Central Retinal Vascular Trunk Deviation in Unilateral Open-angle Glaucoma

Patients with unilateral open-angle glaucoma (OAG) have suffered glaucomatous optic neuropathy in one eye only despite shared systemic factors between two eyes. It suggests a locoregional susceptibility factor associated with glaucoma development. In this study, we measured the distance of the central retinal vascular trunk from the Bruch’s membrane opening (BMO) center relative to that of the BMO margin: the shift index, since it can be used as a surrogate of lamina cribrosa (LC) shift caused by different growth between retinal and scleral layers during eyeball expansion. The shift index was compared between OAG and fellow control eyes within individuals (129 OAG patients). Although OAG eyes also had higher baseline IOP, a larger β-zone parapapillary atrophy area, a larger shift index was the only risk factor of OAG diagnosis in a generalized linear mixed-effects model. Further, a generalized estimating equation regression model revealed that the shift index was larger in the OAG eyes than in the control eyes for all ranges of axial length, while it was the smallest for the axial length of 23.7 mm. Thus, a larger shift index and LC shift may act as a locoregional susceptibility factor for unilateral OAG eyes.


Introduction
Glaucoma is characterized by progressive axonal loss of retinal ganglion cells [1][2][3] . Many factors such as mechanical, ischemic, metabolic, and immunologic insults have been nominated as candidate sources of axonal damage [1][2][3] . So far, intraocular pressure (IOP) is the only controllable factor, but it cannot explain every aspect of glaucoma. To be speci c, in many cases, glaucomatous damage starts with a localized retinal nerve ber layer (RNFL) defect, whereas IOP affects the optic nerve head (ONH) universally. This suggests a loco-regional susceptibility factor that makes some parts of the ONH more vulnerable to damage.
The lamina cribrosa (LC) is the principal site of axonal injury in glaucoma 4 . Therefore, LC change might be involved in increased loco-regional susceptibility. In the recent Boramae Myopia Cohort Study, we found that axial elongation evoked LC shift, in contrast to the relative preservation of the Bruch's membrane opening (BMO) [5][6][7] . In subsequent studies with myopic open-angle glaucoma (OAG), the direction of LC shift from the BMO center showed a strong spatial correlation with the initial hemispheric location of glaucomatous damage 8, 9 . This indicated that LC shift might re ect the loco-regional susceptibility of the ONH.
Besides the initial asymmetric hemispheric involvement of glaucomatous damage, unilateral involvement of glaucoma is often encountered. Even if fellow eyes are not normal and might, therefore, be affected by glaucoma in the future, unilateral glaucoma eyes are the best subjects to be scrutinized for loco-regional susceptibility factors after adjusting for systemic factors within a subject at the time of study. The purpose of the present study was to determine whether the position of the central retinal vascular trunk, as a surrogate of LC shift, increases susceptibility to glaucomatous damage in unilateral OAG patients while other systemic risk factors are controlled by inter-eye comparison within the same subject.

Results
This study initially involved 154 unilateral OAG patients. Of these, 9 patients were excluded due to poor image quality of radial scans leading to incomplete visualization of the BMO margin, and 6 patients were excluded owing to bifurcation of central retinal vascular trunk on emergence. The emergence of the vascular trunk was not visible in 24 patients on infrared imaging. Among these, 14 patients were proved to not have the vascular trunk within the BMO by uorescein or OCT angiography and they were included, while 6 who had not undergone angiography were excluded. Four patients who had the vascular trunk outside the BMO in both eyes were also excluded, leaving a nal sample of 129 unilateral OAG patients. The subjects were aged 55.6 ± 14.6 years, 62 of whom were female (48%).
The OAG eyes had a larger shift index than did the fellow eyes in cases either of nasally located vascular trunk ( Fig. 2A-C) or temporally located vascular trunk ( Fig. 3A-C). To compare the intra-individual difference of shift index and β-zone PPA area, those parameters of the fellow control eyes were subtracted from those of the OAG eyes (Fig. 1B). In contrast to the β-zone PPA area, the shift index was consistently larger in the OAG eyes than in the fellow control eyes over the entire range of axial length (Fig. 1B).
The angular deviation of the vascular trunk showed a correlation with the angular location of RNFL defect in the total group (r=-0.423, P < 0.001) and in the subgroup with the temporal location of vascular trunk (r=-0.539, P = 0.010).

