A Novel Method for Objective Vision Testing in Canine Models of Inherited Retinal Disease

Abstract

purpose. The use of canine models of retinal disease in the development of therapeutic strategies for inherited retinal disorders is a growing area of research. To evaluate accurately the success of potential vision-enhancing treatments, reliable methods for objectively assessing visual function in canine models is necessary.

methods. A simple vision-testing device was constructed that consisted of a junction box with four exit tunnels. Dogs were placed in the junction box and given one vision-based choice for exit. The first-choice tunnel and time to exit were recorded and analyzed. Two canine models of retinal disease with distinct molecular defects, a null mutation in the gene encoding the α subunit of rod cyclic GMP phosphodiesterase (PDE6A), and a null mutation in the gene encoding a retinal pigment epithelium–specific protein (RPE65) were tested and compared to those in unaffected dogs.

results. With the use of bright light versus dim red light, the test differentiated between unaffected dogs and dogs affected with either mutation with a high degree of certainty. The white-light intensity series showed a significantly different performance between the unaffected and affected dogs. A significant difference in performance was detected between the dogs with each mutation.

conclusions. The results indicate that this novel canine vision-testing method is an accurate and sensitive means of distinguishing between unaffected dogs and dogs affected with two different forms of inherited retinal disease and should be useful as a means of assessing response to therapy in future studies.

Recently, the dog has become an important model in the study of inherited retinal degenerative diseases with several spontaneously occurring retinal diseases identified. ¹ Dog models have advantages over rodent models, including a larger eye that lends itself easily to surgical manipulation and a visual system that is adapted to vision in both day and night, which is more comparable to the human condition than rodent models. ¹ Ultimately, the goal of much research in this field is the development of therapeutic strategies for managing these conditions. Inroads toward successful gene therapy have already been made in the RPE65 mutant canine model, ^{2 3} shown by electroretinography but with subjective visual function assessments. Any progress made toward therapeutic development must show, as an essential measure of success, a quantifiable improvement in a dog’s response to a visual cue. A quantifiable measure of improvement in visual performance in dog models is a current unmet need and represents a significant challenge to researchers in this field. 

Currently, most researchers of canine retinal inherited diseases use subjective observation of a treated dog’s visual function when observed interacting with its environment or passing through an obstacle course compared with untreated dogs. ² Sometimes an attempt is made at a more objective assessment by having an observer blinded to the experimental conditions record the number of times a dog bumps into obstacles such as traffic cones, ^{4 5} but the results are either recorded as the number of collisions or as pluses and minuses, neither of which yields satisfying statistical analyses. Jacobs et al. ⁶ performed ocular motility recordings of the nystagmus associated with the RPE65 phenotype to examine phenotypic rescue after gene therapy, but were forced to correlate this measure with a subjective evaluation of visual performance in an obstacle course. 

Obstacle course studies are limited by several factors. Individual dogs in a colony have significant differences in boldness and intelligence that affect behavior. Some dogs have no motivation to move through the obstacle course, and calling the animal excessively can arguably aid a vision-impaired dog in completing the task by relying on other senses. Sometimes even sighted dogs can be excitable and bump into obstacles, creating false or misleading results. In addition, in our experience, it is difficult even for an observer who is blinded to the experimental conditions to refrain from identifying with the dog and possibly interjecting observer bias. 

Mice and rodent models are also used extensively in vision research and quantitative visual assessment has been developed and used in these species. Prusky et al. ^{7 8} have developed a two-alternative, forced choice task that makes use of a water tank with a semisubmerged platform over which a visual cue is projected. The mice must swim and therefore learn to swim toward the visual cue and its platform. This forced-choice approach results in a test that is quantifiable and generates data that lends itself easily to statistical analysis. The dogs’ large size and behavioral differences from mice limit the ability to use such a forced-choice testing apparatus as a water maze. Optokinetic devices can be devised and used to measure visual function of small subjects such as rodents ⁹ and chickens. ¹⁰ Dogs will exhibit a postural change in response to optokinetic stimulation ¹¹ ; however, these studies used small numbers of dogs that were painstakingly trained to stand still on a platform before testing could take place. 

Training dogs and even larger animals such as horses to a task based on visual cues has been used to study visual behavior and pattern recognition. ^{12 13 14 15} However, these studies either used harsh negative reinforcement ¹² or tended to use small numbers of animals and involve extensive positive reinforcement training, with sometimes thousands of trials necessary to generate meaningful results. ^{13 15} Similar visual-based testing has been used extensively to measure cognitive function in a canine model of aging in which some dogs learn a visual-based task more quickly than others. ^{16 17} This type of training is also very time consuming, with dogs requiring daily training sessions for months in some studies. ^{16 17} A larger number of dogs, usually with differing temperaments and aptitudes, are necessary for therapeutic trials than for visual behavior studies. Considering the need to minimize training time and avoid corporeal negative reinforcement, training for such visual tasks is not a practical option. 

There is a clear need for a vision test that is simple, relatively rapid, and based on a behavior that is either a natural action or learned easily by most dogs. We observed that most of our dogs, when placed in a box or crate, naturally wanted to exit. We have used this natural behavior to create a four-alternative, choice-based visual testing system that offers a means of objectively quantifying visual performance under different lighting conditions in dogs with inherited retinal disease that is rapid, repeatable, quantifiable, and generates data that can be analyzed by statistical methods. 

