The multiple layers of the neocortex show an elaborate connectivity pattern, which predicts a specific directionality of interlaminar communication (Callaway, 1998; Douglas and Martin, 2004; Douglas et al., 1989). Feedforward inputs predominantly terminate in layer 4 of sensory cortical areas, before being passed to layers 2/3 and then onwards to layers 5/6, where recurrent inputs to layer 2/3 arise (Callaway, 1998; Callaway, 2004). Conversely, feedback connections from downstream areas predominantly terminate in layer 1 and 5 (Rockland and Pandya, 1979), from where the information is passed to layers 2/3. Local field potential (LFP) recordings, using different stimulus and task conditions, have provided physiological support for these anatomical pathways. Current source density analysis of the LFP, as well as latency of spiking activity has demonstrated that the earliest activity elicited by stimuli centered on the classical receptive field occurs in layer 4c (Bijanzadeh et al., 2018). Stimuli restricted to the far receptive field surround, which are hypothesized to affect processing in V1 primarily through feedback projections (Bair et al., 2003), initially activate V1 in upper superficial layers and in layer 5 (Bijanzadeh et al., 2018). In addition to this anatomical segregation of feedforward and feedback signal input to V1, it has been proposed that feedforward and feedback signals have separable spectral signatures of the LFP. Feedback processing has been aligned with low frequency (alpha [α] band) oscillatory activity, dominating activity in infragranular layers (Buffalo et al., 2011; Lopes Da Silva and Storm Van Leeuwen, 1977; van Kerkoerle et al., 2014), and with beta (β) band oscillations (Bastos et al., 2015a; Xing et al., 2012). Conversely, feedforward processing has been associated with low frequency theta [θ], and higher frequency oscillatory activity (gamma [γ] band), originating in layer 4, dominating in (or even being largely restricted to) supragranular layers (Bastos et al., 2015a; Bollimunta et al., 2011; Buffalo et al., 2004; Buffalo et al., 2011; Maier et al., 2010; Spaak et al., 2012; van Kerkoerle et al., 2014; Xing et al., 2012). However, the generality of this scenario has recently been questioned (Haegens et al., 2015), whereby the strongest low-frequency (α) oscillations generators were found in supragranular layers of sensory cortex. Moreover, while most studies failed to find prominent γ oscillations outside supragranular layers (Buffalo et al., 2011; Maier et al., 2010; Xing et al., 2012), exceptions do exist (Maier et al., 2011).

To address these controversies and investigate the origin and spectral signature of feedforward and feedback signals in detail, we recorded across cortical layers of V1 of awake macaque monkeys, while manipulating stimulus size systematically. Increasing the stimulus size invokes center surround interactions, which result in specific input patterns across a cortical column (Bijanzadeh et al., 2018), and which strongly alter the spectral signatures LFPs. Increasing the size of a patch of an orientated grating centered on a receptive field (RF) increases γ power in conjunction with a general decrease of neurons firing rate (Bauer et al., 1995; Gieselmann and Thiele, 2008; Jia et al., 2011; Ray and Maunsell, 2010). If this increased γ power was a signature of a feedforward signal, then we would expect it to arise in either layer 4, or layer 2/3 of V1. Conversely, the feedback signals should show spectral signatures in lower frequency bands, such as α (van Kerkoerle et al., 2014), which should arise and dominate in layers 1 and 5 (Bastos et al., 2015a; van Kerkoerle et al., 2014).

While our results support some of the features outlined above, they violate them in multiple aspects, and show that feedforward and feedback signals attain stimulus dependent oscillatory frequencies across cortical layers on V1.

We recorded neuronal activity from area V1 from 16 contact laminar electrodes in two passively viewing male macaque monkeys. Electrodes were inserted perpendicular to the cortical surface, aiming to record from within a single V1 cortical column. To ensure consistency of our analysis we rejected experiments (total n = 25/9, monkey 1/2) from further analysis if the current-source-density (CSD, see Materials and methods) profiles did not meet any of the following three conditions: (1) An identifiable current source as early as 50ms after the stimulus onset (n = 1/1, monkey 1/2). (2) The ‘early’ current source is identified on a channel between the seventh and twelfth contact (from the top of the electrode array, to ensure a minimum coverage of cortical layers 2–6; n = 3/1, monkey 1/2). (3) A pattern of the CSD-profile that is consistent with the majority of experiments (n = 1/1, monkey 4/0). After applying the rejection criteria, we were left with 24 experiments (monkey 1: n = 17, monkey 2: n = 7) from three hemispheres which were included into this study.

Figure 1 shows the alignment of neuronal data from an example experiment. The data were obtained during receptive field (RF) mapping (Figure 1B) and during presentation of gratings (Figure 1C and D). In this recording the eighth channel from the top was defined as the alignment channel indicating putative layer 4c at zero depth (Figure 1A). Correspondingly, the RF maps of the envelope of multi-unit-activity (MUAe, see Materials and methods) showed the early emergence of a visual RF in the same channel between 30 and 50ms (Figure 1B left), demonstrating that the early current sink coincides with the earliest activation from thalamo-cortical afferents. This electrode placement resulted in contacts in supragranular layers (SG), alignment layer 4c, and infragranular layers (IG) (Figure 1A). The gray matter region (from third channel from top to sixths from the bottom in Figure 1A) is flanked by two large current sources at the top and at the bottom (Figure 1D). The perpendicularity of probe orientation relative to the cortical surface is indicated by the good spatial alignment of RF-centers (Figure 1B). The LFP responses in the gray matter region show strong evoked potentials, and sustained oscillations (Figure 1C). The lower five channels in Figure 1 were located in the white matter, and possibly in area V2, as evident by a sudden shift in visual RF’s (Figure 1B, lowest contacts). While the strength of the LFP-signal seems to be unchanged (Figure 1C) for these contacts, current sources and sinks were weaker and less structured than in the gray matter. A detailed analysis of RF sizes for different layers is presented in Figure 1—figure supplement 1.

Figure 1 with 1 supplement see all

Download asset Open asset

We next focus our analysis on basic phenomena of spiking activity and current source density changes with different sized stimuli, as these provide insight into the underlying pathways that mediate changes (feedforward, lateral intra-areal, intracolumnar feedback and feedback from downstream areas), as well as providing a benchmark against previously published results.

Response latency as a function of cortical depth and stimulus size

We calculated spiking latencies by using the MUAe signal using a curve-fitting procedure (Figure 2A, see Materials and methods and Self et al., 2013). The shortest latencies occurred in granular layers, followed by infragranular layers, with supragranular layers generally lagging behind (Figure 2B). This overall pattern was independent of stimulus size used. However, stimulus size significantly altered response latencies in granular (p < 0.01, one-factor-ANOVA, Figure 2C) and in infragranular layers (p < 0.001, one-factor-ANOVA, Figure 2C). In granular layers larger stimuli resulted in up to ~3ms longer latencies, while in infragranular layers they caused up to ~6ms shorter latencies. Thus, when large stimuli were presented the lag between granular and infragranular layers was reduced to ~2ms (from ~7ms for small stimuli, Figure 2C). No significant changes were found for supragranular layers (p = 0.98, one-factor-ANOVA, Figure 2C). Latency changes also appeared to occur in the top-most layer, although this was not significant, possibly because of relatively small sample sizes (not every recording included that layer/contact due to electrode placement).

Figure 2 with 2 supplements see all

Download asset Open asset

The CSD profile equally changed with stimulus size. Figure 2D and E show the average CSD profile across all experiments for a small (0.5°) and a large (15.0°) stimulus. The CSD-profile of each experiment was normalized to the peak of the early onset current sink in the alignment layer (0 µm). Figure 2E shows a surface plot of time depth CSD differences plot (difference between CSD profiles in C and D), where only significant differences are displayed (p < 0.05, cluster-mass-test). In granular and supragranular layers we observed moderate decreases in inward current amplitudes with larger stimuli, which occurred as early as 50ms in granular layers. Most of the changes in and above layer 0 can be described as a general attenuation of current amplitudes. In other words, the overall laminar pattern of current sinks and sources in granular and supragranular layers was similar in the CSD profile for small and large stimuli, but it was overall slightly attenuated for large stimuli, probably due to surround suppression effects at the level of the lateral-geniculate-nucleus (LGN). The currents (sinks) in infragranular layers were also mostly reduced, but in addition changes in the pattern of sinks and sources occurred. Specifically, a current sink emerged ~75ms after stimulus onset at –600 µm (Figure 2D and E), where a source was present when small stimuli were presented. This indicates that stimulating the surround of a V1 RF results in additional input to infragranular layers. Given that RF sizes differed somewhat between layers (Figure 1—figure supplement 1) it could be argued that the analysis should be repeated for different reference sizes (other than e.g. 0.5°). Doing so yielded qualitatively identical results (Figure 2—figure supplement 1; Figure 2—figure supplement 2).

We next focused on surround suppression/enhancement effects on MUAe, which are most profound during later parts of the response. Here we analyzed the activity from 200 to 500ms after stimulus onset. Surround suppression affected all cortical layers. However, it was more pronounced in supragranular and granular than in infragranular layers (Figure 3A and B). Moreover, the stimulus size that resulted in maximal activity equally varied between layers, with slightly larger stimuli (0.75°, 1.5°) eliciting maximal activity in infragranular layers, while the smallest stimulus (0.5°) elicited maximal activity in supragranular layers (Figure 3C).

