Multi-luminance mobility testing after gene therapy in the context of retinal functional diagnostics

Leber congenital amaurosis (LCA) is a severe form of inherited retinal dystrophy. It is hallmarked by a rapidly progressing visual impairment which first presents in early childhood and usually leads to complete blindness by the fourth decade of life [1]. About 6% of all cases of LCA are caused by biallelic mutations in RPE65 [4] which disrupt the recycling of rhodopsin [10]. Nonetheless, photoreceptor morphology may be preserved for several years [5], raising the possibility of functional recovery via gene supplementation.

In late 2017, the first gene therapy for RPE65-linked LCA (voretigene neparvovec, Luxturna®) became available in the USA. Studies have demonstrated that it may not only slow the progression of degeneration but also restore some visual function [6, 9, 12]. Yet, this improvement seems to be mostly limited to younger patients [13]. Nonetheless, many recipients report improved vision in everyday life. To assess this effect objectively, visual navigation tests are increasingly used as endpoints in clinical trials [11]. But as these tests rely on the patient’s active cooperation, their results can be susceptible to confounding factors such as motivation and learning. These factors could undermine the test’s utility for identifying treatment effects in patients with very low vision. Another limitation of the procedure is the inter-individual differences in the patients’ remaining visual abilities which usually requires a higher level of adaptability as compared to other tests.

To investigate the actual meaning of improvements in these tests, we did a multi-luminance mobility test with recipients of voretigene neparvovec and correlated the results with other (objective) clinical parameters of visual function which have already been used to demonstrate the effectiveness of the therapy.

The purpose of this paper is to provide fellow researchers who are currently looking for suitable read-out parameters for their clinical trials with exemplary data for mobility testing in patients with retinal degeneration and to point out a few practical considerations and limitations.

Two patients (P3 and P8) passed the test with 500 Lux LCVNC at screening and thus performed the low-contrast version. The remaining subjects did not pass and were tested with the HCVNC. The median pass level before treatment was 63 Lux. The lowest luminance level at which participants were able to complete the course was 8 Lux. Most participants passed at 178 Lux or below. One subject (P12, left eye) required the highest light level (500 Lux) in the HCVNC at baseline, which differed from the performance of the remaining participants by more than two standard deviations and was therefore excluded from the correlation analysis.

Three months after treatment, all but one participant improved by at least one level, with a median of 2 levels. In the LCVNC, P3 improved by two levels and P8 by one level. One participant passed the HCVNC (P1, LE only) at the lowest luminance level (0.35 Lux), corresponding to an improvement by three levels. Only P13 passed at a higher level after treatment (63 vs 178 Lux).

Three-dimensional visual navigation tests have been used in several studies to evaluate the impact of treatments for retinal degenerative diseases [2, 3, 8]. However, depending on the targeted patient population, these tests may have certain limitations that could affect its applicability as a clinical read-out parameter.

Here, we investigated the dependencies between conventional parameters of visual function and the performance in a mobility test to evaluate its usefulness for identifying improvements in vision following gene therapy with voretigene neparvovec.

A few recipients displayed notable improvements in multiple parameters of visual function after therapy. Specifically, reductions in FST and DAC thresholds (representing rod sensitivity), as well as improved pupil responses (representing the neurons’ sum activation), point to a local reactivation of rods in the treated area in a few patients [13]. Subjects with advanced retinal degeneration, however, often showed no measurable effects on standard parameters.

Since rod vision contributes strongly to orientation at lower luminances, participants with measurable changes in scotopic vision were expected to show stronger changes in the VNC compared to non-responders. Indeed, the correlation between the improvement in the VNC and the change in FST with blue light indicates that improved performance in the VNC is driven by the reactivation of rods. However, almost all subjects improved by one or even two light levels, regardless of their results in other tests. Furthermore, there was no significant correlation with the absolute FST values from the first and second measurement. Hence, individual improvements in the VNC seemingly do not accurately reflect treatment effects in all patients. There are several factors that may explain this apparent lack of specificity.

In general, it needs to be considered that standard ophthalmological procedures measure very specific aspects of rod or cone vision, whereas tests for visual navigation rely on higher visual processes. Hence, their results may not be easily comparable with one another.

But beyond that, the lack of correlation could also be attributed to our data analysis. Specifically, because of the small size of our patient sample, we included all treated eyes as separate samples, which could have compromised the independence of the data and altered the outcome.

