The structure–function relationship between multifocal pupil perimetry and retinal nerve fibre layer in glaucoma

The need for a straightforward, effective, and objective perimeter for assessment of visual function has led to the development of multifocal pupillographic objective perimetry (mfPOP). This technique, which uses relative changes in pupil diameter to generate a functional map of visual field sensitivity, has been shown to be both sensitive and specific to dysfunction in a number of diseases affecting the visual system. These include macular degeneration [1‐3], diabetic retinopathy [4‐6], multiple sclerosis [7], and glaucoma [8‐10]. In glaucoma, the characteristic degeneration of retinal ganglion cells (RGCs) leads to observable changes in the optic nerve head and retinal nerve fibre layer (RNFL). This loss of afferent neurons of the main visual pathway leads to corresponding changes in visual function. The association between these anatomical and perceptual changes, the so-called structure–function relationship, has been extensively investigated using peripapillary RNFL (pRNFL) measurements and maps of visual sensitivity obtained using standard automated perimetry (reviewed Malik et al. [11]). The amount to which localized structural changes in glaucoma are reflected by changes in mfPOP sensitivity has not yet been addressed.

A variety of different RGC classes contribute to different components of the pupillary response [12‐14]. Some of these RGCs do not contribute to visual perception and so it is possible that structure function changes for mfPOP might be different to SAP. Steady-state pupillary diameters are largely mediated by retinotectally-projecting intrinsically photosensitive RGCs (ipRGCs) [14]. We have shown that the ability to detect glaucomatous scotomas is impaired using long-duration specialised mfPOP stimuli targeting the ipRGC melanopsin response, suggesting either that ipRGCs may be less involved or that their large and overlapping receptive fields reduce the spatial resolution of measurements [10]. We have recently shown that the responses to the standard transient mfPOP stimuli are mediated by the extrastriate cortex [15], as expected from the neuro-anatomy [12]. This should mean that, in glaucoma, the pattern of pupillary responses produced by these brief mfPOP onset stimuli is subject to similar factors as standard automated perimetry. Pupillary responses to mfPOP stimuli are, therefore, likely to be directly affected by the loss of RGCs in the main visual pathway. It follows then, that the topography and degree of reductions in mfPOP sensitivity should correlate with reductions in retinal nerve fibre layer thickness in glaucoma.

Published models of nerve-fibre trajectories now allow topographic mapping of pRNFL sectors to circumscribed areas of retina. One such model, by Jansonius et al. [16], will be used here to explore the structure–function relationships of mfPOP and SAP in subjects with and without glaucoma.

Of the 25 glaucoma subjects, 10 were classified as Mild (both eyes SAP MD ≥ -6 dB), 6 as Moderate (worst eye MD < -6 dB but ≥ -12 dB), and 9 as Severe (worst eye MD < -12 dB). No subject had MD < -12 dB in both eyes (Table 1). Median HFA MD, PSD and inter-eye differences for each of the Normal and Glaucoma subgroups are presented in Table 1. Because most glaucoma subjects had asymmetric visual field loss and were classified according to their worst performing eye, the median HFA MD for subjects in the Moderate and Severe glaucoma subgroups was close to the upper limits of these two classifications. The OCT pRNFL clock-hour sectors that most often fell below normal limits in glaucoma subjects were the inferior and superior sectors and those adjacent (Table 1). The median proportion of mfPOP recording lost due to e.g. blinks and fixation losses was 1.5%; the most missing from any single test was 8%.

Positive correlations between pRNFL and pooled arcuate division mfPOP/SAP deviation scores are apparent in the scatterplots of data from all glaucoma subjects (Fig. 2). The principal curves (red dots) shown on these scatterplots reveal a much more linear relationship between the mfPOP and RNFL deviations than that of HFA. Plots of each subject’s Spearman correlation coefficients (ρ) against the MD of their worst-performing eye revealed similar slopes for both mfPOP and HFA tests, with all tests producing stronger correlations in subjects with lower function i.e. more advanced disease (Fig. 3). pRNFL deviations increased systematically with glaucoma severity in each of the mfPOP variants; SAP was slightly less consistent with a stronger correlation for patients with Mild eyes compared to those with at least one eye classified as Moderate (Fig. 4). No consistent differences were seen between mfPOP and SAP across glaucoma subgroups; SAP produced a stronger correlation for Mild eyes whereas mfPOP was somewhat stronger in normal subjects. Since these correlation coefficients represent measurements from a number of subjects they will likely be affected by inter-subject variability in measures of both structure and function. For the bulk of these analyses, Spearman correlation was chosen over Pearson due to the non-linear relationship between SAP and pRNFL thickness (Fig. 2D and [22]). Figure 4 (lower) shows that when Pearson correlation was used this somewhat favored mfPOP over HFA.