Discussion
Newborns have their vascular trunk in the central area of the ONH in most cases 10 . In the Boramae Myopia Cohort Study, we demonstrated actual shifting of vascular trunk in contrast to the preserved BMO during myopic axial elongation [5][6][7] . Because the central retinal vascular trunk is embedded in the dense connective tissue of the LC 11 , shift and deviation of the vascular trunk from the BMO center would be related to shift of the underlying LC. Moreover, vascular trunk deviation from the BMO center is closely related to the location of optic nerve damage in myopic OAG patients 8, 9 . This could be an explanation for the close association between vascular trunk location and the location of glaucomatous damage, which were reported in many other reports [12][13][14][15] . Taken together, LC shift on the en-face plane, possibly as the reason of vascular trunk deviation from the BMO center, could re ect stress exerted on the LC. In the present study, we found that the vascular trunk position was deviated farther from the BMO center in glaucoma eyes than in fellow control eyes in unilateral OAG patients. This implies that a more shifted LC might be more susceptible to glaucomatous damage than a less shifted LC within the same subject.
To evaluate the effect of LC shift on glaucoma development, we included exclusively unilateral OAG patients having one glaucoma eye and one non-glaucoma eye. In this manner, we could exclude the effects of systemic factors such as aging and other general health-related conditions 16 . Subsequently, we focused on the local factors that make eyes more susceptible to glaucomatous damage. Although, the fellow eyes of unilateral OAG patients might differ from the healthy eyes, and unilaterality might exist only within a limited period of time 17 , our study, at the time of its conduct, informed us which eye was more susceptible to damage: the eye with larger LC shift was affected by glaucomatous damage earlier than the fellow eye with smaller LC shift. This might explain why glaucoma occurs in the eye that it does, only under the same systemic risk factors and similar IOP.
Shift of the LC can increase the susceptibility of the ONH to glaucomatous damage, because it re ects the tangential stress that had been applied to the ONH during eyeball growth [5][6][7][8][9] . Interestingly, in the present study, LC shift did not show a linear correlation with axial length (Fig. 1A). Rather, the correlation between LC shift and axial length was a J-shaped curve, and the GEE regression model showed that the shift index was the smallest for the axial length of 23.7 mm (Table 3), from which point, it increases in either direction. This could be understood by the various shapes of ocular expansion during growth 18 . The average axial length of newborns is around 17 mm, which is about 7 mm shorter than that of adults 19 . This means that every eyeball has to grow after birth, regardless of how short it will be in adulthood. Therefore, the process of outer-wall shift would not be limited to myopic eyes only, but would manifest in eyes of larger-than-17 mm axial length as well. In contrast to the prolate (= axial overgrowth) growth of myopic eyes 18,20 , some eyes, especially hyperopic ones, have been reported to have oblate growth 18, 21 . We speculated that oblate growth in hyperopic eyes would lead to the temporal shift of the outer-wall, which could be the cause of shift index increase in the reverse direction in eyes of axial length less than 23.7 mm (Fig. 1A).
Our study showed that such a shift, across the entire range of axial length, was larger in the OAG group than in the control group (Table 3). The GEE regression model revealed that the shift index of the OAG eye was larger than that of the control eye for a given axial length (Fig. 1A). This suggested that the OAG eyes had to endure more shifting than the control eyes with similar axial lengths. Therefore, LC shift could be an indicator of cumulative tensile stress exerted on the LC during growth, and not in myopic eyes only but in any eyes.
It should be noted that the direction of LC shift was associated with the location of glaucomatous damage in cases either of nasal or temporal shift. The large pores of the LC in the superior and inferior regions are considered to be more susceptible to glaucomatous damage 4,11,22 . Moreover, the LC is reported to have a horizontal ridge 11,23 , which might protect the LC from tensile stress acting parallel to it. Therefore, we speculated that tensile stress would converge to the susceptible pores in the superior and inferior regions 8, 9 , even if the direction of LC shift was nearly parallel to the horizontal meridian.
Park et al. showed by means of a subgroup analysis that β-zone-PPA-associated variables could be risk factors for unilateral normal-tension glaucoma 24 . Also in our study, the β-zone PPA area was larger in the OAG group, though it was not signi cant in the multivariable analysis. In our previous study, we showed that a part of β-zone PPA, which is to say, not only γ-zone PPA but also some of β-zone PPA with Bruch's membrane, appeared as the manifestation of LC and scleral shifting beneath the preserved BMO 7 . In this type of β-zone PPA, the extent of LC shift, as measured by vascular trunk dragging, was larger than the extent of β-zone PPA change 7 . Therefore, in this study, the effect of β-zone PPA might have been smeared out by the larger amount of vascular trunk deviation, which is more representative of LC shift (Fig. 1B) 7 .
The following implication of β-zone PPA, however, should be noted: at least in some eyes, larger β-zone PPA would represent a larger LC shift below the BMO, which makes the ONH more susceptible to a second insult such as increased IOP or other tissue-toxic factors 1,3 .
In the present study's multivariable analysis, baseline IOP was not the factor associated with OAG presence. This can be explained in two ways. First, most of the OAG patients in this study had baseline IOP within the normal range: normal-tension glaucoma. Second, even in the presence of unilateral OAG, the baseline IOPs of the subjects were highly correlated with each other (r = 0.883, P < 0.001). Therefore, the intra-individual difference of IOP was too small to make any substantial effect. This, however, should not be interpreted to mean that IOP is not an important factor. LC shift might make the ONH more susceptible to fundamental IOP-mediated insult. Furthermore, the concept of LC shift could explain why IOP reduction is the most effective treatment for OAG, even in cases of normal IOP. IOP lowering can reduce not only the direct IOP-related axial force but also the tensile stress exerted on the outer loadbearing structures, which could act as a tangential force exacerbating, by the shearing effect, damage to the ONH. Since growth-acquired shift cannot be reversed, additional stress in susceptible eyes with large LC deviation should be reduced instead, especially in cases of evident glaucoma progression.
This study has several limitations. First, the study design was cross-sectional. As such, we could not demonstrate actual LC shifting during earlier growth periods. The premise of this study was based on our previous prospective cohort study results [5][6][7] . Thus, we cannot exclude the possibility of confounding effect of LC remodeling on the CRVT position in the OAG eyes. However, glaucomatous ONH change has been reported not to affect the position of the CRVT in the LC portion 9,25 . Second, the location of the vascular trunk within the BMO, which was used as the indicator of LC shift, has a limitation. Since the vascular trunk outside the BMO could not be visualized with angiography, the extent of LC shift could be underestimated in such cases. Moreover, some eyes showed bifurcation of the vascular trunk on its emergence. Since we excluded such cases, we did not have any means of measuring the LC shift in those eyes. Additionally, the initial location of the vascular trunk was presumed to be the BMO center, which would not be certain for all eyes. However, most newborns had the central retinal vascular trunk in the center of the optic disc 10 , and the hyaloid artery is reported to be in the middle of the orbital part of the nerve when the back of the globe is formed in the embryo state 26 . Finally, we exclusively included subjects with unilateral OAG. Unilateral and bilateral OAG patients might have different characteristics 16 .
Patients with systemic risk factors might have bilateral glaucomatous damage 16 , though they were selectively excluded from this study. Therefore, our relevant study results should be interpreted with caution.
In conclusion, the LC had been shifted farther from the BMO center in the glaucoma eye than in the fellow control eye of unilateral OAG patients. Larger LC shift might represent the larger cumulative stress exerted on the ONH during growth, which makes the ONH more susceptible to additional insults induced by increased IOP or other factors. During the acquisition of SD-OCT images, the subjects were asked to xate on the target, and images were acquired with the forehead and chin stabilized by the headrest. Extra care was taken during each exam to con rm that the forehead and chin were correctly positioned and did not move. Prior to treatment, IOP was measured repeatedly (typically ve times) on the same or different days. The average value, which was de ned as the baseline IOP, was used for the subsequent analysis.

