Infra-slow modulation of fast beta/gamma oscillations in the mouse visual system

Infra-slow (<0.02 Hz) and fast beta/gamma (20 – 100 Hz) oscillations in neurophysiological activity have been widely found in the subcortical visual system. While it is well established that fast beta/gamma oscillations are involved in visual processing, the role (if any) of infra-slow oscillations is currently unknown. One possibility is that infra-slow oscillations exert influence by modulating the amplitude of fast oscillations, yet the extent to which these different oscillations arise independently and interact remains unknown. We addressed these questions by recording in vivo spontaneous activity from subcortical visual system of visually intact mice, and animals whose retinal network was disrupted by advanced rod/cone degeneration (rd/rd cl) or melanopsin loss (Opn4-/- We found many neurons expressing only one type of oscillation, and indeed fast oscillations were absent in rd/rd cl. Conversely, neurons co-expressing the two oscillations were also common, and were encountered more often than expected by chance in visually intact but not Opn4-/- mice. Finally, where they co-occurred we found that beta/gamma amplitude was modulated by the infra-slow rhythm. Our data thus reveal that: 1.) infra-slow and beta-gamma oscillations are separable phenomena; and 2.) that they actively co-occur in a subset of neurones in which the phase of infra-slow oscillations define beta-gamma oscillation amplitude. These findings suggest that infra-slow oscillations could influence vision by modulating beta-gamma oscillations, and raise the possibility that disruptions in these oscillatory behaviours contribute to vision dysfunction in retinal dystrophy. KEY POINTS SUMMARY Neurophysiological activity in the subcortical visual system fluctuates in both infra-slow and fast oscillatory ranges, however the level of co-occurrence and potential functional interaction of these rhythms is unknown. Analyzing dark-adapted spontaneous activity in the mouse subcortical visual system, we find that these two types of oscillation interact uniquely through a population of neurons expressing both rhythms. Genetic ablation of rod/cone signaling potentiates infra-slow and abolishes fast beta/gamma oscillations while genetic ablation of melanopsin substantially diminishes the interaction between these two rhythms. Our results indicate that in an intact visual system the phase of infra-slow modulates fast beta/gamma oscillations. Thus one possible impact of infra-slow oscillations in vision is to guide visual processing by interacting with fast narrowband oscillations.

3 KEY POINTS SUMMARY 48  Neurophysiological activity in the subcortical visual system fluctuates in both infra-49 slow and fast oscillatory ranges, however the level of co-occurrence and potential 50 functional interaction of these rhythms is unknown. 51  Analyzing dark-adapted spontaneous activity in the mouse subcortical visual system, 52 we find that these two types of oscillation interact uniquely through a population of 53 neurons expressing both rhythms.   To test this possibility we recorded from mouse subcortical visual system (dLGN, vLGN, and 90 OPN). We found that infra-slow and fast beta/gamma oscillations interact through a population 91 of neurons that co-express both rhythms. Namely in these neurons the phase of infra-slow 92 modulates the amplitude of gamma oscillations. We then asked whether this interaction 93 required intact vision by repeating the same experiments in animals with genetic ablation of 94 rods+cones or melanopsin photoreception. We found that rod+cone loss abolished fast 95 oscillations, but left infra-slow rhythm intact, whereas both oscillations were retained in 96 melanopsin knockouts but their co-expression was substantially reduced. Together these results 97 confirm that infra-slow and fast oscillations are separable phenomena reflecting different 98 network states, and also that in physiological conditions their co-expression in the same neurons 99 is an active consequence of a particular network state that is disrupted by melanopsin loss.  (Grundy, 2015). 106 The following report is in part a re-analysis of previously collected and partially published   Opn1mw R mice were used as 'wild type' animals. All animals were removed from their home  All surgical procedures were conducted under deep urethane anesthesia (1.55 g / kg, 30% w / 125 v; Sigma-Aldrich, Munich, Germany). Animals were injected intraperitoneally and the depth 126 of anesthesia was ascertained by lack of withdrawal and ocular reflexes. If necessary, animals 127 were supplemented with 10% of the initial dose of urethane, however that has never been done 128 during the recordings. Body temperature was automatically maintained at 37 ± 0.5°C by the 129 thermistor with a feedback-controlled heating pad. Anesthetized mice were carefully placed in LGN. The coordinates were assessed based on a stereotaxic brain atlas for mice (Paxinos & 134 Franklin, 2001) and were AP, -2.8; LM, 0.9; DV, -2 to -2.8 mm and AP, -2.5; LM, 2.2; DV, 2.5 135 to 3.5 mm for the OPN and LGN, respectively.  Signals were amplified (3000×), filtered (300 Hz) and digitized at 40 kHz. All data were saved 142 on a computer hard disc for further analysis.  Multi-unit recordings were processed using Offline Sorter (version 2.8.8; Plexon) or Spike2 165 (version 6.08). After removal of cross-channel artifacts, each channel was analyzed separately. 166 Single-units were detected and categorized based on spike waveform via principal component 167 analysis (PCA) and related statistic ( Fig. 2A-B). Moreover, the inter-spike interval (ISI) 168 histograms were computed to monitor unit refractory period. Cross-correlograms for all isolated 169 units were examined to ensure that each unit was only included once in the further analysis 170 (probability of synchronous firing <0.01). In total we detected 2606 neurons and 2372 (those 171 with firing rates >0.02Hz) were further analysed. 2) The relation / between time constant and oscillation period has to be above a fixed 192 threshold (=0.6) in order to reveal regular oscillations (see Fig. 1D-F).

