Fine-scale computations for adaptive processing in the human brain
Abstract
Adapting to the environment statistics by reducing brain responses to repetitive sensory information is key for efficient information processing. Yet, the fine-scale computations that support this adaptive processing in the human brain remain largely unknown. Here, we capitalise on the sub-millimetre resolution of ultra-high field imaging to examine functional magnetic resonance imaging signals across cortical depth and discern competing hypotheses about the brain mechanisms (feedforward vs. feedback) that mediate adaptive processing. We demonstrate layer-specific suppressive processing within visual cortex, as indicated by stronger BOLD decrease in superficial and middle than deeper layers for gratings that were repeatedly presented at the same orientation. Further, we show altered functional connectivity for adaptation: enhanced feedforward connectivity from V1 to higher visual areas, short-range feedback connectivity between V1 and V2, and long-range feedback occipito-parietal connectivity. Our findings provide evidence for a circuit of local recurrent and feedback interactions that mediate rapid brain plasticity for adaptive information processing.
Introduction
Interacting in cluttered and complex environments, we are bombarded with plethora of sensory information from diverse sources. The brain is known to address this challenge by reducing its responses to repeatedly or continuously presented sensory inputs (for reviews: Clifford, 2002; Kohn, 2007). This type of sensory adaptation is a rapid form of plasticity that is critical for efficient processing and has been shown to involve changes in perceptual sensitivity (for review: Clifford, 2002) and neural selectivity (for review: Kohn, 2007). Numerous neurophysiological studies (for review: Kohn, 2007) have shown sensory adaptation to be associated with reduction in neuronal responses that are specific to the features of the adaptor. Functional brain imaging studies in humans have shown functional magnetic resonance imaging (fMRI) adaptation for low-level visual features (e.g. contrast, orientation, motion; for review Larsson et al., 2016) as indicated by decreased BOLD responses in visual cortex due to stimulus repetition. Similar BOLD decreases have been reported in higher visual areas for repeated presentation of more complex visual stimuli (e.g. faces objects), an effect known as repetition suppression (Grill-Spector et al., 2006; Krekelberg et al., 2006). Yet, the fine-scale human brain computations that underlie adaptive processing remain debated.
In particular, neurophysiological studies focussing on primary visual cortex provide evidence of rapid adaptation at early stages of sensory processing (Gutnisky and Dragoi, 2008; Whitmire and Stanley, 2016; Xiang and Brown, 1998). In contrast, fMRI studies have suggested top-down influences on sensory processing of repeated stimuli via feedback mechanisms (e.g. Ewbank et al., 2011; Summerfield et al., 2008). Yet, the circuit mechanisms that mediate adaptive processing in the human brain remain largely unknown, as fMRI at standard resolution does not allow us to discern feedforward from feedback signals.
Here, we capitalise on recent advances in brain imaging technology to determine the contribution of feedforward vs. feedback mechanisms to adaptive processing. Ultra-high field (UHF) imaging affords the sub-millimetre resolution necessary to examine fMRI signals across cortical depth in a non-invasive manner, providing a unique approach to interrogate human brain circuits at a finer scale (for review: Lawrence et al., 2019a) than that possible by standard fMRI techniques (for review: Goense et al., 2016). UHF laminar imaging allows us to test the finer functional connectivity across cortical depth based on known anatomical laminar circuits. In particular, sensory inputs are known to enter the cortex from the thalamus at the level of the middle layer (layer 4), while output information is fed forward from superficial layers (layer 2/3), and feedback information is exchanged primarily between deeper layers (layer 5/6) as well as superficial layers (for review: Self et al., 2019).
Here, we combine UHF laminar imaging with an orientation adaptation paradigm (i.e. observers are presented with gratings at the same or different orientations) to test whether orientation-specific adaptation alters input processing in middle layers of visual areas (V1 and extrastriate visual areas), or feedback processing in superficial and deeper layers (Figure 1A). We demonstrate that adaptation alters orientation-specific signals across cortical depth in the visual cortex with stronger fMRI-adaptation (i.e. BOLD decrease for repeated stimuli) in superficial layers. This layer-specific fMRI adaptation relates to a perceptual bias in orientation discrimination due to adaptation, as measured using a tilt aftereffect paradigm (e.g. Gibson and Radner, 1937). Further, functional connectivity analysis shows that adaptation involves: (a) enhanced feedforward connectivity between V1 superficial layers and middle layers of extrastriate visual areas, and (b) enhanced short-range feedback between V2 and V1 deeper cortical layers. Finally, we test the role of the posterior parietal cortex in adaptive processing, as it is known to be involved in stimulus expectation and novelty detection (de Lange et al., 2018; Garrido et al., 2009; Li et al., 2010; Summerfield and de Lange, 2014). Our results show enhanced feedback connectivity from posterior parietal cortex to V1 deeper layers, suggesting top-down influences on visual processing via long-range feedback mechanisms. Our findings provide evidence for a circuit of local recurrent (i.e. feedforward and short-range feedback) processing across cortical depth in visual cortex, and occipito-parietal feedback interactions that mediate adaptive processing in the human brain.

Functional magnetic resonance imaging (fMRI) laminar circuits and fMRI design.
