Abstract
Purpose.:
In a group of humans with strabismic amblyopia, the relationship was examined between the structure and function of different brain regions. Three question were addressed: (1) Is the lateral geniculate nucleus (LGN) in humans with amblyopia structurally as well as functionally abnormal? (2) Do structural anomalies in the visual cortex correlate with the previously reported cortical functional losses? (3) Is there a link between the functional anomalies in the visual cortex and any structural anomalies in the geniculate?
Methods.:
The structure was compared by using voxel-based morphometry (VBM) and the function by functional magnetic resonance imaging (fMRI).
Results.:
The results showed that the geniculate is structurally abnormal in humans with strabismic amblyopia.
Conclusions.:
These findings add further weight to the role of the LGN in the cortical deficits exhibited in human strabismic amblyopes.
Amblyopia is a condition in which the vision through one eye is permanently reduced due to a disruption in early visual development. This disruption can be loss of form vision (deprivation amblyopia), loss of focus (anisometropic amblyopia), or loss of ocular alignment (strabismic amblyopia). Electrophysiological studies in humans
1,2 and single-cell studies in animals made artificially amblyopic
3–5 suggest that the site of the deficit is not in the retina. Morphologic changes have been reported in the layers of the lateral geniculate nucleus (LGN) that receive input from the deprived eye in animals
6–9 and humans,
10,11 although the functional properties of these cells have been considered to be normal
12,13 in most animal studies. On the basis of these single-cell findings, it has been concluded that the site of the amblyopic deficit is in the input layers of the primary visual cortex.
14 However, there is a body of literature comprising studies in which functional anomalies have been reported in the LGN of deprived animals. These range from selective deficits in X-cells,
15 selective deficits in Y-cells,
16 and more diffuse deficits in all cells.
8 Furthermore, a case study
17 suggested a functional magnetic resonance deficit at the level of the LGN in humans with anisometropic amblyopia. More recently, it has been shown that the functional responses are reduced at the level of the geniculate in humans with strabismic, anisometropic, and deprivation amblyopia
18 and that this deficit may be selective for P-cell function.
19,20
The reduced geniculate response when driven by the amblyopic eye may in principle be due to a reduced input from the eye, anomalous geniculate function per se, or aberrant feedback signals from the primary visual cortex. To better understand the basis for the reduced geniculate response reported in humans with amblyopia, we undertook a structural study of the geniculate to answer the following questions: First,
is the LGN in humans with amblyopia structurally as well as functionally abnormal? This question relates to whether the previously reported response reduction
17–20 is due to reduced geniculate function per se or to a reduction in either the feedforward drive from the retina or the feedback drive from the cortex. Second,
do structural anomalies in the visual cortex correlate with the previously reported cortical functional losses? This question relates to whether the previously reported cortical structural losses
21,22 have any functional significance, as one might expect from a simple cellular loss hypothesis that explains the functional deficit. And finally,
is there a link between the functional anomalies in the visual cortex and any structural anomalies in the geniculate? This question relates to whether any structural deficit in the geniculate is of fundamental importance to the cortical processing deficit in amblyopia or whether it is just an epiphenomenon.
The stimuli in this experiment were conventional retinotopic wedge and annulus checkerboard sections used for retinotopic mapping.
23,24 The abruptly alternating radial square-wave checkerboard had a fundamental temporal frequency of 8 Hz. The fundamental circumferential spatial frequency of the checks varied from 1.0 cyc/deg centrally to 0.1 cyc/deg peripherally. Both stimuli completed a full cycle in 12 time frames (0.03 Hz), giving a total of six cycles per scanning run. The checkerboard had a contrast of 80%. The wedge subtended 90°. The radial checkerboard contained 20 radial spokes, 10 concentric bands, and subtended a visual angle of 34°. The subject was instructed to attend to a central fixation point.