193
3) The autocorrelation model of equation 1 has to reduce (by >22.5%) the relative fitting error  Estimation of coupling between infra-slow and fast beta/gamma oscillations 217 Spike timestamps were binned at 1 s for infra-slow and 5 ms for fast beta/gamma to obtain 218 spike counts. Those counts were filtered by using zero-phase Kaiser filters (infra-slow: low-219 pass filter with cut-off at 2.5/ Hz and stop-band at 5/ Hz; fast beta/gamma: band-pass 220 filter with cut-off at 0.8 and 1.2 times the peak frequency and stop-band at 0.6 and 1.4 times).

221
A Hilbert transformation was then applied to the spike counts filtered in the infra-slow range 222 and the phase of these signals was clipped in the range (-π, π). Finally the phase of infra-slow 223 11 spike counts was used to divide fast beta/gamma band spike counts into the intervals ((0, π/2); 224 (π/2, π); (0, -π/2); (-π/2, -π)). In order to test against the possibility that the amplitude of fast 225 beta/gamma oscillations is modulated by the phase of infra-slow oscillations we compared the 226 amplitude of these oscillations in the range (-π/2, π/2) with the amplitude observed in the 227 complementary range by using a non-parametric sign-test.

229
Permutation test was used to test the null hypothesis that occurrence of fast and infra-slow repeating this procedure 100000 times. An example is shown in Table 1.

241
In the real data the firing rate distribution is different between units expressing infra, fast 242 beta/gamma or no oscillations (see Fig. 4G and Fig. 6G). Therefore, the number of units co-

263
The main aim of this work was to determine the relationship between infra-slow and fast 264 narrowband oscillations in the mouse visual system. We addressed these questions both in 265 image forming and in non-image forming centres in the brain by focussing on the dLGN (image   Fig. 2J, Fig. 3E).

294
Based on that data (in total 640 cells from 30 recording sites from 22 wild type animals) we 295 were able to describe four types of spontaneous activity (in darkness) in the sub-cortical visual  such narrowband oscillations (Fig. 3G-H).

