Changes in Rod and Cone-Driven Oscillatory Potentials in the Aging Human Retina

Abstract

Purpose.: We recorded oscillatory potentials (OPs) to document how age impacts on rod- and cone-driven inner retina function.

Methods.: Dark- and light-adapted electroretinogram (ERG) luminance-response functions were recorded in healthy human subjects aged 20 to 39, 40 to 59, and 60 to 82 years. Raw ERG traces (0.1–300 Hz) were filtered (75–300 Hz) to measure OPs trough-to-peak in the time–amplitude domain. Morlet wavelet transform (MWT) allowed documenting OPs time–amplitude–frequency distribution from raw traces.

Results.: Under dark adaptation, both methods revealed reduced OP amplitudes and prolonged implicit times by 40 years of age. The MWT identified a high-frequency band as the main oscillator, which frequency (150–155 Hz) was unaffected by age. Under light adaptation, most OP peaks were delayed by 40 years of age. Peak–trough measures yielded inconsistent results in relation to luminance. Contrastingly, MWT distinguished two frequency bands at all luminances: high frequency (135 ± 6 Hz) time locked to the onset of early OPs and low frequency (82 ± 7 Hz), giving rise to early and late OPs. By 60 years, there was a consistent power reduction specific to the low-frequency band.

Conclusions.: Age-related OP changes precede those seen with a- (photoreceptoral) and b-waves (postphotoreceptoral). In addition, MWT allows quantifying distinct low- and high-frequency oscillators in the human retina, which complement traditional OP analysis methods. The identification of an age-independent OP marker (light-adapted high frequency band) opens a new dimension for the screening of retinal degenerations and their impact on inner retina function.

Introduction

The full-field electroretinogram (ERG) is used widely to assess retina function. ¹ The ERG trace consists of several distinct light-driven components, with the a-wave, b-wave, and oscillatory potentials (OPs) constituting the major ones. Many lines of evidence support that the a-wave essentially reflects activity at the presynaptic photoreceptor level, ² whereas the b-wave corresponds largely to postsynaptic optic neuropathy (ON) bipolar cell activity. ³ Coinciding with the b-wave are the OPs, which reflect activity from the inner retina. ⁴ Based on pharmacologic manipulations, OPs are generated from the light-driven activation of negative feedback pathways between amacrine cells, retinal ganglion cells, and bipolar cells. ^5,6  

The traditional approach to characterizing OP properties involves band-pass filtering to remove low frequency components, such as the a- and b-waves. Individual OP peaks then can be identified in the time domain and trough-to-peak amplitudes, and implicit times determined. An alternative representation of the power spectrum of OPs in the frequency domain can be achieved by applying Fast Fourier Transform (FFT) analysis. The FFT can be used to convey information about the frequency and phase of a signal. The power spectrum FFT analysis, though, only retrieves the global frequency content of the OP signal, excluding time-related variations. Continuous wavelet transform is a novel application that allows complex information to be dissected into elementary forms at different temporal positions and scales, and then reconstructed with high precision. Wavelet transform of a function can be seen as an “improved” version of the FFT in that it allows analyzing nonstationary signals (such as the ERG) in addition to stationary ones. This approach, when applied to ERG waveforms, opens up a new analytical dimension of the ERG signal that can be applied to age-related changes in the healthy and diseased retina. 

Numerous studies have documented age-related changes in the outer nuclear layer (ONL) of healthy human retinas. ^7–9 Curcio et al. ¹⁰ described a preferential loss of parafoveal rods in aged retinas, which led us to rely on the full-field ERG as a means to assess age-related changes in rod- versus cone-driven retina function in humans. We previously showed that the dark-adapted a-wave, but not b-wave, was reduced in amplitude with age. ¹¹ These results are in agreement with the potential occurrence of compensatory postreceptoral changes, in response to decreased photoreceptoral input, associated with b-wave amplitude preservation. From an anatomical perspective, these functional changes could reflect the sprouting of ON bipolar cell dendrites as characterized in the healthy aging retina; these dendrites expand into the outer nuclear layer, a dendrite-free region in young healthy retinas. ¹²  

It is plausible that the retina undergoes such plasticity in its interplexiform circuitry before the detection of rod and cone dysfunction by using other methods. To validate this hypothesis, we sought to identify precocious compensatory changes in the inner retina of intermediate-aged healthy individuals (40–59 years); for comparison purposes, younger (20–39 years) and older (60–82 years) groups were similarly studied. We extensively characterized OPs over a wide range of flash strengths using traditional and novel analytical methods. In addition, we relied, for the first time to our knowledge, on Morlet wavelet transform (MWT) to simultaneously quantify the frequency, peak time, and power spectrum of the OP components of the human ERG. Wavelet transform has been used before to analyze ERG signals in rodents ¹³ and to study evoked potentials in humans. ¹⁴ Recording and analyzing OPs, as done in this study, provides a highly sensitive approach for the early detection of age-related functional changes in the human retina. If such changes occur, these must be taken into account as being part of an age-dependent baseline when screening for pathologies affecting the inner retina, such as diabetic retinopathy, central vein occlusion, and glaucoma. 

Methods

Subjects

Previously-recorded ERG waveforms from 63 healthy human subjects were reanalyzed for the purpose of this study. Freund et al. ¹¹ reported changes in dark- and light-adapted a- and b-wave properties from the same cohort (Supplementary Fig. S1). Inclusion criteria consisted of no ocular pathology (current or previous) and best corrected visual acuity of 20/20. These individuals were split into three groups according to age: 20 to 39 years (n = 26; mean age ± SD, 24.6 ± 4), 40 to 59 years (n = 21, 52.1 ± 5), and 60 to 82 years (n = 16, 65.8 ± 5). To compare the two OP extraction methods in pathological states, we further included five subjects with unilateral neovascular age-related macular degeneration (AMD) undergoing intravitreal anti-VEGF therapy, as well as two subjects presenting with treatment-naïve unilateral serous pigment epithelial detachment (sPED). In each case, the fellow non-neovascular eye served as an internal control to identify potential inner retina dysfunction in the neovascular eye. All fellow eyes were classified as having intermediate dry AMD, defined by the presence of either many medium-sized drusen, or one or more large drusen. This study received ethics approval from the “Health Research Ethics Board (Biomedical Panel)” of the University of Alberta (license #6194). All procedures conformed to the Code of Ethics of the World Medical Association (Declaration of Helsinki), and were done with the understanding and written consent of each participant. 