Methods

Animals

Dogs used in the study were purpose-bred members of breeding colonies that were either homozygous for one of two mutations that result in retinal diseases (an RPE65 null mutation ¹⁸ or a PDE6A null mutation ¹⁹ ), and unaffected normal or were phenotypically normal carrier animals used as control subjects. Table 1contains a summary of dogs used in the study, their disease states, and which vision tests were performed. The PDE6A mutant dogs were of the Cardigan Welsh Corgi breed, and those with the RPE65 mutation were Beagle-Briard crosses. The dogs were all of a medium size, weighing between 20 and 40 pounds. The Corgi dogs were of shorter stature than the Beagle crosses, but generally weighed about the same. Ages of the 19 dogs used in the study ranged from 7 to 51 months. There were 15 females and 4 males. The dogs were housed in a vivarium under standard 12-hour light and dark cycles and treated in accordance with the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research. 

Testing Apparatus

The testing device consisted of a junction box (3 × 3 × 3 ft) with four opaque canvas tunnels, one extending from each side. Circular openings (28-inches in diameter) were cut into the center of each side of the central box. A wire-reinforced canvas tunnel, 28 inches in diameter and 36 inches long, was inserted into each opening and secured. A top was constructed for the central junction box, and snug-fitting end caps were made for the end of each tunnel (Fig. 1) . 

Testing Conditions

The apparatus was set up in a windowless room with overhead lights, and four lamps equipped with dimmer switches were positioned between the tunnels. Dim red or white light bulbs were used in the lamps. For one experiment, goggles for dogs (Doggles, LLC, Diamond Springs, CA) with blocked side holes and fitted with opaque black lenses were used to blindfold normal dogs, to verify that there were not any nonvisual cues that could indicate which tunnel was open. 

Before an experiment was begun, the random order of open tunnels was predetermined with a random-number generator from 1 to 4 (Excel; Microsoft Corp., Redmond, WA). For each run of the test, the ends of all tunnels except the designated open tunnel were capped with snug-fitting end caps. 

Test dogs were placed in the junction box in the same orientation for each run, with the person placing the dog in the device remaining between two tunnels to observe the dog’s choices without being visible through an open tunnel. A second observer was in place between the opposite two tunnels so that the dogs could not gain visual cues, and each observer monitored two tunnels. A stopwatch was started as the dog’s feet were placed on the floor inside the box, and the top was quickly placed on the box. The first tunnel chosen and time taken for the dog to exit the device were recorded. 

The first-choice tunnel was defined as the first tunnel the dog entered. When a dog entered a tunnel, it caused the tunnel to rustle noticeably, which allowed us to detect the presence of the dog in the tunnel in all lighting conditions. The junction box was large enough for the dogs to turn in circles without rustling a tunnel, allowing us to distinguish easily between movement within the junction box and movement within the tunnels. Timing was stopped when the dog’s nose reached the open end of a tunnel. If the dog had not exited the device in 60 seconds, the run was stopped, and the time recorded as >60 seconds. If the dog remained in the junction box without entering a tunnel, the first choice was recorded as 0 and treated as an incorrect first choice in the analysis. 

Training

Training was often unnecessary, as the dogs tended to exit the device naturally. The dogs were given praise as they exited the device. When dogs were introduced to the device, they were given several practice runs in full room light that were not timed or recorded to explore the tunnels. The tunnels were for exit only, and the dogs were never allowed to enter the device via the open tunnels, only by being picked up and placed in the junction box. During the practice runs, the dog’s name was used to call it from the device, but never from a location where the dog could look down the open tunnel and see the caller. Most dogs began exiting the device via the open tunnel, either immediately or within four to five runs. Dogs that did not feel motivated to exit the device during these training runs were encouraged to do so by the observer tapping on the outside of the box or tunnel, making noise, and creating mild negative reinforcement for staying inside, combined with praise on exit. 

We observed certain behaviors to be indicative of loss of interest during testing for both affected and unaffected dogs. They included jumping upward after being placed in the junction box, scratching at the sides of the device or lying down in a tunnel. If a dog exhibited any of these behaviors during testing, the box or tunnel would be tapped to give negative reinforcement, and that run was dropped from the analysis. If the dog’s attention could not be regained within one or two runs, the dog was taken back to the kennel for a break. If a dog walked into an incorrect tunnel, simply stood still in the junction box, or turned circles without making a tunnel choice, it was not reprimanded with tapping on the outside of the device. Food rewards were not used in this study, partly to reduce the effect dropped crumbs may have on the dogs’ performance and partly because this particular group of dogs seemed to be relatively more motivated by praise than food. 

Bright-Light versus Dim-Light Testing

For the full-light condition, the overhead lights were on, and the dogs were run through the device at least seven times. Overhead lights were then turned off, and the test was run at least seven times in the dim red-light condition. The light detector of the photometer used in this study could not be used to measure the intensity of red light. Red lighting was set up in the identical way for each test dog. Seventeen dogs were tested under bright versus dim conditions, 11 unaffected normal control dogs, 3 affected dogs homozygous for the RPE65 mutation, and 3 affected dogs homozygous for the PDE6A mutation (Table 1) . 