Figure 3

Download asset Open asset

Together the results from Figures 2 and 3 are compatible with recent reports suggesting that stimuli extending into the far surround of neurons invoke center surround interactions through feedback connections from higher cortical areas that terminate predominantly in layers 1 and 5 (Bijanzadeh et al., 2018; Nassi et al., 2013; Nurminen et al., 2018; van Kerkoerle et al., 2014).

LFP spectral power and phase relationship analysis

We now turn to the question how feedback signals impose their effects on local cortical processing in area V1. It has been argued that feedback signals can be characterized in terms of low frequency (α, β) oscillatory activity, while feedforward signals can be characterized in terms of θ- and γ-frequency oscillations (Bastos et al., 2015a; Bosman et al., 2012; Buschman and Miller, 2009; Fries, 2015). A simple prediction would thus be that all stimulus sizes elicit γ frequency oscillations in layers 2/3, as all of the stimuli will invoke feedforward drive. Within this context, two scenarios are possible. Firstly, γ frequency oscillations in L2/3 are slightly reduced when larger stimuli are presented, due to the effects of surround suppression that are inherited from the LGN. Alternatively, their spectral power increases as stimulus size increases, because large stimuli invoke stronger inhibitory surround interactions, which through pyramidal-inhibitory interactions drive γ frequency oscillations (Tiesinga and Sejnowski, 2009). Additionally, infragranular layers should show limited γ frequency power (Dougherty et al., 2017; Maier et al., 2011; Spaak et al., 2012; Xing et al., 2012), but they should show increased low frequency (α/β) oscillations with increasing stimulus size as the level of feedback input increases (Bastos et al., 2015a; Bosman et al., 2012; Buffalo et al., 2011; van Kerkoerle et al., 2014).

To answer these questions, we transformed the raw LFP signal into the bipolar referenced LFP (LFPbp, see Materials and methods). This was necessary, as the raw LFP is not a sufficiently local signal to allow for meaningful layer specific analysis (Trongnetrpunya et al., 2016). The single trial LFPbp of the sustained part of the response (200–500ms after stimulus onset) was then subjected to standard spectral analysis (see Materials and methods). We focused on the sustained part of the neuronal response, as endogenous γ oscillations are most pronounced during this period (Gieselmann and Thiele, 2008). The results are shown in Figure 4. It shows a CSD profile (Figure 4A) and the layer dependent induced spectral power (iSP, z-score relative to spontaneous activity 300ms before stimulus onset) as a function of stimulus size for an example experiment (Figure 4B, for cross-cuts at selected depths see Figure 4—figure supplement 3). When small stimuli were presented, the supragranular layers showed relatively strong iSP across multiple γ frequencies (20–100 Hz). In addition, there was weaker, but nevertheless pronounced broad band iSP (20–100 Hz) in infragranular layers. As stimulus size increased the iSP across all layers became more focused with clear peaks emerging in the low γ frequency range (~40 Hz). This pattern became even more pronounced when analyzing iSP across different laminae averaged across all experiments (Figure 4C–E, Figure 4—figure supplement 3). Small stimuli (0.5°) elicit broad band power changes ranging from 30 to 100 Hz, with increasing iSP toward higher frequencies in supragranular layers (Figure 4D). Infragranular layers showed relatively small, but~ homogenous broad band β/γ iSP in the range from 20 to 80 Hz (Figure 4D). When large stimuli were shown, iSP in the supragranular layers concentrated in more narrow band regions of ~40 Hz. A smaller peak in that frequency range also occurred in granular layers. A surprisingly large γ frequency iSP increase occurred in infragranular layers 5 and 6, at ~40–50 Hz (compare to Xing et al., 2012). Also, the peak frequency of the evoked gamma oscillation at around 40 Hz showed the same inverse relationship to stimulus size (Figure 4—figure supplement 3 and Gieselmann and Thiele, 2008), and is in line with other previous publications using grating stimuli (Bauer et al., 1995; Gray et al., 1989; Jia et al., 2011). Calculating power changes relative to different reference stimuli resulted in similar outcomes.

Figure 4 with 3 supplements see all

Download asset Open asset

We quantified the effects of stimulus size on spectral power by calculating the mean iSP of the LFPbp-signal in discrete frequency bands. This was done for the θ (4–8 Hz, Figure 5A–C), α (8–13 Hz, Figure 5D–F), β (14–25 Hz, Figure 5G–I) and low γ (35–55 Hz, Figure 5J–L) range. Figure 5A, D, G and J show the average depth profiles of band power in response to a small (0.5°, blue) and a large (15°, red) stimulus. For all frequency bands, stimulus size modulated the band, but these modulations differed between layers and frequency bands. For the θ-band, α-band, and β-band, stimulus-size-induced changes are most pronounced for granular and lower supragranular layers (Figure 5C, F 1). For infragranular layers, larger stimuli caused a reduction in θ-band (Figure 5B), no change in α-band or β-band, and strong increases in γ-band. Overall, larger stimuli caused significantly more γ-band across all contacts (Figure 5J), but relative changes are strongest in supragranular and infragranular layers (Figure 5L).

Figure 5

Download asset Open asset

Gamma-connectivity across cortical laminae

Previous studies have argued that phase-locked oscillatory activity can provide insight into the directionality of information flow for a given frequency band. Specifically, it was shown that α-oscillations locked to the α-phase in granular layers appeared to originate in infragranular and upper supragranular layers, progressing toward middle supragranular layers (van Kerkoerle et al., 2014). Conversely, γ-oscillations locked to the γ -phase in granular layers originated in granular or lower supragranular layer progressing toward upper supragranular and infragranular layers respectively. To analyze this for different stimulus sizes, we computed the phase-triggered average (PTA) to estimate the phase-locked oscillations between electrode contacts for different frequency bands. Phase-triggered-averaging describes the alignment of LFPbp signals of a given electrode contact to the phase of an oscillation at a defined frequency at the reference-channel. The aligned LFPbp is then averaged across the number of alignments. Temporal structure in the resulting histogram can be interpreted as a phase-coupled oscillation between the two signals. Figure 6 shows the depth profiles of the four main frequency band PTA-histograms for a reference channel at –150 µm, averaged across all experiments. As expected, the depth used as the reference phase of the oscillatory cycle always showed strong phase coupling, with close to zero phase delays (Figure 6, gray triangles indicate the reference channel). Many other contacts showed coupling to the reference channel for all frequency bands. However, the exact phase coupling varied between frequency bands and between stimulus sizes. Low frequency (θ, α) showed a phase progression from layer 5 to layer 6, and then to granular and supragranular layers for small stimuli. This pattern changed radically for larger stimuli, where phase progression was seen bidirectionally from granular to supragranular and to infragranular layer 5, but also from infragranular layer 6 to layer 5. For γ frequencies, layer 4 and 6 were largely in phase for small stimuli, but almost 90° out of phase with layer 5. Some continuous phase progression could be seen from granular toward upper supragranular layers. However, for large stimuli phase progression occurred from layer 5 to layer 6, onwards to granular layers, and then to supragranular layers. These patterns were not artefacts of selecting a specific reference channel (Figure 6—figure supplements 1–4). These patterns were less pronounced when using the iCSD signal instead of the bpLFP signal but the globally described pattern largely prevailed (Figure 6—figure supplements 1–4).

Figure 6 with 5 supplements see all

Download asset Open asset

Isolated phase relationships per se do not yield insight into causal interactions, as phase lags can equally be interpreted as phase advances due to the periodicity of the signal. Moreover, causal interaction assignment based on phase relationships can be inconsistent with Granger causality analysis (Brovelli et al., 2004). To gain insight into causality, we performed spectral Granger-causality (GC) analysis (Figure 7—figure supplement 1). We then integrated the GC-Index in the four different frequency bands analyzed above (θ, α, β, and γ frequency) and generated connectivity plots using the net GC (difference between directions) for each connection (Figure 7). Hence, Figure 7 shows the direction and associated strength of granger-causal influence of the four different frequency LFP bands between different layers. We only plotted those connections that passed the reverse-granger-test (RGT) to protect against spurious results induced by linear mixed noise (Haufe et al., 2013; Vinck et al., 2015). In Figure 7—figure supplements 2 and 3 we show the same data including the connections which failed RGT and the results of the time reversed GC analysis, respectively. Figure 7 shows the analysis of non-conditional GC. Conditional GC produced qualitatively equivalent results, albeit with overall smaller granger causality indices. A striking feature of the directionality of all GCs (irrespective of the frequency bands analyzed) was its dominance in the 'upward' direction, namely toward supragranular layers. This indicates that communication was directed toward the cortico-cortical feedforward channels in V1 (for related results see Ferro et al., 2021). Sizeable GCs in the downward direction were usually restricted to granular to infragranular interactions, and to layer 5 to layer 6 interactions. Downward GCs were generally more pronounced with larger stimuli.

