Short-term plasticity in the visual thalamus

While there is evidence that the visual cortex retains a potential for plasticity in adulthood, less is known about the subcortical stages of visual processing. Here we asked whether short-term ocular dominance plasticity affects the visual thalamus. We addressed this question in normally sighted adult humans, using ultra-high field (7T) magnetic resonance imaging combined with the paradigm of short-term monocular deprivation. With this approach, we previously demonstrated transient shifts of perceptual eye dominance and ocular dominance in visual cortex (Binda et al., 2018). Here we report evidence for short-term plasticity in the ventral division of the pulvinar (vPulv), where the deprived eye representation was enhanced over the non-deprived eye. This pulvinar plasticity effect was similar as previously seen in visual cortex and it was correlated with the ocular dominance shift measured behaviorally. In contrast, there was no short-term plasticity effect in Lateral Geniculate Nucleus (LGN), where results were reliably different from vPulv, despite their spatial proximity. We conclude that the visual thalamus retains potential for short-term plasticity in adulthood; the plasticity effect differs across thalamic subregions, possibly reflecting differences in their cortical connectivity.

representation was enhanced over the non-deprived eye. This pulvinar plasticity effect was 23 similar as previously seen in visual cortex and it was correlated with the ocular dominance 24 shift measured behaviorally. In contrast, there was no short-term plasticity effect in Lateral 25 Geniculate Nucleus (LGN), where results were reliably different from vPulv, despite their 26 spatial proximity. We conclude that the visual thalamus retains potential for short-term 27 plasticity in adulthood; the plasticity effect differs across thalamic subregions, possibly 28 reflecting differences in their cortical connectivity. 29  (Turrigiano, 2012). 43 Recently, we explored the neural underpinnings of this effect using ultra-high field 44 functional Magnetic Resonance Imaging (fMRI). Although our technique did not directly 45 measure ocular dominance columns, we were able to detect short-term plasticity effects in 46 primary visual cortex V1 that were compatible with a change in ocular drive (Binda et al., 47 2018). 48 While ocular dominance plasticity has been thoroughly investigated in the visual cortex, less 49 is known about its effects on subcortical visual processing. The thalamus is a crucial node of 50 the visual system, with different subnuclei serving complex and diverse functions that we 51 have only begun to uncover. The Lateral Geniculate Nucleus (LGN) receives the largest 52 contingent of retinofugal fibers, and it is the main source of feedforward signals to V1 53 (Blasdel & Lund, 1983;Hendrickson et al., 1978;Hubel & Wiesel, 1972 previous study has tested their potential for short-term reorganization in the human adult. 73 Here we address this question using the short-term monocular deprivation (MD) paradigm 74 applied to normally sighted human adults studied with Magnetic Resonance Imaging (MRI). 75 Mapping thalamic nuclei with MRI is notoriously difficult due to the low signal to noise ratio 76 and the small size of these structures. Here we overcome these limitations using ultra-high 77 field (7T) fMRI and relying on independently defined regions of interest (from the Natural 78 Scenes Dataset, Allen et al., 2021). These label LGN and two subdivisions of the Pulvinar, 79 ventral (vPulv) and dorsal. vPulv is most tightly connected with occipito-temporal visual 80 cortex and it is retinotopically organized; the ROI comprises two complete and precise maps 81 of the contralateral hemifield, sharing the fovea and the vertical meridian; it partially 82 overlaps the anatomically defined lateral and inferior pulvinar subnuclei. Conversely, the 83 dorsal region of the Pulvinar shows rough retinotopy and its responses are better explained 84 by attentional and cognitive phenomena than by the low-level visual features, which is 85 coherent with its preferential connectivity with parietal cortex (Arcaro et al., 2018;Arcaro et 86 al., 2015). Thus, vPulv is most clearly involved in visual processing and should respond most 87 vigorously to the simple visual stimuli used in our study; for these reasons, we focused our 88 analyses on this region and compared its behavior with LGN. 89

90
We measured 7T BOLD responses to monocular visual stimulation delivered before and 91 after 2h of eye patching in 18 adult participants with normal vision (Experimental design is 92 shown in Figure 1A). We previously analyzed responses in visual cortical areas (Binda et al.,  Figure 1C shows the temporal dynamics of BOLD responses (to stimuli in the to-be-deprived 112 eye, before monocular deprivation) extracted from these independently defined ROIs. 113 Responses had similar size in LGN and vPulv. These were clearly weaker than previously 114 measured in V1 (were signals peaked at about 2.5% at 9s from stimulus onset; Binda et al., 115 2018), but reliably larger than 0 at all points between 3s after stimulus onset to 3s after its 116 offset (all t(17) > 4.30 and p < 0.01). Response dynamics was faster than in V1 (the peak 117 here occurs around 6s from stimulus onset), and slightly faster in LGN than in vPulv, as 118 previously reported (Lewis et al., 2018). This led us to quantify BOLD response amplitudes 119 with an approach that makes minimal assumptions on temporal dynamics. Since the visual 120 stimulus was a periodic alternation of stimulus present/absent epochs (ignoring variations 121 in a stimulus dimension that is not relevant here, see methods), the amplitude of visually 122 evoked responses could be extracted simply by Fourier analyses of the fMRI timeseries, 123 taking the amplitude at the stimulus frequency (note that analyses based on General Linear 124 Modelling and Event Related Averaging produced the same pattern of results, as detailed 125 below). 126 With this approach, we compared responses to stimuli delivered to the two eyes. 127 Before monocular deprivation, no systematic differences in eye dominance were expected; 128 therefore, we used BOLD responses to stimuli in the two eyes for estimating the internal 129 consistency of our results. We found that responses to the two eyes were correlated across