White-Light Intensity Series

Light intensity at the opening of each tunnel was measured (IL 1700 Research Radiometer with SED033 silicon light detector; International Light, Inc, Peabody, MA). Under full room light conditions, all lamps were turned to the highest intensity and the overhead lights were kept on. The light level at each tunnel opening was recorded. We chose to measure light intensity at the tunnel opening rather than inside the device, because it could be rapidly and reproducibly measured without disturbing the device’s position relative to all the lamps. The device was not light-tight, but the light level at the tunnel opening was at least 1 order of magnitude brighter than the light level inside the box or inside the tunnel for all conditions (data not shown). The different light intensities were created with the overhead lights off, using four lamps placed between tunnel openings with lights pointing upward. White light bulbs of various wattage and dimmer switches on each lamp were used to create the light intensity conditions. For a given lighting condition the intensities at the tunnel openings were as close as possible to each other. The intensity reading at each opening was recorded and used individually in the analyses. Because the setup and method were being developed during the course of the study, the intensities used and the number of intensities were not the same for each dog. The number of intensities used in these series ranged from 12 to 6. Most dogs were tested with six intensity levels, with intensities falling into the broad ranges of 8 to 26, 1.5 to 4, 0.1 to 0.8, 0.04 to 0.09, 0.01 to 0.03, and 0.001 to 0.006 cd/m². For each intensity series experiment, no matter where in the broad range the intensity was, the light levels at the individual doors were made as close as possible to each other. They were typically within 1 to 3 cd/m² of each other for intensities between 1 to 4 cd/m², 0.1 to 0.3 cd/m² of each other for intensities between 0.9 to 1.0 cd/m², 0.01 to 0.03 cd/m² of each other for intensities between 0.09 to 0.1 cd/m², and so on. The greatest variability in intensities between tunnels was during the full-light condition due to the placement of overhead lights relative to tunnel openings. The intensity at each tunnel opening was used individually in all analyses. The normal dog (G) was subjected to four intensities, from 14 to 0.001 cd/m². 

Seven dogs (dogs A–G) were tested in a light-intensity series. Three dogs with the RPE65 mutation (A–C) and three dogs with the PDE6A mutation (D–F) were tested. Individuals that were used for both bright versus dim testing, and the intensity series are indicated in Table 1 . Dog G in the intensity series testing was an unaffected dog and was also used in the bright and dim testing. 

The dogs were run through the device at least seven times at each light intensity. Occasions when a dog was run through the device more than seven times included dogs that had apparent loss of attention during testing and were given a break (see the Training section), or cases in which in the same tunnel was open more than two times in a row, possibly leading to the subject’s choice being influenced by learning the location of the open tunnel. In these cases, seven additional runs were performed, and all test runs were included in the analysis. The intensity series testing was performed with each light intensity level in the series randomized rather than proceeding from brightest to dimmest or vice versa. The dogs were given from 5 to 10 minutes of adaptation time for each light condition before beginning runs through the device. Two to three runs at full light intensity separated each light condition. 

Data Analysis

Two responses were recorded for each test: whether the first tunnel entered by the dog was the open exit (correct choice) and time to exit. The proportion of correct exit choice for full versus dim light were compared in a two-tailed test of proportions. Exit times at fixed light levels were approximately log-normally distributed. To assess a difference in performance between full- and dim-light conditions, a two-sample t-test was performed using the natural log of exit time. Exit times greater than 60 seconds were not recorded. In the case in which one or more of the dim-light exit times exceed 60 seconds those points were dropped from the time-to-exit analysis, and two-sample t-tests were performed for the reduced data sets. Because the only tests that were terminated occurred in the low-light conditions, the difference in performance estimated by the two-sample t-test with truncated data sets can only be smaller than the actual performance difference. In these cases, we quote the probabilities of the t-tests with the truncated data sets and note that the probabilities would have been even smaller if we had been able to include the longer exit times from the excluded tests. Note that had there been terminated tests in both the full-light and low-light conditions, we would not have been able to make this argument, and a more sophisticated analysis would have been necessary. 

For full-intensity series, the exit time data were analyzed by using a linear regression of the natural log of exit time versus the natural log of light intensity. Correct exit choice as a function of natural log of light intensity was analyzed by binary logistic regression (Minitab ver. 14; Minitab, Inc, State College, PA). 

Electroretinogram (ERG)

Animals were anesthetized under a dim red light. They were premedicated with acepromazine maleate (0.1–0.3 mg/kg) intramuscularly, induced with thiopental sodium (6–12 mg/kg IV) and maintained with isoflurane delivered in oxygen. A pulse-oximeter (Vet/Ox 4400; Heska Corp., Fort Collins, CO) was used to record pulse rate and oxygen saturation for the duration of the procedure. Body temperature was maintained with a heating pad. Pulse rate, oxygen saturation, and body temperature were recorded every 5 minutes during the ERG recording. 

Both pupils were maximally dilated with 1% tropicamide (Mydriacyl; Alcon Laboratories, Honolulu, HI) and 10% phenylephrine hydrochloride (AK-Dilate; Akorn Inc., Buffalo Grove, IL). The globes were positioned in primary gaze with stay sutures of 4-0 silk (Ethicon, Inc., Piscataway, NJ) placed in the conjunctiva adjacent to the limbus. A 2.5% hydroxypropyl methylcellulose solution (Goniosol; Iolab Pharmaceutical Inc., Claremont, CA) was applied to keep the corneas moist. 