Figure 7 with 4 supplements see all

Download asset Open asset

In the upward direction strong θ-GCs originated in layer 6 (lower contacts) affecting predominantly supragranular layers, irrespective of stimulus size. With increasing stimulus size GCs increased in an upward direction between granular and supragranular, and between supragranular-supragranular contacts. Overall, a very similar pattern was present for GCs in the α- frequency band. In the β- frequency band, GCs were fairly weak in the downward direction, when small stimuli were presented, but increased between layer 5 and layer 6 for large stimuli. Upward GCs were pronounced between granular and supragranular, and between supra-supragranular contacts for all stimulus sizes. With increase in stimulus size a pronounced infragranular-supragranular GC occurred. An overall similar pattern was present for γ-frequency GCs, with an even stronger increase in upwards directed supra-supragranular GCs with stimulus size. The results from Figure 7 show that surround modulation results in strongly increased low, mid and high frequency band communication between granular and supragranular compartments, and it results in strongly increased γ-band communication originating from layer 5, which drives γ frequency changes in layer 6, and which in turn changes γ frequency oscillations in upper layer four and in layers 2/3. Performing a similar analysis using MUAe signals instead of LFP signals results in overall similar pattern, with an additional component of γ frequency GC originating in layer 4 (Figure 7—figure supplement 4).

Using stimuli of different sizes in combination with laminar recordings in primary visual cortex of the macaque revealed that narrow band γ frequency oscillations occur across cortical layers and increase across cortical layers with increasing stimulus size. When stimuli encroached on the far surround of the receptive field, the LFP γ oscillations in layer 5, the feedback recipient layer, exerted pronounced granger causal influences on γ oscillations in layer 6, which in turn granger causal influenced γ oscillations in supragranular layers. This pattern of γ frequency granger causal interactions is compatible with the known anatomical termination sites of feedback connections into V1 (Rockland and Pandya, 1979). The flow of information (granger causal interactions in the γ frequency range) occurred in a manner compatible with the canonical microcircuit, but in a largely reverse direction. Here, information terminating in layer 5, is passed on to layers 6 and 4, who in turn send the information to layers 2/3. This demonstrates that γ oscillations are neither restricted to superficial layers, nor are they invariably signatures of feedforward communication. Moreover, low-frequency oscillation GC interactions did not strictly adhere to their proposed role in feedback communication, showing most profound increases in granular to supragranular and supra- to supragranular interactions with stimulus size. Feedback influence terminating in layer five was not dominated by low-frequency oscillations as predicted by recent publications (Bastos et al., 2015a; Bastos et al., 2015b; Bosman et al., 2012; Buffalo et al., 2011; Buschman and Miller; van Kerkoerle et al., 2014), even though GC interactions in low-frequency bands in the intracolumnar feedback direction (infragranular to granular/supragranular layers) were profound (Figure 7—figure supplement 1). This demonstrates that low-frequency oscillations are not restricted (or dominant) to infragranular layer and can attain feedforward roles in V1. As described in Materials and methods, the stimuli were optimized in relation to orientation tuning in one of the monkeys, but not the other. However, the key results reported in the paper did not differ between animals, which suggests that stimulus optimization is not critical for these results to be obtained.

The differences to previous publications could be due to the use of cognitively demanding task engagement (Bastos et al., 2015a; Bastos et al., 2015b; Bosman et al., 2012; Buffalo et al., 2011; Buschman and Miller, 2007; van Kerkoerle et al., 2014) vs. passive fixation as used in our task. Additionally, the use of large grating stimuli might be a contributing factor, as our phase lag and GC results using small stimuli are more similar to those reported in other publications (van Kerkoerle et al., 2014), even though their stimuli are not directly comparable to our small grating stimuli.

Two awake male rhesus monkeys were used for the electrophysiological recordings reported in this study. After initial training, monkeys were implanted with a head holder and recording chambers above V1 under general anesthesia and sterile conditions (for details of surgical procedures, postsurgical analgesics, and general postsurgical treatment, see Thiele et al., 2006). All procedures complied with the European Communities Council Directive RL 2010/63/EC, the U.S. National Institutes of Health Guidelines for the Care and Use of Animals for Experimental Procedures, and the UK Animals Scientific Procedures Act. Animals were motivated to engage in the task through fluid control at levels that do not affect animal physiology and have minimal impact on psychological wellbeing (Gray et al., 2016).

Original source data from all relevant data sets are available at DOI: https://doi.org/10.25405/data.ncl.17136182.v1 in the data depository of Newcastle University (https://data.ncl.ac.uk/ managed by figshare). Data are in binary format (Neuralynx) and are in the form of original recordings, along with relevant event files.

1. 1. Gieselmann M
   2. Thiele A

   (2022) Newcastle University

   Stimulus dependence of directed information exchange between cortical layers in macaque V1.

   https://doi.org/10.25405/data.ncl.17136182.v1

1. 1. Alitto HJ
   2. Usrey WM

   (2008) Origin and dynamics of extraclassical suppression in the lateral geniculate nucleus of the macaque monkey

   Neuron 57:135–146.

   https://doi.org/10.1016/j.neuron.2007.11.019

   • PubMed
   • Google Scholar
2. 1. Angelucci A
   2. Levitt JB
   3. Lund JS

   (2002) Anatomical origins of the classical receptive field and modulatory surround field of single neurons in macaque visual cortical area V1

   Progress in Brain Research 136:373–388.

   https://doi.org/10.1016/s0079-6123(02)36031-x

   • PubMed
   • Google Scholar
3. 1. Bair W
   2. Cavanaugh JR
   3. Smith MA
   4. Movshon JA

   (2002)

   The timing of response onset and offset in macaque visual neurons

   The Journal of Neuroscience 22:3189–3205.

   • PubMed
   • Google Scholar
4. 1. Bair W
   2. Cavanaugh JR
   3. Movshon JA

   (2003) Time course and time-distance relationships for surround suppression in macaque V1 neurons

   The Journal of Neuroscience 23:7690–7701.

   https://doi.org/10.1523/JNEUROSCI.23-20-07690.2003

   • PubMed
   • Google Scholar
5. 1. Barnett L
   2. Seth AK

   (2014) The MVGC multivariate Granger causality toolbox: a new approach to Granger-causal inference

   Journal of Neuroscience Methods 223:50–68.

   https://doi.org/10.1016/j.jneumeth.2013.10.018

   • PubMed
   • Google Scholar
6. 1. Barone P
   2. Batardiere A
   3. Knoblauch K
   4. Kennedy H

   (2000) Laminar distribution of neurons in extrastriate areas projecting to visual areas V1 and V4 correlates with the hierarchical rank and indicates the operation of a distance rule

   The Journal of Neuroscience 20:3263–3281.

   https://doi.org/10.1523/JNEUROSCI.20-09-03263.2000

   • PubMed
   • Google Scholar
7. 1. Bastos AM
   2. Vezoli J
   3. Bosman CA
   4. Schoffelen JM
   5. Oostenveld R
   6. Dowdall JR
   7. De Weerd P
   8. Kennedy H
   9. Fries P

   (2015a) Visual areas exert feedforward and feedback influences through distinct frequency channels

   Neuron 85:390–401.

   https://doi.org/10.1016/j.neuron.2014.12.018

   • PubMed
   • Google Scholar
8. 1. Bastos AM
   2. Vezoli J
   3. Fries P

   (2015b) Communication through coherence with inter-areal delays

   Current Opinion in Neurobiology 31:173–180.

   https://doi.org/10.1016/j.conb.2014.11.001

   • PubMed
   • Google Scholar
9. 1. Bauer R
   2. Brosch M
   3. Eckhorn R

   (1995) Different rules of spatial summation from beyond the receptive field for spike rates and oscillation amplitudes in cat visual cortex

   Brain Research 669:291–297.

   https://doi.org/10.1016/0006-8993(94)01273-k

   • PubMed
   • Google Scholar
10. 1. Bijanzadeh M
    2. Nurminen L
    3. Merlin S
    4. Clark AM
    5. Angelucci A

    (2018) Distinct Laminar Processing of Local and Global Context in Primate Primary Visual Cortex

    Neuron 100:259–274.

    https://doi.org/10.1016/j.neuron.2018.08.020

    • PubMed
    • Google Scholar
11. 1. Bollimunta A
    2. Mo J
    3. Schroeder CE
    4. Ding M

    (2011) Neuronal mechanisms and attentional modulation of corticothalamic α oscillations

    The Journal of Neuroscience 31:4935–4943.

    https://doi.org/10.1523/JNEUROSCI.5580-10.2011

    • PubMed
    • Google Scholar
12. 1. Bosman CA
    2. Schoffelen JM
    3. Brunet N
    4. Oostenveld R
    5. Bastos AM
    6. Womelsdorf T
    7. Rubehn B
    8. Stieglitz T
    9. De Weerd P
    10. Fries P

    (2012) Attentional stimulus selection through selective synchronization between monkey visual areas

    Neuron 75:875–888.

    https://doi.org/10.1016/j.neuron.2012.06.037

    • PubMed
    • Google Scholar
13. 1. Brovelli A
    2. Ding M
    3. Ledberg A
    4. Chen Y
    5. Nakamura R
    6. Bressler SL

    (2004) Beta oscillations in a large-scale sensorimotor cortical network: directional influences revealed by Granger causality

    PNAS 101:9849–9854.

    https://doi.org/10.1073/pnas.0308538101

    • PubMed
    • Google Scholar
14. 1. Buffalo EA
    2. Fries P
    3. Desimone R

    (2004)

    Layer-specific attentional modulation in early visual areas

    Society for Neuroscience, Abstracts 35:717.

