Induction of excitatory brain state governs plastic functional changes in visual cortical topology

The classical method to demonstrate the reorganization of cortical connectivity following retinal lesions (see Box 1) is the electrophysiological recording of single cells and the determination of their receptive field topography while recording within the LPZ and in the normal cortex (Fig. 1).

After recovery periods between 2 weeks and 12 months the lesion-induced cortical changes were investigated in acute preparations under continuous anesthesia. Single cortical cells were recorded in the upper layers of cat area 17 moving from inside the LPZ in the upper part of the medial bank downwards into the visual cortex representing the surrounding normal retina (Fig. 1B). In control animals, RFs shifted from the area centralis progressively towards the periphery of the visual field. In lesioned animals, while moving the electrodes downwards through the LPZ the RFs piled up at ectopic positions next to the retinal lesion and only showed a progressive shift when recording further down from the normal cortex.

The maintained activity (Fig. 1D) and the visual responses (for details of recordings see Box 1) were quantitatively determined (Fig. 1E) and depicted as a function of recording location along the electrode track (Fig. 1B).

In control animals, it was found that the relative recording probability of single cells was constant along the vertical recording tracks, but decreased progressively with distance from the normal cortex in lesioned animals and reached about 15% of normal close to the center of the LPZ with recovery times up to 3 months. A significant increase to more than 50% was observed after 1 year of recovery.

Spatiotemporal peaks of significantly increased maintained activity were discovered at 2 and 4 weeks migrating with time towards the center of the LPZ which disappeared upon completion of the filling-in after 12 weeks to 1 year. At this time normal maintained activity was measured throughout the LPZ, while very low maintained activity was observed in the central LPZ (Fig. 1D). At the same recovery times comparable peaks of visual hyperexcitability characterized the newly emerging visual responses inside the LPZ (Fig. 1E). The rising phase and peak of visual hyperexcitability resulted from strong responses to full-field drifting gratings while no well-defined RFs were detected at these recording sites during early reorganization of the LPZ. This spatiotemporal pattern characterized the forefront of visual re-innervation of previously "blind" cells. The well-defined RFs emerged later in time within the LPZ, i.e. closer towards the normally innervated cortex (see arrows in Fig. 1E).

In normal cats, RF size grows larger at larger eccentricities in the visual field as shown along the recording tracks from the cortical surface downwards (Fig. 1F, control). In contrast, at the shorter survival times after lesioning the width of RFs in the LPZ started at significantly larger values and decreased towards the border of the LPZ (Fig. 1F). The widest RFs were detected close to the excitability peak (Fig. 1D, E). At longer survival times (3 and 12 months), the RF widths inside the LPZ recovered and decreased to the sizes found in normal cells close to the LPZ (Fig. 1F).

It can be concluded that increased excitability in general and the increased size of RFs in particular might indicate an imbalance between glutamatergic excitation and GABAergic inhibition in the LPZ of adult cats in early stages of functional reorganization. Two weeks after retinal lesioning the GAD immunoreactivity is downregulated (Arckens et al. 2000; Rosier et al. 1995) and GABA concentrations are decreased. At the same time, a peak of strongly increased Glutamate immunohistochemistry is detected in the LPZ. This peak moves spatiotemporally within about 3 months together with a peak of hyperactivity (Fig. 1D, E) from the border towards the center of the LPZ (Arckens et al. 2000). This is accompanied by a functional filling-in of the scotoma (Giannikopoulos and Eysel 2006). In early postnatal life increased excitation and reduced inhibition are known to promote synaptic plasticity based on the weak GABAergic inhibition and even excitatory effects of GABA (Guo et al. 1997; Lin et al. 1994). Under these conditions, effective long-term potentiation can be elicited (Fox 1995). Electrical stimulation of horizontal fibers elicited LTP in the cat visual cortex (Hirsch and Gilbert 1993), and horizontal axons responded to retinal lesions with axonal sprouting in the adult cat and monkey (Darian-Smith and Gilbert 1994; Yamahachi et al. 2009). On these grounds, we suggest that the hyperexcitability combined with synaptic use in the horizontal cortical fiber system are key mechanisms for synaptic reconnection and functional filling-in following retinal lesions.