A needle reference electrode (Grass; Grass Technologies, Warwick, RI) was placed in the skin over the zygomatic arch 1 cm from the lateral canthus, and a ground needle electrode was placed in the skin of the dorsal neck. Full-field, short-flash ERGs were recorded (ERG-Jet electrode lenses; Universo, La Chaux-de-Fonds, Switzerland) and an electrophysiology unit (UTAS-E 3000; LKC Technologies Inc., Gaithersburg, MD). The band-pass was set at 1 to 500 Hz; the gain setting varied from 2 × 10³ to 4 × 10⁴. Interstimulus intervals were set and ERG responses averaged based on flash intensities. Dogs were dark adapted for 60 minutes, and ERG responses from a series of 16 white flash stimuli (−3.18, −2.98, −2.79, −2.6, −2.0, −1.6, −1.19, −0.79, −0.39, −0.001, 0.39, 0.85, 1.36, 1.9, 2.38, and 2.82 log cd/m²) were recorded. Interstimulus intervals were increased from 1 second at low intensities to 360 seconds at the highest intensity to avoid light-adapting the rods. Preliminary studies showed that these interstimulus intervals were adequate to prevent rod adaptation in normal dogs (data not shown). The a- and b-wave amplitude (in microvolts) was measured for each averaged response. The a-wave amplitude was measured from the onset of light stimulus to the trough of the first negative wave; b-wave amplitude from the trough of the first negative wave to the peak of the first positive wave. ERG amplitudes were plotted as a function of light stimulus. Light-adapted ERGs were recorded after exposure to a background light of 30 cd/m² for 10 minutes and the ERG in responses to white-light flashes of the following intensities superimposed on the same background light was recorded: −0.39, 0.39, 0.85, 1.36, 1.9, 2.38, and 2.82 log cd/m². 

Results

ERG Assessment of Retinal Function in Tested Dogs

Retinal function in five of the six affected dogs was assessed by means of scotopic and photopic ERG. ERG analysis was not performed on RPE65 mutant dog C because she was pregnant and considered an anesthetic risk. Dark-adapted and light-adapted a- and b-wave amplitudes (average of right and left eye) versus flash intensity were plotted for PDE6A mutant, RPE65 mutant, and unaffected dogs (Fig. 1) . Both a- and b-wave amplitudes for dogs with either mutation were lower than amplitudes for the unaffected dog. The RPE65 dogs had higher dark- and light-adapted amplitudes than the PDE6A dogs. PDE6A dog D had no recordable wave form on ERG. 

Testing Device

The testing device, with its central junction box, tunnels, and end caps is shown in Figure 2 . Nineteen dogs were tested in the device, in two light conditions, in a light-intensity series, or in both. Most dogs (17/19) learned the process easily with only four to five practice runs before data were recorded. An RPE65 mutation dog negotiating the test in both bright and dim conditions is shown in Movie 1, and in Figure 3 . The dog was first placed in the device in full room light and, after choosing the correct tunnel first, exited the device (Fig. 3A) . Next, the dog was placed in the box under dark conditions, in the same orientation, with the same tunnel open for demonstrative purposes (Fig. 3B) . Under actual testing conditions, the open tunnel was randomized for each run through the device. The dog selected an incorrect tunnel first, then proceeded back and forth between two incorrect tunnels before finding the way out of the open tunnel. 

Test’s Accuracy in Distinguishing between Affected and Unaffected Dogs

Whether the test could reliably distinguish between unaffected dogs and dogs affected with either the RPE65 mutation or PDE6A mutation was ascertained by testing dogs in the device in full room light and dim red light. Seventeen dogs were tested in this manner, 11 of which were unaffected, and 6 affected (3 RPE65 mutants and 3 PDE6A mutants). 

The test predicted with a high degree of certainty that all six dogs were affected when either time to exit or first-choice tunnel was measured (Table 2) . All the affected dogs chose the incorrect tunnel first significantly more often and took significantly longer to exit in dim light and than did the unaffected dogs. There was no difference in performance between the two disorders. For the PDE6A mutant dogs, the probabilities for time to exit were P = 0.001 or less. The probabilities for the RPE65 mutant dogs’ time to exit were estimates ranging from P ≤ 0.0005 to P ≤ 0.001. The probabilities were estimates, because the most dim light runs for the RPE65 mutant dogs were truncated at 60 seconds, making the true exit time unknown. When considering the first tunnel chosen, the probabilities for all mutant dogs ranged from P ≤ 0.001 to P = 0.024. 

The test accurately predicted that 9 of 11 unaffected dogs tested in full versus dim light were indeed unaffected by both time to exit and first-choice tunnel (Table 2) . The 2 dogs for which the test did not accurately predict the affected versus unaffected status based on visual performance were very timid animals. One lay down and refused to exit the device at all, making data unrecordable. The other exited and chose the correct tunnel first significantly more often, but was slow to act and exit. Therefore, the test was inaccurate for this subject and predicted unaffected by first-choice tunnel but affected by time to exit (Table 2) . Because of other activities within the colony, it was not possible to repeat testing on either individual to assess the effect that familiarity with the test may have had on the dogs’ timid behavior. 