    • Google Scholar
15. 1. Buffalo EA
    2. Fries P
    3. Landman R
    4. Buschman TJ
    5. Desimone R

    (2011) Laminar differences in gamma and alpha coherence in the ventral stream

    PNAS 108:11262–11267.

    https://doi.org/10.1073/pnas.1011284108

    • PubMed
    • Google Scholar
16. 1. Bullmore ET
    2. Suckling J
    3. Overmeyer S
    4. Rabe-Hesketh S
    5. Taylor E
    6. Brammer MJ

    (1999) Global, voxel, and cluster tests, by theory and permutation, for a difference between two groups of structural MR images of the brain

    IEEE Transactions on Medical Imaging 18:32–42.

    https://doi.org/10.1109/42.750253

    • PubMed
    • Google Scholar
17. 1. Buschman TJ
    2. Miller EK

    (2007) Top-down versus bottom-up control of attention in the prefrontal and posterior parietal cortices

    Science (New York, N.Y.) 315:1860–1862.

    https://doi.org/10.1126/science.1138071

    • PubMed
    • Google Scholar
18. 1. Buschman TJ
    2. Miller EK

    (2009) Serial, covert shifts of attention during visual search are reflected by the frontal eye fields and correlated with population oscillations

    Neuron 63:386–396.

    https://doi.org/10.1016/j.neuron.2009.06.020

    • PubMed
    • Google Scholar
19. 1. Callaway EM

    (1998) Local circuits in primary visual cortex of the macaque monkey

    Annual Review of Neuroscience 21:47–74.

    https://doi.org/10.1146/annurev.neuro.21.1.47

    • PubMed
    • Google Scholar
20. 1. Callaway EM

    (2004) Feedforward, feedback and inhibitory connections in primate visual cortex

    Neural Networks 17:625–632.

    https://doi.org/10.1016/j.neunet.2004.04.004

    • PubMed
    • Google Scholar
21. 1. Dougherty K
    2. Cox MA
    3. Ninomiya T
    4. Leopold DA
    5. Maier A

    (2017) Ongoing Alpha Activity in V1 Regulates Visually Driven Spiking Responses

    Cerebral Cortex (New York, N.Y 27:1113–1124.

    https://doi.org/10.1093/cercor/bhv304

    • PubMed
    • Google Scholar
22. 1. Douglas R.J.
    2. Martin KAC
    3. Whitteridge D

    (1989) A Canonical Microcircuit for Neocortex

    Neural Computation 1:480–488.

    https://doi.org/10.1162/neco.1989.1.4.480

    • Google Scholar
23. 1. Douglas R
    2. Martin KAC

    (2004) Neuronal circuits of the neocortex

    Annual Review of Neuroscience 27:419–451.

    https://doi.org/10.1146/annurev.neuro.27.070203.144152

    • PubMed
    • Google Scholar
24. 1. Ferro D
    2. van Kempen J
    3. Boyd M
    4. Panzeri S
    5. Thiele A

    (2021) Directed information exchange between cortical layers in macaque V1 and V4 and its modulation by selective attention

    PNAS 118:2022097118.

    https://doi.org/10.1073/pnas.2022097118

    • PubMed
    • Google Scholar
25. 1. Freeman JA
    2. Nicholson C

    (1975) Experimental optimization of current source-density technique for anuran cerebellum

    Journal of Neurophysiology 38:369–382.

    https://doi.org/10.1152/jn.1975.38.2.369

    • PubMed
    • Google Scholar
26. 1. Frien A
    2. Eckhorn R
    3. Bauer R
    4. Woelbern T
    5. Kehr H

    (1994) Stimulus-specific fast oscillations at zero phase between visual areas V1 and V2 of awake monkey

    Neuroreport 5:2273–2277.

    https://doi.org/10.1097/00001756-199411000-00017

    • PubMed
    • Google Scholar
27. 1. Fries P

    (2015) Rhythms for Cognition: Communication through Coherence

    Neuron 88:220–235.

    https://doi.org/10.1016/j.neuron.2015.09.034

    • PubMed
    • Google Scholar
28. 1. Gieselmann MA
    2. Thiele A

    (2008) Comparison of spatial integration and surround suppression characteristics in spiking activity and the local field potential in macaque V1

    The European Journal of Neuroscience 28:447–459.

    https://doi.org/10.1111/j.1460-9568.2008.06358.x

    • PubMed
    • Google Scholar
29. 1. Gray CM
    2. König P
    3. Engel AK
    4. Singer W

    (1989) Oscillatory responses in cat visual cortex exhibit inter-columnar synchronization which reflects global stimulus properties

    Nature 338:334–337.

    https://doi.org/10.1038/338334a0

    • PubMed
    • Google Scholar
30. 1. Gray H
    2. Bertrand H
    3. Mindus C
    4. Flecknell P
    5. Rowe C
    6. Thiele A

    (2016) Physiological, Behavioral, and Scientific Impact of Different Fluid Control Protocols in the Rhesus Macaque (Macaca mulatta)

    ENeuro 3:eNeuro2016.

    https://doi.org/10.1523/ENEURO.0195-16.2016

    • PubMed
    • Google Scholar
31. 1. Haegens S
    2. Barczak A
    3. Musacchia G
    4. Lipton ML
    5. Mehta AD
    6. Lakatos P
    7. Schroeder CE

    (2015) Laminar Profile and Physiology of the alpha Rhythm in Primary Visual

    Auditory, and Somatosensory Regions of Neocortex. J Neurosci 35:14341–14352.

    https://doi.org/10.1523/JNEUROSCI.0600-15.2015

    • Google Scholar
32. 1. Hartmann TS
    2. Raja S
    3. Lomber SG
    4. Born RT

    (2019) Cortico-Cortical Feedback from V2 Exerts a Powerful Influence over the Visually Evoked Local Field Potential and Associated Spike Timing in V1

    Neuroscience 10:792010.

    https://doi.org/10.1101/792010

    • Google Scholar
33. 1. Haufe S
    2. Nikulin VV
    3. Müller K-R
    4. Nolte G

    (2013) A critical assessment of connectivity measures for EEG data: A simulation study

    NeuroImage 64:120–133.

    https://doi.org/10.1016/j.neuroimage.2012.09.036

    • PubMed
    • Google Scholar
34. 1. Hupé JM
    2. James AC
    3. Payne BR
    4. Lomber SG
    5. Girard P
    6. Bullier J

    (1998) Cortical feedback improves discrimination between figure and background by V1, V2 and V3 neurons

    Nature 394:784–787.

    https://doi.org/10.1038/29537

    • PubMed
    • Google Scholar
35. 1. Jia X
    2. Smith MA
    3. Kohn A

    (2011) Stimulus Selectivity and Spatial Coherence of Gamma Components of the Local Field Potential

    Journal of Neuroscience 31:9390–9403.

    https://doi.org/10.1523/JNEUROSCI.0645-11.2011

    • PubMed
    • Google Scholar
36. 1. Kraut MA
    2. Arezzo JC
    3. Vaughan HG

    (1985) Intracortical generators of the flash VEP in monkeys

    Electroencephalography and Clinical Neurophysiology/Evoked Potentials Section 62:300–312.

    https://doi.org/10.1016/0168-5597(85)90007-3

    • PubMed
    • Google Scholar
37. 1. Lopes Da Silva FH
    2. Storm Van Leeuwen W

    (1977) The cortical source of the alpha rhythm

    Neuroscience Letters 6:237–241.

    https://doi.org/10.1016/0304-3940(77)90024-6

    • PubMed
    • Google Scholar
38. 1. Lund JS

    (1988) Anatomical organization of macaque monkey striate visual cortex

    Annual Review of Neuroscience 11:253–288.

    https://doi.org/10.1146/annurev.ne.11.030188.001345

    • PubMed
    • Google Scholar
39. 1. Maier A
    2. Adams GK
    3. Aura C
    4. Leopold DA

    (2010) Distinct superficial and deep laminar domains of activity in the visual cortex during rest and stimulation

    Frontiers in Systems Neuroscience 4:31.

    https://doi.org/10.3389/fnsys.2010.00031

    • PubMed
    • Google Scholar
40. 1. Maier A
    2. Aura CJ
    3. Leopold DA

    (2011) Infragranular sources of sustained local field potential responses in macaque primary visual cortex

    The Journal of Neuroscience 31:1971–1980.

    https://doi.org/10.1523/JNEUROSCI.5300-09.2011

    • PubMed
    • Google Scholar
41. 1. Maris E
    2. Oostenveld R

    (2007) Nonparametric statistical testing of EEG- and MEG-data

    Journal of Neuroscience Methods 164:177–190.

    https://doi.org/10.1016/j.jneumeth.2007.03.024

    • PubMed
    • Google Scholar
42. 1. Markov NT
    2. Ercsey-Ravasz M
    3. Van Essen DC
    4. Knoblauch K
    5. Toroczkai Z
    6. Kennedy H

    (2013) Cortical high-density counterstream architectures

    Science (New York, N.Y.) 342:1238406.

    https://doi.org/10.1126/science.1238406

    • PubMed
    • Google Scholar
43. 1. Markov NT
    2. Ercsey-Ravasz MM
    3. Ribeiro Gomes AR
    4. Lamy C
    5. Magrou L
    6. Vezoli J
    7. Misery P
    8. Falchier A
    9. Quilodran R
    10. Gariel MA
    11. Sallet J
    12. Gamanut R
    13. Huissoud C
    14. Clavagnier S
    15. Giroud P
    16. Sappey-Marinier D
    17. Barone P
    18. Dehay C
    19. Toroczkai Z
    20. Knoblauch K
    21. Van Essen DC
    22. Kennedy H