In summary, when visual cortical cells are deprived of retinal input by homonymous retinal lesions in adulthood they form an initially unresponsive “blind” LPZ. This cortical region becomes the stage for subsequent neuronal plasticity with visual reactivation and topographic remodeling (Chino et al. 1992, 1995; Gilbert 1992; Gilbert and Wiesel 1992; Heinen and Skavenski 1991; Kaas et al. 1990). Altogether the literature about lesion-induced plasticity suggests that cortical remodeling is accompanied by a temporal loss of inhibition and increase of cortical excitability that appears to promote functional reorganization of the cortical circuitry.

Single cell recordings as described above provide accurate local information about neuronal output. However, because of limits in the density of spatial sampling and the suprathreshold nature of the signals, the underlying analogue synaptic events that drive far-reaching integrative processes cannot be accessed. Moreover, gradual voltage changes in postsynaptic population activity (Jancke 2017, 2018) within the LPZ that accompany recovery processes remain elusive.

Wide-field optical imaging using voltage-sensitive dye (VSD) (see Box 2) permits the depiction of the ongoing state of the cortex through measurements of changes in membrane voltage across millions of neurons under varying input conditions (Onat et al. 2011a). Furthermore, as the emitted fluorescent signals reflect a continuum of membrane potentials, gradual changes in subthreshold activity that spread via long-range cortical connections (Creutzfeldt et al. 1977; Fisken et al. 1975; Gilbert and Wiesel 1979; Rockland and Lund 1982) become unmasked. Consequently, reorganization processes across widely interconnected neurons, including their nonlinear interactions that are not apparent in extracellular spiking, can be tracked over a wide spatial range as postsynaptic activity at mesoscopic population level (Freeman and Barrie 2000; Jancke et al. 1999; Jancke 2000, 2017; Onat et al. 2011a, 2013). Because the optically measured signals provide a direct link to the functional cortical architecture (Arieli et al. 1996; Blasdel and Salama 1986; Bonhoeffer and Grinvald 1991), potential systematic mappings of stimulus parameters and their combinations become accessible in terms of cortical coordinates (Onat et al. 2011b).

One major strength of VSD imaging is its capability to unravel horizontal influence and propagation of activity across upper cortical layers (Grinvald et al. 1994; Jancke et al. 2004; Jancke 2017; Muller et al. 2014; Rekauzke et al. 2016). For a direct proof Fig. 2 demonstrates an example in rat visual cortex, where horizontal propagation was unmasked using an “artificial scotoma” (i.e., hindering retinal feed-forward input using a local isoluminant gray area that omitted a part of the full-field stimulus). Note again that the visualization of such horizontal spread is of particular importance regarding studies of cortical plasticity, as these follow the hypothesis that horizontal connections are a major source of retinal lesion-induced cortical rewiring (Das and Gilbert 1995; Giannikopoulos and Eysel 2006; Kaas et al. 1990).

In our experiments in rats, a laser was used to coagulate a small region (~ 1 mm in diameter) of the retina eliminating all retinal layers dorsal to the optic disc (Fig. 3A). Hence, this intervention produced a retinal lesion that removed feedforward visual input to the medial part of the primary visual cortex, leading to a temporo-nasal scotoma of ~ 15–20° (Fig. 3A, B, D). In this study, following the retinal lesions, VSD imaging was conducted in three groups of animals: the first group, lesioned at P65, was recorded at P69–P72 (“acute lesion”). The second group was also lesioned at P65 but imaged after an extended period of recovery, that is, P92–P105. A third group of unlesioned rats with matching ages was imaged in order to characterize normal retinotopic cortical representations as a control (Fig. 3C).