Significant Difference in Performance between Affected and Unaffected Dogs in White-Light Intensity Series of Both RPE65 and PDE6A Mutant Dogs

To assess performance in greater detail, a white-light intensity series was performed for one unaffected, three RPE65 mutant, and three PDE6A mutant dogs. Dogs A, B, and C were RPE65 mutation dogs; dogs D, E, and F were PDE6A mutation dogs, and dog G was an unaffected dog. 

Figure 4shows the probability of choosing the correct exit as a function of light intensity based on binary logistic regressions of light intensity versus first-choice tunnel. Because there are four exits from which to choose, a completely blind dog is expected to choose the correct exit tunnel randomly 25% of the time when using this test device. We verified this conjecture by testing an unaffected dog wearing opaque goggles in the device under full room light. The blinded dog chose the correct tunnel one time during seven runs through the device (data not shown). We were therefore interested in identifying the light intensity at which the affected dogs would begin randomly choosing tunnels. The unaffected dog (G) almost exclusively chose the correct tunnel first at all light intensities (Fig. 4 , line G). We were unable to test a normal dog in the device in a condition dark enough for it to begin choosing randomly because we would have been unable to see well enough to determine tunnel choice or exit time. Therefore, the opaque goggles placed on a normal dog were used to show that it chose the open tunnel randomly in the absence of visual stimulation. 

All six affected dogs, whether affected with the RPE65 mutation (dogs A–C) or the PDE6A mutation (dogs D–F), achieved or approached random-choice behavior as the light intensity decreased. There was a significant difference between the first-choice tunnel for each affected dog compared with the unaffected dog (Fig. 4 , lines A–F). 

Linear regressions of the resultant time to exit versus light intensity are shown in Figure 5 . The unaffected dog (G) exited the device quickly at all light intensities used in the study, with no difference in performance as light intensity decreased (Fig. 5 , line G). The RPE65 mutation dogs (A–C) showed a rapid and significant increase in time to exit as the light intensity decreased (Fig. 5 , lines A–C). The PDE6A mutant dogs were also significantly slower to exit than the unaffected dogs, as the light intensity decreased, but were faster to exit than were the RPE65 mutant dogs (Fig. 5 , lines D–F). 

There was no significant effect of the age of any affected dog on performance in the test. In addition, there was no effect of run order on performance, indicating the dogs’ performance did not change with increasing experience in the device. The device was not impenetrable to light, but neither this nor spectral changes as the lights were dimmed affected the performance of the dogs (Gearhart P, et al., unpublished observations, 2004). 

Distinguishing RPE65 and PDE6A Mutation Dogs

There was a significant difference in performance between the PDE6A (dogs D–F) and RPE65 (dogs A–C) mutant dogs in the light-intensity series, whether examining first-choice tunnel (Fig. 4)or time to exit (Fig. 5) . The RPE65 mutant dogs began random selection of the correct open door at brighter light intensities than did the PDE6A dogs. The odds ratio for the disease state from binary logistic regression (Table 3)shows that the odds for the PDE6A mutant dogs selecting the correct tunnel was 2.48 times higher than that of the RPE65 mutation dogs. Similarly, the PDE6A mutation dogs were significantly faster to exit (P < 0.0005) than were the RPE65 mutant dogs (Table 4) . There was no significant difference in performance for either time to exit or first-choice tunnel in the individual dogs with the same gene mutation. 

Discussion

We have successfully developed a four-choice vision-assessment system for use in dogs that is easy to perform and yields repeatable, quantitative results that can be analyzed by standard statistical methods. Until now, the best method available for characterizing vision in dogs was to set up obstacles such as traffic cones and observe the dogs among them. Our device is easy to build and the test is both simple and effective, detecting differences in performance to a high degree of statistical significance. The test is sensitive and can show differences in vision between the two different disease states tested to date. Finally, it is a task based on a natural reaction for most dogs; therefore, very little training is necessary for successful vision testing. 

Experience in the device seemed to have little effect on performance. At no time were run–order effects observed in the testing (data not shown), indicating the dogs’ performances neither improved with experience nor decreased with boredom or fatigue. In addition, when two light-intensity series were run using the same dog 3 months apart, there was good correlation between the regressions for each series (data not shown). Except for two individuals tested to date, variability in dog aptitude had little effect on the experimental results. The two dogs that had difficulties in the test (Table 2)were used in another study and therefore not retested. Any effects further training or experience with the test may have had on their performances is unknown. 

