Convergence of unisensory-evoked signals via multiple pathways to the cerebellum

The cerebellum receives signals directly from peripheral sensory systems and indirectly from the neocortex. To reveal how these different types of signals are processed in the cerebellar cortex, in vivo whole-cell recordings from granule cells and unit recordings from Purkinje cells were performed in mice in which primary somatosensory cortex (S1) could be optogenetically inhibited. Tactile stimulation of the upper lip produced two-phase granule cell responses (with latencies of ∼ 8 ms and 28 ms), for which only the late phase was S1 dependent. Complex spikes and the late phase of simple spikes in Purkinje cells were also S1 dependent. These results indicate that individual granule cells integrate convergent inputs from the periphery and neocortex, and send their outputs to Purkinje cells, which then combine those signals with climbing fiber signals from the neocortex.


27
The cerebellum and neocortex are interconnected by a large fiber system (Schmahmann et al., 28 2019; Watson and Apps, 2019). Although the basal ganglia similarly connect with the neocortex, 29 the cerebellum receives inputs not only from the neocortex but also from the peripheral sensory 30 systems, including tactile, proprioceptive, and vestibular systems (Bostan et al., 2013). How the 31 cerebellum integrates these signals is not well understood.

32
The cerebellum receives inputs via two types of projection fibers, namely, mossy fibers that 33 project to granule cells and climbing fibers that project to Purkinje cells. A major subgroup of the 34 mossy fibers projects from the pontine nuclei (basilar pontine nuclei and nucleus reticularis 35 tegmenti pontis), which relay signals from the neocortex (Kratochwil et al., 2017;   upper lip area of the contralateral S1. S1 was optogenetically suppressed in alternating trials.

135
The above-described results were consistent with the established theory that at least two distinct 136 groups of mossy fibers project to the GCL of crus II: one directly from the trigeminal nuclei and the 137 other via the cerebropontine pathway (Morissette and Bower, 1996;Steindler, 1985;Woolston et 138 al., 1981). To determine whether these two inputs converge onto individual granule cells, we 139 performed whole-cell patch-clamp recordings in anesthetized mice. In 50% (12/24) of granule cells, 140 excitatory postsynaptic currents (EPSCs) evoked by stimulation of the upper lip had two 141 components (as in the representative cell in Figure 2A), indicating that two types of mossy fibers 142 converge on some individual cells. However, other granule cells had conspicuous EPSCs with only 143 early (12.5%) or late (25%) timing, and the remainder (12.5%) had no response (cutoff, 0.5 144 events/trial) ( Figure 2B). The numbers of EPSC events for the two components did not correlate (r 145 = 0.028, n = 24), suggesting that these connections were established independently. In these 146 recordings, it was noted that each component often had multiple EPSC events with very short 147 intervals (around 3 ms on average; Figure 2C). This is in line with previous studies showing that a 148 single mossy fiber can fire high-frequency bursts of action potentials that trigger high-frequency 149 EPSCs in granule cells (Chadderton et al., 2004;Rancz et al., 2007). Indeed, in our occasional 150 whole-cell recordings from putative mossy fiber boutons (n = 3), action potentials occurred in highfrequency bursts with either early or late timing ( Figure S3). Thus, the multiple EPSCs in each component were likely derived from a single mossy fiber. Furthermore, as the interevent intervals

156
(A) Representative recording; simultaneous field potential recording from S1 (top) and whole-cell        176 177 pontine mossy fibers can fire similar high-frequency bursts. However, the fluctuation of the timings e e g reflecting a longer multistep pathway for transmission of the late response. However, the amplitude of individual EPSC events was larger for the early components ( Figure 2E), suggesting that the expected, the late component of EPSCs was mostly eliminated by optogenetic suppression of S1 obtained when synaptic charge (the area over the curve), instead of event number, was measured 186 as an indicator of synaptic strength ( Figures S4A and S4B). Furthermore, spontaneous EPSCs 187 were partially blocked by the suppression of S1 ( Figure S5). These results indicate that granule 188 cells receive convergent inputs from two types of mossy fibers, although the balance of these 189 inputs varies between cells. Additionally, we recorded inhibitory postsynaptic currents (IPSCs) in a 190 small number of granule cells (n = 4) to examine feed-forward inhibition from Golgi cells. IPSCs 191 also exhibited two components, and the late component was suppressed by light illumination of S1