ERG Recordings

Subjects underwent bilateral full-field ERG recordings. Stimulus presentation, amplification (0.1–300 Hz band pass, without notch filtering), and digital data acquisition were provided by the Espion E ² system (Diagnosys LLC, Lowell, MA, USA). Digitization rate was set at 1 kHz for all tests, with the exception of 5 kHz for scotopic intensity responses for which a higher rate was used to optimize definition of fast events, such as the leading edge of the saturated a-wave. Although the sample rate of 1 kHz suggested by the International Society for Clinical Electrophysiology of Vision (ISCEV) was adequate for the purposes of this study, a higher sample rate is worth considering when analyzing OPs using the MWT method. Dawson-Trick-Litzkow (DTL) type recording electrodes were placed below the limbus between the inferior eyelid and sclera in both eyes; reference electrodes (gold surface electrode, F-E5GH-60; Grass Technologies, West Warwick, RI, USA) were placed on the lateral edge of the orbital bone after using a mild abrasive gel (NuPrep Skin Prep Gel; Weaver and Company, Aurora, CO, USA) to exfoliate and remove sebum at the area of the electrode contact. Impedance was compliant with ISCEV guidelines (lower than 5 K Ohm, as measured by the Espion E2 system by passing a very weak current, 500 nA). A ground electrode (gold surface electrode) was placed in the center of the forehead, 1 inch above the eyebrows. Pupils were dilated by applying two drops of tropicamide 1% on each cornea 20 minutes before starting the recording session; pupil size (average diameter ± SD) remained unchanged between age groups (20–39 years, 6.4 ± 0.4 mm; 40–59 years, 6.5 ± 0.5 mm, and 60–82, 6.3 ± 0.7 mm). Full dilation was confirmed after the end of the testing period. The visual stimuli consisted of flashes administered in a full-field Ganzfeld by a xenon bulb of 6500° K color temperature (10 μs duration); light levels were calibrated with an IL1700 photometer (International Light Technologies, Inc., Peabody, MA, USA) equipped with either a photopic or scotopic filter. 

Photopic ERG

Subjects were initially light-adapted for 10 minutes (background of 30 cd/m², measured at the corneal surface) with the purpose of saturating rods to record cone-driven ERG responses. White flashes were presented with the following increasing time-integrated luminance along 6 stepwise increments: −0.02, 0.38, 0.88, 1.37, 1.89, and 2.39 photopic log cd·s/m² (logarithm of candela s/m²). The interval between each step was set at 10 seconds and each stimulus was presented 6 times at 5-second interstimulus intervals. Responses were averaged at each step. 

Dark-Adapted ERG

Following ISCEV standard 20-minute dark adaptation, ¹⁵ white flashes were presented with the following increasing time-integrated luminance along four stepwise increments: −0.02, 0.38, 0.88, and 1.37 scotopic log cd·s/m². Depending on the stimulus intensity used, responses were averaged either three or five times. For mid-intensity stimuli, five sweeps per result were used. For high-intensity stimuli (0.88 and 1.37 scotopic log cd·s/m²), only three sweeps per result were used to reduce subject discomfort to bright light exposure. To allow for maximal rod recovery between consecutive flashes, interstimuli intervals were increased from 15 seconds at the lowest stimulus strength up to 60 seconds at the highest stimulus strength. 

OP Extraction and Analysis

Isolation of OPs According to ISCEV Standards.

As specified by the ISCEV “guidelines for calibration of stimulus and recording parameters,” ¹⁶ a 75 to 300 Hz digital band pass was applied to isolate OPs. The digital filter used was a 2-pole Bessel filter with a 3-dB cutoff frequency. Trough to peak measurement was used to assess OP amplitudes. Under dark-adapted conditions, five peak OP amplitudes (OP1–OP5) and their corresponding implicit times (t1–t5) were measured from the filtered traces (Fig. 1A). The sum of the first four OP peaks was used to calculate the “summed OP amplitude.” ¹⁵ Under light-adapted conditions, four peak OP amplitudes (OP1–OP4) and their corresponding implicit times (t1–t4) were measured from the filtered traces (Fig. 1B). Calculation of summed photopic OPs, always included OPs 1 to 3. For stimuli higher than 0.88 log cd·s/m², we assessed OP4 and included its amplitude in the summed OPs. 

Figure 1

 

Representative band pass-filtered (75–300 Hz) luminance-response ERG waveforms. (A) Dark-adapted responses showing the five distinct OP peaks (OP1–OP5) and corresponding implicit times (t1–t5). (B) Light-adapted responses showing the four distinct OP peaks (OP1–OP4) and corresponding implicit times (t1–t4). The number of measured OP peaks changes as a function of flash strength. Two distinct OP peaks are identified (OP1 and OP2) at lower stimulus strengths (−0.02, 0.38 log cd·s/m²). A third OP (“OP3–4 complex”) becomes apparent as the stimulus strength is increased. This OP complex splits and gives rise to two distinct OP peaks (OP3 and OP4) at higher stimulus strengths (>1.37 log cd·s/m²). The horizontal line starts at time point zero (stimulus onset) and represents amplifier calibration at zero microvolt amplitude. Scale bar represents time (x-axis) and amplitude (y-axis).

Quantification of OPs With Continuous MWT Analysis.

The MWT analysis was adapted from a previous study of rat ERG signals described by Forte et al. ¹³ The MWT was used to divide a continuous-time function into smaller or shorter waveforms (wavelets). Unlike Fourier transform, which cannot capture changes in frequency response with respect to time, the continuous MWT allows generating the time-frequency representation of a signal with high resolution in the frequency and time domains. In general, it has proven particularly useful for the analysis of transient, aperiodic, and nonstationary signal features, for ERG and ECG. The equation of the continuous MWT is given below:  

 

where ψ(t) is the fitted mother wavelet (the Morlet function in this case, see below), t the time, a the position parameter, and b the scaling parameter of the wavelet.  

