 2 (for S-cones) and 6 (for rods). The upper plot shows the estimated Weber threshold of the model for L-, M-, and S-cones as a function of temporal frequency (L-cones, red squares; M-cones, green circles; S-cones, blue triangles). The lower plot shows the slope of the ascending portion of the combined model from L-, M- and S-cones and of a simple linear model for rods (black diamonds). The latter was chosen because the rod data were not well described by the combined linear model under most circumstances.

Download Full Size | PDF

B. Rods

1. Dark Adaptation after Bleaching

The dark adaptation curves for subject JK are shown in Fig. 4. He was unable to see the test field for a short time span after bleaching. Shortly after reappearance of the test field, he was able to perceive L- and M-cone flicker. Between 4 and 10 min after the start of dark adaptation, contrast thresholds for L-cone and M-cone modulation were close to baseline. In contrast, rod flicker was visible only after 8 min of dark adaptation and had not reached baseline after 15 min.

 

Fig. 4. Dark adaptation after bleaching. The contrast threshold to a sinusoidal stimulus at low mesopic retinal illuminance (mean: 0.59 phot Td) is shown as a function of time after bleaching with an intense (50,000 phot Td) white light for 4 min (subject JK; L-cones, red squares; M-cones, green circles; rods, black diamonds). Exponential functions were fit to the threshold data. The baseline thresholds before bleaching are shown as horizontal lines (L-cones, red solid line; M-cones, green dashed line; rods, black broken line).

Download Full Size | PDF

 

Fig. 5. Recovery of contrast threshold with time for rod-isolating stimuli at 4 Hz and 10 Hz and at different retinal illuminances (mean: 0.59 phot Td) after bleaching with a 20,000 phot Td white light for 2 min (subject CH). Again, exponential functions were fitted to these measurements.

Download Full Size | PDF

In subject CH, dark adaptation for rod-isolating stimuli at different illuminances of the stimulus were measured at 4 and 10 Hz. Generally, dark adaptation was slower for the 10 Hz stimuli than for the 4 Hz stimuli. Furthermore, dark adaptation became faster with increasing time-averaged retinal illuminance of the stimulus. Higher time-averaged retinal illuminances result in much higher absolute modulation at the same contrast levels. As anticipated for rod-isolating stimuli, there was no evidence of a rod-cone break in any of the dark-adaptation curves.

Owing to the absence of biphasic curves, all dark adaptation curves were fitted with a simple exponential model [Eq. (3)]:

The data from subject JK revealed differences in the coefficient $\tau$ between the different photoreceptor types (see Table 3). The coefficient $\tau$ was 2.37 for rods, 2.51 for M-cones, and 1.01 for L-cones. Table 3 also shows data from subject CH at higher retinal illuminance. In contrast, Table 4 shows rod adaptation data of subject CH at different retinal illuminances and temporal frequencies. However, the model was not well constrained under most conditions. Therefore, the number of free parameters had to be reduced and a $\tau$ coefficient of 2.42 was chosen, because it was the result of the fit to the rod data at 9.3 phot Td and 4 Hz, where the curves were well constrained. Under this assumption, it was possible to make satisfactory fits with initial and final thresholds as free parameters.

Table 3. Parameters of the Exponential Functions That Were Fitted to the Dark Adaptation Thresholds after Bleaching: Comparison between Photoreceptor Types

View Table | View all tables in this article

Table 4. Parameters of the Exponential Functions That Were Fitted to the Dark Adaptation Thresholds after Bleaching: Rod Thresholds at Different Conditions^a

View Table | View all tables in this article

2. tCSFs and SVI Curves

Rod-isolating stimuli were characterized by different tCSFs compared with cones and by different adaptational behavior (Fig. 1). With decreasing retinal illuminances, the subjects became less sensitive to rod-isolating stimuli, but this decrease was much less pronounced than for cone-isolating stimuli. Furthermore, the rod-driven SVI curves were quite different from the cone-driven ones (Fig. 6). First, the sensitivity changed less strongly with retinal illuminance. Second, a Weberian portion was absent. These differences are also reflected in the poor fits of our combined linear model. Therefore, we used a simple linear regression to calculate the slopes of the rod-driven SVI curves. At temporal frequencies below 10 to 12 Hz, the slopes of the SVI curves for rod-isolating stimuli were much lower ($<0.25$) than those of cones ($>0.5$; see Fig. 3: lower plot), whereas they were similar to slopes of M-cone-driven SVIs at higher frequencies. At frequencies between 8 and 12 Hz, the SVI curves had a small horizontal portion at the low-luminance end of the curve, suggesting a Weber region at low retinal illuminances.