25 The mapping stimulus was viewed alternately with each eye, the other eye being patched.
Stimuli were presented in a phase-encoded paradigm, always alternating runs between the left and right eyes of normal subjects or the fixing and amblyopic eyes of amblyopic subjects, while the subject attended to a central fixation spot and performed a visual task designed to control for attention. This task involved the detection of a coherent patch of checkerboard within the checkerboard stimulus as a whole that appeared at random times and positions. The responses were recorded via an optically isolated mouse. This task maintained the subject's attention at an engaged and constant state throughout the scans. In two previous studies using this stimulus, we have shown that amblyopes can maintain central fixation
25 and that any fixation instability does not correlate with reduced BOLD (blood-oxygen-level–dependent) response.
26
We identified retinotopic visual areas by using the methods of Dumoulin et al.
27 A 1.5-T scanner (Sonata; Siemens Medical Systems, Erlangen, Germany) was used to collect both anatomic and functional images. Anatomic images were acquired by using a head coil (circularly polarized transmit and receive) and a T
1-weighted sequence (TR, 22 ms; TE, 10 ms; flip angle, 30°) of 180 sagittal slices of 256 × 256 image voxels were acquired that provided a voxel size of 1 mm
3. Functional scans for each subject were collected via a surface coil (circularly polarized, receive only) positioned beneath the subject's occiput. Each functional imaging session was preceded by a surface coil anatomic scan (identical with the head coil anatomic sequence, except that the number of sagittal slices was reduced to 80 with a resolution of 256 × 256 and a slice thickness of 2 mm), to co-register the data later with the head-coil image. Functional scans were multislice T
2*-weighted, gradient-echo, planar images (GE-EPI; TR, 3.0 seconds; TE, 51 ms; flip angle, 90°). Image volume consisted of 30 slices orthogonal to the calcarine sulcus. The field of view was 256 × 256 mm, the matrix size was 64 × 64 with a thickness of 4 mm, giving voxel sizes of 4 × 4 × 4 mm. Each experiment consisted of four acquisition runs for each eye (two eccentricity runs, two polar angle runs). Eccentricity runs consisted of both expanding and contracting directions and polar angle runs consisted of both clockwise and counterclockwise directions. Each run consisted of 128 image volumes acquired at 3-second intervals (TR). Fixing and amblyopic eye information was averaged separately across the two eccentricity runs and across the two polar angle runs. Runs were alternated between the eyes in each case.
We used software developed at the Montreal Neurologic Institute (
http://www.bic.mni.mcgill.ca/software/) to estimate gray and white matter densities. For each anatomic image, we corrected for intensity variation
28 and transformed the image into standard stereotaxic space.
29,30 We used a nonparametric classification algorithm that incorporates prior tissue probability maps in standard space to label each voxel as gray or white matter or cerebrospinal fluid (CSF).
31 Next, we extracted the three binarized images corresponding to gray matter, white matter, or CSF, which were in each case premultiplied by a whole-brain template, to remove skull and scalp artifacts. These normalized images were then smoothed with an isotropic 6-mm full width at half-maximum (FWHM) Gaussian kernel. Within an individual gray matter image therefore, every point in space corresponded to a weighted local (within the nearby 6 mm) gray matter concentration estimate.
32 Asburner and Friston
32 point out that such estimates should not be confused with cytoarchitechtonic cell-packing density, as the value of the concentration will be determined by both structure (e.g., local cortical curvature
33 and density). That is, the metric is dimensionless and quantifies the number of local voxels classified as gray ranging from 0 to 1.0 in regions where all surrounding voxels are classified as gray.
To compare anatomy across subjects, we used both standardized and individual anatomic templates. We used standard space templates for the occipital and temporal lobes as well as a region defining the LGN. The template lobes were created using mri3dX (
http://cubric.psych.cf.ac.uk/Documentation, provided in the public domain by University of Cardiff, Wales, UK). The LGN templates were constructed based on published stereotaxic coordinates.
34 Based on anatomic scans, the estimates of Kastner et al.
34 estimates of LGN location (±SD) are 22.88 ± 1.8, −21.3 ± 1.49, and −4.63 ± 2.13 and −23.33 ± 1.41, −21 ± 1.6, −4.66 ± 1.33 mm for the right and left LGN, respectively. Postmortem data from Andrews et al.