By applying changes in the location (a) and scaling (b) of the mother wavelet, short duration-high frequency and long duration-low frequency information can be captured simultaneously. 

The Morlet wavelet (generated from raw “unfiltered” traces) provides a robust statistical model of the filtered OP waveform. ¹⁷ It is a complex Gabor function composed of a real cosine component and an imaginary sine component multiplied by a Gaussian window (envelope) and is defined by the following equation:  

 

where f_c is the fundamental frequency of the wavelet determined by the scaling factor b in the wavelet transform, i the square root of −1, and t the time.  

Continuous MWT was applied to each of the raw “unfiltered” ERG traces (0.1–300 Hz) to generate a topographical map of the ERG power spectrum in the frequency and time domains and to quantify the three following variables: relative amplitude, oscillation frequency, and implicit time values for the OPs. The wavelet function software used consisted of the Matlab complex Morlet (Mathworks, Natick, MA, USA). Raw data were multiplied by half Welch window to smooth the trailing edge of the ERG waveform after 118 ms from flash onset. This modification was adopted previously by Forte et al. ¹³ to remove terminal artifacts that might contaminate the wavelet analysis. Modified ERG signals were correlated with the Morlet wavelet at every time point. Scaling values were chosen to encompass the number of steps in frequency measurement and in the envelope of frequencies at which OPs occur (50–300 Hz). Scaling values differed for 1 kHz (light-adapted) and 5 kHz (dark-adapted) samples as scale to frequency conversions are dependent on sample rate. The magnitude of correlation at each point was measured as the complex modulus of the complex wavelet response. Local maxima were located using nearest-neighbor comparison. The largest local maximum in the 50 to 300 Hz range was considered to be the peak of the OP for dark-adapted responses. For light adapted responses, largest local maxima were searched in two frequency ranges: 50 to 100 and 100 to 300 Hz. The complex modulus of the maxima were recorded as peak power (arbitrary units). The use of the complex modulus reduces the phase-dependence of the power measurement. The locations of these maxima in the time and frequency domains were recorded as peak time and peak frequency. To generate a wavelet scalogram, magnitudes of correlation coefficients across the ERG signal were expressed as a grayscale, with relative amplitude normalized to the peak value for each exposure. Maximum response appears white. 

Statistics

To ensure independence of samples, a single eye per subject was included in the analysis. The included eye was chosen based on the highest recorded a-wave amplitude in the scotopic intensity response series except where there was a clear indication (e.g., excessive voltage noise due to poor electrical contact) to discard the recording in favor of the contralateral eye. Statistical significance between age groups was assessed using repeated-measures ANOVA with the Greenhouse-Geisser correction for sphericity. Post hoc analyses were done between the individual groups and at individual stimulus strengths or time points using the Bonferroni technique for multiple comparisons. GraphPad Prism was used for 1-way ANOVA, linear regression, and correlation analyses (GraphPad Software, Inc., La Jolla, CA, USA), and SPSS was used for the repeated-measures ANOVA (SPSS, Inc., Chicago, IL, USA). Significance was set at P < 0.05. Data points on graphs represent mean ± SEM. 

Results

Scalogram Features

Figure 2 displays representative examples of wavelet scalograms normalized to a local peak between 50 and 300 Hz for a period of 250 ms after flash presentation at t = 0 ms. Raw ERG traces have been superimposed on the bottom row scalograms (Figs. 2C, 2D) to determine the origin of each peak. Figure 2A shows a representative wavelet trace with a high signal-to-noise ratio. The red triangle corresponds to the peak with the maximum amplitude. In this example, peak OP amplitude occurs at 28 ms with a frequency of 153 Hz. The a-wave component occurs at 55 Hz. In Figure 2B, a photomyoclonic reflex is seen at t = 125 ms, with the remaining of the trace having low baseline noise level. Figure 2C provides an example of wavelet amplitudes in the 50 to 250 Hz spectrum obtained when presenting a short duration (10 μs) full-field flash of 0.88 log cd·s/m² strength in the light-adapted human retina. Two distinct peaks of comparable power can be identified. Superimposition of the raw ERG trace reveals that the low frequency band is in close proximity to the peak of the b-wave, while concomitantly the high frequency band peaks during the rising edge of the photopic b-wave. 

Figure 2

 

Representative examples of ERG traces (0.3–300 Hz bandpass) with corresponding MWT scalograms displayed along the same time axis (x-axis); the y-axis for the scalogram and the ERG trace respectively represent “frequency” and “amplitude” (time zero corresponds to stimulus onset). Correlation magnitudes of the ERG signal are represented by shades of gray in terms of a scalogram, with relative amplitude normalized to the peak value for each exposure. (A) A wavelet trace essentially deprived of background retina activity. Peak OP amplitude occurs at 28 ms with a frequency of 153 Hz. An a-wave–related component is observed at 55 Hz. (B) The occurrence of a photomyoclonic reflex (extraocular origin) occurring at 125 ms. Of note, in (A) and (B), the scalogram representation of the OPs is distinct of that from the a-wave, and the timing of the peak of the OP train (red triangle) is synchronous with the b-wave ascending phase. (C) The scalogram representation of a light-adapted response to a flash of 0.88 log cd·s/m² strength. Two distinct peaks of comparable power can be identified on the scalogram. Superimposition of the ERG trace reveals that the low frequency band occurs at the same time as the peak of the b-wave. The scalogram allows separation of this low frequency event (bottom red triangle) from a higher frequency event (top red triangle) that occurs during the rising phase of the photopic b-wave. (D) The scalogram representation of a dark-adapted response to a mid- strength stimulus (−0.02 log cd·s/m²). Two oscillators are superimposed in time: one corresponding to the second trough of the a-wave (lower triangle with frequency peaking at ∼75–80 Hz) and the other one corresponding to OPs on the ascending limb of the b-wave (upper triangle with frequency peaking at ∼150–160 Hz). The presence of a photomyoclonic reflex can be noted at 140 ms.