195
To investigate how synaptic inputs trigger action potentials in granule cells, we recorded 196 membrane potentials in current clamp ( Figure 3A). The resting potentials of granule cells varied 197 widely (from -96.0 to -50.8 mV; mean, -72.9 ± 3.4 mV, n = 16). In cells that had a relatively 198 hyperpolarized resting potential, the excitatory postsynaptic potential (EPSP) did not reach the 199 spike threshold ( Figure 3B), indicating that a single tactile stimulus generates a spike in only a 200 subset of the granule cells. As the resting potentials of granule cells are modulated by multiple 201 factors, such as tonic inhibition from Golgi cells and potassium channel function (Duguid et al., 202 2012;Hamann et al., 2002;Millar et al., 2000), we applied steady depolarizing current (up to 20 203 pA) to 9 of the 16 cells, thereby raising the average resting potential to -57.2 ± 1.9 mV (n = 16, 204 including seven cells without adjustment) and increasing their propensity to fire in response to 205 tactile stimulation of the upper lip (11/16 cells fired action potentials; cutoff, 0.5 spikes/trial)

206
( Figures 3A and 3B). In this condition, the granule cells fired during the early or late phase ( Figure   207 3C). As expected, action potentials in the late phase were eliminated by photoinhibition of S1 208 ( Figure 3F). Investigation of the relationship between synaptic inputs and firing outputs showed 209 that the number of action potentials evoked in the early phase correlated with the number of EPSC 210 events during the same period (r = 0.77, n = 13) ( Figure 3D, left). However, the correlation was 211 less clear in the late phase (r = 0.53, n = 13) ( Figure 3D, right), suggesting that factors other than 212 instantaneous synaptic inputs may be involved. Indeed, depolarization during the early phase was

252
Consistent results were obtained with field potential recordings ( Figure S7A). These results

253
indicate that the putative trigeminal and corticopontine mossy fiber inputs are variably distributed 254 over multiple bands in crus II. Furthermore, trial-by-trial analysis revealed a correlation between the 255 amplitude of the S1 response and the number of EPSC events in the late component but not in the 256 early component (Figures 4D and 4E), confirming that only the late component is S1 dependent.

257
Similar to that observed in VGAT-ChR2 mice, the mean EPSC amplitude was larger and the jitter 258 of the first event was smaller in the early response than in the late response, and the interevent 259 intervals did not differ (Figures S7C-E).

262
We next investigated how the suppression of S1 affects the firing of Purkinje cells, which are 263 downstream of granule cells in the cerebellar circuit (Bower and Woolston, 1983). In general, 264 simple spikes (SS) in Purkinje cells are generated spontaneously but are affected by excitatory 265 synaptic inputs from parallel fibers (i.e., granule cell axons) and inhibitory synaptic inputs from 266 molecular layer interneurons, whereas complex spikes (CS) are triggered solely by climbing fiber 267 inputs. Under control conditions (without S1 inhibition), there were four phases of SS in response 268 to stimulation of the upper lips of VGAT-ChR2 mice, namely, early excitatory (within 10 ms after

272
(A) Representative recordings; simultaneous field potential recordings from S1 (top) and whole-cell

283
Means ± SEMs are presented as black bars and lines, respectively. See also Figure S7.

285
Photoinhibition of S1 did not affect the early phases of SS but eliminated the late excitatory and 286 inhibitory phases ( Figures 5A-5D). Interestingly, the CS response was also abolished. These 287 results suggest that the direct trigeminal inputs to granule cells trigger the early excitatory and 288 inhibitory phases of SS, the latter of which presumably results from the activation of molecular 289 layer interneurons (Blot et al., 2016;Jelitai et al., 2016). These results also suggest that the late 290 excitatory responses of SS reflect inputs from S1. However, the late inhibitory response could 291 reflect either an interneuronal effect or a pause after CS. As CS did not always occur under the 292 control condition, we were able to separate traces with CS from those without ( Figure S8). The late 293 inhibition phase of SS was larger in traces with CS than in those without CS, suggesting that the 294 late inhibition largely reflects the pause after CS. We also tested the effects of photoinhibition at 295 multiple loci in the neocortex and found that suppression of the upper lip area of the contralateral 296 S1 was most effective in blocking CS in Purkinje cells ( Figure 5E). Furthermore, suppression of S1 297 with a long (10 s) light stimulus inhibited spontaneous CS firing and triggered rebound activation 298 after the light was turned off, suggesting that the spontaneous activity of the inferior olive is under 299 strong control of S1. However, this photoinhibition had no effect on SS firing ( Figure S9), 300 suggesting that S1 activity is not directly linked to the spontaneous activity of granule cells ( Figure   301 S5) in line with a previous report (Ros et al., 2009).