To establish which of the pRNFL clock-hour sectors were associated most strongly with the functional measures, correlations between OCT sector deviations and those of the pooled test-regions/point were performed by pRNFL clock-hour sector (Fig. 5). Although these results were somewhat variable, some clear patterns did emerge. Correlations in normal subjects were weak and mostly non-significant, particularly for SAP. Those of Mild glaucoma subjects were a little stronger, with SAP having the highest value in the largest number of sectors. The strongest structure–function relationships were observed in glaucoma subjects in the superotemporal sectors, e.g. mfPOP LumBal 0.93, Lum + Col 0.91 (Severe group), and in the inferotemporal sectors, SAP 0.89 (Moderate). The sectors with the strongest correlations were also outside normal OCT pRNFL limits in the largest number of subjects (Table 1, far-right column).

Sectoral correlations in the Moderate glaucoma group were, in general, between those of Mild and Severe, however null or negative correlations were present in each of the tests in some sectors in which a strong relationship would be expected e.g., superior and temporal sectors in SAP and superotemporal in mfPOP. The small size of the Moderate subset may have contributed to these results. In Severe glaucoma subjects the mfPOP LumBal test far outperformed the other mfPOP tests as well as SAP.

Correlation coefficients were also estimated between individual, rather than pooled, mfPOP test-region/HFA test-point deviations and each of the pRNFL sectors. These are shown in Figs. 6 (mfPOP) and 7 (SAP). This analysis provided an insight into the integrity of the sectoral divisions and showed which visual-field locations were most strongly correlated with their corresponding pRNFL sector. MfPOP LumBal test-regions and HFA test-points in congruent visual field locations produced correlations that were strongest in the same pRNFL sector over much of the visual field, although this was sometimes in a sector adjacent to that expected e.g. both LumBal test-regions and HFA test-points in the inferior field surrounding the midline produced strongest correlations in the superior-superotemporal sector, however in some of these regions the expected pRNFL sector was superior. Correlations tended to be stronger and more in accord with the expected pRNFL sector in more eccentric and more nasally-located visual field test-regions.

While the majority of variation in perimetric visual-field sensitivities is due to the number of functioning ganglion cells [24], the efficacy of perimetry in glaucoma is still largely reliant on the ability of the patient to accurately complete the procedure. Patients in poor health may find longer tests, or those requiring repeated responses, quite difficult to complete. In addition, the higher presence of media opacities in older patients make the use of tests utilizing short-wavelength light, or stimuli containing high spatial frequencies, problematic [25]. For these reasons, a test that can measure visual function quickly and objectively using longer-wavelength light, such as mfPOP, should have advantages in glaucoma assessment. Contradictory findings in rodent glaucoma models have led to some uncertainty regarding the relative effect of glaucoma on steady-state pupillary responses [26‐28]. The use in mfPOP, however, of transient pupil responses to very short duration impulse stimuli, understood to be the result of visual signal mediated by ganglion cells and interneurons of the main visual pathway [10] and mediated through the extrastriate cortex [12, 15] allows us to objectively measure responses which should be affected by disease in a similar way to a patient’s visual percept. This study lends support to this hypothesis, with mfPOP producing comparable structure–function correlations to those derived using HFA.

Prior to this study, published findings of the relationship between pRNFL thickness and the pupillary response have largely been limited to investigations using ganzfeld stimuli and inter-eye asymmetry (see e.g. [29‐31]). The only published report in which comparisons are made between the pRNFL and focal pupil stimuli failed to identify significant correlations [32]. Part of the reason for this may relate to a recent finding by our group that pupillary responses arising from the melanopsin-mediated ipRGC signal appear to be impaired in their ability to resolve glaucomatous scotomas [10]. The pretectal olivary nuclei of the pupillary pathway receive two distinct types of visual signal: ipRGC-mediated direct input which dominates the response to long duration stimuli, and input from various regions of visual cortex involved in the processing of conscious vision [12]. It is this latter cortical input that is understood to mediate pupillary responses to transient stimuli such as used in mfPOP [10, 15]. This divergent neural circuitry also leads to profound differences in the inferences that can be made using long-duration as opposed to transient stimuli. Steady-state ipRGC-mediated responses can only provide an indirect estimate of RGC-related visual loss since ipRGCs are not known to be involved in conscious vision in primates. This study thus provides the first evidence of a relationship between structural changes in the pRNFL and the amplitude of pupillary responses to stimulation of discrete areas of retina. Given our previous findings, this is not a result that would be expected from stimuli targeting very slow ipRGCs, with their low spatial-resolution [10].