(A) Schematic representation of feedforward (superficial – middle layers; green), feedback (deeper – deeper layers; red), and feedforward plus feedback (superficial – deeper layers; yellow) anatomical connectivity between V1 and higher cortical regions. Here, we focussed on feedforward vs. feedback connections. (B) fMRI design. Adaptation blocks comprised 16 sinewave gratings presented at the same orientation. Non-adaptation blocks comprised 16 gratings presented at different orientations. During stimulus presentation (1900 ms stimulus on, 100 ms stimulus off), participants were asked to perform an Rapid Serial Visual Presentation (RSVP) task; that is, count the number of times a target letter (e.g. X) was displayed in the stream of distracters and report it at the end of each stimulus block. Each letter was displayed for 150 ms and participants had 2000 ms to give their response.
Results
fMRI adaptation across cortical depth in visual cortex
To test whether adaptation alters visual orientation processing, we measured fMRI responses when participants (N = 15) were presented with gratings (N = 16 per block) either at the same orientation (adaptation) or different orientations (non-adaptation) in a blocked fMRI design (Figure 1B). Participants were asked to perform a Rapid Serial Visual Presentation (RSVP) task (i.e. detect a target in a stream of letters presented in the centre of the screen) to ensure that they attended similarly across conditions (Larsson et al., 2006).
We tested for fMRI adaptation in visual cortex due to stimulus repetition by comparing fMRI responses for adaptation (i.e. the same oriented sinewave grating presented repeatedly within a block) vs. non-adaptation (i.e. gratings of varying orientation presented in a block). To test for differences in orientation-specific fMRI adaptation across cortical depth, for each participant we mapped the retinotopic areas in the visual cortex (V1, V2, V3, V4), assigned voxels in three cortical depths (deeper, middle, superficial layers) using an equi-volume approach (see Methods, MRI data analysis: Segmentation and cortical depth sampling; Figure 2D), and extracted fMRI responses across cortical depths. To control for possible differences in thermal noise, physiological noise, or signal gain across cortical depths (Goense et al., 2012; Havlicek and Uludağ, 2020), we (a) matched the number of voxels across cortical depths for each participant and region of interest (ROI), and (b) z-scored the laminar-specific time courses to control for differences in variance across cortical depths, while preserving condition-dependent differences within each cortical layer.

Functional magnetic resonance imaging (fMRI) data analysis, layer segmentation, and vascular contribution correction.
(A) Sagittal brain view of representative participant: red insert highlights region of interest (ROI, early visual cortex). Structural (B) and functional (C) images of the ROI showing activation maps for stimulus vs. fixation. Activation is well confined within the grey matter borders. (D) Mapping of cortical layers within the ROI: deeper layers shown in red, middle layers in green, superficial layers in blue. (E) Voxels confounded by vasculature effects (in red) overlaid on mean functional image. (F) Mean BOLD (per cent signal change from fixation baseline) across participants for V1 across cortical depth. Comparison between BOLD signal before (blue) and after temporal signal-to-noise ratio (tSNR) and t-value correction (red), and 3D GRASE BOLD signal (black). The superficial bias observed in the BOLD signal is reduced after correction and matches closely the laminar profile of the 3D GRASE data.
Our results showed laminar-specific fMRI adaptation (i.e. decreased fMRI responses for adaptation compared to non-adaptation) in visual areas (Figure 3A). In particular, a repeated measures ANOVA showed significant main effects of ROI (V1, V2, V3, V4; F(3,39)=23.276, p=0.001), cortical depth (deeper, middle, superficial; F(2,26)=4.942, p=0.034), and condition (adaptation non-adaptation; F(1,13)=11.872, p=0.004). There was no significant ROI × condition × cortical depth interaction (F(6,78)=1.949, p=0.131), suggesting similar fMRI adaptation across visual areas. A significant condition × cortical depth interaction (F(2,26)=9.506, p=0.002) indicated stronger fMRI adaptation in superficial and middle than deeper layers. We observed a similar pattern of results when we controlled for signal contribution from voxels at the border of adjacent layers, using a spatial regression analysis (Kok et al., 2016; Koster et al., 2018; Markuerkiaga et al., 2016). To unmix the signal, we regressed out the time course of voxels assigned to middle layers and adjacent to the superficial layers from the time course of voxels assigned to superficial layers. We applied the same approach to voxels assigned to the deeper layers and adjacent to the middle layers. We observed stronger fMRI adaptation in superficial layers across visual areas following this correction (Figure 3A), as indicated by a significant condition × cortical depth interaction: F(2,26)=5.996, p=0.022.

Laminar BOLD and fMRI adaptation index for V1, V2, V3, and V4.
(A) Mean BOLD across V1, V2, V3, and V4 cortical layers. Bar-plot shows z-scored BOLD signal from fixation baseline for adaptation (grey) and non-adaptation (white) across cortical layers of V1, V2, V3, and V4. Error bars indicate within-subject confidence intervals (Cousineau, 2005) (N = 15 for V1, V2, and V3; N = 14 for V4). (B) fMRI adaptation index across cortical layers (D: deeper, M: middle, S: superficial) for V1, V2, V3, and V4. Bar-plots show difference in z-scored BOLD signal between the non-adaptation and the adaptation conditions. Error bars indicate within-subject confidence intervals (Cousineau, 2005) (N = 15 for V1, V2, and V3; N = 14 for V4). Stars indicate statistically significant comparisons for p<0.05.
-
Figure 3—source data 1
Tables for mean BOLD responses to adaptation, non-adaptation, and adaptation index across cortical layers of areas V1, V2, V3, and V4.