304
Simultaneous field potential recordings from S1 are not shown. SS (black) and CS (red) are plotted

312
(E) Bubble size indicates magnitude of inhibition of the cerebellar CS when the blue light was 313 applied to each spot in the right cerebral hemisphere. Only five spots were tested. Data from six 314 animals were combined. Each spot is the average from 3-6 animals.
By taking advantage of the high temporal resolution of electrophysiology and optogenetics, we 319 obtained functional evidence that cerebellar granule cells in crus II receive inputs directly from the 320 periphery and indirectly via S1. We found that approximately half of the granule cells receive 321 convergent signals from both pathways. We also showed that the olivocerebellar inputs to these Optogenetic analysis of the neural pathways 325 We adopted an optogenetic method in which photostimulation of GABAergic neurons expressing 326 ChR2 can suppress the activity of specific areas of the neocortex (Guo et al., 2014;Li et al., 2015,  pons relays signals from S1 to the cerebellar cortex (Leergaard and Bjaalie, 2007;

333
2006; Morissette and Bower, 1996;Watson and Apps, 2019;Welker, 1987). Therefore, it is highly 334 likely that the late component of the cerebellar response in our study was transmitted via the 335 cortico-ponto-cerebellar pathway.

338
In contrast to the conventional view of Marr and Albus (Albus, 1971;Marr, 1969) that granule cells 339 receive inputs from mossy fibers carrying different types of information, it was recently proposed 340 that individual granule cells receive only one type of input (Bengtsson and Jörntell, 2009;Dean et 341 al., 2010;Spanne and Jörntell, 2015). Although both trigeminal and corticopontine signals were 342 evoked by the same tactile stimulation in our study, these signals are fundamentally different 343 because they can be differentially modulated on route to the cerebellum. For instance, neocortical 344 states (such as anesthesia and wakefulness) affected them differently (Figure 1), and they 345 exhibited different trial-to-trial fluctuations (Figure 4). Furthermore, as the S1 in mice sends efferent 346 motor signals directly to the brain stem (Matyas et al., 2010), the corticopontine signals we

370
An electrophysiological study using extracellular recording suggested that the GCL has a 371 stratified organization in which the trigeminal mossy fibers are located deeper than the 372 corticopontine mossy fibers (Tahon et al., 2011). Although our results do not directly contradict this 373 idea, they strongly suggest that the trigeminal and corticopontine mossy fibers largely overlap in 374 the GCL even if they tend to be distributed at different depths. Future studies, such as those using 375 functional cellular imaging (Giovannucci et al., 2017;Wagner et al., 2017Wagner et al., , 2019, may elucidate the 376 detailed organization of the GCL.

377
As there are extensive studies regarding inhibitory inputs to granule cells from Golgi cells 378 (Duguid et al., 2012Tabuchi et al., 2018;Vos et al., 1999), we performed only a few 379 experiments on IPSCs in this study. We found that, like EPSCs, the IPSC events had two phases.

380
These currents were delayed in relation to EPSCs by only a few milliseconds, reflecting rapid feed-

Convergence of inputs to Purkinje cells 385
We found that Purkinje cells receive signals from the neocortex not only via the mossy fiber-386 parallel fiber pathway but also via climbing fibers. This finding is consistent with anatomical 387 observations that Purkinje cells in the 6+ compartment (D1 band) receive inputs from climbing 388 fibers from the ventral principal olive, which receives inputs from the area parafascicularis is also in line with classical physiological studies showing a convergence of mossy and climbing fiber signals originating from the same neocortical areas in cats and monkeys  evoked climbing fiber responses. In contrast to what we observed, a recent study (Kubo et al., 397 2018) reported that inhibition of the somatosensory cortex in mice did not inhibit CS. The reason 398 for this discrepancy is unclear but may involve minor differences in sensory stimulation, 399 optogenetic methods, or the sites recorded.