    (2014) A weighted and directed interareal connectivity matrix for macaque cerebral cortex

    Cerebral Cortex (New York, N.Y 24:17–36.

    https://doi.org/10.1093/cercor/bhs270

    • PubMed
    • Google Scholar
44. 1. Mitzdorf U
    2. Singer W

    (1979) Excitatory synaptic ensemble properties in the visual cortex of the macaque monkey: a current source density analysis of electrically evoked potentials

    The Journal of Comparative Neurology 187:71–83.

    https://doi.org/10.1002/cne.901870105

    • PubMed
    • Google Scholar
45. 1. Mitzdorf U

    (1987) Properties of the evoked potential generators: current source-density analysis of visually evoked potentials in the cat cortex

    The International Journal of Neuroscience 33:33–59.

    https://doi.org/10.3109/00207458708985928

    • PubMed
    • Google Scholar
46. 1. Nassi JJ
    2. Lomber SG
    3. Born RT

    (2013) Corticocortical feedback contributes to surround suppression in V1 of the alert primate

    The Journal of Neuroscience 33:8504–8517.

    https://doi.org/10.1523/JNEUROSCI.5124-12.2013

    • PubMed
    • Google Scholar
47. 1. Nowak LG
    2. Munk MHJ
    3. Girard P
    4. Bullier J

    (1995) Visual latencies in areas V1 and V2 of the macaque monkey

    Visual Neuroscience 12:371–384.

    https://doi.org/10.1017/s095252380000804x

    • PubMed
    • Google Scholar
48. 1. Nurminen L
    2. Merlin S
    3. Bijanzadeh M
    4. Federer F
    5. Angelucci A

    (2018) Top-down feedback controls spatial summation and response amplitude in primate visual cortex

    Nature Communications 9:2281.

    https://doi.org/10.1038/s41467-018-04500-5

    • PubMed
    • Google Scholar
49. 1. Peter A
    2. Uran C
    3. Klon-Lipok J
    4. Roese R
    5. van Stijn S
    6. Barnes W
    7. Dowdall JR
    8. Singer W
    9. Fries P
    10. Vinck M

    (2019) Surface color and predictability determine contextual modulation of V1 firing and gamma oscillations

    eLife 8:e42101.

    https://doi.org/10.7554/eLife.42101

    • PubMed
    • Google Scholar
50. 1. Pettersen KH
    2. Devor A
    3. Ulbert I
    4. Dale AM
    5. Einevoll GT

    (2006) Current-source density estimation based on inversion of electrostatic forward solution: effects of finite extent of neuronal activity and conductivity discontinuities

    Journal of Neuroscience Methods 154:116–133.

    https://doi.org/10.1016/j.jneumeth.2005.12.005

    • PubMed
    • Google Scholar
51. 1. Ray S
    2. Maunsell JHR

    (2010) Differences in gamma frequencies across visual cortex restrict their possible use in computation

    Neuron 67:885–896.

    https://doi.org/10.1016/j.neuron.2010.08.004

    • PubMed
    • Google Scholar
52. 1. Roberts MJ
    2. Lowet E
    3. Brunet NM
    4. Ter Wal M
    5. Tiesinga P
    6. Fries P
    7. De Weerd P

    (2013) Robust gamma coherence between macaque V1 and V2 by dynamic frequency matching

    Neuron 78:523–536.

    https://doi.org/10.1016/j.neuron.2013.03.003

    • PubMed
    • Google Scholar
53. 1. Rockland KS
    2. Pandya DN

    (1979) Laminar origins and terminations of cortical connections of the occipital lobe in the rhesus monkey

    Brain Research 179:3–20.

    https://doi.org/10.1016/0006-8993(79)90485-2

    • PubMed
    • Google Scholar
54. 1. Sceniak MP
    2. Hawken MJ
    3. Shapley R

    (2001) Visual spatial characterization of macaque V1 neurons

    Journal of Neurophysiology 85:1873–1887.

    https://doi.org/10.1152/jn.2001.85.5.1873

    • PubMed
    • Google Scholar
55. 1. Schroeder CE
    2. Tenke CE
    3. Givre SJ
    4. Arezzo JC
    5. Vaughan HG Jr

    (1991) Striate cortical contribution to the surface-recorded pattern-reversal VEP in the alert monkey

    Vision Research 31:1143–1157.

    https://doi.org/10.1016/0042-6989(91)90040-c

    • PubMed
    • Google Scholar
56. 1. Self MW
    2. van Kerkoerle T
    3. Supèr H
    4. Roelfsema PR

    (2013) Distinct roles of the cortical layers of area V1 in figure-ground segregation

    Current Biology 23:2121–2129.

    https://doi.org/10.1016/j.cub.2013.09.013

    • PubMed
    • Google Scholar
57. 1. Spaak E
    2. Bonnefond M
    3. Maier A
    4. Leopold DA
    5. Jensen O

    (2012) Layer-specific entrainment of γ-band neural activity by the α rhythm in monkey visual cortex

    Current Biology 22:2313–2318.

    https://doi.org/10.1016/j.cub.2012.10.020

    • PubMed
    • Google Scholar
58. 1. Supèr H
    2. Roelfsema PR

    (2005) Chronic multiunit recordings in behaving animals: advantages and limitations

    Progress in Brain Research 147:263–282.

    https://doi.org/10.1016/S0079-6123(04)47020-4

    • PubMed
    • Google Scholar
59. 1. Thiele A
    2. Delicato LS
    3. Roberts MJ
    4. Gieselmann MA

    (2006) A novel electrode-pipette design for simultaneous recording of extracellular spikes and iontophoretic drug application in awake behaving monkeys

    Journal of Neuroscience Methods 158:207–211.

    https://doi.org/10.1016/j.jneumeth.2006.05.032

    • PubMed
    • Google Scholar
60. 1. Tiesinga P
    2. Sejnowski TJ

    (2009) Cortical enlightenment: are attentional gamma oscillations driven by ING or PING?

    Neuron 63:727–732.

    https://doi.org/10.1016/j.neuron.2009.09.009

    • PubMed
    • Google Scholar
61. 1. Trongnetrpunya A
    2. Nandi B
    3. Kang D
    4. Kocsis B
    5. Schroeder CE
    6. Ding MZ

    (2016) Assessing Granger Causality in Electrophysiological Data: Removing the Adverse Effects of Common Signals via Bipolar Derivations

    Frontiers in Systems Neuroscience 9:189.

    https://doi.org/10.3389/fnsys.2015.00189

    • Google Scholar
62. 1. Vaknin G
    2. DiScenna PG
    3. Teyler TJ

    (1988) A method for calculating current source density (CSD) analysis without resorting to recording sites outside the sampling volume

    Journal of Neuroscience Methods 24:131–135.

    https://doi.org/10.1016/0165-0270(88)90056-8

    • PubMed
    • Google Scholar
63. 1. van Kerkoerle T
    2. Self MW
    3. Dagnino B
    4. Gariel-Mathis MA
    5. Poort J
    6. van der Togt C
    7. Roelfsema PR

    (2014) Alpha and gamma oscillations characterize feedback and feedforward processing in monkey visual cortex

    PNAS 111:14332–14341.

    https://doi.org/10.1073/pnas.1402773111

    • PubMed
    • Google Scholar
64. 1. Vinck M
    2. Huurdeman L
    3. Bosman CA
    4. Fries P
    5. Battaglia FP
    6. Pennartz CMA
    7. Tiesinga PH

    (2015) How to detect the Granger-causal flow direction in the presence of additive noise?

    NeuroImage 108:301–318.

    https://doi.org/10.1016/j.neuroimage.2014.12.017

    • PubMed
    • Google Scholar
65. 1. Xing D
    2. Yeh CI
    3. Burns S
    4. Shapley RM

    (2012) Laminar analysis of visually evoked activity in the primary visual cortex

    PNAS 109:13871–13876.

    https://doi.org/10.1073/pnas.1201478109

    • PubMed
    • Google Scholar

Decision letter after peer review:

Thank you for submitting your article "Stimulus dependence of directed information exchange between cortical layers in macaque V1" for consideration by eLife. Your article has been reviewed by 2 peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Joshua Gold as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Scott Lowe (Reviewer #1).

Below you will find a set of recommendations, which, if completed, would lead to publication in eLife.

Essential revisions:

1) GC Analysis using MUA (see below, rev1.1).

2) A more extensive of the distribution of receptive field sizes, and if appropriate, repeating the analysis when using 1 deg as reference (rev 1.2, editor comments).

3) Clarify text regarding the difference between lateral intracolumnar feedback and top-down intra-area feedback (rev 2).Reviewer #1:

In this work, the authors study neural oscillations induced in macaque V1 by square-wave stimuli of varying sizes. The results suggest that α/β frequencys bands arise in L6, with long-range interactions to L2/3; and γ frequency arise in L4 or lower L3, with a series of short range interactions up the cortical column, all the way to L1. There is also an interaction from ~L4C to L6, for which the frequency changes depending on the size of the stimulus (low frequency for small and high frequency for large stimuli).