- https://cdn.elifesciences.org/articles/57637/elife-57637-fig3-data1-v2.xlsx
To further quantify fMRI adaptation across cortical depths, we computed an fMRI adaptation index (i.e. fMRI responses for non-adaptation minus adaptation) per ROI and cortical depth (Figure 3B). A repeated-measures ANOVA showed a significant main effect of cortical depth (F(2,26)=9.506, p=0.002), ROI (F(3,39)=5.858, p=0.003), and no significant ROI × cortical depth interaction (F(6,78)=1.949, p=0.131). Post-hoc comparisons showed significantly stronger fMRI adaptation across visual areas in superficial compared to deeper layers (V1: t(14)=−2.556, p=0.023; V3: t(14)=−2.580, p=0.022; V4: t(13)=−2.091, p=0.012; with the exception of V2: t(14)=−0.881, p=0.393) and middle compared to deeper layers (V1: t(14)=−2.429, p=0.029; V2: t(14)=−2.524, p=0.024; V3: t(14)=−3.528, p=0.003; V4: t(13)=−2.519, p=0.026). No significant differences were observed between fMRI adaptation in superficial and middle cortical layers across visual areas (V1: t(14)=−2.093, p=0.055; V2: t(14)=0.331, p=0.746; V3: t(14)=0.942, p=0.362; V4: t(13)=−2.096, p=0.056).
Control analyses
To control for potential confounds due to the contribution of vasculature-related signals to BOLD, we conducted the following additional analyses.
First, it is known that BOLD measured by GE-EPI is higher at the cortical surface due to vascular contributions (Uğurbil et al., 2003; Uludağ et al., 2009; Yacoub et al., 2005). To ensure that the fMRI adaptation we observed in superficial layers was not confounded by this superficial bias, we identified and removed voxels with low temporal signal-to-noise ratio (tSNR) and high t-statistic for stimulation contrast (see Materials and methods: ROIs analysis). Figure 2F shows that following these corrections the superficial bias in the GE-EPI acquired BOLD was significantly reduced. That is, the magnitude and variance of GE-EPI BOLD signals from voxels closer to the pial surface were reduced, as indicated by a significant interaction between GE-EPI acquired BOLD signal from different cortical depths (deeper, middle, superficial) before vs. after correction (F(2,28)=58.556, p<0.0001). That is, the superficial bias corrections resulted in decreased BOLD signal mainly in middle and superficial layers as indicated by post-hoc comparisons (middle: t = 7.992, p<0.0001; superficial: t = 11.241, p<0.0001).
Second, we scanned a subset of participants (N = 5) with a 3D GRASE sequence that is known to be sensitive to signals from small vessels and less affected by larger veins, resulting in higher spatial specificity of the measured BOLD signal (e.g. De Martino et al., 2013; Kemper et al., 2015). Consistent with previous studies (De Martino et al., 2013), the 3D GRASE data showed: (a) overall lower BOLD signal in V1 compared to the GE-EPI acquired BOLD data and (b) similar BOLD amplitude across V1 cortical depths. Figure 2F shows that the corrected GE-EPI BOLD signal in V1 follows a similar pattern across cortical depth as the 3D GRASE BOLD. In particular there were no significant differences in BOLD acquired with 3D GRASE vs. the corrected GE-EPI BOLD signal across cortical depths (i.e. no significant sequence × cortical depth interaction: F(2,6)=2.878, p=0.187), suggesting that our superficial bias corrections reduced substantially the contribution of vasculature-related signals in GE-EPI measurements. The small sample size does not allow further statistical analyses of the 3D GRASE data; yet, comparing the 3D GRASE and GE-EPI data provides a reproducibility test across MRI sequences. We observed similar fMRI adaptation patterns between sequences (Figure 3—figure supplement 2), suggesting that fMRI adaptation in superficial layers could not be simply attributed to vasculature-related confounds.
Taken together our results demonstrate fMRI adaptation across cortical depths of the visual cortex with stronger effects in superficial and middle than deeper layers. Comparing performance on the RSVP task during scanning across conditions showed that it is unlikely that these fMRI adaptation effects were due to differences in attention, as the RSVP task was similarly difficult across conditions. In particular, the mean performance across participants (adaptation condition: 62.7 ± 0.3%; non-adaptation condition 60.15 ± 0.4%, SEM) did not differ significantly between conditions (t(12)=0.312, p=0.76). Finally, we asked whether this layer-specific fMRI adaptation relates to perceptual bias in orientation discrimination due to adaptation (Figure 3—figure supplement 1), as measured by a tilt-aftereffect paradigm (e.g. Gibson and Radner, 1937). Correlating a perceptual adaptation index (i.e. difference in the perceived orientation of a test grating between adaptation and non-adaptation conditions) and fMRI adaptation (i.e. mean fMRI adaptation index across V1, V2, V3, and V4) showed a significant correlation for superficial (r = 0.656, p=0.039) but not middle (r = 0.562, p=0.091), nor deeper (r = 0.567, p=0.087) layers. These results suggest that laminar-specific fMRI adaptation in visual cortex relates to perceptual bias in orientation discrimination due to adaptation.