35 suggest that the LGN (parvo plus magno, across left and right structures) volume in humans is 118.5 ± 19.5 mm
3, when approximating the LGN as a cube gives a mean side length of 4.89 ± 0.27 mm. This closely accords with anatomic MRI estimates from Gupta et al.
36 of 4.74 ± 0.54 and 4.83 ± 0.95 mm for right and left LGN, respectively. Taking the mean SD across all dimensions gives 1.62 mm (i.e., 95% of all LGNs will be centered within approximately 3.24 mm of these locations). Approximating the LGN as a cube with a side length of 5 mm, to fit this variability we must accommodate 5 + 3.24 · 2 mm in each dimension within the template. With this location variability in mind, we constructed two anatomic templates of different cubic volume (5 × 5 × 5 and 12 × 12 × 12 mm) both centered on the mean locations from Kastner et al.
34 Unless otherwise stated, all results shown are for the 12-mm cubic template.
To rigorously test whether the differences observed between the control and amblyopic subjects were due to chance, we constructed 100 randomly placed pseudo-LGN structural templates. These templates were the same size as the original anatomic templates (5- or 12-mm cube side) and were also symmetrically placed about the midline by randomly selecting a cube within the right hemisphere and mirroring it to the left.
For each individual, we also constructed individual anatomic templates based on the retinotopic maps (V1,V2, V3, VP, V3A, and hV4) defined from the functional retinotopy scans. For each template, we calculated the total gray matter by simply multiplying the smoothed binary classified images by the binary template volume and averaging the image values remaining.
The retinotopy scans also yielded BOLD signal change measures for each eye (published elsewhere
25 ). fMRI time series were normalized, and the design matrix for the general linear model was constructed by means of the inverse Fourier transformation.
26 A first-order autoregressive model was used to fit the temporal correlation and then, the mean squares of regression (MSR) and errors (MSE) were calculated, where MSR constitutes the amount of variance predicted by the model and MSE the unexplained variance. BOLD signal activation was quantified by means of an
F ratio where
F = MSR/MSE.
37 We then computed Spearman rank correlations between the amount of LGN gray matter and the difference in
F value between eyes (
F good −
F bad). We also tested whether the functional difference observed within a visual area could be explained by the amount of gray matter within that same area.
Is the LGN in Humans with Amblyopia Structurally as Well as Functionally Abnormal?
Our main finding was that there was significantly (LGN size 12, Wilcoxon's sign rank test,
z = 2.1;
P < 0.04, two-tailed) less gray matter in the LGNs of the amblyopic group than in the control group (
Fig. 1). For LGN size 12, the mean gray matter concentration in the amblyopic group was 0.5231 (SE 0.0197) and in the normal group, 0.6012 (SE 0.0308).
These results were found no matter which LGN template we used (for LGN size 5, z = 2, P < 0.05). For LGN size 5, the mean gray matter concentration in the amblyopic group was 0.5034 (SE 0.0202) and in the normal group, 0.5793 (SE 0.0320).
However, we did not find a difference in gray matter between the amblyopes and the control groups in any of the other visual areas identified or within the occipital or temporal lobes.
We were concerned that the differences observed at the LGN could be due to the relatively small generic anatomic template, which gave an anomalous result by chance. To more rigorously test this possibility, we produced 100 templates containing randomly but symmetrically placed masks of both sizes (either 5 or 12 mm). The locations of these randomly chosen masks are shown in
Supplementary Figure S1.
Figure 2 shows the difference in gray matter within these masks (in the 12-mm case) in the amblyopes compared with the control group. The distribution of differences obtained suggests that our results are unlikely to have arisen by chance (
P < 0.01 and
P < 0.02 for the 12- and 5-mm masks, respectively).
Do Structural Anomalies in the Visual Cortex Correlate with the Previously Reported Cortical Functional Losses?
Is There a Link between the Functional Anomalies in the Visual Cortex and Any Structural Anomalies in the Geniculate?