I find the paper is worthy of publication in eLife as it stands, and could be made stronger with some additional work.

1. The GC analysis was applied to four frequency-bands: theta, α, β, and low-γ. The same analysis could easily be applied to the MUA. Also, there is the possiblity of cross-frequency coupling (Spaak et al., 2012): high-frequency oscillations in one region may induce low-frequency oscillations in another (and vice-versa), and this could be studied with GC between different frequencies. Why is such analysis excluded from the paper? It would seem to be a natural fit, and I think this additional work would make the paper stronger.

2. Much of the discussion throughout the paper focuses on the difference of effects with small centre-only stimuli and large stimuli which stimulate the surround. However, the distribution of RF size of the cortical columns being studied is not reported. This should be included so the size of the stimuli can be contextualised. I also posit that correcting for the size of the RF might lead to a clearer transition from the centre to surround recording condition. If there is a large range of RF sizes between sessions (a factor of 2 difference, or more) I ask that the authors consider this, or explain why correcting for the RF size would not be desirable.

3. On L160, the authors cite Einevoll to justify using the LFPbp, but I am sure he would instead advocate to use the CSD instead. Why use the LFPbp instead of the CSD?

4. There are more accurate methods of estimating the CSD than the double-spatial derivative. There is the inverse-CSD (iCSD; Pettersen, 2006) and kernel-CSD (kCSD; Potworowski, 2012). Problems can arise with the double-spatial derivative when the current sources are small compared to the electrode measurement interval. These new methods take into account the electrode sampling interval, and the expected spatial extent of sources, in order to estimate the CSD from the LFP. Why use the double-spatial derivative instead of iCSD or kCSD?

5. With regards to determining the spectral power (L483), why not use the Welch method, or similar, instead of just the FFT? This would give a more stable estimate of power, especially in the high frequencies.

6. It is curious that you find stimulus-driven changes centred at a frequency of 40Hz (Figure 4). Earlier works have found that stimulus-driven information is isolated into two compartments, one below 40Hz and the other above 40Hz (Belitski, 2008) and that frequencies below and above 40Hz encode information about different spatio-temporal components of stimuli (Lowe, 2017), whereas (in these works) frequencies around 40Hz do not contain information about the stimulus. Furthermore, they appear to be driven by dopaminergic neuromodulation (Zaldivar, 2018). Also, it is noteworthy that the stimulus presentation spatial frequency of 1.5cpd is the point where γ frequencies (<40Hz) become more informative about the stimulus than theta and α (Lowe, 2017).

7. I am curious as to what other CSD-profiles were seen (L72). Were they just noisy, or indications that the probe was not aligned with cortical column? Or is there some (albeit weak) evidence CSD profiles are not uniform across V1? I assume the former, but ask in case there is a genuine minority CSD profile which multiple labs are seeing and rejecting. I have rejected a session before which was perfect, except for the inconvenient fact that the sign of all CSD currents were inverted.

Refs:

Belitski et al., (2008). doi:10.1523/JNEUROSCI.0009-08.2008.

Logothetis et al., (2007). doi:10.1016/j.neuron.2007.07.027.

Lowe (2017). https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.738787

Pettersen et al., (2006). doi:10.1016/j.jneumeth.2005.12.005.

Potworowski et al., (2012). doi:10.1162/NECO_a_00236.

Zaldivar et al., (2018). doi: 10.1016/j.cub.2017.12.006.Reviewer #2:

Anatomical tracing studies have established that the connectivity between visual areas abides by a global hierarchy, such that between any two visual areas, there is a graded hierarchical relation. Prior studies (Bastos, Neuron, 2015; Michalareas, Neuron, 2016) have shown that α and β GC is stronger in the top-down direction, whereas theta and γ GC is stronger in the bottom-up direction. Within a cortical column, the concept of a canonical microcircuit suggests a flow of information from granular to supragranular and then to infragranular layers. Prior studies (Livingstone, J. Neurophys., 1996; van Kerkoerle, PNAS, 2014) have shown that γ oscillations have characteristic laminar phase relations, with systematic lags as function of laminar distance from granular, and that α oscillations can show the opposite, i.e. systematic leads. These two sets of results, i.e. interareal and intracolumnar, are consistent, but need to be regarded separately. The hierarchical ordering of visual areas is supported by unequivocal evidence, whereas a columnar information-flow pattern is merely suggested, and different models are consistent with the empirical evidence. The current manuscript speaks to the columnar level and presents interesting evidence, partly consistent with van Kerkoerle, and partly extending the van Kerkoerle study to different stimulus sizes. It does not speak directly to the interareal level. The authors need to revise their manuscript to avoid confusion.

Figure 6 and related figures and text: This is almost the same analysis as performed by van Kerkoerle, with the important difference that Kerkoerle used CSD, whereas the present manuscript uses bipolar LFP. The present manuscript makes direct comparisons to Kerkoerle and partly reports conflicting results. For a sound and fair comparison, the authors of the current manuscript need to provide these analysis based on the CSD.

The phase-lag and GC results are reported as mainly contradicting previous results. However, if only the results are considered that use exactly the same methods as previous studies (e.g. Kerkoerle with the same stimulus size), results are not so differet. Results obtained with other stimulus sizes are not in conflict with this. And all results of those intra-V1 recordings are not in conflict with interareal GC analyses. Thus, the authors should limit reported conflicts to aspects that can really be compared.

The same holds for the reporting of power. For example line 307 – 309: Xing, 2012 did show γ power in many layers, with a distinct peak in deep L5/upper L6; and Buffalo, 2011 focused on spike-field coherence. SFC is a very different metric from LFP power. It is possible that a particular layer, for a particular stimulus size, shows clear γ LFP power, but maybe the spikes there do not phase-lock to the LFP γ phase, such that γ SFC is low. Thus, again, the authors should limit reported conflicts to aspects that can really be compared.

I am somewhat concerned about the upwards dominance of GC (Figure 7 supp 1). Could there be any trivial reason for this?

Line 326: Early work from the Eckhorn lab showed clear γ in V2. A recent preprint from the Richard Born lab shows strong reductions in V1 γ upon V2 cooling. Also, Bastos, Neuron, 2015 showed clear V2-to-V1 γ GC; note that this paper relates the DAI to the SLN in a graded manner, and the known feedforward-type (supragranular originating) projections from V2 to V1 are expected to carry γ in this top-down direction. Thus, we do know that the input feedback signal shows γ.

Figure 4D: It would be very helpful to see spectra (cross-cuts through the presented spectro-depth plots), at the depths with strongest γ, i.e. around 450 and -450

Figure 7 and similar plots: This is a matter of taste, but I find a matrix version of such results much easier to parse.

Line 37,38: Interareal theta-band GC is stronger in the feedforward, not the feedback, direction (Bastos, Neuron, 2015; Spyropoulos, PNAS, 2018). The feedforward nature of theta is actually correctly referred to in line 144.

Line 44: Classical work by Lopes da Silva located the α source rather deep (Lopes da Silva, Neuroscience Letters, 1977, pages 237 – 241).

Lines 69-74: Clarify whether all three conditions have to be met, or whether one suffices.

Line 104: How was the ANOVA done? Is it suitable to provide the p-value for the statement in the preceding sentence on 3 ms longer latencies?

Line 107: The figure lists p=0.98, whereas the text lists p=0.59. Also, the text specifies a one-way ANOVA. Please clarify.

Line 169: Gieselmann and Thiele previously showed an actual γ peak for small stimuli, whose peak frequency decreased with stimulus size. This should be related to the current results.

Figure 4B, D and Figure 5A, D, G, J: The text and figure legends are quite clear on that these plots show power changes due to stimulation. I am surprised that changes are exclusively positive. Previous studies have found stimulus induced power decreases, e.g. in α. Are these rectified power changes? In any case, the authors should address this excplicitly, simply to avoid confusion. Also, where the text refers to those panels, it should be careful to actually refer to power changes (not just power).

Line 177 and a few other places: the Xing, Shapley, PNAS, 2012 paper showed several γ peaks along cortical depth, including a weak one at lower layer 5/upper layer 6.

Line 193: I was surprised by the use of the term phase-amplitude-coupling for this analysis. Usually, PAC is used for the relation of low-frequency phase to higher-frequency power, not the amplitude of the raw signal.

Line 220-221: This statement is true for one isolated phase lag value. However, a pattern of phase lags as analyzed by Kerkoerle or in the present manuscript (or in the context of phase-lag spectra) resolves these uncertainties. However, the statement can be made that phase relations are sometimes not consistent with GC (explicitly mentioned by Brovelli, PNAS, 2002, start of page 9853), and this can be resolved with GC methods (Witham, Riddle, Baker, J Physiol, 2011).

Line 264: Figure 7 supp 1 B shows the dominant influence of infra onto supra to be around 20 Hz, i.e. in the same β band that Bastos, Neuron, 2015 found to be feedback related. Figure 7 supp 1 G shows infra to very supra and shows a broad α peak that is at least compatible with Buffalo, PNAS, 2011.

Line 301, 302: Also Peter, Fries, Vinck, eLife, 2019.

Line 335: Barone, JNS, 2000 shows an SLN of 47 %, i.e. not predominantly supragranular, and most likely not predominantly from just layer 3.