fMRI adaptation in intraparietal cortex
We next tested for adaptive processing in posterior parietal cortex regions (IPS1 IPS2; Benson et al., 2014; Benson et al., 2012; Wang et al., 2015) that have been shown to be involved in processing expectation due to stimulus familiarity (de Lange et al., 2018; Garrido et al., 2009; Li et al., 2010; Summerfield and de Lange, 2014). A repeated measures ANOVA (ROI [IPS1 IPS2], condition [adaptation non-adaptation], and cortical depth [deeper, middle, superficial]) showed a significant main effect of condition (i.e. decreased fMRI responses for adaptation compared to non-adaptation [F(1,14)=7.994, p=0.013]) and cortical depth (F(2,28)=25.824, p<0.0001). We did not observe any significant interactions between condition and cortical depth (F(2,28)=0.575, p=0.511), nor between ROI, condition, and cortical depth (F(2,28)=0.639, p=0.510), suggesting similar fMRI adaptation effects across cortical layers (Figure 4), rather than layer-specific fMRI adaptation, in posterior parietal cortex. Spatial regression analysis showed similar pattern of results (main effect of condition: F(1,14)=6.149, p<0.05, cortical layer: F(2,28)=22.359, p<0.0001), suggesting that our results were unlikely to be due to vasculature-related confounds.

Laminar BOLD for IPS1 and IPS2.
Mean BOLD in IPS1 and IPS2 across cortical layers. Bar-plots show z-scored BOLD signal for adaptation (grey) and non-adaptation (white) across cortical layers of IPS1 and IPS2. Error bars indicate within-subject confidence intervals (Cousineau, 2005) (N = 15).
-
Figure 4—source data 1
Tables for mean BOLD responses to adaptation and non-adaptation across cortical layers of IPS1 and IPS2.
- https://cdn.elifesciences.org/articles/57637/elife-57637-fig4-data1-v2.xlsx
Functional connectivity
UHF fMRI allows us to interrogate the finer functional connectivity across areas based on known anatomical models of connectivity across cortical layers. Recent work (for review: Lawrence et al., 2019a) has proposed that anatomical connections between superficial V1 layers and middle layers of higher areas relate to feedforward processing, while anatomical connections between deeper V1 layers and deeper layers of higher areas relate to feedback processing (Figure 1A). We tested functional connectivity in these circuits to discern feedforward vs. feedback processing for orientation-specific adaptation. We did not test connectivity between superficial V1 layers and deeper layers of higher areas, as these connections are known to relate to both feedback and feedforward processing (Maunsell and van Essen, 1983; Rockland and Virga, 1989). Despite the fact that the UHF imaging resolution does not support one-to-one mapping between MRI-defined cortical depths and cyto-architectonically defined layers, Figure 1A provides a framework of feedback vs. feedforward connections across superficial, middle, and deeper cortical depths, as proposed and tested by previous UHF imaging studies (e.g. Huber et al., 2017; Kok et al., 2016; Moerel et al., 2020; Sharoh et al., 2019).
Using this framework, we computed functional connectivity within visual cortex and between visual and posterior parietal cortex. We used independent component analysis (ICA)-based denoising and finite impulse response (FIR) functions to denoise and deconvolve the fMRI time course data per cortical depth, controlling for noise and potential task-timing confounds. We then conducted Pearson correlations between the fMRI eigenvariate time courses across cortical depths. Our results (Figure 5) showed stronger feedforward connectivity for adaptation within visual cortex (i.e. V1 superficial layers and middle layers of higher visual areas), stronger short-range feedback connectivity between V2 and V1 deeper layers, and stronger long-range feedback occipito-parietal connectivity (i.e. V1 deeper layers and IPS1). In particular, a repeated measures ANOVA showed a significant three-way interaction (F(4,52)=3.574, p=0.027) between connections (V1–V2, V1–V3, V1–V4, V1–IPS1, V1–IPS2), pathways (feedforward feedback), and condition (adaptation non-adaptation). Further, we tested whether the difference in connectivity between adaptation and non-adaptation (i.e. difference in Fisher z-transformed values) differed between pathways (feedforward feedback) and across connections (V1–V2, V1–V3, V1–V4, V1–IPS1, V1–IPS2). A repeated measures ANOVA showed a significant pathway (feedforward feedback) × connection (V1–V2, V1–V3, V1–V4, V1–IPS1, V1–IPS2) interaction (F(4,52)=3.945, p=0.022).

Functional connectivity.
Bar-plots show difference (adaptation minus non-adaptation) in Fisher z-transformed r values for connectivity between V1 and V2, V1 and V3, V1 and V4, and V1 and IPS1. Feedforward connections were tested between: (a) V1 superficial and V2, V3, V4 middle layers (b) V1 superficial and IPS1. Feedback connections were tested between: (a) V1 deeper and V2, V3, V4 deeper layers (b) V1 deeper layers and IPS1. Error bars indicate within-subject confidence intervals (Cousineau, 2005) (N = 15 for V1, V2, V3, and IPS1; N = 14 for V4). Stars indicate statistically significant comparisons for p<0.05.
-
Figure 5—source data 1
Tables for cortical depth dependent values of feedforward and feedback functional connectivity.
- https://cdn.elifesciences.org/articles/57637/elife-57637-fig5-data1-v2.xlsx
For functional connectivity within visual cortex, post-hoc comparisons showed significantly higher connectivity between V1 superficial and middle layers of higher visual areas for adaptation compared to non-adaptation (V1–V2: t(14)=2.324, p=0.036; V1–V3: t(14)=2.778, p=0.016; V1–V4: t(13)=2.778, p=0.0157), suggesting enhanced feedforward processing for adaptation within visual cortex. In contrast, no significant differences between conditions were observed in functional connectivity between deeper layers in V1 and higher visual areas (V3: t(14)=0.703, p=0.494; V4: t(13)=0.813, p=0.431) with the exception of V1–V2 connectivity (t(14)=2.223, p=0.043), suggesting enhanced short-range feedback connectivity between V2 and V1 deeper layers for adaptation.
