Directionality of low and high frequency oscillations in feed-forward and feedback processing.

Directionality of low and high frequency oscillations in

feed-forward and feedback processing.

Auteur: Quirine Tordoir

Supervisor: Timo van Kerkoerle

Introduction

Cognitive functions like attention, working memory and consciousness depend on the integrated processing between multiple brain areas. It is unknown how different areas communicate with each other to give rise to these functions. A recent study by van Kerkoerle and colleagues (van Kerkoerle et al., 2011) showed that high frequency oscillatory activity (gamma) flows in the feed-forward direction while low frequency oscillatory activity (alpha) is propagated in the feedback direction. This was observed in the laminar profile in V1 and between V1 and V4 at unspecified depths. If alpha indeed shows a signature of feedback and gamma of feed-forward processing this could be exploited to gain a deeper understanding of the role of these processing modes in visual cognition.

The current study analysed a dataset with laminar probes in monkey V1 and V2 giving the opportunity to strengthen these findings. Using trough triggering analysis, a time delay was found for low and high frequencies between V1 and V2. Gamma oscillatory activity in V2 shifted in timing when it was triggered in V1 compared to V2, meaning that this oscillation propagates from V1 to V2 following feed-forward directions. The same was found for low frequency oscillatory activity but then in opposite direction, from V2 to V1 following feedback connections. These finding confirm the possible role for low and high frequency oscillatory activity as signature for feedback and feed-forward processing.

Theoretical background

There are many connections within the visual cortex (Felleman and Van Essen, 1991). What remains unclear is how all these lines of communication are used to give rise to cognitive functions such as attention and consciousness. There have been implications that oscillatory activity reflect different modes of communications between cortices (Siegel et al., 2012).

When a stimulus is perceived, frequency activity below 15 Hz (in particular in the alpha band 8-12 Hz) goes down while simultaneously inducing gamma oscillations (30-100 Hz), attention modulates this effect (Fries et al., 2001;Buffalo et al., 2011). Moreover, alpha appear to rise when a distracting stimulus is presented in the receptive field suggesting an active suppression of distracting information (Jensen and Mazaheri, 2010). Some preliminary results indicated that gamma and alpha oscillatory activity could reflect feed-forward and feedback processing respectively (von Stein et al., 2000). Irrespective of the functional role of alpha and gamma oscillatory rhythms they have been suggested to indicate feedback and feed-forward processing, respectively (Buffalo et al., 2011;Siegel et al., 2012).

Feed-forward connections account for the direct input information from low to high visual areas while feedback connections account for the information flow back from higher to lower levels to modulate neural activity (Lamme and Roelfsema, 2000;Gazzaley and Nobre, 2012).

On the synaptic level in cortical layers feed-forward connections target layer 4 (and layer 6 to a lesser degree), from here connecting to more superficial layers whereas feedback connections target layer 1 and 5, avoiding layer 4 (Lund, 1988;Rockland and Virga, 1989).

Laminar recordings of area V1 gave direct insight into the current flow across cortical layers (van Kerkoerle et al., 2011) showing frequency specific laminar profiles for alpha and gamma consistent with feedback and feed-forward connections. An analysis of functional connectivity between V1 and V4 also showed that the gamma rhythm propagates from V1 to V4 whereas the alpha rhythm propagates in the opposite direction. However, the cortical depth of these recordings could not be established.

In this paper we try to replicate earlier findings and support the claim that low frequency oscillations can serve as a signature for feedback processing and gamma oscillations for feed-forward processing. Earlier findings between V1 and V4 can be reproduced by recording in V1 and V2. V1 and V2 are close to each other, have strong connections and a straight forward relationship in terms of cortical hierarchy and laminar input profile (Rockland and Pandya, 1979; Markov et al., 2011). Finding a difference in timing for alpha and gamma between these areas would make the previous findings more robust. Moreover, by using laminar probes the cortical depth can now be investigated. We indeed find that in the middle layers, V1 leads V2 for gamma while V2 leads V1 for the low frequencies.

