Outcome measure for the treatment of cone photoreceptor diseases: orientation to a scene with cone-only contrast

Subjects and clinical data

Patients (n?=?7) with GUCY2D-LCA (Table 1) and subjects with healthy vision (n?=?8) participated in this study. GUCY2D-LCA patients enrolled in the current study were different than those included in
our previous publication 17]. All subjects had complete clinical ocular examinations, including best-corrected
visual acuity. The tenets of the Declaration of Helsinki were followed, and informed
consent and assent were obtained from all patients. The research was approved by the
institutional review board at the University of Pennsylvania.

Table 1. Clinical and genetic characteristics of the GUCY2D-LCA patients

Orientation device: discrimination of door on wall by rods and cones

Fifteen tricolor LED strips (Neopixel RGB-PID1460, Adafruit Industries, NY) were hung
with 0.1 m lateral separation to result in a vertical plane resembling a wall that
is 1.4 m wide by 2 m high. The device subtends 34° horizontally and 48° vertically
when observed from the test distance of 2 m. The LED strips were left free to flow
vertically permitting traversal in a way similar to walking through a beaded curtain
(Fig. 1a, left). The direction of the maximum emission of the LEDs was kept perpendicular
to the plane by the montage to assure uniformity of light output across strips. Each
vertical segment contained 60 triplets of individually addressable red, green and
blue LEDs with peak wavelengths of 630, 518 and 460 nm, and FWHM of 14, 34 and 22 nm,
respectively (Fig. 1a, upper right). LED spectra were measured with a spectrometer calibrated with a wavelength
scale reference (USB2000 and Hg-1, Ocean Optics, Dunedin, FL). Each individual LED
was controlled by a dedicated constant-current driver (WS2812B, WorldSemi, China)
with 8-bit pulse-width-modulation (PWM) resolution permitting electronic luminance
adjustment over a 2.41 log unit range. All strips were controlled using a single microcontroller
(Arduino Mega, Ivrea, Italy) with an USB connection to a PC. Custom software was written
for both the microcontroller (Wiring) and the PC application (Java, Linux). The device
was programmed to produce a rectangular pattern target, 3-strip wide, defining the
“door”, surrounded by the rest of the strips defining the “wall”. Two color combinations
were used: green door on a blue wall (Fig. 1a, left), and a green door on a red wall (not shown). The perception of the door on
the wall would be expected to depend on the luminance of each color and luminance
difference between the colors. Also contributing to the perception are the adaptation
state of the subject and the availability of functioning photoreceptor populations
in the retina. For the main test trials, the intensities of the two lights used in
each combination were chosen to be scotopically-matched and produced the same relative
effectiveness for the rod-photoreceptor based vision. Due to the differences in the
spectral sensitivity curves between scotopic and photopic systems, the scotopic-match
would result in mismatched effectiveness for photopic cone-photoreceptor based vision
(Fig. 1a, lower right). Specifically, scotopically matched combinations of blue/green and
green/red produce a luminance contrast for cones of 0.8 and 2.5 log units, respectively
(Fig. 1a, lower right). The scotopic matches were preliminarily determined using a radiometer
calibrated for scotopic luminance (IL1700 with ZCIE filter, International Light, Peabody,
Massachusetts) and the matches were fine-tuned near the normal rod absolute threshold
based on results from dark-adapted normal observers using the final device implementation.
Based on previous dark-adapted sensitivities obtained with two-color methods 17], we assumed scotopic matching in GUCY2D-LCA to be similar to normal subjects.

Fig. 1. Presentation of scotopically matched scenes on an LED-lit wall. (a) Left, LED-lit wall and door device used for the three-alternative forced-choice
design. A mix of light intensities for the “door” (three stripes at right, center
or left) and “wall” (elsewhere) is calculated to be visible to cone-based vision but
unable to be differentiated with rod-based scotopic vision. Right, spectral content
of the LED lights compared to the spectral sensitivities of the rod (scotopic, V’
_?
) and cone (photopic, V
_?
) vision. All curves are normalized to unity at maximum. The relative intensities
of the three lights are set such that they have the same effectiveness for rods (lower
left) but substantially different effectiveness when perceived by cones (lower right).
These differences permit differentiation of door and wall, and enhance visual orientation
performance. (b) Rod vision is expected to perceive the scene as uniform stripes with no features
(upper panels), and cone vision is expected to perceive the “door” as green stripes
on a blue background (lower panels). (c) Appearance of unmatched control trials that should be visible to either rod or cone
systems. (d) Departures of more than approximately 1.5 dB (0.15 log) from the scotopically matched
mix of lights result in successful door-wall differentiation in normal subjects. (e) Representative examples of scotopically-matched test trials interleaved with unmatched
control trials. For a three-alternative forced-choice setup, a subject with rod vision
but no cone vision would be expected to get 33 % correct for matched-stimuli on average,
whereas a subject using cone vision would be expected to get 100 % correct. Unmatched
control presentations with door dimmer (1) or brighter (2) than wall are used to determine
whether the subject has any vision (rod or cone or both) under the testing conditions