For occipito-parietal connectivity, we tested differences in connectivity between V1 layers and IPS subregions (IPS1 IPS2) as there were no significant differences in fMRI adaptation across IPS layers (F(2,28)=0.575, p=0.511). Our results showed significantly higher functional connectivity between V1 deeper layers and IPS1 for adaptation compared to the non-adaptation (t(14)=3.014, p=0.009). This result is consistent with fMRI adaptation in V1 deeper layers (t(14)=3.438, p=0.004) and suggests enhanced feedback processing for visual adaptation. In contrast, no significant differences between conditions were observed for functional connectivity between: (a) V1 superficial layers and IPS1 (i.e. functional connectivity related to feedforward processing) (t(14)=1.242, p=0.235), (b) V1 and IPS2 (V1 deeper layers and IPS2: t(14)=1. 243, p=0.234; V1 superficial layers and IPS2: t(14)=0.629, p=0.539), and (c) extrastriate visual areas and IPS regions (all p>0.146), suggesting specificity of feedback connectivity for adaptation between V1 and IPS1. Further, a regression analysis showed a significant relationship (r = 0.715, p=0.003) between fMRI adaptation (i.e. difference in z-scored BOLD between adaptation and non-adaptation) in deeper layers of V1 and deeper layers of IPS1. These results suggest that fMRI adaptation in deeper layers of IPS1 predicts fMRI adaptation in deeper layers of V1, consistent with enhanced feedback occipito-parietal connectivity for adaptive processing (i.e. top-down influences from IPS1 to processing of visual information in deeper V1 layers).
Finally, to test whether our results were specific to functional connectivity within the visual cortex and between V1 and IPS, we tested connectivity with hMT+ as a control region that is known to be anatomically connected to V1 (Felleman and Van Essen, 1991) but its function relates to motion rather than orientation processing. We defined hMT+ per participant using a probabilistic atlas (Rosenke et al., 2020) and segmented three cortical layers (superficial, middle, and deeper) (see Methods, MRI data analysis: Segmentation and cortical depth sampling). Analysing BOLD responses across cortical depths did not show significant fMRI adaptation. A repeated measures ANOVA did not show a significant main effect of condition (F(1,14)=3.874, p=0.069), cortical depth (F(2,28)=2.898, p=0.101), nor a significant cortical depth × condition interaction (F(2,28)=2.392, p=0.138). We then conducted functional connectivity analyses across cortical depths of V1 and hMT+. That is, we tested for differences between conditions (adaptation vs. non-adaptation) in functional connectivity between: (a) V1 superficial layers and hMT+ middle layers (i.e. feedforward connectivity) and (b) V1 and hMT+ deeper layers (i.e. feedback connectivity). A two-way repeated measures ANOVA did not show any significant differences in these functional connectivity pathways between conditions (i.e. no significant pathway × condition interaction: F(1,14)=0.475, p=0.502). These results suggest that changes in functional connectivity due to adaptation are specific to connectivity within visual cortex (early and extrastriate areas) and between V1 and IPS.
Discussion
Here, we exploit UHF laminar fMRI to interrogate adaptive processing across cortical depth at a finer scale than afforded by standard fMRI methods. Previous studies have focussed on fMRI adaptation as a tool for interrogating selectivity at the level of large-scale neural populations for a given stimulus dimension. This is typically measured by recording fMRI responses to a test stimulus following the repeated presentation of stimuli that have similar or different dimensions to the test stimulus (e.g. Engel, 2005; Fang et al., 2007; Fang et al., 2005). In contrast, our study interrogates the mechanisms underlying adaptive processing, as a signature of short-term sensory plasticity, by measuring fMRI responses during stimulus repetition rather than responses to a test stimulus following adaptation. Employing this paradigm in combination with UHF laminar fMRI allows us to interrogate fMRI responses during stimulus repetition at a sub-millimetre resolution to gain insights into the circuit processes (feedforward vs. feedback) underlying adaptive processing in the human brain.
Our results demonstrate that visual adaptation involves recurrent processing of orientation information in visual cortex, as indicated by orientation-specific fMRI adaptation (i.e. BOLD decreases due to stimulus repetition) across cortical depths with stronger effects in superficial and middle than deeper layers. Our findings are consistent with a recent study (Ge et al., 2020) showing stronger fMRI responses in superficial V1 layers for visual adaptation in the context of the Flash Grab aftereffect. Our study extends beyond this finding to provide further insights into the functional circuit underlying visual adaptation. In particular, we provide evidence for distinct functional connectivity mechanisms for adaptive processing: feedforward connectivity within the visual cortex, indicating inherited adaptation from early to higher visual areas (e.g. Larsson et al., 2016; Solomon and Kohn, 2014), short-range feedback connectivity from V2 to V1, and long-range feedback connectivity from posterior parietal cortex to V1, reflecting top-down influences (i.e. expectation of repeated stimuli) on visual processing. Further, we demonstrate that orientation-specific fMRI adaptation in superficial layers of visual areas relates to perceptual bias in orientation discrimination due to adaptation, suggesting a link between recurrent adaptive processing and behaviour.