There was a dramatic difference in performance between unaffected dogs and affected dogs of either mutation. The unaffected dogs rapidly exited the device with an average exit time in the light-intensity series of 2.28 seconds (Fig. 5) . We also observed that affected dogs accurately exited the device within seconds when the room lights were turned on, even if the dog was unable to exit the device by 60 seconds under dim light (data not shown). The significant change in performance of all affected dogs with decreasing light level is consistent with the nature of the disease states in these dogs. Each diseased dog used in this study was affected with one of two retinal degenerations. The Cardigan Welsh Corgi dog, a model for autosomal recessive retinitis pigmentosa, has a null mutation in the gene encoding the α subunit of rod cyclic GMP phosphodiesterase (PDE6A) ¹⁹ and the Briard-mix dog, a model for Leber congenital amaurosis, has a null mutation in the gene encoding the retinal pigment epithelium specific protein RPE65. ¹⁸  

The PDE6A mutant dogs have a lack of pde6 activity, an absence of detectable rod-mediated ERG responses (Petersen-Jones SM, et al., unpublished findings, 2003) and undergo relatively rapid rod photoreceptor degeneration (Tuntivanich N, et al. IOVS 2005;46:ARVO E-Abstract 5238). The PDE6A mutation does not directly affect cone function, and the loss of rods is followed by a much slower secondary degeneration of cone photoreceptors that eventually leads to blindness. At the ages of PDE6A, mutant dogs used in this study there will have been loss of some, but not all cone photoreceptors. Because cones make up less than 5% of total photoreceptors in dogs, cone ERG waveforms are considerably lower in amplitude than are rod ERG waveforms. The dogs have no measurable rod function and the residual ERG responses originate from cones (Petersen-Jones SM, et al. IOVS 2003;44:ARVO E-Abstract 4537). They subjectively have vision under photopic conditions (not shown), yet very low to nonrecordable a- and b-wave ERG amplitudes (Fig. 1) , and in our test showed quantifiable visual function that declined in low lighting conditions (Figs. 4 5) . 

The RPE65 mutant dogs have an absence of rpe65 protein in the retinal pigment epithelium. ² Extrapolation from studies in RPE65 mutant mice shows that it is very likely that in the RPE65 mutant dogs there is a lack of 11-cis retinal supply to the photoreceptors, resulting in very reduced function of both and rods and cones. ^{20 21} The RPE65 mutant dog has a greatly elevated ERG threshold, ² which is illustrated in Figure 1 . This finding is similar to that in RPE65 mutant mice, in which the predominant remaining ERG responses originate from rods with markedly reduced sensitivity. ²² RPE65 mutant dogs have severely reduced scotopic and photopic vision, ^{23 24} consistent with the known effect of lack of RPE65 function in other species. ^{25 26} We were able to show that the RPE65 dogs had a more severe reduction in vision than did the PDE6A mutant dogs at the ages used in this study (Figs. 4 5) . However, they did have higher ERG amplitudes than did the PDE6A mutant dogs (Fig. 1) . This apparent contradiction in which visual performance of the RPE65 dogs was worse than that of the PDE6A dogs and yet the ERG responses were of greater amplitude is readily explained when the effects of the two mutations are considered. Assuming that the RPE 65 mutant dogs are similar to the RPE65 mutant mouse, most of the recordable ERGs in the RPE65 mutant dog should originate from rods, which make up more than 95% of the total photoreceptors. It is therefore not surprising that the ERG amplitudes from RPE65 mutant dogs are greater compared with the low-amplitude ERGs derived from the low number of cones in PDE6A mutant dogs, even though the RPE65 mutant dog rods have a marked loss of sensitivity and function. These results show that the test is sensitive enough to detect vision differences between animals with mutations causing different phenotypic diseases. 

It is noteworthy that we observed variability in performance between individual affected dogs of the same mutation when comparing first-choice exit (Fig. 4)and time to exit (Fig. 5) . For example, dog C (RPE65 mutation) showed a precipitous decline in the probability of correct tunnel choice, choosing randomly at the highest light intensity observed for any dog, yet was exiting the device much faster than the other RPE65 mutation dogs. Sometimes the opposite was observed (dog B). Similar variations occurred with the PDE6A mutation dogs. Testing of a greater number of dogs is currently being performed. It is possible that these differences are the result of personality differences between the dogs. We have found various personality differences among the dogs in our colony that confound the subjective obstacle course testing, including slow dogs that refuse to walk through the cones, or energetic dogs that knock down all the cones surrounding them. Unlike the obstacle course, however, the results generated from the vision-testing device are interpretable even in the face of apparent behavioral differences between individual dogs. 

It is possible that the differences in performance between dogs affected with the same mutation are the result of differences in the ages and therefore severity of disease. In this small sample, we found no significant difference in performance with age in dogs of either mutation. However, the ages of the PDE6A and RPE65 mutation dogs were 1.7, 3.3, and 3.4 years and 1.2, 1.7, and 2.2 years, respectively, making the age distribution too narrow to draw meaningful conclusions regarding age as a variable. Studies are currently under way to characterize further the possible effects of age on visual performance of dogs with either mutation. 

In this investigation, we truncated the exit time data for dim-light conditions by terminating the tests if the dog had not exited by 60 seconds. For the specific case presented in this article, the dogs exited the device in full light quickly and consistently. Only in the dim-light conditions did the dogs take longer than a minute to exit the apparatus. Because the truncated data were restricted to dim-light conditions, we can state with confidence that the average exit time during dim-light conditions would have been longer had we not truncated the data and the average exit time under the full-light condition was not changed by the truncation. For this reason we were able to truncate the data and quoted an upper limit on the probability. However, for a general research procedure in which researchers must more accurately estimate visual performance, elimination of some points based on exit time will overestimate the visual performance. We find that the first-choice tunnel result is less ambiguous and lends itself to analysis because it eliminates this ambiguity. 

