Layer-dependent loss and enhancement of geniculostriate and retinotectal pathways in adult human amblyopia

Abnormal visual experience during critical period leads to reorganization of neuroarchitectures in primate visual cortex. However, developmental plasticity of human subcortical visual pathways remains elusive. Using high-resolution fMRI and pathway-selective visual stimuli, we investigated layer-dependent response properties and connectivity of subcortical visual pathways of adult human amblyopia. Stimuli presented to the amblyopic eye showed selective response loss in the parvocellular layers of the lateral geniculate nucleus, and also reduced the connectivity to V1. Amblyopic eye’s response to isoluminant chromatic stimulus was significantly reduced in the superficial layers of the superior colliculus, while the fellow eye’s response robustly increased in the deeper layers associated with increased cortical feedbacks. Therefore, amblyopia led to selective reduction of parvocellular feedforward signals in the geniculostriate pathway, whereas loss and enhancement of parvocellular feedback signals in the retinotectal pathway. These findings shed light for future development of new tools for treating amblyopia and tracking the prognosis. Highlights Amblyopia impairs feedforward processing in the P layers of the LGN Layer-dependent loss and enhancement of cortical feedback signals in the SC Pathway-specific abnormalities explain amblyopic deficits in visual acuity and attention Significance statement How abnormal visual experiences during critical period shape the function and wire the neural circuits of human subcortex remains largely unknown. With high-resolution fMRI and visual stimuli to preferentially activate layer-dependent response in human subcortical pathways, the current study clearly demonstrates that amblyopia shifts the homeostatic interocular balance of human subcortex in a pathway-specific manner. Amblyopia led to selective loss of parvocellular feedforward signals in the geniculostriate pathway, whereas deficit and enhancement of parvocellular feedback signals in the retinotectal pathway. These pathwayspecific functional abnormalities provide the neural basis for amblyopic deficits in visual acuity, control of eye movement and attention. It sheds light for future development of new tools for treating amblyopia and tracking the prognosis.


Introduction 1. Amblyopic eye's response was selectively reduced in the P layers of the LGN
In the 3T fMRI experiment (2mm isotropic voxels), M and P biased visual stimuli 139 ( Figure 1A) differed in spatial frequency, temporal frequency, contrast and chromaticity were 140 used to selectively activate the M and P layers of the LGN. The amblyopic eye (AE) and the 141 fellow eye (FE) of unilateral amblyopia patients were tested in separate sessions. The non-142 dominant eye of normal controls (NE) was also tested. We found selective loss of amblyopia 143 eye's response to the chromatic P stimulus in the P layers of the LGN. In order to reduce the 144 partial volume effect from ventral pulvinar and to further confirm the P layer deficits from the 145 3T experiment, we performed a 7T fMRI experiment at a higher spatial resolution (1.2mm 146 isotropic voxels). The P cells of the LGN are responsive to both chromatic stimulus and high 147 spatial frequency achromatic stimulus. If there is a general functional loss for the P cells, the 148 P layers' response to high spatial frequency achromatic stimulus should also be reduced. To 149 test this hypothesis in the 7T experiment, high contrast and high spatial frequency achromatic 150 stimulus (achromatic P stimulus, Figure 2A) was monocularly presented to the amblyopic eye 151 and to the fellow eye of amblyopia patients. The achromatic P stimulus presented to the 152 fellow eye strongly and selectively activated the P layers of the LGN. Compared to the fellow 153 eye, the amblyopic eye's response was significantly reduced in the P layers of the LGN. 156 chromatic stimulus in the P layers of the LGN 157 As shown by Figure 1A, the chromatic P stimulus was a high spatial frequency, 158 isoluminant red/green checkerboard pattern, reversing contrast at 0.5 Hz. The M stimulus was 159 a low contrast (30%), low spatial frequency achromatic checkerboard, counter-phase 160 flickering at 10 Hz. The right side of Figure 1A shows the M and P layers of the LGN from 161 Nissl stained histology (Amunts et al., 2013) and the simulated M-P fMRI pattern (see 162 methods for details about the simulation). Figure 1B shows the group averaged M-P beta 163 maps of the LGN for normal, amblyopia and fellow eye groups. Different eye groups 164 consistently showed a ventral cluster with response bias to the M stimulus, and a dorsal 165 cluster preferred the P stimulus. For each cluster, 30 mm 3 voxels with the strongest response 166 bias were shown (the highest and lowest M-P beta values for the M and P dominant clusters, 167 respectively). For all three eye groups, the M dominant clusters were located at the ventral 168 LGN, while the P dominant clusters were located at the dorsal LGN. Such spatial pattern was 169 consistent with the simulated M-P pattern from the histology of the human LGN, and was 170 also consistent with the findings from previous studies (Zhang, Zhou, et al., 2015). For the 171 amblyopia eye, response bias to the P stimulus in the P cluster was much weaker compared to 172 the normal and fellow eye groups, suggesting a selective response loss to sustained chromatic 173 stimulus in the P layers of the LGN.

C.
Layer-dependent Loss and Enhancement of Subcortical Pathways in Human Amblyopia selected FOV for slices showing in C. Lower: significant T maps from a representative subject (p < 0.05 250 uncorrected). C. Left: The left three columns show the group averaged beta maps for the amblyopic eye (AE), the 251 fellow eye (FE), and the response difference between the AE and NE conditions. The right column shows the T map 252 of significant AE-NE voxels (p < 0.05 uncorrected). Maps were up-sampled to 0.6mm isotropic resolution. Black LGN into an M and a P layer compartments (right panel of Figure 2A, and also Figure S2).