We interpret our results within a framework of feedback vs. feedforward connectivity across cortical depths (Figure 1A), as proposed by previous UHF imaging studies (e.g. Huber et al., 2017; Kok et al., 2016; Moerel et al., 2020; Sharoh et al., 2019). In particular, sensory inputs are known to enter the cortex at the level of the middle layer (middle layer 4) and output information is fed forward through the superficial layer (superficial layer 2/3). In contrast, feedback information is thought to be exchanged mainly between deeper layers (deep layer 5/6) (Larkum, 2013; Markov et al., 2014), as well as superficial layers (Rockland and Virga, 1989).
Neurophysiological studies have shown that this micro-circuit is involved in a range of visual recognition (Self et al., 2013; van Kerkoerle et al., 2014) and attention (Buffalo et al., 2011) tasks. Recent laminar fMRI studies provide evidence for the involvement of this circuit in the context of sensory processing (De Martino et al., 2015) and visual attention (Fracasso et al., 2016; Lawrence et al., 2019b; Scheeringa et al., 2016). Our results show fMRI adaptation across layers in visual cortex with stronger effects in superficial and middle than deeper layers. It is likely that adaptation alters processing of orientation information in middle layers that is then forwarded to superficial layers, consistent with recurrent processing for sensory adaptation. These signals are then forwarded to higher visual areas, as indicated by increased functional connectivity between V1 and higher visual areas. These results are consistent with previous neurophysiological studies showing that sensory adaptation is a fast form of plasticity (Gutnisky and Dragoi, 2008; Whitmire and Stanley, 2016; Xiang and Brown, 1998) and brain imaging studies showing that adapted BOLD responses in higher visual areas are inherited from downstream processing in V1 (Ashida et al., 2012; Larsson et al., 2016).
It is possible that orientation-specific adaptation is implemented in visual cortex via recurrent processing of signals across V1 columns (Self et al., 2013). Horizontal connections across V1 columns are known to mediate iso-orientation inhibition (Malach et al., 1993) that is suppression of neurons that are selective for the same orientation across columns. Iso-orientation inhibition is shown to be more pronounced in superficial layers and support orientation tuning (Rockland and Pandya, 1979). In particular, previous work has shown that horizontal connections between V1 columns primarily terminate in middle and superficial layers (Rockland and Pandya, 1979) and pyramidal cells in superficial layers make extensive arborisations within the same layer (Douglas and Martin, 2007). Consistent with this interpretation, previous neurophysiological studies have shown stronger decrease in neural population responses due to stimulus repetition in superficial layers of V1, while delayed adaptation effects in middle and deeper levels (Westerberg et al., 2019).
An alternative explanation is that BOLD effects in superficial layers reflect feedback processing (e.g. Gau et al., 2020; Muckli et al., 2015). Previous work has shown that synaptic input to superficial layers may result due to increase in feedback signals carried by neurons that have dendrites projecting to the superficial layers and their cell bodies in deeper layers (Larkum, 2013). Our results showing increased functional connectivity between V1 and V2 deeper layers suggest that short-range feedback from V2 contributes to orientation processing in V1, consistent with recurrent processing within visual cortex. Further, our results showing fMRI adaptation in deeper layers in V1 and increased functional connectivity between IPS and deeper V1 layers suggest that long-range feedback from the posterior parietal cortex contributes to adaptive processing in V1, consistent with the role of parietal cortex in expectation and prediction due to stimulus repetition. Recent fMRI studies focussing on higher visual areas have investigated the role of expectation in repetition suppression that – similar to sensory adaptation for simple stimulus features in early visual areas – is characterised by decreased BOLD responses to more complex stimuli (i.e. faces objects) in higher visual areas (Grill-Spector et al., 2006). In particular, Summerfield et al., 2008 showed stronger repetition suppression in the lateral occipito-temporal cortex for identical stimulus pairs that were repeated frequently, providing evidence for a role of top-down influences (i.e. expectation) in repetition suppression and visual processing.
A framework for linking adaptive processing within visual cortex and feedback repetition suppression mechanisms due to expectation is proposed by the predictive coding theory (Friston, 2005; Rao and Ballard, 1999; Shipp, 2016). According to this framework, perception results from comparing feedback expectation and prediction signals in upstream regions with feedforward signals in sensory areas. When these signals match, the error (i.e. the difference between the prediction fed back and the incoming sensory input) is low; in contrast, when the expectation does not match with the sensory input, the prediction error is high resulting in increased neural responses for unexpected compared to expected (i.e. repeated stimuli). Bastos et al., 2012 have proposed a microcircuit model of predictive coding that combines excitatory and inhibitory properties of pyramidal neurons across cortical layers to account for prediction encoding, prediction errors, and modulation of incoming sensory inputs to minimise prediction error. Considering our findings in light of this model provides insights in understanding the circuit underlying adaptive processing in the human brain. It is likely that long-range top-down information (e.g. expectation signals from posterior parietal cortex) is fed back to the deeper layers of V1 and it is then compared with information available at the superficial layers (i.e. iso-orientation inhibition). A mismatch (i.e. prediction error) of signals (i.e. expectation of a repeated stimulus compared to the presentation of an unexpected stimulus) results in decreased fMRI responses for expected compared to unexpected stimuli. This is consistent with previous laminar imaging studies showing stronger fMRI responses in deeper or superficial V1 layers for perceptual completion tasks and suggesting top-down influences in visual processing (Kok et al., 2016; Muckli et al., 2015).