The effect of dark adaptation time was not specifically addressed in the present study. Each light condition was tested in random order, rather than structuring the intensity series from brightest to dimmest light or vice versa. All dogs were given from 5 to 10 minutes of adaptation at each light intensity before the testing began. There was a lack of run-order effects within each lighting condition, indicating that the dogs’ performances were not improving as the number of runs (and therefore time) increased for a given light intensity. This suggests that increased dark adaptation time does not affect performance for the duration of our testing. However, a detailed analysis of the effect dark adaptation times may have on test performance for dogs with each mutation would be worthwhile. 

Our test does not make an attempt to measure visual acuity in the dogs. In force-choice visual testing methods for mice, a grating can be projected over the submerged platform to which the mice must swim. After many trials, the mice learn to swim toward the grating to find the platform. By changing the projected grating, a measure of visual acuity can be made. ^{7 8} Unlike the mouse test, the target to which our dogs were moving was large (the 28-inch diameter tunnel opening) without fine features or detail. The addition of an obstacle for the dog to avoid as it exited the tunnel was considered, but was rejected on the basis that it would interject the same kind of observer bias experienced in the obstacle course testing, and a single obstacle would not be as beneficial to visual acuity assessment as the projected gratings used in the other species. 

One goal in designing an objective test of visual function for dogs was its use in determining the response to therapeutic intervention. The test described herein objectively measures visual function and can distinguish not only between affected and unaffected, but between two different retinal disease processes. It therefore holds promise for use in therapeutic response trials in which light intensity series performed before and after treatment would be compared. In addition, the task is learned easily by most dogs, decreasing the need for extensive training of individual animals and making the system suitable for vision assessment of colony members or a study cohort. 

 Supported by National Eye Institute Grant R01 EY014160; Midwest Eye Banks and Transplantation Center, Fight For Sight, Michigan State University Companion Animal Funds and PureBred Dog Endowment Fund, and the Glassen Foundation.

 Submitted for publication May 25, 2007; revised November 9, 2007; accepted May 30, 2008.

 Disclosure: P.M. Gearhart, None; C.C. Gearhart, None; S.M. Petersen-Jones, None

 The publication costs of this article were defrayed in part by page charge payment. This article must therefore be marked “advertisement” in accordance with 18 U.S.C. §1734 solely to indicate this fact.

 

Table 1.

 

Summary of Dogs Used in the Study

	Disease Status	Bright vs. Dim Testing	Intensity Series Testing
1	U	✓
2	U	✓
3	U	✓
4	U	✓
5	U	✓	✓ (Dog G)
6	U	✓
7	U	✓
8	U	✓
9	U	✓
10	U	✓
11	U	✓
12	A (RPE65)	✓	✓ (Dog A)
13	A (RPE65)	✓	✓ (Dog B)
14	A (RPE65)		✓ (Dog C)
15	A (PDE6A)	✓	✓ (Dog D)
16	A (PDE6A)		✓ (Dog E)
17	A (PDE6A)	✓	✓ (Dog F)
18	A (RPE65)	✓
19	A (PDE6A)	✓

 Dogs are numbered so as to correlate with Table 2 . Tests that each dog underwent are indicated by a ✓ in the appropriate column. The designation of each dog in the white-light intensity series (Figs. 4 5)are indicated in parentheses (last column). U, unaffected; A, affected; RPE65, RPE65 mutant dog; PDE6A, PDE6A mutant dog.

Figure 1.

 

Scotopic and photopic ERG a- and b-wave amplitudes for unaffected, PDE6A mutation, and RPE65 mutation dogs. ERG analyses were performed on the unaffected dog in the light-intensity series (dog G), two of the RPE65 mutation dogs (dogs A, B), and all three of the PDE6A mutation dogs (dogs D–F) in the light-intensity series. Amplitudes in millivolts are plotted against the log light intensity (log cd/m²).

Figure 2.

 

Schematic diagram and photograph of the testing device. (A) The device from above showing the central junction box with four tunnels. For each run, the device is set up with one tunnel open and three tunnels capped, and the dog is placed in the junction box in the same orientation. (B) Photograph of the device.

Figure 3.

 

Example of a dog in the device. Still frames from the Vision Test Movie (Movie 1) show a PDE6A mutant dog in the device in light, then dim conditions. (A) In the light, the dog is placed in the box (frame A1), the top is placed on the box (A2) and the dog chooses the correct open tunnel first and exits the device (A3). (B) In the dark, the dog is placed in the box in the same orientation (frame B1), the top is placed (B2), the dog chooses the incorrect tunnel first, and wanders back and forth between incorrect tunnels before finding the way out (B3-B6). Note that the dog is filmed from the side of the open tunnel in the bright-light condition so that the videographer is not inadvertently influencing the dog with her presence. In the filmed dim-light condition, however, the videographer is outside the open tunnel so that the dog’s movements within the device can be documented. This run was not included in the overall analyses that are presented in this study. With an infrared video recorder (NightShot, Sony, Tokyo, Japan) an image of the dog was visible through the tunnel in the dim-light condition in the video; however, it was not possible to see through the tunnel walls, either from inside or outside the tunnel with the naked eye.

Table 2.