Methods

The experiment was designed and recorded by Mark Roberts and Peter de Weerd at Donders Institute for Brain, Cognition and Behaviour. For their experiment they used two rhesus macaque monkeys, monkey S and monkey K, who had to perform a fixation task using Gabor drifting gratings variable in contrast. The monkey fixated its gaze on central fixation spot (yellow annulus) which started the trial (Figure 1). The V1 and V2 receptive fields (RF) were mapped (dashed rectangles). Stimulus presentation was preceded by 1000ms baseline. Stimuli were circular patches of static square-wave gratings (2 cycles/degree) presented at the location of the V1 and V2 RFs. Stimulus presentation time was randomized between 750 and 4000ms. If the monkey’s eye position did not move from the fixation point during the time of the trial the monkey received a juice reward at the end of the trial and was free to start a new trial after a minimum period of 1000ms.

Figure 1: Task paradigm. Trial epochs are depicted as grey squares.

Data acquisition

Neuronal activity was measured in V1 and V2 using two multi-contact ‘U’ probes (Plexon, Inc.) with 8 channels spaced 200µm apart. Both monkeys were recorded for 12 days each. From all layers in V1 and V2 local field potential (LFP) and multi-unit activity (MUA) were recorded simultaneously.

Current source density

The current source density (CSD) is a measure of the currents flowing in and out of neurons. In the laminar recordings, CSD was calculated from the LFP as:

CSD x( ) (x h) 2 ( )₂x (x h)

h

φ φ φ

σ − − + +

= ⋅

The φ stands for the voltage, the x for the point at which the CSD is calculated, the h for the spacing of channels for the computation (here 200μm) and σ is the conductivity of cortical tissue with a value of 0.4 S.m-1 (Mitzdorf, 1985).

The probes used for the recording had 8 channels with a spacing of 200µm, 1400µm in total meaning it didn’t cover all layers of the visual cortex. The visual cortex has a thickness of about 1.7 mm (Lund, 1973), so we had to combine several recording at different depths to get the data for the whole laminar profile. To determine in which layers of the visual cortex the probes were placed we used a ‘flash evoked’ CSD, with this method it was possible to localize the current flows contributing to the generation of the field potentials (Schroeder et al., 1991).

Frequency analysis

To determine the frequency spectrum we performed a Fourier analysis. Because we are interested in the sustained activity and not the event related activity we subtracted the stimulus evoked potential from the raw signal leaving us with the induced potential of the task. To derive a time-frequency spectrum from the data we convolved the signal with a set of Morlet-wavelets and calculate the Fourier transform, to investigate the specific frequency bands of low and high frequency oscillatory activity.

Based on high power for both frequency bands, a time window was selected 1-2 seconds after stimulus onset. This window was then used to visualize the laminar profile for these frequencies. Here we encountered some channels that showed some irregularity. This irregularity was consistent over days and channels but changing between V1 and V2 signal. Several tests were performed on the amplifiers and the two probes used for the recordings. This resulted in the finding that one of the probes had two contact points (namely 5 and 7), that were less sensitive then the other contact points. Therefore they did record signal, although the power was much less than from the other contact points. To try and resolve this problem we used a 4th degree polynomial fitting to derive corrections coefficients. These coefficients were then used to correct the signal. This correction worked well for the last recordings but unfortunately resulted in an overcorrection for early recording days. This due to the fact that the insulation of the probe worsened after use and therefore the fit for the correction coefficient was best for the last recordings. To use this correction method successfully we would have needed “day specific correction” coefficients which unfortunately are impossible to measure afterward.

Because this correction was the best method that we tried but still not sufficient we decided to only use the laminar profile of the good probe, resulting in loss of the layer specific V1 data for monkey K and layer specific V2 data for monkey S.