Figure 2D consists of a scalogram in which the appearance is representative of dark-adapted recordings, where ERG waveforms with “double-trough” a-waves are present. A low frequency band with a peak at 75 to 80 Hz is noted in close proximity to the second trough of the a-wave. A second high frequency peak at 150 Hz follows, which coincides with the ascending limb of the b-wave. The early peak has been attributed previously to the first OP (OP1) that coincides with the a-wave and splits it into a double-trough a-wave. ^18,19 To remove the potential influence of a-wave onset on the low-frequency OPs, we conditioned the original ERG waveform (Fig. 3A) to the lowest voltage before the first a-wave trough (Fig. 3B) or before the second a-wave trough (Fig. 3C). We repeated this exercise for representative individuals in each age group at flash intensities where “double-trough” a-waves were present (Supplementary Fig. S3). Wavelet analysis of such conditioned waveforms revealed that removal of the a-wave before the first trough had no effect on the power of the low frequency OPs (Fig. 3B), whereas removal of the first a-wave trough resulted in a 40% power reduction in the low-frequency band (Fig. 3C). High frequency OPs remained unaffected. Therefore, the low-frequency peak should be attributed to the oscillatory system without contamination from the a-wave. As such, two distinct oscillators can be identified in the dark-adapted retina. To avoid the issue of potential a-wave contamination with the low frequency OPs, we limited our aging analysis to the high-frequency dark-adapted OPs, which could be clearly separated from the slow a-wave components in the frequency domain. 

Figure 3

 

Effect of the a-wave onset on dark-adapted OPs. Wavelet analysis prior to (A) and after preconditioning the ERG waveform to the lowest voltage before the first (B) and the second a-wave trough (C). Power of the low frequency band (76 Hz) is not altered after removal of the a-wave leading edge (B), but becomes significantly reduced once the second a-wave trough has been removed (C). The high frequency band remains unaffected. Power units are arbitrary units (complex modulus) and frequency is expressed in Hz.

Dark-Adapted Responses

Band-Pass Filtering and Trough-to-Peak Analysis.

For all age groups, the first four OP peaks (OP1–OP4) grew in amplitude and occurred earlier as stimulus strength increased (Fig. 4). Aging was associated with a reduction of amplitudes of the first four OP peaks (OP1–OP4, Figs. 4A–D) in addition to a prolongation of implicit times for all OP peaks (Figs. 4F–I). The calculated summed OP amplitude showed a similar dependency with increased stimulus strength (Fig. 4E). 

Figure 4

 

Effect of age on amplitude and implicit time of the respective OP components as a function of stimulus strength, under dark adaptation. In all three age groups, amplitude increases and implicit time decreases, respectively, as a function of stimulus strength increments. Amplitudes of OP1 to OP4 are shown in (A–D), and the summed amplitudes of these respective OPs are shown in (E). There is a significant amplitude reduction with age for all individual OP components and summed OP amplitudes. Post hoc analysis shows differences between the young (blue) and older (red and green) groups, but not between the middle aged and old groups. Implicit times of OP1 to OP4 (F–I) show a similar trend as for amplitudes: there is a significant implicit time prolongation in middle-aged and old individuals (red and green) compared to young (blue) for all individual OP components. Statistically significant differences (P < 0.05) †between the young (20–39) and middle aged (40–59), and ‡between the young (20–39) and old (60+) groups.

Wavelet-Derived Parameters (Amplitude, Peak OP Time, and Frequency).

The MWT confirmed the relationship (described above) between stimulus strength and peak OP amplitude for all age groups (P < 0.0001, Fig. 5); see Figure 5A for representative scalograms. Relative amplitudes increased 3-fold for the young and 2-fold for the older groups from −0.02 to 1.36 log cd·s/m² flash strength. Peak OP time, defined as the location of the peak OP amplitude in the time domain, was decreased as a function of flash strength (P < 0.0001). Caution was taken to include as OP peaks only local maxima that occurred after the trough of the a-wave. Based on dark-adapted a-wave analysis undertaken by Freund et al. ¹¹ using the same ERG waveforms, such trough was estimated to occur before 27.76 and 14.27 ms (90th percentile) for the lowest (−0.02 log cd·s/m²) and highest (1.36 log cd·s/m²) strengths, respectively. Significantly lower peak OP amplitudes (P < 0.001, Fig. 5B) and higher peak OP times (P = 0.0028, Fig. 5C) were calculated for the two older groups. Figure 5D shows that the peak OP frequency remained unchanged as a function of flash strength for all age groups. One main source of oscillation could be identified for stimulus strengths higher than 0.38 log cd·s/m² with a peak at 145 to 155 Hz. At moderate strengths (−0.02 log cd·s/m²), a second peak of oscillation with a mean frequency of 77 Hz and onset at 29.1 ± 3.1 ms could be distinguished. This OP accounted for 60% of the total observed power in the 50 to 300 Hz frequency range. Potential a-wave contamination was excluded, as discussed above. As flash strength increased, its peak time occurred earlier and in close proximity to the a-wave trough, making quantification unreliable without preconditioning ERG waveforms to remove the a-wave. 

Figure 5

 

Analysis of dark-adapted OPs using MWT. (A) Representative examples of MWT scalograms for increasing stimulus strengths; time zero represents visual stimulus onset. Isolation of OP trains from other oscillators allows quantifying distinct peak amplitude, timing, and frequency (orange triangle) for all stimulus strengths. For all age groups, peak OP amplitudes increase (B) and peak OP times decrease (C) as a function of stimulus strength. Peak frequency (D) is independent of stimulus strength, regardless of age. At the lowest stimulus strength (−0.02 log cd·s/m²), a second peak of oscillation can be recorded in synchrony with the a-wave peak (D). Lower peak OP amplitudes with delayed implicit times are characteristic of the older groups (red and green) when compared to the young group (blue) No differences are noted between the middle-aged (green) and old group (red). Statistically significant differences (P < 0.05) †between the young (20–39) and middle aged (40–59), and ‡between the young (20–39) and old (60+) groups.