 

Fig. 6. Averaged SVI curves of the three color-normal observers for rod-isolating stimuli for different temporal frequencies (black diamonds). The combined linear model did not describe these data well. Therefore, we show a simple linear model that has been fit to the rod data. Error bars represent the standard deviation. As in Fig. 2, the SVI curves for L- and M-cones are shown for reference (L-cones, red squares; M-cones, green circles) [3].

Download Full Size | PDF

C. Measurements in an S-Cone Monochromat

The S-cone monochromat (Fig. 7) was unable to detect L-cone and M-cone-isolating stimuli under most conditions [3]. Despite minor differences (lower S-cone sensitivities at low temporal frequencies, rod tCSFs always bandpass), the tCSFs for the S-cone- and rod stimuli were comparable to those in normal subjects. In contrast to the normal subjects, sensitivities to rods decreased with increasing retinal illuminance. The S-cone monochromat’s tCSFs showed gaps at intermediate temporal frequencies (often around 4–6 Hz; Fig. 7). A secondary analysis of the raw data showed that the measurements were terminated by the software due to the observer reporting seeing flicker at 0% contrast more than three times.

 

Fig. 7. tCSFs to S-cone and rod-isolating stimuli at three different retinal illuminances in one subject with S-cone monochromacy (S-cones, blue triangles; rods, black diamonds). With our protocol, good measurements were possible despite severe visual disability, glare sensitivity, and fixation nystagmus. For comparison, a smoothed curve and the standard error of the mean (SEM) fitted to the tCSFs of the normal subjects is also shown.

Download Full Size | PDF

4. DISCUSSION

In order to be able to perform measurements using the silent substitution paradigm in a clinical setting in the future, we used novel stimulus conditions. The silent substitution paradigm requires relatively exact estimations of the rod and cone fundamentals [41], which may show intraindividual variability due to prereceptoral filtering or polymorphisms in the cone opsin genes [9,26]. Methods that have been developed to correct for such variations and to calculate individual cone fundamentals through linear transformation (observer calibration) [19,20] are time-consuming and difficult to perform for untrained observers. We used a relatively large annular test field (similar to Sun et al. [42]) to minimize the influence of the macular pigment, which is the most important prereceptoral filter and varies considerably between subjects [43,44]. Our calculations (see Table 2) show that for S-cone- and rod-isolating stimuli, the prereceptoral filters are relevant because of their strong absorption at the short-wavelength end of the spectrum, in contrast to L- and M-cone-isolating stimuli, where variability in the spectral sensitivity functions is the main cause of residual responses in nondesired photoreceptors. The annular test field also has the advantage that rod sensitivities can be measured in the perifovea without the need for eccentric fixation, which is another challenge for untrained observers. Our results support the validity of this approach.

A. Identifying Intrusion

It is important to identify intrusion by the supposedly silenced photoreceptors. The effect of intrusion by a nondesired photoreceptor type depends on the type of interaction between the photoreceptor types. There are three possible types of interactions: linear summation [45–47], probability summation [48], and peak detection. In the case of peak detection or probability summation, intrusion by other photoreceptor types would probably be negligible because modulation contrast would be much smaller in the intruding photoreceptors than in the targeted photoreceptors. However, if there is linear summation, the signal from the intruding photoreceptor type would be vector-added to the signal from the isolated photoreceptor type. This would have a stronger effect and may be much harder to detect. Linear summation has been demonstrated for the interaction between rod and cone signals that are transported via the magnocellular pathway [42,49]. We hypothesize that such a mechanism may be causative for the detection of rod-isolating stimuli at high retinal illuminances and high temporal frequencies. Rod signals are also conveyed via the parvocellular pathway, and their contribution to color perception is linear, with rod contrast [50]. Rod inputs produce a percept that is more similar to M-cones than to those of L-cones, but the exact mechanism of this interaction is not known [19]. We assume that cone intrusion will be more likely to occur at higher temporal frequencies, where perception is mediated by the magnocellular pathway, because residual contrast in L- and M-cones are usually in phase (see Table 2). However, we conclude from our data that perception is mediated by the desired photoreceptor type under most conditions.

Detection of intruding signals is much more relevant under conditions where the targeted photoreceptor type is absent (such as in the validation of our L- and M-cone-isolating stimuli in dichromats [3]) or severely damaged. The tCSF will run parallel below those of the photoreceptor type actually mediating perception, and the amount of “cross talk” can be deducted from the ratio of the sensitivities of both curves. This is the case because the threshold of a given photoreceptor type will be identical regardless of the stimulus (owing to the identical adaptation ensured by the silent substitution paradigm), but the actual residual modulation contrast for the unintendedly stimulated photoreceptor will usually be much lower than the theoretical modulation contrast calculated for the targeted photoreceptor.

