Retinal nerve fiber layer is associated with brain atrophy in multiple sclerosis

Much of the permanent disability experienced by patients with multiple sclerosis (MS) is thought to be attributed to axon degeneration secondary to inflammation and demyelination. Conventional MRI techniques are designed to be largely sensitive to inflammation (e.g., T2-weighted lesions) and may not specifically reflect axonal damage. Although quantification of lesions on MRI has been a sensitive tool for monitoring disease activity in MS, it has shown only modest correlation with clinical disability.^1,2 Alternatively, whole brain atrophy and T1 black holes have been linked with both disability progression and gross axon loss in MS.^1,3–5

Consequently, high-resolution anatomic MRI has been used to derive a measure of whole brain atrophy called the brain parenchymal fraction (BPF), which has been used as a measure of disease progression in patients with MS.^1,6–8 This method employs automated normalization and segmentation techniques to compute the volumes of various intracranial compartments and total brain parenchyma.^5,9 This metric and similar techniques have been applied to standardized disability measures but with mixed results.^7,10–14

Although MRI has become the most widely accepted marker of disease progression, it is both expensive and time consuming. Moreover, in part because of the potential confounding factors of volume loss due to atrophy and volume increase due to inflammation, optimal MRI surrogate measures have yet to be established. Therefore, there remains a need to validate novel methods of monitoring disease progression. Ideal methods are specific for axon damage and can be performed quickly, frequently, conveniently, and inexpensively.

Optical coherence tomography (OCT), a relatively new imaging technique, is a promising and sensitive tool that measures the thickness of the retinal nerve fiber layer (RNFL)^15,16 and costs approximately 10 to 15% that of MRI. Recent studies have shown that OCT can detect RNFL thinning, possibly due to axon damage, within the retinas of patients with MS regardless of clinical history of optic neuritis,^17–19 suggesting that it may be an effective biomarker of destructive disease.^20,21

In this cross-sectional case-control study, we sought to examine whether two OCT measures, RNFL thickness and total macular volume (TMV), are correlated with MRI-derived assessment of brain compartment volumes and BPF, as well as clinical disability.

METHODS

RESULTS

Our study cohort included 40 subjects with MS and 15 controls, all of whom had whole-brain anatomic MRI and OCT scans. The demographics of our study cohort are summarized in table 1. Correlations between brain atrophy (assessed with MRI) and RNFL thickness (assessed with OCT) are also visually striking (figure 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 MR images and optical coherence tomography (OCT) results for three patients with multiple sclerosis (MS) depicting the relationship between these two measures

MRI magnetization-prepared rapid gradient echo sequences with the subjects’ corresponding brain parenchymal fraction values are shown above. Retinal nerve fiber layer (RNFL) thickness measures for each subject are shown below for the right (OD) and left (OS) eyes. Circular images below are OCT images. Green quadrants indicate thickness within the 5th to 95th percentile range of the subject’s age group, yellow quadrants fall into the 1st to 5th percentile range, and red quadrant is less than the 1st percentile. The figure shows a spectrum of brain atrophy, from minimal (left) to severe (right). This corresponds to OCT measures that show minimal RNFL thinning on the left (as indicated by green and yellow quadrants and low normal RNFL thickness values) to severe thinning on the right (indicated by more red quadrants and markedly low RNFL thickness values).

Correlations for minimum RNFL thickness and minimum TMV across significant tissue segmentation classifications are presented in table 2 and figure 2. We found that minimum RNFL thickness and age predict a significant amount of the variance (adjusted R-squares of 21% to 23%) in BPF, CSF volume, and EDSS, with absolute partial correlation coefficients for minimum RNFL thickness ranging from 0.35 to 0.47. White matter volume was significantly predicted by age but not by minimum RNFL thickness, and gray matter volume was not associated with these measures. Minimum TMV was not significantly associated with any of the MRI measures or with clinical disability.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2 Association between optical coherence tomography and MRI measures or Expanded Disability Status Scale (EDSS)

Patients with multiple sclerosis (MS) are shown on the left and controls on the right. (A) MS brain parenchymal fraction (BPF) vs minimum retinal nerve fiber layer (RNFL) thickness; (B) control BPF vs minimum RNFL thickness; (C) MS CSF volume vs minimum RNFL thickness; (D) control CSF volume vs minimum RNFL thickness; (E) MS BPF vs minimum total macular volume (TMV); (F) control BPF vs minimum TMV; (G) MS EDSS vs minimum RNFL thickness. Regression lines control for age, which for the purposes of this plot is assumed to be held constant at its mean value.

Among controls, there was no significant association between minimum RNFL thickness or TMV and any of the MRI measures, although there was an association between age and gray matter volume that was not found among the patients with MS.