This investigation was based on OAG patients included in the Boramae
Glaucomatous optic nerve damage was de ned as rim thinning, notching and the presence of RNFL defects, and was evaluated by a glaucoma specialist (SHK). OAG was de ned as glaucomatous optic nerve damage and associated visual eld defects, and an open iridocorneal angle (in the case of cataract surgery, the angle was con rmed preoperatively). Glaucomatous visual eld defect was de ned as (1) outside normal limits on glaucoma hemi eld test, or (2) three abnormal points, with a p value less than 5% probability of being normal and one with a p value less than 1% by pattern deviation, or (3) pattern standard deviation of less than 5%. Visual eld defects were con rmed on two consecutive reliable tests ( xation loss rate of ≤ 20%, and false-positive and false-negative error rates of ≤ 25%).
The inclusion criteria were unilateral OAG patients. To be speci c, the contralateral eye for each patient had to have a normal optic disc appearance, an open iridocorneal angle, a normal red-free fundus photograph, and a normal visual eld. The exclusion criteria were BCVA of < 20/40, a sharply de ned posterior staphyloma (which can deform the contour of the eyeball) on funduscopic examination, a history of ocular surgery other than cataract extraction or corneal refractive surgery, retinal or neurologic disease other than glaucoma that could cause visual eld defect, a poor-quality image (i.e., quality score <15) of any section on enhanced depth imaging (EDI) SD-OCT radial scans, and a central retinal vascular trunk position located within the BMO but impossible to determine clearly due to vessel bifurcation. OAG eyes were compared with their fellow control eyes.