 

Predictive Value of the Vision Test

Dog	Frequency of Correct Choice			Prediction of Disease Status			Actual Disease Status	Comments
		Full Light	Dim Light	By Choice	By Time
1	5/7	4/7	U P = 0.573	U P = 0.996	U
2	6/7	5/7	U P = 0.508	U P = 0.417	U
3	4/7	5/7	U P = 0.573	U P = 0.359	U
4	6/7	5/7	U P = 0.508	U p = 0.359	U
5	6/7	7/7	U P = 0.280	U P = 0.471	U	Dog G
6	9/10	6/10	U P = 0.099	U P = 0.732	U
7	5/8	6/8	U P = 0.586	U P = 0.383	U
8	6/7	4/7	U P = 0.212	U P = 0.932	U
9	6/7	5/7	U P = 0.508	U P = 0.850	U
10	6/7	7/7	U P = 0.280	A P = 0.017	U	Timid dog
11	N/A	N/A	N/A	N/A	U	Fearful dog; test aborted
12	6/7	1/7	A P = <0.0005	A P = <0.001	A (RPE65)	Time truncated at 60 seconds
						Dog A
13	12/13	6/11	A P = 0.024	A P = <0.0005	A (RPE65)	Time truncated at 60 seconds
						Dog B
14	7/7	0/7	A P = <0.0005	A P = <0.001	A (RPE65)	Time truncated at 60 seconds
15	6/7	1/7	A P = <0.0005	A P = <0.0005	A (PDE6A)	Dog D
16	7/7	4/7	A P = 0.022	A P = 0.001	A (PDE6A)	Dog F
17	6/7	3/9	A P = 0.011	A P = <0.0005	A (PDE6A)

 The frequency of correct choices in full light and in dim light and the disease status as predicted by the test for correct choice and time to exit are shown, along with the actual disease status of the tested animals. Dogs that were also used in the intensity series are indicated in italics in the comments column. U, unaffected; A, affected; N/A, not applicable; RPE65, RPE65 mutant dog; PDE6A, PDE6A mutant dog.

Figure 4.

 

Fraction of correct first-choice tunnel versus light intensity. The fraction of correct first-choice tunnel was plotted against the light intensity on a log scale for the PDE6A mutation dogs (D–F), the RPE65 mutation dogs (A–C), and the unaffected dog (G). There was a significant correlation between log light intensity and correct first-choice tunnel (P < 0.005).

Figure 5.

 

Regression of exit time versus light intensity. Regressions of the time to exit versus the light intensity on a log scale for the PDE6A mutation dogs (D–F), RPE65 mutation dogs (A–C), and the unaffected dog (G). Each line is the representative regression resulting from the plots of the individual time to exit and light intensity data points for each dog. Test runs truncated at 60 seconds were dropped from the analysis, making some regression lines shorter than others. There was a significant correlation between log light and exit time with P < 0.05. R ² = 26.2–41.5.

Table 3.

 

Binary Logistic Regression of First Choice Tunnel as a Function of Light Intensity and Mutation

Predictor	Coef.	SE Coef.	z	P	Odds Ratio	95% CI
												Lower	Upper
Constant	0.046361	0.162056	0.29	0.755
PDE6A vs. RPE65	0.907591	0.237645	3.82	<0.0005	2.48	1.56	3.95
Log Light	0.250756	0.043666	5.74	<0.0005	1.28	1.18	1.40

 Binary logistic regression provides a prediction equation for the odds of selecting the correct exit (O) from the maze as a function of the predictor variable. The predictor log light is the natural log of light intensity (I) at the maze exit. The predictor PDE6A vs. RPE65 (M) is a binary variable that is 0 for RPE65 mutations and 1 for PDE6A mutations. The prediction equation for these results is ln (O) = 0.046361 + 0.90759M + 0.250756 ln (I). The Coef. column contains the best-fit coefficients for the prediction equation, the SE Coef. column contains the statistical uncertainty in the best-fit coefficients. z and P are the z statistic and associated P-values for the best-fit coefficients. The odds ratio is the increase in the odds of selecting the correct exit associated with unit increase in either M or I, where the odds are defined as the probability of making a correct choice divided by the probability of making an incorrect choice.

Table 4.

 

Linear Regression of Log Exit Time (T _exit) as a Function of Light Intensity and Mutation

Predictor	Coef.	SE Coef.	t	P
Constant	2.61373	0.05968	43.79	<0.0005
PDE6A vs. RPE65	−0.65330	0.08220	−7.95	<0.0005
Log light	−0.18929	0.01440	−13.15	<0.0005

 The predictor log light is the natural log of light intensity (I) at the maze exit and PDE6A vs. RPE65 (M) is a binary variable that is 0 for RPE65 mutations and 1 for PDE6A mutations. The prediction equation for these results is ln (T _exit) = 2.61371 − 0.65330M − 0.18929 ln (I). The values in the Coef. column are the best-fit coefficients for the prediction equation, the numbers in the SE Coef. column are the statistical uncertainty in the best-fit coefficients. t and P are the Student t statistic and associated P-values for the best-fit coefficients.

Supplementary Materials

Movie 1 - 6MB (.wmv) 

RPE65 mutant dog in the vision testing device under bright and dim light. 

The authors thank Janice Forcier for technical assistance in performing vision and ERG testing.