262
The visual stimulus was a full contrast, high spatial frequency checkerboard pattern, 263 reversing contrast at 4Hz (Figure 2A, left). According to our recent study (Qian et al., 2020), this 264 achromatic P stimulus should strongly activate the P layers of the human LGN. Indeed, when the 265 fellow eye was stimulated, the achromatic P stimulus strongly activated the dorsal part of the 266 LGN ( Figure 2B), corresponding to the P layers. Compared to the fellow eye, P layers' response 267 to the amblyopic eye was much reduced (as shown by the AE-FE beta map, Figure 2C   LGNs were averaged for each subject. The LGN volume was 132.82 ± 43.20 mm 3 (Mean ±  As shown by Figure 3A (upper panel), the M-and P-biased visual stimuli mainly 300 activated the superficial SC, with weaker BOLD response in the deeper part of the SC. This 301 finding is consistent with the consensus that superficial SC mainly processes visual sensory 302 information. The lower panel of Figure 3A show that compared to normal controls, the 303 amblyopia eye's response to the P stimulus was significantly reduced in the superficial SC,  During ROI analysis, we first generated a normalized depth map of the SC (from 0 to 1, 308 Figure S3). Since the nominal spatial resolution of the fMRI voxel is at 2mm isotropic which 309 may not be high enough to resolve each individual layers of the SC, and that the functions of 310 the superficial and deeper (middle and deep) layers of the SC is more distinguished in term of 311 visual sensory processing and premotor control and attention, the SC was divided into a 312 superficial and a deeper part (at normalized depth = 0.5). In Figure 3B   Two-way ANOVA showed that the eyes × layers interaction was significant for the P 321 stimulus (F(1,32) = 6.7, p = 0.014), but not for the M stimulus (F(1,32) = 1.78, p = 0.19).

322
Post-hoc t-test showed significant difference between the AE and NE response to the P 323 stimulus in the superficial SC (t(32) = 2.2, p = 0.036), but not in the deep SC (t(32) = 0.69, p 324 = 0.49). These results indicate that compared to normal controls, amblyopia eye's response to 325 the chromatic P stimulus was selectively reduced in the superficial but not in the deep SC.

326
Then we compared the responses between the fellow and normal eye groups. Three-way ANOVA of the FE and NE response showed a significant interaction of eyes × stimuli 328 (F(1,32) = 5.28, p = 0.028), and significant interaction of eyes × stimuli × layers (F(1,32) =    Levi and Harwerth found contrast sensitivity loss for amblyopia eye most pronounced at high 476 spatial frequencies (Levi & Harwerth, 1977). However, using a luminance pedestal 477 discrimination paradigm, Zele and colleagues found that contrast sensitivity was similarly  (Hess, Li, et al., 2009;Li et al., 2013). A recent fMRI study showed more reduced response to 483 chromatic stimulus than to achromatic stimulus in the LGN of amblyopia patients (Hess,484 Thompson, et al., 2010), but the M and P layer-specific responses were not distinguished.

485
Since the P cells respond well to both chromatic and achromatic stimuli, it is unclear whether 486 this result was due to selective response loss to chromatic than to achromatic stimuli in the P 487 layers, or due to selective loss to chromatic response in the P layers than to achromatic  In the current study, we found very robust amblyopic response loss to the sustained chromatic 507 (P) stimulus in the inferior pulvinar, while the response to the transient achromatic (M) 508 stimulus was almost unaffected. Effective connectivity from pulvinar to V4 was also 509 significantly reduced for stimuli presented to the amblyopic eye. These findings demonstrate 510 parvocellular response deficits in the visual pulvinar and abnormal information transmission through the pulvino-cortical pathway. However, amblyopic eyes' input produced significant magnocellular feedforward connectivity from the SC to ventral pulvinar, suggesting intact 513 magnocellular processing through tectal-pulvinar pathway.

514
To the best of our knowledge, the influence of amblyopia to the function of 515 retinotectal pathway has never been investigated in primate. A previous study on rodent 516 found reduced glucose metabolism in the superficial layers of SC by monocular lid suture

722
A subject-level bootstrapping procedure was used to calculate the statistics of 7T 723 response in Figure 2C (right), the normalized fMRI response in Figure 5, and test the difference 724 in driving input strength between the AE and NE/FE groups ( Figure 6). For each permutation, 725 the data for a group of subjects were resampled with replacement. Then the group averaged 726 fMRI responses for different eye groups were calculated. The bootstrapping procedure was 727 repeated 10 6 times. A null distribution was generated by shifting the mean of bootstrapped 728 distribution to zero, and a two-sided p value was derived from the null distribution.

M-P pattern simulation for the 3T BOLD fMRI experiment
As shown by Figure S1, image was resampled to the resolution of fMRI measurement (2mm isotropic) and up-sampled to 0.6mm isotropic with cubic interpolation as in the fMRI data analysis, finally a differential image was generated by subtracting the patterns between the M and P stimulus conditions. LGN V1 A.

B.
Selective Loss and Enhancement of Parvocellular Response in Subcortex of Human Amblyopia correct refractive errors was tested, and the opposing eye was covered by a black eye patch.
Examples of simulated eye movement traces were shown in Figure S6A. The simulation matched the characteristics of eye movements of strabismus from a previous study (Chung,