A summary of the cortical retinotopic map obtained after lesion is shown in Fig. 3, where with the exception of the LPZ (stippled line marks LPZ border) that was “cortically blind”, activity in response to all stimuli outside the projection of the retinal lesion was retinotopically arranged in overlapping fashion (compare colors of sketch of stimulus positions in the visual field and contours encircling cortical activation in Fig. 3D, left and right panels, respectively). Recordings of cortical activity in response to stimuli that covered the entire visual field (full-field gratings), including the scotoma, revealed significant differences between groups (Fig. 4), 6 days after lesion (“acute”, upper row) and 28 days post-lesion (“recovery”, lower row). At response onset, cortical regions that were unaffected by the lesion showed a similar instantaneous and broadly distributed emergence of activity in both groups. This reflected most likely maintained early subcortical input to the parts of the cortex outside the LPZ (marked-off by black line). Indeed, the latency maps in Fig. 4B confirmed that the earliest responses were found along the lateral posterior–anterior axis (green colors) where direct thalamic input remained functionally intact. From then on, in both cases (i.e., after short and longer recovery times) latencies increased systematically towards medial regions (red/gray), demonstrating delayed spread of activity crossing the LPZ border via horizontal axons (black line). Thus, lateral input became gradually effective across regions within the LPZ, which can be seen by the systematic shift of the curves in Fig. 4C towards the LPZ [each of the traces depicts spatial averages across unaffected regions (green), around LPZ border (red), and within the LPZ (gray)]. Noteworthy, also the slopes of latency gradients throughout the LPZ became steeper for 28 days post-lesion cases when compared with acutely lesioned animals (Fig. 4C, lower and upper plot, respectively), demonstrating a more rapid succession of activation across the LPZ after longer recovery. Furthermore, amplitudes of activation inside the LPZ reached higher values in animals after long recovery than in acute animals that had only little time for functional retrieval (Fig. 4D).

In summary, this study showed maintained ability of the adult rat visual cortex for plastic reorganization. In particular, VSD imaging enabled in this case the direct visualization of the functional strengthening of lateral connectivity during the time course of recovery from retinal lesion. That is, after retinal lesion (i.e., local removal of direct retinal input), visually evoked activity of neuronal populations inside the LPZ remained initially subthreshold indicated by propagation of low-level activity and further verified through additional extracellular recordings (not shown). Contrastingly, after longer recovery times, the emergence of stimulus-evoked suprathreshold activity within the LPZ was revealed, most likely due to reorganization processes that strengthened horizontal cortical input.

These findings are corroborated and extended by retinal lesion studies by Keck et al. in adult mice, where within 1–2 days the number of inhibitory cell boutons and spines dropped sharply and inhibitory synaptic density and function declined, while the initially depressed excitatory activity returned to normal values due to synaptic scaling. This was followed by a massive restructuring of excitatory connectivity (within 3–63 days), accompanied by stabilization of new connections and functional filling-in of the scotoma (Keck et al. 2008, 2011, 2013); see also review by Sammons and Keck (2015). Altogether, our results in rodents suggest that horizontal connectivity can undergo gradual reinforcement that counteracts lesion-induced loss of visual thalamic projections, similarly as found in our cat experiments (Giannikopoulos and Eysel 2006). Of note, using the retinal lesion model in monkey revealed only little reorganization effects at the border of the cortical LPZ (Smirnakis et al. 2005). These results were obtained with fMRI recordings and may therefore be not sensitive enough to capture post-lesion neuronal organization processes (Calford et al. 2005). On the other hand, it needs to be considered that projections of the length of axons into visual space coordinates are significantly different between rodents and primates. This may facilitate reorganization processes over larger distances in functional space for rodents as compared to primates. Interestingly, Russo et al. (2021) showed recently in humans that focal cortical lesion can result in the deafferentation of connected areas promoting the emergence of sleep-like slow-waves. Thus, besides local state changes in the region around the LPZ, additional long-range effects may contribute to the observed reorganization properties in V1 following retinal lesion.

The crucial point of this review is to present the idea that an induced excitatory cortical state can trigger functional cortical rewiring in the adult. So far, we showed that retinal lesions produce such an excitatory state in the LPZ in the affected cortical region. However, the question arises whether an excitatory cortical state can also be produced locally through noninvasive methods? If this is the case, applications in humans might become feasible that could initiate neuronal rewiring to confined cortical regions.