A three alternative forced-choice (3-AFC) experimental design was implemented by presenting
the “door” at three locations (left, center or right) on the “wall” (Fig. 1b). The percent correct door position identification was estimated by conducting a
set of at least ten trials, and required n???7 correct determinations to reject the
null hypothesis of no discrimination at the ??=?0.05 level, yielding a power of 88 %.
Several sets of trials were performed sequentially to determine the percent correct
as a function of scene luminance. Specifically, experiments started with fully-dark-adapted
subjects and a very dim scene luminance, in which perception is rod-mediated in normal
subjects, and it was increased until the dynamic range limit of the device was reached.
The relative door/wall intensities were kept scotopically matched at all scene luminances.
As full mobility course traversals for this number of trials would become prohibitive,
we divided the assessment into two stages. A first stage was designed to determine
the orientation capacity of subjects, and to obtain the contributions of rod and cone
photoreceptor systems towards orientation capacity. The orientation sessions reported
in the current work consisted of verbal assessments of the door location on the wall
while the patient is located at the start of the course. The trials could proceed
rapidly and total session durations were kept below 30 min. A second stage designed
to obtain mobility-specific metrics and consisting of active course traversals may
be developed in the future as treatment initiatives become available. Such a second
mobility stage would piggyback on the orientation stage reported here by using a carefully
selected subset of scene luminance parameters appropriate to each subjects’ level
of vision. Also, the test paradigm is flexible enough to be adapted for use in the
future in CPDs that differ in the relative losses of each cone subsystem, or in treatment
approaches that may target preferentially one of the cone classes.

Detailed methodology to determine orientation capacity

Each dark-adapted (45 min.) subject is placed at the start of the course wearing
a pair of goggles fitted with neutral density filters (Roscolux R98, Rosco Laboratories,
Sun valley, CA). The goggles are used to adjust the scene luminance over a greater
dynamic range than possible electronically, and do not interfere with the scene boundaries
when looking straight ahead. The different levels of scene luminance are obtained
by varying the number of neutral density filter sheets in the goggles. As the spectral
absorbance of the individual sheets is not perfectly flat, the combined spectral characteristic
of the stack of filters changes slightly as more sheets are added. To compensate for
this effect, the resulting light output of the combination of device plus filter stack
is kept scotopically matched at all times by small electronic adjustments of the LED
PWM level.

A set of trials consists of no less than 10 scotopically-matched test trials plus
a number of scotopically- and photopically-unmatched control trials (Fig. 1c, left). One of these sets is performed for each scene luminance, proceeding from
dim to bright. The unmatched trials are used for control (should produce a success
fraction close to 100 % for subjects having any form of orientation vision at the
tested scene luminance) and to maintain orientation during the trials, and also to
explain the test to subjects with very low vision. The contrast level for the control
trials greatly exceeds the range of scotopic matches in normal subjects, which is
approximately 0.4 log wide (Fig. 1d). Matched and mismatched trials are randomly interleaved by the software, with control
(unmatched) trials being 20 % of the total trials on average. For each trial, a sound
is produced after the adjustment of the wall and door intensities by the computer,
and the subject is asked to verbally report the apparent door position as “Left”,
“Center” or “Right”; they are encouraged to guess one of the three choices if unable
to determine where the door is. The responses are recorded by the computer. Figure 1e illustrates representative examples of three sets producing different success rates
(33, 50 and 100 %) for test trials (filled circles) with instances of interleaving
control trial sequences (open triangles). Failure to discriminate the door position
is signaled by a yield of ideally 33 % of correct answers to the 3-AFC design. Full
discrimination implies near 100 % answers being correct. Results are acquired under
free viewing conditions, except for illuminations requiring attenuating goggles.

All normal data were fit with a logistic function of the form

where p is the probability of correct, x is the scene luminance, ? is guessing probability (fixed to 0.3) and ? is the lapsing probability (fixed to 0) 20]. The spread parameter, ?, was allowed to vary for normal results but held constant
at the normal value for the patient results. The threshold parameter, ?, was allowed to vary for both normals and patients.