Trough triggering analysis

Oscillatory activity is not directly linked to an external event. Unlike for example ERP there is no specific time point to use as a reference point to compare waves of activity. Therefore we used the troughs of the LFP as reference points, to be able to discriminate propagation of oscillatory activity throughout the layers of visual cortex (figure 2A). Troughs were selected from the filtered LFP signal in the low and high frequency band. The troughs where defined in one channel, namely the first channel above the CSD defined reversal between layer 4 and layer 5. The time points obtained by these troughs were then placed back to the original induced LFP signal (figure 2B) of all channels from 1 to 2 seconds after stimulus onset, cutting out window of LFP signal using the trough time points as the centre of the window (figure 2C&D). Window size was 1.5 cycle (3п) of the mean filtered frequency.

D

A

C

B

Filtered High Frequency

Laminar profile

Window per channel

LFP Signal with Troughs

Figure 2: A) Filtered high frequency signal with the troughs depicted in red dots. B) Trough time points placed back to the induced LFP signal. C) Windows of averaged LFP signal for each channel with troughs taken as centre of the 1.5 gamma cycle window. D) The same data as panel C but then represented as a surface plot with the different channels on the y axis.

Stimulus Evoked CSD

Results

CSD

Visual input enters the visual cortex in layer 4C, causing a strong source just after stimulus onset in this layer that can be made visible by CSD analysis. Figure 3 shows an example recording session, with red indicating a sink and blue indicating a source. The reversal between the sink at channel 2 and the source at channel 3 reflects the border between layer 4C and 5. Because there was no flash evoked potential recorded with the purpose of defining the reversal we used the onset of the grating with the highest contrast.

Figure 3: Stimulus Evoked CSD Frequency analysis

We did a frequency analysis using Morlet wavelets on the LFP data of V1 and V2. We found a peak in the low frequencies indicating an oscillatory rhythm with a peak centre frequency at around 4 Hz in both V1 and V2. This low frequency oscillatory activity for the high contrast stimulus followed the expected down going power after stimulus presentation compared to prestimulus baseline, similar to the pattern that is shown for alpha in previous studies (Fries et al., 2001; Buffalo et al., 2011). Surprisingly, the low contrast stimulus showed an enhancement of power compared to baseline. The high frequencies show a nice increase of oscillatory activity in the gamma range for the high contrast stimulus versus the prestimulus baseline. This gamma band is almost absent for the low contrast stimulus.

V1

V2

Low

High

A

C

B

D

According to these finding we determined the low frequency oscillatory band in range from 2 to 6 Hz and high frequency band in the range from 30 to 60 Hz as used for further analysis.

Figure 4: A-B) The low frequency power for condition 1 with a low contrast stimulus is shown in red, the high contrast in blue (calculated between 1-2 seconds after stimulus presentation) and prestimulus baseline in grey (calculated between 500ms before stimulus presentation) for A) V1 and B) V2. C-D) The high frequency power for C) V1 and D) V2.

Tr

igg

er

in

V

1

Tr

igg

er

in

V

2

V1: High

V1: Low

D

A

B

C

With the trough triggering analysis it is possible to visualize the propagation of oscillations throughout the layers of the visual cortex. We first looked at the activity within V1 in monkey S. When triggered within V1 the LFP during a low frequency cycle is in phase across the depth of the cortex (figure 5A), while the high frequency cycle shows a slight phase shift in both the deep and the superficial layers (figure 5B), confirming previous results from van Kerkoerle et al 2011. We can also look at the activity within V1 while triggering in V2. This shows a comparable profile (figure 5C & D) indicating that both low and high frequency oscillations are coherent between V1 and V2. When we look at the LFP in V2 for monkey K we saw a similar profile as in V1 for the low frequencies (figure 6A & C) but also the gamma cycle was in phase across the depth of the cortex (figure 6B & D). This shows that the laminar profile of gamma is specific to V1.

Figure 5: Laminar profile of the LFP in V1 for monkey S. A-B) Activity in V1 triggered in V1 and C-D) triggered in V2 for a cycle of A & C) low frequency activity and B & D) high frequency activity.

Tr

igg

er

in

V

1

Tr

igg

er

in

V

2

D

A

B

C

V2: High

V2: Low

Figure 6: Laminar profile of the LFP in V2 for monkey K. A-B) Activity in V2 triggered in V1 and C-D) triggered in V2 for a cycle of A & C) low frequency activity and B & D) high frequency activity.