One of the major points of this review is to outline the idea that an induced excitatory cortical state, here generated through the use of 10 Hz repetitive transcranial magnetic stimulation (rTMS), may directly be exploitable to target cortical rewiring processes. Insight into the mechanisms underlying such reorganization processes will crucially help creating settings, in which cortical functional loss in humans may partially be restored (or recovery supported) using noninvasive applications that externally stimulate the brain. Next, we therefore describe our experiments that aimed at elucidating cortical reorganization processes after application of TMS followed by subsequent sensory (in our case visual) stimulation during post-TMS excitatory state.

Using fluorescent light to track changes in neuronal activity, VSD imaging avoids signal artifact contamination due to the strong electric field induced by the TMS pulses. Indeed, this optical approach allowed to accomplish, for the first time, artifact-free visualization of TMS-induced activity over several millimeters of cortex in the anesthetized cat with a single image-frame resolution of 10 ms (Fig. 5). Using VSD imaging it could be shown that each TMS pulse (Fig. 5B, left trace) produces a brief focal spot of activation with highest amplitudes evoked close to the coil (cf. Figure 5A, C, first 2 frames after TMS onset, reddish colors). This initial activation was succeeded by strong and widespread suppression lasting up to ~ 300 ms (bluish colors). Thus, a single TMS pulse caused a brief ~ 20 ms period of excitation succeeded by an instantaneous decrease in the postsynaptic potential of a large amount of neurons below baseline (i.e., producing a so-called “vertical lesion”). Following suppression, activity showed a rebound (Moliadze et al. 2003) during which locally rising activity was surrounded by cortical regions that remained suppressed. Altogether, these observations suggest that TMS releases a powerful volley that may initially drive interneurons (Lenz et al. 2016; Murphy et al. 2016; Werhahn et al. 1999; Ziemann 2003), most likely inhibitory neurons expressing parvalbumin (PV) (Benali et al. 2011). As PV neurons target pyramidal neurons on the soma (Markram et al. 2004) and given their far-reaching axonal-dendritic structure (Chung et al. 2013), which may produce enhanced TMS-sensitivity (McAllister et al. 2009), PV interneurons likely impose synchronized inhibition in the cortical circuitries (Pascual-Leone et al. 2000). In turn, excitatory networks could be activated that counterbalance such initially TMS-evoked inhibition, presumably involving activation of NMDA-receptors (Huang et al. 2007).

How are further consecutive TMS pulses represented in the cortex and how do they influence the progression of activity dynamics? After the early phase of suppression following the first TMS pulse, high-frequency TMS (10 Hz) applied over the visual cortex (Aydin-Abidin et al. 2006; Bohotin et al. 2002) induced consecutive boosts in activity in close synchrony with each TMS pulse. Hence, in this case, activity builds-up to high levels with widespread activation (see right trace in Fig. 5C). In Fig. 5D time traces of spatially averaged activity are shown. It can be noticed that in the 10 Hz condition (red trace) the strength of suppression after the first-pulse appears moderately smaller than for the single-pulse protocol (Fig. 5D, compare red and green trace, respectively). We speculate that multiple repetitions of TMS trials lead to enhanced activity that accumulates during high-frequency trials, possibly driven by reduced efficacy of inhibition (Lenz et al. 2016). Overall, repetitive 10 Hz TMS-pulses increased step-wise the level of postsynaptic potentials, starting from initial suppression up to high amplitudes producing an excitatory state across a large pool of cortical neurons (Kozyrev et al. 2014).