306
Firing rate of neurons expressing only infra-slow were significantly correlated between each 307 other in all the structures investigated (except the OPN, Table 2 for Opn1mw R ) as revealed by 308 the cross-correlation analysis (see Material and Methods section). Significant correlations were 309 also observed in neurons co-expressing both infra-slow and fast oscillations (Table 2 for   310 Opn1mw R ). Moreover, these two populations were significantly correlated with each other (Table 2 for Opn1mw R ) suggesting that a single infra-slow rhythm coordinates firing rate in both population-level infra-slow oscillations (Fig. 4A,B). We then asked whether this modulation 326 could instead occur within the same population of neurons. Therefore we repeated the same 327 analysis on populations co-expressing both infra-slow and fast beta gamma oscillations. We 328 found that within these populations fast beta/gamma rhythms were strongly modulated 329 according to the phase of infra-slow oscillations (Fig. 4D,E). These results were consistent 330 across the whole dataset indicating that co-expression of both rhythms at the level of individual 331 unit is required for infra-slow modulation of fast narrowband oscillations (Fig. 4C,F).

332
Finally, we measured the extent to which the expressions of infra-slow and fast oscillations 333 were coordinated at population level.

334
To address this question, we tested the possibility that co-expression of fast and infra-slow Kruskal-Wallis test, p = 0, n = 822). However the distributions of the four types as function of 338 firing rate were largely overlapping (Fig. 4H)  structures. We found that the number of neurons co-expressing infra-slow and fast oscillations 346 was larger than expected by chance, indicating that these rhythms tended to co-occur at the 347 level of individual units (Fig. 4I).  Table   362 3. Importantly, there were no differences in the mean firing rates between genotypes across the Kruskal-Wallis test, Kruskal-Wallis statistics = 5.740, p = 0.0567).

366
In both groups of mice we found significant differences in the prevalence and the features of  (Fig. 5H).

383
Since both rhythms were retained following melanopsin loss, we next asked whether this 384 manipulation had altered the interaction between infra-slow and fast beta/gamma oscillations.

385
Similarly to what we found in visually intact animals, neurons recorded from Opn4 -/mice could  (Table 2 for Opn4 -/-). However, differently from 392 visually intact animals, the firing rates were not correlated between these two populations 393 (Table 2 for Opn4 -/-). Infra-slow modulation of fast oscillations was observed, but as in visually 394 intact mice, this only occurred in neurons expressing both rhythms (Fig. 6).

395
Finally we asked whether the rate of co-expression of both oscillations was also preserved. We 396 found that this was not the case since co-expression of infra-slow and fast beta/gamma 397 oscillation was not significantly different from chance level in Opn4 -/mice ( Fig. 6G-I). This  We find that these two types of oscillation are distinct phenomena, but that they do co-occur  (Fig. 2). Differences between these species in their frequency could have different 429 potential origins. On one hand, it could be a genuine species difference in visual physiology.

430
On the other, it may be connected to the fact that albino Wistar rats carry dysfunctions in the  type of oscillation, others express either both or none (Fig. 3).

444
In principle, in such population in which not all neurons expressing one type of oscillation also 445 have the other, slow oscillations could still modulate fast beta/gamma oscillations by two types 446 of circuitries. On the one hand, the population of infra-slow neurons could provide a modulation 447 of beta/gamma oscillations, including neurons that did not themselves have infra-slow 448 oscillations ( Fig. 7A-F). Alternatively, such modulation could be restricted to neurons in which 449 the two types of oscillations could interact within single neurons (Fig. 7H-J).

450
Our analyses revealed that only neurons co-expressing both oscillations are responsible for 451 mediating infra-slow modulation of fast beta/gamma oscillations (Fig. 4D-F). That is, when 452 they co-occur the amplitude of fast modulations is modified according to the infra-slow rhythm.

453
Importantly, the level of co-expression of these rhythms was higher than expected by two 454 independent events suggesting that the interaction between infra-slow and fast beta/gamma 455 occurs "by design" rather than simply by chance (Fig. 4G).

456
It is currently not possible to map the circuits displayed in Fig.7   Infra-slow oscillations were retained in all genotypes. In fact, they were most prevalent in the 486 rd/rd cl (see Table 3 Table 2 Cross-correlation between infra-slow oscillatory neurons, neurons co-expressing  Table  Permutation Table   #unit Infraslow Fast Co-expression #unit Infra-slow Fast Co-expression