Phase of neural oscillations as a reference frame for attention-based routing in visual cortex

Selective attention allows the brain to efficiently process the image projected onto the retina, selectively focusing neural processing resources on behaviorally relevant visual information. While previous studies have documented the crucial role of the action potential rate of single neurons in relaying such information, little is known about how the activity of single neurons relative to their neighboring network contributes to the efficient representation of attended stimuli and transmission of this information to downstream areas. Here, we show in the dorsal visual pathway of monkeys (medial superior temporal (MST) area) that neurons fire spikes preferentially at a specific phase of the ongoing population beta (~20 Hz) oscillations of the surrounding local network. This preferred spiking phase shifts towards a later phase when monkeys selectively attend towards (rather than away from) the receptive field of the neuron. This shift of the locking phase is positively correlated with the speed at which animals report a visual change. Furthermore, our computational modelling suggests that neural networks can manipulate the preferred phase of coupling by imposing differential synaptic delays on postsynaptic potentials. This distinction between the locking phase of neurons activated by the spatially attended stimulus vs. that of neurons activated by the unattended stimulus, may enable the neural system to discriminate relevant from irrelevant sensory inputs and consequently filter out distracting stimuli information by aligning the spikes which convey relevant/irrelevant information to distinct phases linked to periods of better/worse perceptual sensitivity for higher cortices. This strategy may be used to reserve the narrow windows of highest perceptual efficacy to the processing of the most behaviorally relevant information, ensuring highly efficient responses to attended sensory events.


Introduction
Spike counts and patterns are used to represent sensory information in neural systems [1][2][3][4][5][6]. 44 Independent of the magnitude of spike rate, the temporal structure of the individual spikes relative but also the performance of subjects' perceptual decisions [32,33]. There have been increasing 59 evidence suggesting that the brain uses the phase of oscillatory population activities to distinguish 60 information items via separate phase alignment of spikes. By shifting the phase relationship 61 between the spike trains and the underlying oscillatory population activity in a systematic way, 62 known as the phase shift phenomenon, the brain classifies spikes according to their information. 63 Data from rodents indicate that spatial information is encoded at specific phases of the ongoing 64 population oscillations for the hippocampal pyramidal cells [27,34,35]. Similarly, prefrontal 65 neurons encode different memory contents at distinct phases, allowing the brain to store and 66 separate different objects in memory concurrently [24]. This strategy may be exploited by the 67 visual cortex to differentiate behavioral relevance of different information items. 68 Due to the brain's limited resources to process the environmental information, selective 69 information processing is crucial for an efficient processing of behaviorally relevant information, 70 a cognitive function of the mammalian brain through which some aspects of environment gain 71 advantage over nearby distractors, known as selective attention [36][37][38][39]. Attention has been shown 72 to affect visual neurons which represent the attended location/feature in a systematic way to 73 enhance the neural representation and consequently the perception of behaviorally relevant (over 74 irrelevant) sensory inputs which benefits behavior [40][41][42][43][44][45][46][47][48][49][50][51][52][53][54]. More recently, several studies have 75 suggested a temporal structure for spatial/feature-based attention, i.e., attention samples the visual 76 environment rhythmically, that is the brain rhythms govern the visual target's detection 77 performance [55][56][57][58][59][60][61][62][63]. This rhythmic sampling is an indication of alternating periods of 78 better/worse perceptual sensitivity (assigned to 'good' and 'poor' phase of collective neural 79 activity, respectively) [57,61,64]. 80 Here we hypothesize that the phase of ongoing population oscillations may provide an internal 81 reference frame for discriminating sensory information of different behavioral relevance (see [21] 82 for consistent observations for working memory). We conjecture that the preferred phase of 83 spiking may be used as reference when encoding stimuli at the attended location to allow for a 84 selective routing to downstream areas.

85
To test this hypothesis, we recorded from the medial superior temporal cortex (MST), an area in 86 the extrastriate visual cortex of monkeys, with neurons selective to spiral motion patterns [65,66]. 87 We simultaneously recorded local field potentials (as a proxy of the collective synaptic activity of 88 local neural populations [67-69]) and single cell activity from two rhesus monkeys engaged in a 89 visual attention task. We compared the neurons' preferred spiking phase when they were encoding 90 relevant vs. irrelevant sensory information. Our data and the results of our computational 91 modelling support our hypothesis that the temporal pattern of single neuronal spikes is shaped by 92 the phase of beta oscillations in the LFP, and importantly, spatial attention organizes spike times 93 in order to discriminate relevant from irrelevant sensory inputs by aligning the spikes to different 94 phases of beta oscillations. These results provide insight into how the information of attended 95 stimuli, rather than unattended stimuli might be preferentially read out by downstream brain areas.

97
It has been shown that single neurons spike at specific phases of their neighboring network's 98 activity. This spike-phase coupling is indicative of a neuron's activity relative to its surrounding 99 network. Here, we hypothesize that the location/feature towards which attention is allocated may 100 be encoded by the coupling of the spikes' timing to different phases of the local network's 101 oscillatory activity (Fig 1). spikes between the two corresponding phase alignments. Therefore, changes in similarity between 109 the focus of attention to the position/feature to which the sensory neuron is tuned, shifts the 110 preferred spiking phase. Here, attention towards vs. away from the receptive field of a given neuron 111 is depicted by different colors (orange and blue). Note, that the choice of preferred phase depicted 112 here for the two conditions is arbitrarily selected.