In our experiments in the cat visual cortex, the baseline status of orientation map layout was determined before application of TMS (pre-treatment maps). These maps were then re-evaluated directly after rTMS and after a subsequent “passive visual training” protocol lasting ~ 30 min. The visual stimulation protocol consisted of repeated presentations of gratings (phase-shifted) with a single orientation (Fig. 6A, sketch on top). Two examples of orientation maps before (pre) and after treatment with high-frequency rTMS (10 Hz) and after visual stimulation (post) are shown in Fig. 6A. Pre-TMS maps were characterized by a systematic and balanced representation of orientation angles centered around so-called “pinwheel” centers (Fig. 6A, left column). Thus, the maps before application of the protocol displayed their typical and well-known layout (Bonhoeffer and Grinvald 1991) with roughly equal incidence of all orientations. Next, the maps after 10 Hz rTMS and the following prolonged exposure to a selected orientation (visual “training”) were re-evaluated. The timespan directly after visual stimulation was excluded from analysis to avoid confound with early adaptation effects (Dragoi et al. 2000). We discovered that the maps were now dominated by regions representing the stimulated orientation (see Fig. 6A upper row after “training” with horizontal (0°) orientation). In fact, the cortical area encoding the “trained” orientation, here horizontal (see orange-reddish colors) increased by 28.9%, as measured in a time interval of 1–2.5 h after visual stimulation. This plasticity effect was independent of the orientation used for “training”, as verified by the use of different orientations in the individual experiments. For instance, in the experiment depicted in Fig. 6A bottom row, the stimulus orientation was 90°. In this case, orientation dominance shifted towards vertical (see bluish-greenish colors), revealing an increase (18.6%) in the representation of orientations that again matched closely the stimulated orientation.

The relative enlargement of regions encoding the “trained” orientation was on average 19.0 ± 4.0% SEM (7 experiments). Conversely, a significant reduction of the number of pixels with orientation preferences that were orthogonal to the stimulated orientation was observed (see Fig. 6B, red bar at 90°). Small but no significant changes of orientation maps were detected when the visual stimulation was carried out in combination with sham TMS (Fig. 6B, black bars, 3 experiments). Therefore, the reorganization of the cortical maps was evident by an increased occurrence of orientations that represented the visually “trained” and neighboring orientations when 10 Hz rTMS was coupled to subsequent visual stimulation (Fig. 6C, thin red lines in left graph show distribution of orientation preference after 10 Hz TMS for the individual experiments). Additional analysis showed that the TMS- and “training”-induced bias in the remodeled map layouts was not randomly recruited. Instead, remodeling was characterized by a systematic shift of orientation preferences towards the targeted visually “trained” orientation (cf. Fig. 3 in Kozyrev et al. 2018).

The above observations clearly suggest that the key mechanisms underlying significant remodeling of the functional cortical architecture are generated by means of high-frequency 10 Hz rTMS. Of note, no plastic changes were detected after application of 1 Hz rTMS. Therefore, further analysis of the time window directly after the 10 Hz rTMS treatment (“post-10 Hz TMS”) was performed. The left image in Fig. 7A depicts the layout of an orientation map, this time assigning reproducibility values to all pixels (brightness in maps indicates strength in reproducibility). In brief, if a pixel response was consistently strongest for a certain orientation in each trial (i.e., displaying low single-pixel trial-to-trial circular variance; for in depth explanations of the procedure see Grabska-Barwinska et al. 2009, 2012), a high reproducibility score is assigned. Thus, in Fig. 7A low values (dark) indicate that preferred orientation was inconsistent across trials. Interestingly, and as a proof of concept, this analysis also locates the position of pinwheel centers (cf. dark small regions in the center of encircled areas). Around pinwheel centers (Bonhoeffer and Grinvald 1991) neurons are tuned to different orientations in close vicinity. Hence, their spatial average across pixels leads to low reproducibility values. Using this approach, it could be revealed that the reproducibility of orientation preference for each image pixel was strongly declined after 10 Hz rTMS as compared to the map obtained before TMS (Fig. 7A, compare right (post) and left frame (pre), respectively).