We also analyse the laminar profile in the CSD. Unfortunately, the probes used for these recording had quite a large spacing between channels, causing a very low laminar resolution for CSD analysis (data not shown).

Time delay

We can now also look at the phase shift between area V1 and V2, investigating the flow of oscillatory activity in the feed-forward and feedback direction. We averaged the activity of layer 4, being channel 9 to 12. A time delay was found for both low and high frequency oscillatory activity in opposing direction. When looking at the high frequency activity you see that oscillations shift depending on the area in which the troughs were defined. The trough of the oscillation in V1 is nicely at zero when triggered within V1, the same is true for an oscillation in V2 when triggered in V2 (blues lines in figure 7). By looking at the oscillation within an area while choosing the trigger in another area you can investigate the time delay between those areas

V1

V2

Lo

w

H

igh

D

A

B

C

(red lines in figure 7). When looking at the gamma oscillation in V1 while triggering in V2 it shifts earlier in time (figure 7A) which means that gamma is earlier in V1 then in V2, in accordance with the direction of feed-forward connections. As you might expect, the shift should move in the opposite direction when we look at V2 triggered within itself versus in V1 (figure 7B). This indeed happens; the oscillation gets shifted later in time conforming the directionality of gamma in the feed-forward direction.

Figure 7: Time delay for both low and high frequency oscillatory activity in V1 of monkey S and V2 of monkey K. A) High frequency oscillations in V1 while being triggered within itself (blue) and in V2 (red). B) High frequency oscillations in V2 with trigger within itself (blue) and in V1 (red). C) Low frequency oscillations V1 triggered within itself (blue) and in V2 (red). D) Low frequency oscillations in V2 triggered within itself (blue) and in V1 (red).

Also for low frequency oscillations a time delay is found but then in the opposing direction, following feedback connections. Here the oscillatory activity moves later in time when

looking in V1 triggered in V2 (figure 7C), meaning that alpha arrives later in V1 compared to V2. Also findings in V2 confirm that alpha moves in feedback direction (figure 7D).

Discussion

First of all, we reproduced the laminar pattern of the LFP in V1 as shown by van Kerkoerle et al. 2011. In addition we showed that in V2 for the low frequencies the laminar profile of alpha is in phase across the depth of the cortex equal to findings in V1. This suggests that the layer in which you trigger is not crucial for the low frequencies. For the gamma frequencies, the phase only shifts between layers in V1 but remains in the same phase across layers in V2.This can be due to the fact that the layer division in V2 is slightly different from V1 namely that layer 4 covers a larger part. Also, the laminar divisions in V1 are much more differentiated than in V2 (Rockland and Virga, 1989).

Secondly, we investigate time delays between V1 and V2. We compared oscillatory activity that was aligned to the troughs in the low and high rhythms as seen in the LFP. Depending on trigger location, being in either V1 or V2, we were able to determine the

directionality of these oscillations indicating that low frequencies travel in the feedback direction from V2 to V1 while high frequencies travel in the feed-forward direction from V1 to V2. This confirms earlier results from van Kerkoerle et al. showing frequency specific phase lags between V1 and V4. We were now able to determine the layers in which we recorded. Therefore we could choose the activity in the same layer within each area. These results are in line with previous studies in cat visual cortex (Castelo-Branco et al., 1998;von Stein et al., 2000).

During the analysis we encountered many problems, of which not all were solvable. Especially the bad channel caused a lot of trouble resulting in a big loss of data. It took away the opportunity to look more precisely into the laminar profile of V1 and V2 connections and into the propagation of the low and high frequency oscillation throughout the different layers. Furthermore, the probes used for these recording had too large spacing between channels to show an accurate CSD profile.

A point of attention remains the defined ‘alpha’. A possible reason for it to be so low was the length of the trial which in this case was quite long. Previous studies have shown that the trial structure can entrain the low frequencies (Lakatos et al., 2008). This could entail that the trial length of the task has influence on the band of the low frequencies with longer trials causing a lower dominant frequency band. We also investigated the time delays for the classically defined alpha band (8-12Hz). We did not find a phase shift at this frequency, possible due to the low power in these frequencies.