It is important to note that despite the advances afforded by UHF imaging, GE-EPI remains limited by vasculature-related signals contributing to BOLD at the cortical surface, resulting in loss of spatial specificity (Kay et al., 2019). To reduce this superficial bias, we removed voxels with low tSNR (Olman et al., 2007) and high t-statistic for stimulation contrast (Kashyap et al., 2018; Polimeni et al., 2010). Further, we applied a signal unmixing method (Kok et al., 2016; Koster et al., 2018) to control for draining vein effects from deep to middle and middle to superficial layers. We compared BOLD signals across conditions (adaptation vs. non-adaptation) and cortical depths after z-scoring the signals within each cortical depth to account for possible differences in signal strength across cortical layers (Goense et al., 2012; Havlicek and Uludağ, 2020). Following these corrections, we observed stronger fMRI adaptation (i.e. stronger BOLD response for non-adapted than adapted stimuli) in superficial layers, suggesting that our results are unlikely to be confounded by vasculature-related superficial bias. Further, using a 3D GRASE sequence that measures BOLD signals that are less affected by macro-vascular contribution showed similar results of layer-specific fMRI adaptation. Our findings on orientation-specific adaptation in superficial layers are consistent with previous laminar imaging studies showing BOLD effects in superficial layers in a range of tasks (De Martino et al., 2015; Olman et al., 2012). Recent advances in cerebral blood volume (CBV) imaging using vascular space occupancy (VASO) (e.g. Beckett et al., 2020; Huber et al., 2019) could be exploited in future studies to enhance the spatial specificity of laminar imaging in the human brain.
In sum, exploiting UHF imaging, we provide evidence that adaptive processing in the human brain engages a circuit that integrates recurrent processing within visual cortex with top-down influences (i.e. stimulus expectation) from posterior parietal cortex via feedback. This circuit of local recurrent and feedback influences is critical for rapid brain plasticity that supports efficient sensory processing by suppressing familiar and expected information to facilitate resource allocation to new incoming input. Combining laminar imaging with electrophysiological recordings has the potential to shed more light on the dynamics of this circuit, consistent with recent evidence (Buffalo et al., 2011; Self et al., 2013; van Kerkoerle et al., 2014) that gamma oscillations are linked to feedforward processing in input layers, while alpha/beta oscillations are related to feedback mechanisms in superficial and deeper cortical layers. Understanding these circuit dynamics is the next key challenge in deciphering the fast brain plasticity mechanisms that support adaptive processing in the human brain.
Materials and methods
Participants
Eighteen healthy volunteers (11 females and 7 males) participated in the study. Seventeen participants were scanned with a Gradient Echo-Echo Planar Imaging (GE-EPI) sequence (main experiment). Due to the lack of previous 7T fMRI adaptation studies, we determined sample size based on power calculations following a 3T fMRI study from our lab using the same paradigm (Karlaftis et al., 2019) that showed fMRI adaptation for effect size of Cohen’s f2 = 0.396 at 80% power. Data from two participants were excluded from further analysis due to excessive head movement (higher than 1.5 mm) and technical problems during acquisition, resulting in data from 15 participants for the main experiment (mean age: 24.44 years and SD: 3.83 years). Five participants (four who participated in the main experiment and an additional participant) were scanned with a 3D GRASE EPI sequence. Twelve of the participants who took part in the fMRI experiment completed an additional psychophysical experiment. All participants had normal or corrected-to-normal vision, gave written informed consent, and received payment for their participation. The study was approved by the local Ethical Committee of the Faculty of Psychology and Neuroscience at Maastricht University and the University of Cambridge Ethics Committee (ethics number PRE2017.057).
Stimuli
Request a detailed protocolStimuli comprised sinewave gratings (one cycle per degree) of varying orientations (Figure 1B). Stimuli were presented centrally within an annulus aperture (inner radius: 0.21°; outer radius: 6°). The outer edge of the aperture was smoothed using a sinusoidal function (standard deviation: 0.6°). Experiments were controlled using MATLAB and the Psychophysics toolbox 3.0 (Brainard, 1997; Pelli, 1997). For the main fMRI experiment, stimuli were presented using a projector and a mirror setup (1920 × 1080 pixels resolution, 60 Hz frame rate) at a viewing distance of 99 cm. The viewing distance was reduced to 70 cm for the control experiment, as a different coil was used, and adjusted so that angular stimulus size was the same for both scanning sessions.
Experimental design
fMRI session
Request a detailed protocolBoth the main and control fMRI experiments comprised a maximum of eight runs (13 participants completed eight runs for each experiment; two participants in the main experiment and one participant in the control completed six runs). Each run lasted 5 min 6 s, and started with 14 s fixation, followed by six stimulus blocks, three blocks per condition (adaptation non-adaptation) and ended with 14 s fixation. The order of the blocks was counterbalanced within and across runs. Each block comprised 16 stimuli followed by 2 s for response to the RSVP task. The orientation of the gratings was drawn randomly from uniform distributions, ranging from −85° to −5° and +5° to +85° in steps of 7.27°, excluding vertical (i.e. 0°). The same orientation was presented across adaptation blocks per participant. Sixteen different orientations were presented per block for the non-adaptation condition. Each stimulus was displayed for 1900 ms with a 100 ms inter-stimulus interval for both the adaptation and non-adaptation conditions to ensure similar stimulus presentation parameters (e.g. stimulus transients) between conditions. During scanning participants were asked to perform an RSVP task. A stream of letters was presented in rapid serial order (presentation frequency: 150 ms, asynchronous to the timings of grating presentation) within an annulus at the centre of the screen (0.5° of visual angle). Participants were asked to fixate at the annulus and report, by a key press at the end of each block, the number of targets (one to four per block). No feedback was provided to the participants.