In our subgroup analysis (table 3), significant associations between minimum RNFL thickness were found among patients with RRMS, but not among patients with SPMS. Thus, minimum RNFL thickness could predict up to 43% of the variance in BPF in RRMS. Because the PPMS subgroup contained only five subjects, it was not analyzed separately.

To assess the degree to which multiple linear regression modeling predicts BPF, CSF volume, and EDSS among patients with MS, we calculated the median percent difference between predicted and actual values, as well as the percentage of individuals whose MRI measures and EDSS were predicted within one SD of the actual values. For BPF, median difference was 2.1% (95% predicted within one SD); for CSF volume, 8.8% (95%); and for EDSS, 40% (97%). The larger median difference for EDSS reflects the larger coefficient of variation for that variable (table 1).

DISCUSSION

We report a significant association between RNFL thinning and global brain atrophy (primarily manifested by increasing CSF volume) in a cohort of 40 patients with MS. RNFL thinning also correlates significantly with EDSS. The lack of association in the control group may be an epiphenomenon of a small sample size and the relative youth of the controls we studied: both brain volume loss and RNFL thinning are known to occur in healthy people later in life, and they may be associated with one another. In MS, however, RNFL thickness and age are independently associated with brain atrophy and are not significantly correlated with one another, suggesting that the result is driven at least in part by disease-related pathology. Definitively separating the effects of pathology and normal aging would best be accomplished by longitudinal studies.

Potential explanations for the less than perfect correlations between OCT and MRI volumes may relate to variability in biologic processes and technological limitations of our measures as well as nonlinear relationships between our variables. Biologic processes include 1) brain atrophy is attributed to both myelin as well as axon loss; 2) global brain atrophy may follow years after RNFL thinning caused by an acute episode of optic neuritis; 3) severe damage to only a single eye (with a contralateral healthy eye) may not be representative of severity of damage in the brain; 4) approximately 1/3 to 1/2 of the human brain is devoted to vision and we did not determine whether RNFL thickness correlates with atrophy in that portion of the brain as opposed to overall atrophy. Additionally, the specific mechanism by which RNFL thinning occurs is unknown. Thinning may be partially related to retrobulbar demyelination and loss of phosphorylated neurofilaments leading to decreased axon caliber.²¹ Further, a cross-sectional relationship may not yield the strongest correlations since brain atrophy may lag behind RNFL changes or vice versa.

On the technical side, although automated segmentation methods reduce user bias, SIENAX imperfectly segments the brain into its components, in part due to subtle differences between signal intensity in brain tissue. This can lead to imperfect estimation of BPF and consequent reduced correlations with OCT measures. Additionally, OCT may itself be an imprecise measure.

Lack of correlation between TMV and BPF may be partially explained by the fact that TMV has been found to change only after many years following disease onset.^20,21 Volume loss in the macula has been documented in MS and may be a result of axon loss in the RNFL by retrograde degeneration of retinal ganglion cells. In a cohort of patients with MS with a mean duration of disease of approximately 10 years, TMV may be less informative than in a cohort with a longer history of MS. An alternate explanation may be that TMV measures by OCT are more imprecise than axonal loss measures or have greater variance within a given cohort.

There are a number of reasons that might explain why RNFL thickness correlates with brain atrophy more strongly in RRMS than in SPMS, even when adjusted for age. First, we examined the distributions and the characteristics of the average RNFL within the SPMS cohort. It is possible that there might be a basement effect in the progressive group, where RNFL or brain tissue may have reached its lowest levels, at which the extent of change is minimal. Indeed, the SPMS group was more skewed than the RRMS group.

Another explanation might be related to the phenotype of patients categorized as SPMS. Most patients with SPMS have clinical decline due to cumulative spinal cord disease rather than accumulation of brain disease. Some of these patients might have less cerebral involvement and thus less brain atrophy. In order to identify if any of these hypotheses hold up, a larger cohort and longitudinal studies are necessary, as well as a more detailed analysis of the degree to which each subject’s brain is affected by MS.

This small study presents strong and encouraging data based on a modest sample size with limited numbers of each MS subtype. There remain questions about comparisons between progressive vs relapsing remitting MS. Nonetheless, even with a small number of subjects, the preliminary correlations between OCT and MRI tissue classifications are promising and provide a rationale for future longitudinal studies linking these techniques. Our findings also encourage further assessment of the utility of OCT, which is a rapid and relatively inexpensive test, as a potential outcome measure in clinical trials of neuroreparative and neuroprotective drugs. Overall, these data suggest the possibility that damage to axons in the retina is related to similar changes in the brain and may be able to predict cerebral axon damage. Longitudinal studies and pathologic correlation are necessary to confirm this hypothesis.

ACKNOWLEDGMENT

The authors thank Terri Brawner, Deanna Cettomai, Karen DeBusk, Kathleen Kahl, and Ivana Kusevic for their assistance with this project.