Overall these results show a clear temporal relationship between V1 and V2 at specified depth, confirming previous results between V1 and V4 (van Kerkoerle et al. 2011). This shows the generality of this mechanism in monkey visual cortex. Our data therefore confirm a possible role for high and low frequency oscillatory activity as signatures for feed-forward and feedback processing respectively. These rhythms can now be used as fingerprints to disentangle the direction of communication between different cortical levels, aiding the study of cognitive functions when using EEG and MEG in humans (Siegel et al., 2012).

Reference List

Buffalo EA, Fries P, Landman R, Buschman TJ, Desimone R (2011) Laminar differences in gamma and alpha coherence in the ventral stream. Proc Natl Acad Sci USA 108:11262-11267. Castelo-Branco M, Neuenschwander S, Singer W (1998) Synchronization of visual responses between the cortex, lateral geniculate nucleus, and retina in the anesthetized cat. J Neurosci 18:6395-6410.

Felleman DJ, Van Essen DC (1991) Distributed hierarchical processing in the primate cerebral cortex. Cereb Cortex 1:1-47.

Fries P, Reynolds JH, Rorie AE, Desimone R (2001) Modulation of oscillatory neuronal synchronization by selective visual attention. Science 291:1560-1563.

Gazzaley A, Nobre AC (2012) Top-down modulation: bridging selective attention and working memory. Trends Cogn Sci 16:129-135.

Jensen O, Mazaheri A (2010) Shaping functional architecture by oscillatory alpha activity: gating by inhibition. Front Hum Neurosci 4:186.

Lakatos P, Karmos G, Mehta AD, Ulbert I, Schroeder CE (2008) Entrainment of neuronal oscillations as a mechanism of attentional selection. Science 320:110-113.

Lamme VAF, Roelfsema PR (2000) The distinct modes of vision offered by feedforward and recurrent processing. Trends Neurosci 23:571-579.

Lund JS (1973) Organization of neurons in the visual cortex, area 17, of the monkey (Macaca mulatta). J Comp Neurol 147:455-496.

Lund JS (1988) Anatomical organization of macaque monkey striate visual cortex. Annu Rev Neurosci 11:253-288.

Markov NT, Misery P, Falchier A, Lamy C, Vezoli J, Quilodran R, Gariel MA, Giroud P, Ercsey-Ravasz M, Pilaz LJ, Huissoud C, Barone P, Dehay C, Toroczkai Z, Van Essen DC, Kennedy H, Knoblauch K (2011) Weight consistency specifies regularities of macaque cortical networks. Cereb Cortex 21:1254-1272.

Mitzdorf U (1985) Current source-density method and application in cat cerebral cortex: investigation of evoked potentials and EEG phenomena. Physiol Rev 65:37-100.

Rockland KS, Pandya DN (1979) Laminar origins and terminations of cortical connections of the occipital lobe in the rhesus monkey. Brain Res 179:3-20.

Rockland KS, Virga A (1989) Terminal arbors of individual "feedback" axons projecting from area V2 to V1 in the macaque monkey: a study using immunohistochemistry of anterogradely transported Phaseolus vulgaris-leucoagglutinin. J Comp Neurol 285:54-72.

Schroeder CE, Tenke CE, Givre SJ, Arezzo JC, Vaughan HG, Jr. (1991) Striate cortical contribution to the surface-recorded pattern-reversal VEP in the alert monkey. Vision Res 31:1143-1157.

Siegel M, Donner TH, Engel AK (2012) Spectral fingerprints of large-scale neuronal interactions. Nat Rev Neurosci 13:121-134.

van Kerkoerle TJ, Self M, Poort J, van der Togt C, Roelfsema PR (2011) High frequencies flow in the feed-forward direction through the different layers of monkey primary visual cortex while low frequencies flow in the recurrent direction. pp no 270.08.

von Stein A, Chiang C, Konig P (2000) Top-down processing mediated by interareal synchronization. Proc Natl Acad Sci U S A 97:14748-14753.