For clinical applications, identification of signals from supposedly silenced photoreceptors is crucial. The clinician has to be able to identify such intrusion by signatures in the shapes of the tCSFs or the SVI curves. This may be rendered difficult by noisy data. In the dark adaptation experiments, we expected a break in the curve at low retinal illuminances in the case of significant residual detection by cones, similar to the classic dark adaptation curve. Such a break was not observed.

B. S-Cones

The good isolation that we achieved for the S-cone-isolating stimuli is demonstrated by the differences in light adaptation (slope and $I_{\text{Weber}}$; see Fig. 3) compared to L-/M-cones and rods, as well as the similarity between the tCSFs of the normal subjects and the S-cone monochromat (Fig. 7). The robustness of the S-cone-isolating stimuli is also supported by our calculations (Table 2). They show that L- or M-cone contrasts exceeding 2% are highly unlikely for our S-cone-isolating stimuli. Only frequencies above 16 Hz should probably be avoided, where our data indicate that intrusion of other cone types cannot be ruled out completely, although the reason for the large increase in the slopes of the ascending portion of the SVI curves at high temporal frequencies (Fig. 3, lower plot) may also have been that the model fit was not well constrained by the data owing to the relatively small number of points available for modeling.

The isolation that can be achieved with our protocol is an advance compared with earlier methods that have been introduced to produce isolating stimuli using either selective adaptation (including short-wavelength automated perimetry) or double silent substitution with a CRT screen. The use of primaries with a spectrally narrow bandwidth in the LED stimulator prevents the relatively high residual contrasts of up to 5% in L- and M-cones that may occur when using CRT screens [21]. Intrusion by rods is more likely to occur, according to our calculations. However, the data presented here (particularly the SVI curves with S-cone-isolating stimuli (see Fig. 2) did not indicate rod response intrusions even at low retinal illuminances.

Our results agree with earlier data [51], showing that the S-cone tCSFs had the lowest sensitivities of all photoreceptors and showed low-pass characteristics with a sharp decrease in sensitivity at frequencies above about 2 to 4 Hz. This is consistent with the paucity of the S-cones and mediation by the koniocellular system. Although S-cones can feed into the faster magnocellular (MC) system under specific conditions [9,51,52], this mechanism does not seem to be relevant for the detection thresholds as measured in our experiments. However, although it is difficult to draw firm conclusions from the data of only one observer with S-cone monochromacy, the increased sensitivity to the S-cone-isolating stimuli at high temporal frequencies at 294 phot Td may indicate that the S-cone signals are transported by the luminance channel in the absence of L- and M-cone signals.

S-cone mediated vision is frequently more severely affected by retinal diseases than L- and M-cone mediated vision. Different mechanisms have been implicated and are discussed in detail here [10,21]. Damage does not seem to be limited to one retinal locus, because different diseases including retinitis pigmentosa (photoreceptors), diabetic retinopathy (bipolar cells), and glaucoma (retinal ganglion cells) have been associated with disruption of the S-cone pathways [53]. Our measurements indicate that our paradigm may be useful as a clinical tool. Particularly, the data from the S-cone monochromat indicate that, with our setup, good results can be obtained despite severely reduced visual acuity, nystagmus, and glare sensitivity.

C. Rods

The validity of our rod-isolating stimuli was confirmed by the bleaching adaptation experiments and by demonstrating that the SVI curves had quite distinct slopes and did not show regions where Weber’s law applies, compared with SVI curves based on cone-mediated detection. Cone intrusion did not seem to be a problem except at temporal frequencies of 12 Hz and higher, where the SVI curves had steeper slopes (see Fig. 3) and were similar to those of L- and M-cone-isolating stimuli. Thus, our data support the fact that with the chosen stimulus configuration, isolation of rod responses can be achieved at temporal frequencies $\le 12\mathrm{Hz}$ for all used retinal illuminances (up to 587 Td). This extends earlier psychophysical studies that have proven that the rod system may contribute to visual perception at higher light levels than expected from the classic experiments by Aguilar and Stiles [12,54]. Therefore, rod-driven flicker perception may be tested at relatively high retinal mesopic illuminances, possibly eliminating the need for time-consuming dark adaptation. The time of 15 to 30 min that is needed in order to achieve reliable dark adaptation is part of the (ISCEV) standard ERG, and it may not be a problem when performing an extensive workup in a patient with hereditary retinal disease, but it strongly limits the routine use of measuring rod function in areas where it is currently not standard, such as in age-related macular degeneration [15,16]. Furthermore, just sitting in a dark room for a prolonged time is considered a discomfort by many patients. Furthermore, the visual system will be most sensitive immediately after dark adaptation, but thresholds may increase after repeated stimulation [17]. In contrast, using silent substitution, reliable rod isolation may be achieved with a constant adaptation. Finally, cone-driven psychophysical sensitivities can be measured at an identical state of adaptation and thus be directly compared with the rod-driven sensitivities. Comparable results have been found in electrophysiological studies, where the authors found that rod-driven ERGs can be reliably measured without dark adaptation at retinal illuminances up to about 500 phot Td [17].