400
The physiological significance of this cortio-olivo-cerebellar connection is intriguing,  One limitation of the present study is that the data were largely obtained from animals anesthetized 411 with ketamine/xylazine, which has a suppressive effect on parallel fiber-Purkinje cell synapses 412 (Bengtsson and Jörntell, 2007;Duguid et al., 2012). However, the sensory stimulation applied was 413 able to evoke SS in Purkinje cells. Ketamine/xylazine also synchronizes activity in the neocortex 414 (Pachitariu et al., 2015), which may enhance the size and distribution of cortical responses to a 415 punctuate stimulus (Harris and Thiele, 2011;Poulet and Crochet, 2018). The fact that S1-mediated 416 signals observed in the cerebellum were clearer (larger and better isolated) in anesthetized 417 animals than in awake animals ( Figure 1) may be attributable to this effect. As the precise 418 mechanisms of action of the anesthetics are not known, further studies are required to understand 419 how brain states, including natural sleep, affect cerebrocerebellar communication.

420
A second limitation is that we did not have a method to specifically block the trigemino-421 cerebellar pathway. A future challenge is to manipulate the trigeminal neurons projecting to the 422 cerebellum without affecting those projecting to the thalamus. This may be possible in principle as 423 those neurons may be different populations in the trigeminal nuclei (Bennett-Clarke et al., 1992; 424 Zhang et al., 2018). On a related note, we do not know whether the early phase inputs from the 425 trigeminal nuclei to the cerebellar granule cells were monosynaptic or polysynaptic. Although these 426 responses had very short latencies, it is possible that rapid polysynaptic inputs, relayed within the mice had two clear peaks. In this study, we collectively referred to these early responses as the 429 direct trigeminal response to distinguish them from the indirect S1-mediated response.

430
Finally, in this study, the only sensory stimulus was tactile stimulation (air puff) of the upper 431 lip. We adopted this form of stimulation because it gave large responses in both S1 and the 432 cerebellum. However, tactile stimulation of other body parts or stimulation of other modalities may 433 similarly evoke multipathway responses in the cerebellar cortex. Given the extensive connections 434 between the neocortex and the cerebellum (Leergaard and Bjaalie, 2007;Proville et al., 2014)  The mice were anesthetized with an initial dose of a mixture of ketamine (86 mg/kg body weight) and xylazine (10 mg/kg), and supplemented with a continuous infusion of ketamine (70 mg/kg/h) removing the overlaying skin and muscles, a craniotomy was performed over the left cerebellar 457 crus II area and the right cerebral somatosensory area. After removing the dura, the exposed brain 458 surface was kept moist with a HEPES-buffered saline containing (in mM) NaCl 150, KCl 2.5,

466
Whole-cell in vivo patch-clamp recordings were performed via a resistance-guided (blind) method 467 as previously described (Ishikawa et al., 2015;Rancz et al., 2007). Voltage clamp and current

499
Aichi, Japan) controlled via the USB-6259 interface as described above.

502
To analyze field potential recordings, the peak amplitude from the baseline was measured. In

545
Black bar indicates the duration of air puff, whose onset is indicated by the vertical dotted line.

546
(C-E) The peak amplitudes of S1 field potentials (C) and the early (D) and the late (E) components of 547 cerebellar GCL field potentials (n = 5). Means ± SEMs are presented as black bars and lines, 548 respectively.

558
MFBs were categorized as the early or the late type on the basis of their timing of firing.

559
(B) Spike patterns were analyzed in the same manner as described for Figures 2D, 2E, and 2F.

571
Means ± SEMs are presented as black bars and lines, respectively. 572 in e Figure S5. Effect of S1 photoinhibition on spontaneous EPSC in granule cells.

573
(A) Representative recordings; simultaneous field potential recordings from S1 (top) and whole-cell

609
(A) Representative recordings; simultaneous field potential recordings from S1 (top) and unit recordings