159
We performed several control analyses to support these conclusions. response amplitude from the event-related average curve, which we averaged in the 171 interval between 3s and 12s after using the 0s timepoint for baseline correction (essentially: 172 integrating the response in Figure 1C over the 3-12s interval and dividing by the duration of 173 this interval). Again we found a reliable time by eye interaction in vPulv (F(1,17) (Figure 1-supplement 1). 188

195
Using this alternative definition of vPulv, we still found a significant time by eye interaction 196 (F(1,17) = 13.34, p = 0.002, Figure 2-supplement 1, panel A), confirming the reliable 197 monocular deprivation effect in the ventral (or visual) Pulvinar. We followed a similar 198 strategy to obtain an alternative definition of LGN. We located it based on the histological 199  (Figure 1-supplement 1), thereby equating ROI size between LGN and 201 vPulv. With this alternative definition of LGN, we still found no significant time x eye 202 interaction in LGN (F(1,17) = 0.10 p = 0.756, Figure 2-supplement 1, panel B).

209
Our study is the first to show evidence for short-term plasticity in the adult human thalamus. 210 We found that 2h of monocular deprivation, besides shifting ocular dominance as assessed 211 from behavior and from V1 responses (Binda et al., 2018), also affects ocular drive in a specific 212 subregion of the visual thalamus, ventral Pulvinar or vPulv. 213 With a series of controls, we obtained strong evidence against the possibility that this is an 214 artifact of BOLD analyses or region labelling; we confirmed the results with three different 215 approaches -we cross-checked them with two independent atlases and reached the same 216 conclusion, that the plasticity effect was clearly observed in vPulv. 217 In contrast, the adjacent LGN region was reliably unaffected by monocular deprivation. Note 218 that we did not attempt to separate magnocellular and parvocellular subdivisions of the LGN. cortical territory where short-term plasticity is the strongest; if short-term plasticity effects 252 are carried through feedback signals, it is reasonable to assume that these will be stronger, 253 more stable, and ultimately easier to detect in vPulv than in LGN. 254 The second hypothesis, that plasticity effects are (at least in part) generated within the 255 thalamus, is in line with growing evidence on the importance of the thalamus in active vision 256 (Saalmann & Kastner, 2011). The traditional view of this nucleus as a passive relay of 257 peripheral information has been overruled by evidence that the thalamus actively regulates 258 information transmission to the cortex and between cortical areas (Saalmann & Kastner, 259 2011). This is particularly true for the pulvinar, which has been involved in a variety of Although these two hypotheses (that short-term plasticity affects feedback or feedforward 269 connections between the thalamus and the visual cortex) are equally compatible with the 270 bulk of our data, the concept of plasticity originating in the pulvinar may be better suited to 271 explain the correlation between BOLD modulations in this subcortical area and perceptual 272 modulations. Interestingly, the concept that the visual pulvinar plays a fundamental role in 273 short-term plasticity is also supported by a recent human neuroimaging study, where pulvinar 274 was suggested to gate GABAergic inhibition in the cortex and the associated short-term 275 learning effect (Ziminski et al., 2021). 276 In conclusion, the present study showed that short-term monocular deprivation effects, We acquired four BOLD time series per participant, two before and two after monocular 329 deprivation. In each series, only one eye was stimulated, and the other viewed a mid-level 330 gray image. Stimuli consisted of bandpass filtered, dynamic noise images presented in a block 331 design, with 9 sec long periods of stimulation (during which the noise stimulus was refreshed 332 at a rate of 8 Hz) separated by 12 sec of rest (mid-level gray screen), repeated 10 times. Across 333 blocks, the spatial frequency cut-off of the bandpass filter was varied. Unlike in Binda et al. 334 (2018), here we pooled across spatial frequencies, for both theorical (spatial frequency tuning 335 in the thalamus is not expected to be as sharp as in the cortex) and practical reasons (pooling 336 across repetitions compensates for the lower SNR of the subcortical regions). This turned our 337 stimulus into a periodic alternation of ON (9 sec) and OFF periods (12 sec), expected to 338 generate periodic visually evoked responses, the amplitude of which can be efficiently 339 estimated with Fourier analysis, extracting the amplitude at the stimulus frequency (1 cycle 340 every 21 sec or 0.047 Hz). The advantage of this method is that it does not make assumptions 341 on the latency of the response, which is captured by the phase parameter, and it is free to 342 vary across regions. For visualization purposes ( Figure 1A and Figure 1-supplement 1) 343 amplitude estimates were computed for individual voxels, after averaging fMRI time series 344 across conditions and participants (amplitude estimates were divided by the root mean 345 squared error of the corresponding sinusoidal function, yielding a value that is conceptually 346 similar to a t-statistics). 347 We complemented this analysis with two other methods that, contrary to the Fourier 348 approach, do make assumptions on response latencies. 349 First, we used an Event related averaging approach to estimate the profiles of fMRI responses. 350 We selected 21 sec long (7TRs) BOLD epochs following each stimulus onset and averaged 351 across epochs (of which we had 10 per acquisition). We assumed that the response occurs 352 between 3 sec and 12 sec from stimulus onset, and we used the average over this interval to 353 estimate its amplitude.