We measured thresholds to rod-isolating stimuli during adaptation after a bleach using sine-wave modulation around different time-averaged retinal illuminances. One has to keep in mind that these dark adaptation curves can be compared with the well-known dark-adaptation curves (where absolute thresholds are measured after bleaching) only when the mean retinal illuminance is low. In stimuli with higher time-averaged retinal illuminance, as used in the present experiments, the absolute modulation of photon catches is much higher at the same Michelson contrast. Gaining a more thorough understanding of the mechanisms that underlie the dark adaptation to our isolating stimuli is a worthwhile goal for further research. In the context of this discussion, the value of the bleaching experiments lies in the validation of the rod-isolating stimuli.

In subject JK, adaptation was much slower for rod-isolating than for cone-isolating stimuli at low, nearly scotopic retinal illuminances. Interestingly, thresholds for rod-isolating stimuli were still higher than at baseline even after 15 min of adaptation, whereas thresholds for L- and M-cone stimuli were lower than at baseline. A change in response criterion cannot be ruled out completely. However, subject JK is an experienced observer and the randomly interleaved staircase algorithm is not very prone to bias, in our experience. Thus, we think that reduced lateral suppression by dark-adapted cones [55–57] is a more likely explanation.

In subject CH, mean retinal illuminances after the bleach were higher and recovery of sensitivity to rod-isolating stimuli was faster. It cannot be fully excluded that this indicates perception by cones. However, we did not observe biphasic curves at intermediate retinal illuminances, as would be expected if there were significant cone intrusion. Time constants $\tau$ were much shorter for the cone types at intermediate retinal illuminances. Our exponential model suggests that differences in the dark-adaptation curves with rod-isolating stimuli at different retinal illuminances may be explained exclusively by differences in absolute thresholds without a change in parameter $\tau$, although the model was not well constrained by the limited number of data points.

The rod-driven tCSFs were different from those driven by cones, and they are consistent with the literature. Rod tCSFs are reported to be bandpass, with a maximum at intermediate frequencies between 6 and 8 Hz [20,58]. At higher retinal illuminances, we found that the tCSFs were low-pass with a very flat tCSF at low and intermediate frequencies and a sharp decline in sensitivity at high frequencies (above 15 Hz) [12,14]. The presence of a second rod pathway, where signals are transmitted via rod-cone gap junctions [14], did not manifest itself at higher retinal illuminances. Especially, there was no break in the rod-driven SVI curves. At high temporal frequencies and high mesopic retinal illuminances, the SVI curves for rod-isolating stimuli were more cone-like and superimposed on the M-cone-driven curves, suggesting cone intrusion. This cannot be clearly attributed to the second rod pathway. Sharpe et al. [14,59] have observed a region where flicker was invisible at frequencies around 15 Hz, threshold intensities around 0 log10 scot Td, and background intensities below 0 log10 scot Td. They have attributed this to destructive interference between signals from the two pathways. Due to the instrument gamut, which limits Michelson contrast to 27% for rod-isolating stimuli, it was not possible to reproduce similar conditions with our setup. We measured thresholds for stimuli that probably activated the fast rod pathway of Sharpe et al.

The quality of isolation under most conditions is further demonstrated by the considerable differences between rods and cones in the SVI curves (see Figs. 3 and 6). With increasing retinal illuminances, rod-driven contrast sensitivities increased less strongly, resulting in a shallower slope of the SVI curves (less than 0.25 for rod-isolating versus larger than 0.5 for cone-isolating stimuli (see Fig. 5, right). The rod curves were almost linear, especially at low frequencies. This shallow increase is consistent with a slope of approximately 0.8 previously reported for rod TVI curves [54,60]. At frequencies between 8 and 12 Hz, the rod SVI curves were in the Weber range at retinal illuminances below 1 phot Td. In contrast, cone-driven SVI curves increased steeply at lower retinal illuminances and had a plateau at high light levels (Weber region).