Light-Adapted Responses

Band-Pass Filtering and Trough-to-Peak Analysis.

A more complex picture of OP generation was observed for light-adapted responses (Fig. 6). The number of measured OP peaks varied as a function of flash strength. At lower strengths (−0.22 and 0.38 log cd·s/m²), two OP peaks could be distinguished (OP1 and OP2; Figs. 6A, 6B). A third OP was evident at 0.88 log cd·s/m², which we refer to as “OP3-4 complex” following previous description by Rufiange et al. ²⁰ (Fig. 6C). This OP complex would split and give rise to two separate OP peaks (OP3 and OP4; Figs. 6C, 6D) at higher strengths (>1.37 log cd·s/m²). Initially, there was a 2-fold increase in the amplitude of OP1 and OP2 for the first two strengths, followed by a plateau phase for OP1, and a decline and plateau phase for OP2 at higher flash strengths. Increased age was associated with amplitude reduction of OP1 and OP2 in the old (60-+) compared to the young (20–39) and middle-aged (40–59) groups (repeated measures ANOVA, P = 0.006 for OP1 and P = 0.0029 for OP2). For OP2 Bonferroni adjustment for multiple comparisons limited the difference at the 0.38 log cd·s/m² flash strength. The amplitudes of the third and fourth OPs (OP3 and OP4) were not affected by either flash strength or age. The calculated summed OP amplitude showed an initial positive linear growth with increasing flash strength, followed by a plateau phase for higher strengths (>0.88 log cd·s/m²). Differences between the old (60+) and the two other age groups were only significant at the 0.38 log cd·s/m² flash strength after Bonferroni adjustment (Fig. 6E). Peak implicit times for the first two OPs (OP1 and OP2) were reduced as flash strengths increased (P < 0.0001). Such dependence was not noted for the late OPs (OP3 and OP4). Peak delays were recorded in both older groups compared to the young group for the first three OP components (P < 0.0001, Figs. 6F–H), but not for OP4 (Fig. 6I). 

Figure 6

 

Effect of age on amplitude (A–E) and implicit time (F–I) of the respective OP components as a function of stimulus intensity, under light adaptation. The number of measured OP peaks changes as a function of flash strength. There is a significant difference between the old (green) and the two other (blue and red) age groups for OP1 and OP2 amplitudes. For OP2 Bonferroni adjustment for multiple comparisons limited the difference at the 0.38 log cd·s/m² flash strength The relationship between summed OP amplitude and stimulus strength is best described as an initial linear increase followed by a plateau beginning at 0.88 log cd·s/m² (E). A reduction in summed OP amplitude characterizes the older group (60+) compared to the other two. Bonferroni adjustment for multiple comparisons limited the difference at the 0.38 log cd·s/m² flash strength (asterisk in [E]). OP1 and OP2 implicit times are inversely proportional to stimulus strength. When comparing the two older groups with the younger one, the first three OP peaks are delayed (F–I). Statistically significant differences (P < 0.05) †between the young (20–39) and middle aged (40–59), and ‡between the young (20–39) and old (60+) groups.

Wavelet-Derived Parameters (Amplitude, Peak OP Time, and Frequency).

Figure 7A shows representative wavelet scalograms for different flash strengths under light-adapted conditions. More than one distinct band of oscillations can be identified in the frequency domain (white arrows). The red triangle highlights the peak with the maximum amplitude. By applying strict power criteria (local peaks within at least 1/3 of the maximum amplitude) and time domain filtering, candidate OP peaks could be defined more accurately. A frequency-domain representation of all candidate OP peaks as a function of flash strength is given in Figure 7B. At lower strengths (−0.02 and 0.38 log cd·s/m²) one oscillatory peak at approximately 90 to 100 Hz was the main source of OP power. At higher strengths (0.88 and 1.37 log cd·s/m²) two distinct bands of oscillations could be noted: a set of high frequency OPs with a peak at 135 ± 6 Hz and a set of low frequency OPs with a peak at 82 ± 7 Hz. As flash strength increased further, the low frequency band could be recorded only in a subset of young individuals. The high frequency band remained the main source of OP power for the rest of the subjects. The location of the OP peaks in the time domain is illustrated in Figure 7C. High frequency OP peaks (>120 Hz) were time locked to the onset of early OPs, whereas low frequency OPs (70–100 Hz) gave rise to early and late OPs. ANOVA was used to analyze differences in OP peak time and amplitude among groups. For older subjects (middle-aged and old), a significant delay was recorded for most OP peaks (Fig. 7C, asterisk denotes significance at P = 0.05). Amplitude analysis with frequency segregation revealed a pronounced reduction in peaks of the low frequency spectrum (Fig. 8A). Contrastingly, the amplitudes of the high frequency OP peaks remained unaffected by age (Fig. 8B). ANOVA results of wavelet-derived amplitude parameters are summarized in the Table for all flash strengths. Significant differences were found between the young (20–39) and old (60+) groups only. 

Figure 7

 

Analysis of light-adapted OPs using MWT. (A) Representative examples of Morlet wavelet scalograms for increasing stimulus strengths; time zero represents visual stimulus onset. For a subset of stimulus strengths (middle two scalograms), two distinct oscillatory peaks are identified along the frequency domain (white arrows). The red triangle highlights the peak with the maximum power. (B) Segregation of high (hollow circles) and low (filled circles) frequency OP peaks as a function of stimulus strength; aggregated data from all subjects, independently of age. (C) Time domain of OP peaks as a function of stimulus strength. High frequency OP peaks (>120 Hz) are elicited before their low frequency counterparts (70–100 Hz). When comparing the two older groups (red and green) with the younger group (blue), most of the OP trains are delayed regardless of their peak frequency. Statistically significant differences (P < 0.05) †between the young (20–39) and middle aged (40–59), and ‡between the young (20–39) and old (60+) groups.