One major cause of increased response variability after 10 Hz rTMS could be reduced inhibition. Note that before application of rTMS, the response to visual stimulation displayed a “notch”, i.e., a small downturn during the early rising phase of activity (Fig. 7B, gray time course). Such notch can typically be found in visually evoked VSD signals and is considered as a signature of inhibition, which may sharpen neuronal tuning to orientation (Sharon and Grinvald 2002). We found that after 10 Hz rTMS the notch was attenuated and activity continued rising towards higher amplitudes as compared to pre-TMS (Fig. 7, see encircled region and further progression of time courses, bluish and gray traces, respectively). Taken together, these results indeed suggest that following 10 Hz rTMS inhibition is reduced, which may facilitate an excitatory cortical state, where amplitudes of sensory-evoked activity are increased and orientation-specific response constituents are weakened (Kozyrev et al. 2014). Consequently, orientation selectivity, calculated as the difference between preferred and orthogonal responses (i.e., “modulation depth”), was significantly reduced for pixels with low as compared to pixels with high reproducibility (Fig. 7C; for a similar observation using intrinsic optical imaging and electrophysiology see (Kim et al. 2015). Moreover, the rTMS-induced loss in orientation selectivity manifested in decorrelated activity across the neuronal circuitries. Correlating pre-and post-TMS orientation maps in response to various stimulus orientations revealed that correlations after 10 Hz rTMS strongly declined (Fig. 7E, left and right plots in upper row). In summary, our 10 Hz rTMS experiments in cats suggested that high-frequency TMS stimulation produces a brain state that sets the ground for subsequent remodeling of functional cortical connectivity. In the visual cortex, such state was characterized by decreased orientation selectivity, increased excitability along with signatures of reduced inhibition, and decorrelation of neuronal responses.

How sustained were perturbation effects and the subsequent cortical remodeling after the solitary TMS intervention? Because of dye bleaching reliable VSD imaging could be performed maximally ~ 6 h after application of TMS (median data acquisition time was 2.5 h post TMS). Within this time window the observed reorganization of cortical orientation maps stayed stable. However, it appears reasonable to assume that a long-lasting establishment of rTMS-induced functional cortical remodeling requires repeated sessions (Cirillo et al. 2017; Gersner et al. 2011). One possible explanation for the need of multiple TMS sessions might be that adult neuronal networks need to overcome “plasticity thresholds” to prevent reestablishment of their pre-treatment properties (Rose et al. 2016). It also should be considered that the described experiments were performed under anesthesia, which can critically change neuronal responsiveness (Niell and Stryker 2010). Thus, results could deviate when subjects receive repeated TMS interventions in the awake state. Nonetheless, several findings speak to the fact that remodeling of cortical maps depends also on structural plastic changes. For instance, de novo formation of axonal boutons along with spine growth and their retraction occur over only tens of minutes (Holtmaat and Svoboda 2009). Moreover, in retinal lesion experiments it was demonstrated that such intensified dendritic synaptic spine turnover could contribute to restructuring of cortical circuitries on a relatively large spatial scale (Keck et al. 2008).

Finally, it remains to be clarified whether TMS-induced cortical remodeling includes additional plastic changes in cortico-thalamic connections (Jaepel et al. 2017) and to which extent remodeling may uncover existing intrinsic connectivity by a potential unmasking of latent inhibitory connections (Barron et al. 2017; Calford et al. 1999; Gilbert and Wiesel 1992). High-frequency TMS is well-known to facilitate cortical disinhibition (Cash et al. 2010, 2016) which in turn likely causes increased excitability as seen in our experiments. In entorhinohippocampal slices 10 Hz magnetic stimulation was shown to decrease the strength of GABAergic synaptic input (Lenz et al. 2016). Thus, the 10 Hz rTMS-induced shift in orientation preference observed in our studies towards the “trained” orientation could emerge from functional weakening of lateral suppression (Patterson et al. 2013). Furthermore, based on the observation that cortical activity displays a strong rebound and excitation phase after prolonged TMS application (Kozyrev et al. 2014; Li et al. 2017), it was recently suggested that the rTMS-induced increase in excitability also involves strengthening of cortico-subcortical loops (Li et al. 2017). Thus, we propose that cortical remodeling by rTMS initially triggers increased cortical excitability in parallel with a cascade of changes in functional strength of connectivity including cortico-thalamic input.