D. S-Cone Monochromacy

The data from the S-cone monochromat (Fig. 7) confirmed the findings obtained in the color normal observers that the SVI slopes of rod mediated thresholds were shallower than those of S-cone mediated detection. However, we did not estimate these slopes because of the sparsity of the data. Still, we have made two interesting observations in the S-cone monochromat: higher sensitivities for rod-isolating stimuli at low retinal illuminances and higher sensitivities for S-cone-isolating stimuli at intermediate temporal frequencies at high mesopic light levels. These findings are interesting, although it is difficult to draw firm conclusions from the findings in only one observer.

Hesset al. have measured CFF-log I functions [61]. They have found that these functions had a plateau at 40–45 Hz, which is much higher than S-cone-driven CFF in normal subjects. The authors referred to an experiment by Stockman and coworkers, who have demonstrated the ability of the S-cones to detect flicker at higher frequencies (between 40 and 45 Hz) than was observed in psychophysical experiments before, but this flicker was not perceived under normal conditions [9]. The higher sensitivity of the S-cone monochromat at frequencies between 8 and 12 Hz may support the notion that the S-cone monochromats have higher S-cone-driven CFFs.

Concerning the increased rod sensitivities, our data are in agreement with those of Luo and coworkers, who, based on extensive psychophysical measurements in 25 S-cone monochromacy (SCM) patients, found that rods in SCM patients continued to “signal under conditions normally associated with daylight vision” [62]. However, they did not find increased rod sensitivities under dark-adapted conditions, as we did. Likewise, Shapiro has also found suppression of rod detection when L- or M-cone excitation of the background was increased [54]. Finally, Haegerstrom-Portnoy and Verdon [63] have raised the possibility of stronger rod activity due to the absence of an inhibitory cone-rod interactions, because they observed stronger rod-mediated transient tritanopia in S-cone monochromats compared with trichromats. Their work also suggests that rod-cone or cone-rod interactions can be altered in subjects with S-cone monochromacy. Further research is needed to clarify this.

E. Intrinsically Photosensitive Retinal Ganglion Cells (ipRGCs)

The melanopsin photopigment of the ipRGCs is not silenced with our protocol. However, several observations argue against a prominent role of ipRGCs in our experiments. The bleaching experiments are not compatible with a significant influence of ipRGCs in the perception of rod-isolating stimuli, because ipRGCs are very resistant to bleaches [64]. Only at relatively high retinal illuminances (294 phot Td), recovery of temporal contrast sensitivity was relatively fast. In addition, we measured relatively high sensitivities at high temporal frequencies where ipRGCs are very insensitive because of their slow and sustained responses [65]. Also “giant” ipRGCs, which project to the lateral geniculate nucleus and therefore may play a role in a conscious flicker perception [66], do not respond at these high temporal frequencies.

F. Limitations

Apart from the effects of the absence of an observer calibration, there are some limitations of the employed procedure. The first limitation is the use of a staircase paradigm to determine the thresholds. An n-alternative forced-choice paradigm may be superior in avoiding biased response criteria of the observer. Our assumption that the response criterion for seeing flicker does not vary across frequencies may not always hold. Varying response criteria were also a concern in the S-cone monochromat, who was not experienced with these kinds of experiments and who had difficulties in perceiving the test field due to the presence of severe nystagmus. A second limitation is the variability of the measurements and the relatively small sample size. However, under most conditions our model fits were well constrained by the data, and we are confident of our conclusions. A third limitation is the characterization of the pattern of the SVI curves by a mathematical model that does not correspond to a known physiological process. This makes the interpretation more difficult, but it still allows the demonstration of differences between the SVI curves and helps to identify possible cone intrusion for the rod-isolating stimuli at high temporal frequencies. Lastly, we cannot be sure that rods adapted fully to the decreased light levels when the neutral density filters were changed. Although it seems likely from our experiments, we may have underestimated the rod sensitivity slightly at low light levels.

5. CONCLUSIONS

We have shown that a technique using perifoveal silent substitution stimuli to measure temporal contrast sensitivities of the S-cone and rod-mediated flicker detection is generally feasible without observer calibration. Using this technique, we were able to demonstrate differences in temporal characteristics and adaptational behavior between cones and rods. The measurements of thresholds during adaptation after a bleach in normal observers and of the tCSFs in an S-cone monochromat further validate our method. This method has the potential to be clinically useful. We, however, found that satisfactory rod isolating is not warranted for temporal frequencies of 12 Hz and above. At other temporal frequencies, the possible intrusion of cone signals is probably too small to be of relevance.