Furthermore, some improvements in the VNC of patients with no measurable change in other measures may be explained by learning effects. Although the layout of the course was replaced after each attempt, some subjects still learned to avoid mistakes after a few trials (e.g., by walking slowly and pausing more often to scan the environment). In accordance with this, Cideciyan et al. [3], who used the Ora-VNC™ in a trial for oligonucleotide therapy for LCA, found that improvements were largely symmetrical between treated and untreated eyes. On the other hand, their setup also deviated slightly from the recommended procedure. Specifically, they used a smaller layout, and their light levels were not spaced logarithmically. In combination, these deviations may have reduced the VNC’s repeatability. Similarly, we only used a simplified version of the original Ora-VNC™ protocol for this study that included less randomizations to reduce the time needed for the procedure. In the original version, the participant should complete all levels of both the high- and low-contrast course at screening, which in turn has to be repeated for each eye. Though this procedure may prevent learning effects in the course of follow-up visits, a major disadvantage is that it takes many hours to complete an entire screening test, and follow-up measurements still often take a few hours depending on the subject’s performance.

Another aspect which may undermine the test’s outcome is the low difficulty of the high-contrast test variation, designed for patients with very low vision. Specifically, most of the obstacles in the HCVNC are positioned next to the main path and do not necessarily have to be seen in order to be evaded as long as the participant stays in the middle of the path. Consequently, it was difficult to determine if the subject passed because he actually recognized the obstacles or simply by chance. Only two RP patients (P3, P8) in the present study were able to pass the LCVNC at screening. Both showed similar improvements in FST and DAC thresholds, indicating a reactivation of rods. Despite that, they did not display stronger improvements in VNC levels than other patients (though postoperative foveal scarring could have interfered with P8’s visual navigation [13]). In sum, the change in pass levels of subjects who completed the LCVNC is not comparable with those who did the HCVNC. As to account for this difference between test versions, Cideciyan et al. [3] did not evaluate the absolute luminance levels but instead used a continuous scale from the darkest level of the LCVNC to the brightest level of the HCVNC. Nonetheless, they found no significant differences between treated and untreated eyes (potentially due to the previously mentioned deviations from the original protocol). Other studies on visual navigation assessed the time that their participants needed to complete the obstacle course, showing that RP patients and healthy controls can best be distinguished based on their preferred walking speed [8]. However, our own experience was that speed mostly depends on the subject’s approach to the test,i.e., while some patients proceeded very cautiously to avoid errors as much as possible, others would just walk along the path at their normal walking speed to reach the end of the course as fast as possible, even at the risk of hitting more obstacles.

Related to that, another issue is that the course requires more active participation than other procedures. Appropriately, performance can be biased more easily by fatigue or motivation. P13, for example, whose pass level had increased after treatment, became more reluctant to do the test after initial failures at lower luminances and asked to skip levels.

In summary, the results of the VNC are not as easy to interpret as conventional measures of vision. Especially in the late disease stages, it is difficult to assess whether improvements are caused by the treatment or by subjective factors. To be used as a clinical endpoint, both the outcome parameters and the specific details of the testing procedure should be chosen carefully depending on the patients’ stage of retinal degeneration.

Following treatment with voretigene neparvovec®, most patients showed improvements in visual navigation. This change was correlated with the improvement in FST with blue light stimuli, but other clinical parameters of visual functions, including DAC, CPC, and BCVA, were largely uncorrelated to the pass level. The results indicate rod sensitivity as a main predictor for the performance in the VNC. However, patients with no measurable improvements in standard tests also usually improved in the VNC. This lack of specificity may have been caused by several factors, including learning effects and poor comparability between VNC versions. To achieve higher reliability in the VNC, several points regarding the implementation and interpretation should be considered: Firstly, future trials should include training prior to the first measurement so as to minimize learning effects. Secondly, both the treated and untreated eye should be measured to control for subjective influencing factors. On the other hand, the reliability may be compromised if the complete protocol cannot be performed due to spatial or time limitations. Unfortunately, the inclusion of all these steps makes the test even more time-consuming. Finally, if patients in different stages of retinal degeneration complete very different versions of the mobility test, their improvements may no longer be comparable with one another and should be analyzed separately.