Figure 8

 

Amplitude of the two distinct OP trains segregated by their respective frequency. (A) Low frequency OP trains have reduced amplitudes in the old (green) compared to the young group only (blue). (B) Amplitudes of high frequency OP peaks are unaffected by age. ‡Statistically significant differences (P < 0.05) between the young (20–39) and old (60+) groups.

Table.

 

Comparison of Morlet Wavelet OP Peak Amplitudes (Segregated by Frequency) Among the Three Age Groups

Intensity,	Low Frequency, 70–100 Hz		High Frequency, >120 Hz
Log cds/m2	F Statistic	P Value	F Statistic	P Value
−0.02	6.745	0.0022*
0.37	11.92	<0.0001*
0.88	13.23	<0.0001*	0.5840	0.5601
1.37	4.071	0.0236*	0.4582	0.6343
1.89			1.040	0.3586
2.36			1.163	0.3194

*  Significant differences.

Comparison of the Two OP Extraction Methods in Pathologic States

The above findings underline the notion that aging can introduce strong confounding factors when assessing inner retina function through OPs. To compare the ability of the two OP extraction methods to identify defects within the context of an aging-associated retinal disease, we recorded light-adapted ERG luminance-response functions and analyzed OPs from five subjects with unilateral neovascular (wet) age-related macular degeneration treated with intravitreal anti-VEGF agents and two subjects with unilateral large sPED. The sPED cases were included as examples of pathology confined to the outer retinal layers. There was no sign of intraretinal or subretinal fluid and, therefore, anti-VEGF therapy had not been initiated. 

Figure 9 shows representative examples of the MWT scalograms with corresponding unfiltered and filtered ERG traces obtained from three different patients with unilateral neovascular AMD under anti-VEGF treatment. The fellow non-neovascular (dry) eye served as an internal control. Figures 9A, 9D, and 9G illustrate the MWT scalogram of the fellow non-neovascular (dry) eye and Figures 9B, 9E, and 9H show the MWT scalogram of the neovascular (wet) eye. Figures 9C, 9F, and 9I show unfiltered and filtered ERG traces from both eyes of each patient. Black traces correspond to the non-neovascular (control) eye and red to the neovascular eye. Patient #1 (top row) presented with subretinal and intraretinal fluid. For responses elicited by a flash stimulus of 0.38 log cd·s/m², MWT detected a significant reduction in the low frequency OP band power (Fig. 9A, 21.66; Fig. 9B, 13.85). Time–amplitude analysis detected no difference in summed or individual OP measurements (summed OP amplitude Fig. 9A, 24.1; Fig. 9B, 24.4). Patient #2 (middle row) presented with subretinal hemorrhage. For responses elicited by a flash stimulus of 0.88 log cd·s/m², MWT analysis of the neovascular eye revealed an extinguished high-frequency OP band with a slight decrease in the low-frequency OP power (Fig. 9D, 22.07; Fig. 9E, 19.36). Filtered OP analysis (Fig. 9F) found significant reduction in the amplitudes of OP2 (almost extinguished) and OP3/4 (−47%). Patient #3 (bottom row) presented with subretinal and intraretinal fluid. For responses elicited by a flash stimulus of 0.88 log cd·s/m², MWT detected a preferential loss of the high-frequency band (Fig. 9G, 14.06; Fig. 9H, 7.27), with relative preservation of the low-frequency OP power (Fig. 9G, 28.57; Fig. 9H, 24.41). Time–amplitude analysis (Fig. 9I) revealed reduction in the amplitudes of OP2 (−50%) and OP3/4 (−40%). Scalograms generated with MWT revealed very close proximity of the two bands in the time domain. Their timing (26–32 ms) coincided with the occurrence of the filtered OP2 and OP3/4 complex. Amplitude reduction in the neovascular eyes was confined to the filtered OP2 and OP3/4 complex peaks. 

Figure 9

 

Morlet wavelet transform scalograms with unfiltered and filtered ERG traces obtained from patients with unilateral neovascular (wet) AMD under anti-VEGF treatment. (A, D, G) The MWT scalogram of the fellow non-neovascular (dry) eye; (B, E, H) show the MWT scalogram of the neovascular (wet) eye. A representative example of a normal MWT scalogram for each studied flash strength is given at the top right-hand corner of (A, D, G). (C, F, I) Respective unfiltered and filtered ERG traces from both eyes of each patient.

Figure 10 shows MWT scalograms with corresponding unfiltered and filtered ERG traces obtained with different flash strengths from a patient diagnosed as having unilateral untreated sPED. Figures 10A, 10D, and 10G illustrate the MWT scalogram of the fellow non-neovascular (dry) eye, whereas Figures 10B, 10E, and 10H show the MWT scalogram of the eye with PED. Figures 10C, 10F, and 10I show unfiltered and filtered ERG traces from both eyes of the same patient. The top row illustrates responses elicited by a flash stimulus of 0.38 log cd·s/m². Morlet wavelet transform detected a significant reduction in the low frequency OP band power (Fig. 10A, 25.5; Fig. 10B, 17.8). Time–amplitude analysis detected a 25% reduction in OP2. The middle row illustrates responses elicited by a flash stimulus of 0.88 log cd·s/m². The MWT analysis of the eye with PED revealed a well-preserved high-frequency OP band (Fig. 10D: 9.96; Fig. 10E, 8.18) with significant reduction in the low-frequency OP power (Fig. 10D, 22.71; Fig. 10E, 16.06). Filtered OP analysis (Fig. 10F) identified OP3/4 as the main source of amplitude reduction (−30%). The bottom row illustrates responses elicited by a flash stimulus of 1.36 log cd·s/m². Power of the high-frequency band showed no difference between the two eyes. The low frequency OP power was primarily affected (Fig. 10G, 20.37; Fig. 10H, 14.67). Time-amplitude analysis (Fig. 10I) revealed a universal decline in individual and summed OP amplitudes. In cases of sPED, all of the OP power reduction in MWT originated from the low-frequency oscillators. In terms of filtered OP analysis, the OP2 and OP3/4 complex peaks remained the main source of amplitude reduction. 