Line 339-342: Top-down projections, even if short-range, always have a component of feedback-type connectivity, targeting deep layers. Thus, no reason to invoke longer-range feedback connections.

Line 483: How do the authors obtain 1000 samples in 100 ms? Are they transforming a 10 kHz signal? If all FFT analys were based on 100 ms windows, this would be problematic for lower frequencies. Please clarify.

Line487: Why is only a filter for γ specified? What is the exact meaning of n=3?

Line 489: The windows were wider for lower frequencies.

Line 490-491: Are these analyses actually reported?

Line 493 and following: Is the conditional GC analysis actually reported?

Essential revisions:

1) GC Analysis using MUA (see below, rev1.1)

We performed the same GC analysis as in figure 7 using the MUA signal. The results are added as “Figure 7—figure supplement 4”. We added a section in the manuscript (line 264-266).

2) A more extensive of the distribution of receptive field sizes, and if appropriate, repeating the analysis when using 1 deg as reference (rev 1.2, editor comments).

We analyzed the RF diameter relative to their depth and plotted the results in “Figure 1—figure supplement 1”. We repeated the analysis using different stimulus sizes as reference (0.75 deg, 1.5 deg), and present the data in Figure 2 and Figure 4—figure supplements 1 and supplement 2. We edited a section in the manuscript where we link to these analyses (lines 93-94, lines 128-131, lines 188-189).

3) Clarify text regarding the difference between lateral intracolumnar feedback and top-down intra-area feedback (rev 2).

We have specified it more (line 27, 98-99, 287).

Reviewer #1:

In this work, the authors study neural oscillations induced in macaque V1 by square-wave stimuli of varying sizes. The results suggest that α/β frequencys bands arise in L6, with long-range interactions to L2/3; and γ frequency arise in L4 or lower L3, with a series of short range interactions up the cortical column, all the way to L1. There is also an interaction from ~L4C to L6, for which the frequency changes depending on the size of the stimulus (low frequency for small and high frequency for large stimuli).

I find the paper is worthy of publication in eLife as it stands, and could be made stronger with some additional work.

1. The GC analysis was applied to four frequency-bands: theta, α, β, and low-γ. The same analysis could easily be applied to the MUA. Also, there is the possiblity of cross-frequency coupling (Spaak et al., 2012): high-frequency oscillations in one region may induce low-frequency oscillations in another (and vice-versa), and this could be studied with GC between different frequencies. Why is such analysis excluded from the paper? It would seem to be a natural fit, and I think this additional work would make the paper stronger.

We have now applied the analysis to MUA and added the results as “Figure 7—figure supplement 4” and added a brief text section to the manuscript (line 264-266) (see also Essential Revisions). Regarding a GC analysis between different frequencies, we have discussed this with Scott (rev#1) and external colleagues. Based on the discussion, the reviewer agreed that this is not necessary.

2. Much of the discussion throughout the paper focuses on the difference of effects with small centre-only stimuli and large stimuli which stimulate the surround. However, the distribution of RF size of the cortical columns being studied is not reported. This should be included so the size of the stimuli can be contextualised. I also posit that correcting for the size of the RF might lead to a clearer transition from the centre to surround recording condition. If there is a large range of RF sizes between sessions (a factor of 2 difference, or more) I ask that the authors consider this, or explain why correcting for the RF size would not be desirable.

We added an analysis of the RF diameters to the manuscript (Figure 1—figure supplement 1, see Essential Revisions). The median RF diameters across cortical depths ranged from ~0.7° to ~1.2°. An explicit correction for RF size would make sense if we would have aimed at describing the spatial integration properties of each cortical layer in isolation. However, we aimed at investigating the columnar oscillation pattern across layers. To nevertheless address the query in relative terms we performed key analyses with 2 additional reference sizes, namely 0.75 and 1.5 deg. The results are now shown in Figure 2 and Figure 4—figure supplements 1 and supplement 2. We added sections to the manuscript where we link to these analyses (lines 93-94, lines 128-131, lines 188-189). Based on these analyses we argue that the exact reference size is not key to the main results we try to convey.

3. On L160, the authors cite Einevoll to justify using the LFPbp, but I am sure he would instead advocate to use the CSD instead. Why use the LFPbp instead of the CSD?

We chose to use the bipolar LFP based on the results of Trongnetrpunya et al., (2016). They showed that connectivity analysis based on the bipolar referenced signal is less prone to noise and nearer to the ground truth of network connectivity than current source density signals. This is mentioned in the methods section in line 479-481.

We added the reference to Trongnetrpunya et al., (2016) Line 166 and Line 480

4. There are more accurate methods of estimating the CSD than the double-spatial derivative. There is the inverse-CSD (iCSD; Pettersen, 2006) and kernel-CSD (kCSD; Potworowski, 2012). Problems can arise with the double-spatial derivative when the current sources are small compared to the electrode measurement interval. These new methods take into account the electrode sampling interval, and the expected spatial extent of sources, in order to estimate the CSD from the LFP. Why use the double-spatial derivative instead of iCSD or kCSD?

We used the “standard” method from the iCSD-toolbox but did not mention it in the main text. We changed the text accordingly (line 463-473). We tested other methods from that same toolbox, but the results were not qualitatively different. The standard method was sufficient to extract the alignment criterion.

5. With regards to determining the spectral power (L483), why not use the Welch method, or similar, instead of just the FFT? This would give a more stable estimate of power, especially in the high frequencies.

We tested multiple spectral methods aiming at extracting the prominent ~40Hz oscillation evoked by spatial integration. We tried to avoid any methods that would result in a smoothing over spectral frequency, i.e. multitaper methods. On the other hand, in comparison with other methods, FFT seemed sufficient in quantifying the power of the 40Hz component.

6. It is curious that you find stimulus-driven changes centred at a frequency of 40Hz (Figure 4). Earlier works have found that stimulus-driven information is isolated into two compartments, one below 40Hz and the other above 40Hz (Belitski, 2008) and that frequencies below and above 40Hz encode information about different spatio-temporal components of stimuli (Lowe, 2017), whereas (in these works) frequencies around 40Hz do not contain information about the stimulus. Furthermore, they appear to be driven by dopaminergic neuromodulation (Zaldivar, 2018). Also, it is noteworthy that the stimulus presentation spatial frequency of 1.5cpd is the point where γ frequencies (<40Hz) become more informative about the stimulus than theta and α (Lowe, 2017).

Grating induced γ oscillations with a peak around 40Hz are a well described feature of V1 and have been frequently reported in the literature (Gray et al., 1989; Bauer et al., 1995; Gieselmann and Thiele, 2008; Jia et al., 2011). We agree that visual stimulation via natural images, as they have been used in Belitski et al., (2008), do not induce such a prominent 40Hz peak. However, we were particularly interested in the oscillatory interactions between cortical layers in the context of spatial integration which strongly modulates the 40Hz peak. A detailed analysis of frequency band specific information is beyond the scope of this study.

7. I am curious as to what other CSD-profiles were seen (L72). Were they just noisy, or indications that the probe was not aligned with cortical column? Or is there some (albeit weak) evidence CSD profiles are not uniform across V1? I assume the former, but ask in case there is a genuine minority CSD profile which multiple labs are seeing and rejecting. I have rejected a session before which was perfect, except for the inconvenient fact that the sign of all CSD currents were inverted.

This is an excellent question. But with the current dataset it is impossible to address it in a meaningful manner.

Reviewer #2:

Anatomical tracing studies have established that the connectivity between visual areas abides by a global hierarchy, such that between any two visual areas, there is a graded hierarchical relation. Prior studies (Bastos, Neuron, 2015; Michalareas, Neuron, 2016) have shown that α and β GC is stronger in the top-down direction, whereas theta and γ GC is stronger in the bottom-up direction. Within a cortical column, the concept of a canonical microcircuit suggests a flow of information from granular to supragranular and then to infragranular layers. Prior studies (Livingstone, J. Neurophys., 1996; van Kerkoerle, PNAS, 2014) have shown that γ oscillations have characteristic laminar phase relations, with systematic lags as function of laminar distance from granular, and that α oscillations can show the opposite, i.e. systematic leads. These two sets of results, i.e. interareal and intracolumnar, are consistent, but need to be regarded separately. The hierarchical ordering of visual areas is supported by unequivocal evidence, whereas a columnar information-flow pattern is merely suggested, and different models are consistent with the empirical evidence. The current manuscript speaks to the columnar level and presents interesting evidence, partly consistent with van Kerkoerle, and partly extending the van Kerkoerle study to different stimulus sizes. It does not speak directly to the interareal level. The authors need to revise their manuscript to avoid confusion.

Figure 6 and related figures and text: This is almost the same analysis as performed by van Kerkoerle, with the important difference that Kerkoerle used CSD, whereas the present manuscript uses bipolar LFP. The present manuscript makes direct comparisons to Kerkoerle and partly reports conflicting results. For a sound and fair comparison, the authors of the current manuscript need to provide these analysis based on the CSD.

We produced a new Figure 6—figure supplement 5 from our data to compare to figure 3 in Kerkoerle et al., (2014). And while these results are somewhat different to the LFPbp results, they still differ to those reported by van Kerkoerle et al., We believe that these differences arise from stimulus differences and possibly the use of a cognitively demanding task in van Kerkoerle's study. We now mention these differences in the text (lines 293-298, see also line 374-380).