113
To evaluate our hypothesis, we trained two rhesus monkeys to perform a visual attention task, 114 where they had to direct their attention towards either of the two spiral motion patterns (SMP) 115 presented on the screen. Monkeys N and W correctly performed the task successfully in 91% and 116 89% of those trials where they maintained their eye fixation. We recorded local field potentials 117 (LFP) and single unit activity from 90 motion-selective neurons in visual cortical area MST of the 118 monkeys while they performed the task. Depending on which of the two stimuli was cued and its 119 direction of motion relative to the recorded neuron's properties, trials came from one of three 120 different attention conditions; They either attended to the preferred stimulus inside the receptive 121 field (RF) (attend-in pref), preferred stimulus outside the RF (attend-out pref), or the anti-preferred 122 stimulus outside the RF (attend-out null) (note that in all conditions, the stimulus inside the RF is 123 the preferred stimulus) (Fig 2A). We compared the single unit responses between two of these 124 attentional conditions (attend-in (Pref) versus attend-out (Null)) which differ in the attended 125 stimulus' location and direction, while the stimulus inside the RF is kept the same (preferred).

126
Spike rate was significantly higher when attention was directed to the RF and the preferred 127 stimulus, compared to when attention was directed outside the RF and the null stimulus. This 128 confirmed that the animals did indeed selectively attend to the cued stimulus ( Fig 2B). All of the 129 following analyses focus on the spiking activity and LFP during the 500 ms interval starting from    We next asked if the magnitude of the phase shift is linked to the behavioral response of monkeys.

196
To this end, we examined the relationship between the attentional phase shift and the monkeys'  subnetwork's beta rhythm. We observed that, by increasing the synaptic delay, sensory neurons 218 start to shift their preferred spiking phase relative to modulatory beta oscillations (Fig 5A), 219 presumably due to a delayed effect of the modulatory subnetwork on the sensory neurons. These 220 results suggest that the neural mechanism whereby attention shifts the preferred phase of neurons, 221 may involve controlling the synaptic delay of inter-neuronal connections within a sensory area.  (Fig 6). Specifically, we observed that with the shift of spatial attention from the outside to the 253 inside of the RFs, spikes tend to fire at a later phase of the beta oscillations (Fig 3A). Given the  the 'poor' beta phase, less likely to reach downstream cortical areas. (Fig 6).
Our data further indicate that the attentional phase shift is linked to behavioral performance, i.e.,  Consistently, our results also indicate that MST neurons transmit the information of the attended 276 stimulus to higher cortices through oscillatory beta-aligned periods of enhanced coding for the 277 attended stimulus (Fig 6). Future studies need to examine the existence of a beta-aligned perceptual 278 sensitivity for motion recognition. shift. This is particularly important when asking if beta phase can serve as an internal reference 296 frame for distinguishing the attended stimulus from other stimuli. This could not be addressed in 297 the current study due to lack of enough data and remains a question for future studies.

314
Research with non-human primates represents a small but indispensable component of  All animal procedures of this study were approved by the responsible regional government office Surgeries were performed aseptically under gas anesthesia using standard techniques, including 343 appropriate peri-surgical analgesia and monitoring to minimize potential suffering. The two 344 animals were healthy at the conclusion of our study and were subsequently used in other studies. 345 We have established a comprehensive set of measures to ensure that the severity of our The two monkeys were trained to perform a visual attention task (Fig 2A)   Here, we recorded the activity of 40 MST neurons from monkey N and 50 single-units from 382 monkey W while they were performing the above-mentioned visual attention task.   each of them we determined a mean attentional phase shift within each frequency band (Fig 3). 436 We examined for the similarity of the phase shift across neurons, by applying a Rayleigh test for 437 non-uniformity of the phase shift distributions in each frequency band (gray filled areas in Fig 3   438 indicate frequency bands in which the phase shift data is sampled from a von Mises distribution).

439
Afterwards we examined each of the frequency bands to see if there was a phase shift distribution 440 with a mean unequal to zero by applying a paired Watson-Williams test (as a one-way ANOVA 441 test for circular data).

443
Each of the subnetworks used to construct intra-regional neuronal circuits (i.e., sensory and  gating variable is the same as that for excitatory synaptic inputs, however here we have 1 = 4 482 ms and 2 = 1 ms.

483
The system of differential equations was solved in MATLAB using the midpoint method with time 484 step ∆ = 0.01 ms. The system was initialized asynchronously [83], i.e., we chose the initial state 485 of neurons driven above their spiking threshold in a way that in the absence of synaptic currents, 486 it will take each neuron a time span of to fire its first spike, where is the intrinsic spiking 487 period and ∈ [0,1] is a uniformly-distributed random variable, and the remaining neurons (those 488 driven below their spiking threshold) are initialized close to their stable state (on their phase plane).

489
The system was solved for 500,000 steps (i.e., 5000 ms) and as a convention, spike times were 490 taken to be the moments at which the membrane potential crosses -20 mV from above. Here, we  To examine if there exists any systematic relationship between synaptic delay and the preferred 496 phase of the spiking activity, we considered the synaptic delay as a Gaussian-distributed random 497 variable with a variance of 0.05 ms. We then increased its mean in the range of 0 to 1.5 ms by 498 steps of 0.1 ms, carried out the simulation and finally calculated the preferred spiking phase of the pooled spike times of the sensory subnetwork relative to beta oscillations of the regulatory 500 neuronal population (Fig 5A). Beatrix Glaser for her administrative support.

508
The authors declare no competing interests.