Figure 10

 

Morlet wavelet transform scalograms with unfiltered and filtered ERG traces obtained at different flash strengths from a patient with unilateral pigment epithelial detachment (PED). (A, D, G) The MWT scalogram of the fellow non-neovascular (dry) eye; (B, E, H) show the MWT scalogram of the eye with PED. A representative example of a normal MWT scalogram for each studied flash intensity is given at the top right-hand corner of (A, D, G). (C, F, I) Respective unfiltered and filtered ERG traces from both eyes of the same patient.

Discussion

Previous Work on Age-Related Changes in Human OPs

To our knowledge, age-related changes in human OPs have been addressed in two main previous studies. ^21,22 In the first study, Kergoat et al. ²¹ evaluated individuals aged 75 and older to find amplitude reduction and prolonged implicit times for most OPs recorded under light- and dark-adapted conditions. Our study provided evidence that the inner retinal circuitry undergoes changes much earlier in life (40 years of age). In fact, changes in amacrine cell function may even precede detection of rod and cone dysfunction by other conventional approaches. As such, recording and analyzing OPs, as done in this paper, provides a highly sensitive approach for the early detection of age-related functional changes in the human retina. In the second study, Sannita et al. ²² also examined OPs in a large population of subjects (aged 1–84 years) using Ag/AgCl dermal electrodes. The authors reported an initial increase of the OP amplitude from childhood to adulthood, followed by a decrease above 50 years of age. Implicit times for most OPs also increased with age. Although in agreement with our findings, results from this study should be interpreted cautiously, since all full-field ERG recordings were performed in nondilated eyes. 

OP Response Attenuation With Age

In a previous study, reduction in outer retinal a-waves with preserved postphotoreceptoral b-waves were noted for individuals above 60 years of age (Ref. 11). There was no difference in light-adapted a- and b-wave amplitudes among the three age groups for low and mid intensity stimulus strengths, where OP amplitude reduction was detected for the older cohort (Ref. 11). In terms of sensitivity, both dark- and light-adapted a- and b-wave implicit times were delayed in the middle aged and older groups (Ref. 11). We cannot exclude the fact that delays in peak OP times simply reflect age-related changes in pre- and postsynaptic kinetics. On the contrary, the observed OP amplitude attenuation in dark-adapted retinas of middle-aged individuals and light-adapted retinas of older individuals seems to be independent of photoreceptor and/or bipolar cell amplitude reductions. To further support the precociousness of age-related OP changes, we correlated dark-adapted summed OP, a-wave, and b-wave amplitudes elicited with the brightest flash of 1.37 scotopic log cd·s/m² with age. Results are summarized in Supplementary Figure S2. We found a strong linear association between dark-adapted summed OP amplitude (as assessed with the traditional trough to peak measurement) and age (Fig. C, r ² = 0.54; slope = −0.73, P < 0.001), while there was only a weak association between a-wave amplitude and age (Fig. A, r ² = 0.07; slope = −0.27, P = 0.04). B-wave amplitude showed no association with age (Fig. B, r ² = 0.0017; slope = −0.042, P = 0.76). 

Age-related alterations in crystalline lens density, which reduce retinal luminance in older individuals, could possibly explain the recorded attenuated OP responses, especially under dark adaptation. Even though some people do develop cataracts during their middle-age years (40s and 50s), these cataracts tend to be very minimal to affect ERG responses. In the present study, attenuated OP responses were recorded already by 40 years of age. In addition, a subanalysis restricted to older pseudophakic individuals (n = 6, 68–80 years of age) revealed reduced OP amplitudes and delayed implicit times, suggesting that changes are most likely of retinal origin. Although underpowered to extract statistically significant conclusions, these data reinforce the lack of potential contribution of cataracts to the age-related changes in OP properties reported herein. 

Affected parameters of the oscillatory response have been linked to underlying inner retina dysfunction, ⁴ especially loss of integrity in the retinal microcirculation. ²³ Such dependence has become clear through studying human retinal disorders that show attenuated OP responses as a characteristic ERG feature: diabetic retinopathy, ^24–27 retinopathy of prematurity, ²⁸ central retinal vein occlusion, ^29,30 and primary open-angle glaucoma. ^31–34 The documented impairment of inner retinal cell function during the early stages of these diseases seems to closely resemble the age-related pattern of decline seen with our study. The OP response attenuation with age, therefore, could reflect functional changes resulting from either dysfunction/loss of inner retina cells, ^35,36 and/or changes in the inner retinal and choroidal vasculature. ^37,38  

Evidence of Two Distinct OP Peaks in the Human Retina

The MWT analysis of the human ERG signal allowed simultaneous segmentation of the OP response in the frequency and time domains. This method revealed the existence of more than one oscillator in the dark- and light-adapted human retina. Similar findings were reported in rats by Forte et al., ¹³ the original group to detail the MWT application to analyze the dynamics of rat OPs. In their experiments, two clusters of dark-adapted OP bands at 70 and 130 Hz were identified. For moderate flashes, OP oscillators overlapped in time (50 ms). At the highest stimulus strengths, the clusters were 20 ms after the flash in the 70 Hz band, and 50 ms after the stimulus in the 130 Hz band. A similar pattern characterized human dark-adapted OP responses, with bands at 77 and 155 Hz. At higher strengths, peak time of the low frequency band occurred earlier than the high frequency and in close proximity to the larger and steeper a-wave, making quantification unreliable without preconditioning ERG waveforms to remove the a-wave. 