The phase-lag and GC results are reported as mainly contradicting previous results. However, if only the results are considered that use exactly the same methods as previous studies (e.g. Kerkoerle with the same stimulus size), results are not so differet. Results obtained with other stimulus sizes are not in conflict with this. And all results of those intra-V1 recordings are not in conflict with interareal GC analyses. Thus, the authors should limit reported conflicts to aspects that can really be compared.

We now mention possible sources for these differences and acknowledge that the use of large grating stimuli could be a key element (line 293-298). However, van Kerkoerle used large stimuli (full field structured background, and elongated lines (curve tracing task)) so comparing that directly to small grating stimuli is equally not really compatible.

The same holds for the reporting of power. For example line 307 – 309: Xing, 2012 did show γ power in many layers, with a distinct peak in deep L5/upper L6; and Buffalo, 2011 focused on spike-field coherence. SFC is a very different metric from LFP power. It is possible that a particular layer, for a particular stimulus size, shows clear γ LFP power, but maybe the spikes there do not phase-lock to the LFP γ phase, such that γ SFC is low. Thus, again, the authors should limit reported conflicts to aspects that can really be compared.

We have changed the wording and the associated referencing, to account for those differences (line 332-333).

I am somewhat concerned about the upwards dominance of GC (Figure 7 supp 1). Could there be any trivial reason for this?

We do not think this to be the case. We have reported a similar pattern in a recent publication (Ferro et al., 2021, PNAS), where animals engaged in an attention demanding task, and where we simultaneously recorded in V1 and V4. Here V1 showed the pattern, while V4 did not. We now cite that study (line 248). It has been reported that granger causal interactions depend on absolute power of the signal, where the more powerful signal could be seen as the driver. However, this is not the case for our data, where absolute power dominates in supragranular layers.

Line 326: Early work from the Eckhorn lab showed clear γ in V2. A recent preprint from the Richard Born lab shows strong reductions in V1 γ upon V2 cooling. Also, Bastos, Neuron, 2015 showed clear V2-to-V1 γ GC; note that this paper relates the DAI to the SLN in a graded manner, and the known feedforward-type (supragranular originating) projections from V2 to V1 are expected to carry γ in this top-down direction. Thus, we do know that the input feedback signal shows γ.

We now reference these studies and list them as a potential main source of the layer 5 γ frequency sources with larger stimuli (lines 350-354).

Figure 4D: It would be very helpful to see spectra (cross-cuts through the presented spectro-depth plots), at the depths with strongest γ, i.e. around 450 and -450

We added Figure 4—figure supplement 3 and referred to it in the respective Results section (line 164-189).

Figure 7 and similar plots: This is a matter of taste, but I find a matrix version of such results much easier to parse.

Newly added Figure 7—figure supplement 4 shows a matrix view of the data.

Line 37,38: Interareal theta-band GC is stronger in the feedforward, not the feedback, direction (Bastos, Neuron, 2015; Spyropoulos, PNAS, 2018). The feedforward nature of theta is actually correctly referred to in line 144.

We changed this (line 41)

Line 44: Classical work by Lopes da Silva located the α source rather deep (Lopes da Silva, Neuroscience Letters, 1977, pages 237 – 241).

We now added this reference (line 39)

Lines 69-74: Clarify whether all three conditions have to be met, or whether one suffices.

We edited that section in the text (line 68-74) in reply to a point raised by reviewer#1.

Line 104: How was the ANOVA done? Is it suitable to provide the p-value for the statement in the preceding sentence on 3 ms longer latencies?

We performed a one-factor-ANOVA across stimulus sizes to estimate if there was a main effect of stimulus size. We rephrased the section describing the results to avoid the impression that we statistically tested the time differences indicated by the plot (line 105-108).

Line 107: The figure lists p=0.98, whereas the text lists p=0.59. Also, the text specifies a one-way ANOVA. Please clarify.

The p-value refers to the one-factor-ANOVA across all stimulus sizes (see previous comment). We changed the p-value to the correct value of p=0.98 as shown in the figure (line 111).

Line 169: Gieselmann and Thiele previously showed an actual γ peak for small stimuli, whose peak frequency decreased with stimulus size. This should be related to the current results.

In response to the comments for Figure 4D of reviewer#2 we plotted the spectral power at selected depth positions (new Figure 4—figure supplement 3). These plots qualitatively show an inverse relationship between the γ peak frequency and stimulus size similar to the dataset described in Gieselmann and Thiele (2008). See references to the new figure in (line 164-189).

Figure 4B, D and Figure 5A, D, G, J: The text and figure legends are quite clear on that these plots show power changes due to stimulation. I am surprised that changes are exclusively positive. Previous studies have found stimulus induced power decreases, e.g. in α. Are these rectified power changes? In any case, the authors should address this excplicitly, simply to avoid confusion. Also, where the text refers to those panels, it should be careful to actually refer to power changes (not just power).

Line 177 and a few other places: the Xing, Shapley, PNAS, 2012 paper showed several γ peaks along cortical depth, including a weak one at lower layer 5/upper layer 6.

In our passively fixating monkeys, we indeed rarely found stimulus induced α decreases, that is below levels of spontaneous activity (negative z-score). However, large stimuli induced less power for low frequencies while higher frequencies were simultaneously induced with more power (Figure 4, 5 and Figure 4-supplement 3).

We introduced the term “induced spectral power” (iSP) in line 171 and used it in the section describing the results if figure 4 and 5 (line 164-189).

We agree that Xing et al., (2012) have also shown increased γ power in infragranular layers and added a reference in line 185.

Line 193: I was surprised by the use of the term phase-amplitude-coupling for this analysis. Usually, PAC is used for the relation of low-frequency phase to higher-frequency power, not the amplitude of the raw signal.

We agree that we misused the term phase-amplitude-coupling and changed the terminology throughout the paper, using the terms “phase-triggered-average” (PTA) or “phase-coupled” were appropriate.

Line 220-221: This statement is true for one isolated phase lag value. However, a pattern of phase lags as analyzed by Kerkoerle or in the present manuscript (or in the context of phase-lag spectra) resolves these uncertainties. However, the statement can be made that phase relations are sometimes not consistent with GC (explicitly mentioned by Brovelli, PNAS, 2002, start of page 9853), and this can be resolved with GC methods (Witham, Riddle, Baker, J Physiol, 2011).

We have changed the text accordingly (lines 232-235)

Line 264: Figure 7 supp 1 B shows the dominant influence of infra onto supra to be around 20 Hz, i.e. in the same β band that Bastos, Neuron, 2015 found to be feedback related. Figure 7 supp 1 G shows infra to very supra and shows a broad α peak that is at least compatible with Buffalo, PNAS, 2011.

We have changed our discussion in line with these comments (lines 283-288).

Line 301, 302: Also Peter, Fries, Vinck, eLife, 2019.

We have added the reference (line 328).

Line 335: Barone, JNS, 2000 shows an SLN of 47 %, i.e. not predominantly supragranular, and most likely not predominantly from just layer 3.

We now acknowledge this and cite Barone et al., (line 361-362).

Line 339-342: Top-down projections, even if short-range, always have a component of feedback-type connectivity, targeting deep layers. Thus, no reason to invoke longer-range feedback connections.

While we do not dispute the short-range feedback terminating in deeper layers (and we now make this explicit, line 361-362, the feedback from V2 should predominantly affect supragranular layers), which is not what we see. Also, the size effects seen in our data suggest that areas with larger RFs are involved.

Line 483: How do the authors obtain 1000 samples in 100 ms? Are they transforming a 10 kHz signal? If all FFT analys were based on 100 ms windows, this would be problematic for lower frequencies. Please clarify.

Data from 100 ms sliding window were padded to 1024 points by MATLAB’s “spectrogram” function to achieve a high spectral resolution. We agree that the window width may cause absolute power estimates in lower frequencies (< 10 Hz) to be inaccurate. However, we would argue that the conclusions drawn from Figure 5 C, F still hold as we were comparing induced power levels between different stimulus sizes, and our focus is generally on frequencies >10Hz.

Line487: Why is only a filter for γ specified? What is the exact meaning of n=3?

We used the same procedure to extract the troughs of oscillations in the four frequency bands shown in figure 6 (4-8,8-13,14-25,30-55 Hz). N is the order of the butterworth filter we used as a bandpass filter. We edited the respective method section accordingly (line 515-521).

Line 489: The windows were wider for lower frequencies.

Yes, they were. We edited the respective method section accordingly (line 515-521).

Line 490-491: Are these analyses actually reported?

No, we didn’t report quantification of PTA coupling and removed the respective text in the methods (line 515-521).

Line 493 and following: Is the conditional GC analysis actually reported?

No, it is not reported. We removed the respective text from the method section (line 522-528).

Author details

1. Marc Alwin Gieselmann

   Biosciences Institute, Newcastle University, Newcastle upon Tyne, United Kingdom

   Contribution

   Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Software, Validation, Visualization, Writing – review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-8034-5134

2. Alexander Thiele

   Biosciences Institute, Newcastle University, Newcastle upon Tyne, United Kingdom

   Contribution

   Conceptualization, Funding acquisition, Project administration, Resources, Supervision, Writing - original draft, Writing – review and editing

   For correspondence

   alex.thiele@newcastle.ac.uk

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-4894-0213

• 679

  Page views

• 117

  Downloads

• 4

  Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.