Assessment of the central retinal vascular trunk position
Our strategy for demarcation of BMO and vascular trunk position has been described previously 8,9 . The peripapillary area was imaged by SD-OCT (Spectralis, Heidelberg Engineering) using the EDI technique. Possibility of the magni cation error was prevented by entering the corneal curvature of each eye into the SD-OCT system before scanning. The BMO was demarcated using the Glaucoma Module Premium Edition of the Spectralis machine. With 24 high-resolution radial scan images of the ONH, 15° apart from each other, each averaged from 24 individual B-scans, SD-OCT automatically detects the margin of the BMO. Every detected BMO margin was reviewed by one of the authors (KML), and errors were corrected manually. Based on the edited BMO margin, the Spectralis machine calculated the area and center of the BMO.
The vascular trunk position was determined in the same way of our previous studies 8, 9 . First, the location of the vascular trunk was demarcated on fundoscopic infrared images and color-disc photography (Fig.  4A&B). Then, its location was con rmed by cross-sectional SD-OCT imaging in all cases. If a vascular trunk was not visible on either infrared fundus photographs or B-scan EDI SD-OCT images, uorescein or OCT angiography (Spectralis) was used to determine the presence of the vascular trunk within the BMO. The position of the vascular trunk was de ned in two aspects: 1) its angular deviation (Fig. 4B, α), and 2) the extent of shift (Fig. 4B, a). The angle was measured based on the right-eye orientation, with the nasal horizontal midline as 0° (a positive value indicating a vascular trunk located superiorly, and a negative value indicating a vascular trunk located inferiorly). To evaluate the extent of shift, the distance of the vascular trunk from the center of the BMO (a) was divided by the distance of the BMO margin from the center of the BMO in that direction (b), and de ned as 'shift index' (Fig. 4B, a/b). In cases of invisible vascular trunk due to being located outside the BMO, the shift index was de ned as 1.0, and the angular deviations were not determined. Using the Image J program (version 1.51, National Institutes of Health, Bethesda, MD, USA), one of the authors (KML), who as blinded to the participants' clinical information, measured the distances and angles. The reproducibility of the locating of the central retinal vascular trunk was excellent, as we had stated in the previous study 8 .

Assessment of RNFL defects
The angular location of RNFL defect was de ned in the same image used for the vascular trunk localization based on the right-eye orientation 8 . First, an infrared fundus image was overlapped with a red-free fundus photograph using commercial software (Photoshop; Adobe, San Jose, CA, USA) (Fig. 4C).
The points where the boundaries of the RNFL defect meet the BMO margin were determined. The angular location of the midpoint of the RNFL defect on the BMO margin was measured from the center of the BMO, with the temporal horizontal midline as 0° (a positive value indicating an RNFL defect located superiorly, and a negative value indicating an RNFL defect located inferiorly) (Fig. 4B, β). In cases of RNFL defect in both hemispheres, the angular location was measured to the larger RNFL defect based on the angular width.
Assessment of parapapillary atrophy β-zone parapapillary atrophy (PPA) was de ned as the area without any retinal pigment epithelium (RPE) adjacent to the ONH. The β-zone PPA area was measured on the same infrared OCT fundus images by one observer (KML) blinded to the subjects' information. Using the built-in caliper tool of the Spectralis OCT system, two boundaries were drawn: 1) the RPE opening (RPEO), which is the area without the RPE, and 2) the clinical disc margin (CDM). The β-zone PPA area was de ned as the RPEO area minus the CDM area 7 .

Data analysis
The paired t-test was used for the intraindividual comparison between the OAG and fellow control eyes. Subgroup analyses were performed by the Mann-Whitney U test and Wilcoxon signed-rank test. To account for the paired-eye correlation, a generalized linear mixed-effects model (GLMM) was used for univariable and multivariable analyses. Parameters with a P value less than 0.20 in the univariable analysis were included in the subsequent multivariable analysis. A locally weighted scatterplot smoothing (LOESS) curve was tted to the data. The LOESS curve uses iterative weighted least squares to determine values that best t the data 27