In the light-adapted human retina, MWT revealed a dual-band oscillatory system for specific flash strengths. High frequency peaks occurred earlier than their low frequency counterparts. Several studies have provided support for an intraretinal source differentiation of generators of early and late OPs. Pharmacologic studies in the mudpuppy retina have related the earlier OPs to the ON component and the later ones to the OFF component of the ERG. ^39,40 Patients with congenital stationary night blindness, who have defective ON pathways, preferentially lose the first two OPs (early OPs) on the rising edge of the photopic b-wave. ⁴¹ The last OP wavelet of the human ERG is time-locked to stimulus offset. ⁴² It has been suggested previously that the last major OP of the photopic response as well as the i-wave (a post b-wave component recorded under light adaptation in humans, originating, at least in part, from retinal ganglion cells) are generated by the retinal OFF pathway. ⁴³ In addition, pharmacologic disruption of ON and OFF pathways has been shown to differentially affect the OPs in the amphibian retina; GABA-receptor antagonists modulate early OPs, whereas glycine-receptor antagonists affect the late OPs. ⁴⁴ Zhou et al. ⁴⁵ matched the slow-sequence multifocal ERG from the macular region of the retina of primates with Gabor functions and showed that OPs fell into two distinct frequency bands: a high frequency band peaking at approximately 150 Hz that contributes to early OPs, and a low frequency band peaking at approximately 80 Hz that contributes to early and late OPs. Based on the above, the segregation of light-adapted OP responses into two frequency bands separated in time suggests the possibility of ON and OFF system representation. Even in dark-adapted (rod-driven) OP responses, the OFF system still may be represented, since there has been recent evidence of a direct excitation of cone OFF bipolar cells by rod photoreceptors in rodents. ⁴⁶  

Luminance Dependence of the Dual-Band Frequency Oscillatory Pattern

In the present study, MWT analysis not only confirmed the existence of a dual-band frequency oscillatory pattern in the light-adapted human retina, but also revealed the luminance dependence of this pattern. A plausible explanation for the presence of a stimulus strength threshold for the appearance of the high frequency band could be sought at the complex intensity-response function of the cone-dominated human ERG per se (note that in the text, the term “intensity” is also referred to as “strength”). In response to progressively brighter stimuli, the b-wave of the light-adapted ERG gradually increases in amplitude, reaches a plateau for a narrow range of stimulus strengths and then rapidly decreases with further increments in the flash strength. This unique phenomenon is known as photopic hill. ^20,47 According to Ueno et al., ⁴⁸ the characteristic shape of the photopic hill results from the summation of two events: at higher flash luminance, the ON response amplitude decreases and the positive peak of the OFF response becomes delayed. This is supported by the fact that mathematical modeling of a combination of ON and OFF cone response curves can produce a hill effect. ⁴⁹ Intensity–response analysis of light-adapted wavelet derived parameters showed that for low to moderate light stimuli (<0.88 log cd·s/m²) only the low frequency (80–100 Hz) band could be identified. At these stimulus strengths, OFF bipolar cells (hyperpolarizing type) are known to contribute mostly to the generation of the cone a-wave. ⁵⁰ Interestingly, the stimulus strength threshold for the appearance of the dual-band frequency pattern coincided with the occurrence of the plateau and decline phase of the photopic hill function (Supplementary Fig. S1G). ¹¹ The relative amplitude of the high frequency band remained stable with further increments of flash strength (∼10 arbitrary units of power), whereas the relative amplitude of the low frequency band yielded a characteristic “photopic hill” function (see Fig. 5). Rufiange et al. ²⁰ found that the OPs previously associated with the retinal ON pathway (OP1 and OP2) do not appear to show any deterioration as a result of progressively brighter flashes. Contrastingly, brighter flashes led to a complete abolition of the late OPs (mainly OP4) that are generated by the retinal OFF pathway. ⁵¹ Although not proven directly, it can be postulated that the ON and OFF pathways are not only expressed in OP generation, but also may possess different frequency spectrum characteristics. In support of this hypothesis, only OP peaks in the low frequency band showed vulnerability to age-associated decline. Suzuki et al. ⁵² reported a higher age-related vulnerability of the OFF compared to ON bipolar cells. Further work will be required to determine whether the distinct groups of oscillators identified by wavelet analysis can be attributed to pharmacologically distinguishable retinal pathways. 

Advantage of Morlet Wavelet Transform Over Time–Amplitude Analysis

As a potential age-independent marker of inner retina function, the high frequency OP band could be of significant clinical utility, as one could exclude age as a confounder and attribute with higher confidence deterioration to a disease state per se. We compared the two OP analytical methods in the context of an age-associated disease, to investigate what additional information MWT could provide over traditional time–amplitude analysis. Our findings show that eyes with neovascular (wet) AMD treated with anti-VEGF agents had selectively reduced OP power in the high frequency compared to the low frequency band, when compared to the untreated fellow eye. In cases of treatment-naïve sPED, all of the OP power reduction in MWT originated from the low-frequency oscillators. Filtered OPs failed to provide a similar differentiation; in neovascular AMD and sPED cases, the OP2 and the OP3/4 complex peaks accounted for the recorded amplitude reduction. A possible explanation behind this observation is that, as an approach, time–amplitude analysis relies on the summed amplitude of multiple oscillators. For instance, based on timing, the OP2 and the OP3/4 complex may reflect activity from the low and high frequency oscillators. Scalograms generated with MWT in the neovascular AMD eyes show the occurrence of low and high frequency bands at the very same time (30 ms) of occurrence of the OP3/4 complex for a specific flash strength (Fig. 8). 

In conclusion, MWT provides complementary information not obtainable through traditional analytical methods and, as such, optimizes the differential diagnosis of retinal disorders. Application of this approach to retinal dystrophies with distinct pathway deficits not only would further validate the clinical utility of MWT as a tool, but also provide further insight into the physiological origin of multiple oscillators in the human retina. 

Supplementary Materials

pdf S1 

Acknowledgments

The authors thank Sharee Kuny for proofreading the final version of the manuscript. 

Supported by Canadian Institutes of Health Research (CIHR) Grants 151145 and 192321; Alberta Innovates Health Solutions (AIHS) Establishment Grant 200700584; Canadian National Institute for the Blind; Olive Young Foundation, and The Lena McLaughlin Foundation (Mona & Rod McLennan); by an Alexander S. Onassis Foundation Scholarship and an AIHS Graduate Studentship (ISD); and by an AIHS Senior Scholarship (YS). 

Disclosure: I.S. Dimopoulos, None; P.R. Freund, None; T. Redel, None; B. Dornstauder, None; G. Gilmour, None; Y. Sauvé, None