The large size of the mfPOP test-regions and extent of retina assessed using mfPOP is in stark contrast to the limited coverage of the 0.43° Goldmann size III stimuli used in HFA (Fig. 1). The effect of this is apparent in Figs. 6 and 7 where correlations were performed for each mfPOP test-region and HFA point. In SAP, mostly only single pRNFL sectors originated at each test-point, however for mfPOP test-regions this ranged between 1 and 5. It might have been expected then that the correlation between mfPOP test-regions and each individual pRNFL sector would be lower than SAP due to more than one clock-hour sector having axons originating in that test-region. However, this is not the case in the majority of regions, where mfPOP correlations in sectors where damage would be expected according to axon trajectories (blue arcs, Figs. 6 and 7) are equivalent to, and, in some cases, stronger than the single sectors of HFA test-points. This effect could potentially have been due to correlation in pRNFL damage between adjacent arcuate divisions of retina; however, this would have produced a similar pattern in the adjacent non-corresponding sectors of HFA test-regions, which is not apparent. One interpretation of this is that the large extent of mfPOP test-regions gives access to additional, rather than averaged, information regarding function. This is perhaps not unexpected, given the difference between the proportion of retina assessed using the two methods. The use of Goldmann size V stimuli, or SITA Standard or Full-Threshold strategy rather than SITA FAST, could potentially have provided stronger results for SAP, however the difference in test–retest variability between these strategies is only a fraction of the extent of the overall variability of either method [33], so any improvement would be expected to be minimal.

As might be anticipated, and as is reflected in studies of the diagnostic utility of the pupillary response in glaucoma [8, 9, 34], the strongest correlations were observed in the Severe subset of subjects. Although the greatest magnitude mfPOP correlations in Severe glaucoma were produced using the LumBal test, in Mild and Moderate disease overall structure–function correlations (Fig. 3) were strongest in the mfPOP tests which incorporated color-signal as well as luminance. This pattern mirrors the higher sensitivity and specificity we have previously observed using this type of composite stimulus in early glaucoma [8] adding conviction to the idea that the addition of color information in the Lum + Col and LumBal + Col stimuli has advantages in early disease.

The relatively small number of subjects in each of the glaucoma subgroups is a likely source of variability in this study. Perhaps the best example of this can be seen in the sectoral correlations shown in Fig. 4. In the six glaucoma subjects classified as Moderate, zero and negative SAP correlations in superonasal sectors were accompanied by positive mfPOP correlations; in the same subjects, zero and negative mfPOP correlations in the superior superotemporal sector accompanied a positive SAP correlation. In contrast to this variability, however, is the presence of clear and consistent strengthening of inferotemporal correlations between the Mild and Moderate groups. Inaccuracies in the mapping of clock-hour sectors to retinal locations and inter-individual differences in axon trajectories will also have introduced variability into this analysis. The use of individualized retinal maps incorporating simple parameters like optic nerve head position and axial length, such as proposed by Denniss et al. [35], would likely produce more accurate correlations than those obtained using population averages as we have in this study. However, the aims of this project were to compare SAP structure–function relationships with those of mfPOP, and different mfPOP methods with each other, rather than to provide absolute estimates of the degree of correlation, so given that both test types were subject to the same constraints, this should not affect the conclusions drawn from this data.

The strong structure–function relationship between pRNFL thickness and mfPOP response amplitudes in Severe disease, consistently better than that of SAP, makes a convincing argument for the mfPOP method’s further development and utility in late-stage disease. Recent developments in the mfPOP method, so called clustered volleys presentation, have improved signal to noise ratios by 40% compared to the methods used here [1]. Thus further improvements in structure–function correlation may be possible for mfPOP. The relatively linear scatterplots of Fig. 2 and the predominantly positive correlations, even in eyes of normal subjects, suggest that mfPOP tracks pRNFL thickness relatively well. In summary, this study has further demonstrated the viability of mfPOP as an emerging perimetric method and lent further evidence to the contribution of neurons in the main visual pathway to mfPOP responses.