In the same scanning session, anatomical data and fMRI data for retinotopic mapping were collected following standard procedures (e.g. Engel et al., 1997).
MRI acquisition
Request a detailed protocolImaging data were acquired on a 7T Magnetom scanner (Siemens Medical System, Erlangen, Germany) at the Scannexus Imaging Centre, Maastricht, The Netherlands. Anatomical data were acquired using an MP2RAGE sequence (TR = 5 s, TE = 2.51 ms, FOV = 208×208 mm, 240 sagittal slices, 0.65 mm isotropic voxel resolution).
For the main experiment (N = 17), we used a 32-channel phased-array head coil (NOVA Medical, Wilmington, MA, USA) and a 2D Gradient Echo, Echo Planar Imaging (GE-EPI) sequence (TE = 25 ms, TR = 2 s, voxel size = 0.8 mm isotropic, FOV = 148×148 mm, number of slices = 56, partial Fourier = 6/8, GRAPPA factor = 3, Multi-Band factor = 2, bandwidth = 1168 Hz/Pixel, echo spacing = 1 ms, flip angle = 70°). The field of view covered occipito-temporal and posterior parietal areas; manual shimming was performed prior to the acquisition of the functional data.
For the control experiment (N = 5), participants were scanned with a 3D inner-volume gradient and spin echo (GRASE) sequence with variable flip angles (Feinberg and Oshio, 1991; Kemper et al., 2016). This sequence is largely based on a spin echo sequence for which the measured T2-weighted BOLD signal has higher spatial specificity and is less confounded by large draining veins near the pial surface (e.g. Duong et al., 2003; Goense et al., 2007; Kemper et al., 2015; Uludağ et al., 2009). We used a custom-built surface-array coil (Sengupta et al., 2016) for enhanced SNR of high-resolution imaging of visual cortex (TR = 2 s, TE = 35.41 ms, FOV = 128×24 mm, number of slices = 12, echo-spacing = 1.01 ms, total readout train time = 363.6, voxel size = 0.8 mm isotropic, 90° nominal excitation flip angle, and variable refocussing flip angles ranging between 47° and 95°). The latter was used to exploit the slower decay of the stimulated echo pathway and hence to keep T2-decay-induced blurring in partition-encoding direction at a small, acceptable level, that is, comparable to the T2*-induced blurring in typical EPI acquisition protocols for functional imaging (Kemper et al., 2016).
MRI data analysis
Segmentation and cortical depth sampling
Request a detailed protocolT1-weighted anatomical data was used for coregistration and 3D cortex reconstruction. Grey and white matter segmentation was performed on the MP2RAGE images using FreeSurfer (http://surfer.nmr.mgh.harvard.edu/) and manually improved for the ROIs (i.e. V1, V2, V3, V4, IPS, and hMT+) using ITK-Snap (http://www.itksnap.org/pmwiki/pmwiki.php, Yushkevich et al., 2006). The refined segmentation was used to obtain a measurement of cortical thickness. Following previous studies, we assigned voxels in three layers (deeper, middle, and superficial) using the equi-volume approach (Kemper et al., 2018; Waehnert et al., 2014) as implemented in BrainVoyager (Brain Innovation, Maastricht, The Netherlands). This approach has been shown to reduce misclassification of voxels to layers, in particular for ROIs presenting high curvature. Information from the cortical thickness map and gradient curvature was used to generate four grids at different cortical depths (ranging from 0, white matter, to 1, grey matter). Mapping of each voxel to a layer was obtained by computing the Euclidean distance of each grey matter voxel to the grids: the two closest grids represent the borders of the layer a voxel is assigned to (Figure 2D). Note that due to limitations in the UHF imaging resolution these MRI-defined layers indicate distance (i.e. cortical depth) from the grey matter/white matter and the grey matter/cerebrospinal fluid boundaries rather than one-to-one mapping to the cyto-architectonically defined layers of the human neocortex.
For the 3D GRASE control experiment, we used the LAYNII tools (Huber et al., 2020), as they provided better segmentation for images with a limited field of view.
GE-EPI functional data analysis
Request a detailed protocolThe GE-EPI functional data were analysed using BrainVoyager (version 20.6, Brain Innovation, Maastricht, The Netherlands) and custom MATLAB (The MATHWORKS Inc, Natick, MA, USA) code. Preprocessing of the functional data involved three serial steps starting with correction of distortions due to non-zero off-resonance field; that is, at the beginning of each functional run, five volumes with inverted phase encoding direction were acquired and used to estimate a voxel displacement map that was subsequently applied to the functional data using COPE, BrainVoyager, Brain Innovation. The undistorted data underwent slice-timing correction, head motion correction (the single band image of each run was used as reference for the alignment), high-pass temporal filtering (using a GLM with Fourier basis set at two cycles), and removal of linear trends. Preprocessed functional data were coaligned to the anatomical data using a boundary-based registration approach, as implemented in BrainVoyager (Brain Innovation, Maastricht, The Netherlands). Results were manually inspected and further adjusted where needed. To validate the alignment of functional to anatomical data, we calculated the mean EPI image of each functional run for each ROI and estimated the spatial correlation between these images (e.g. Marquardt et al., 2018). We performed manual adjustment of the alignment if the spatial correlation was below 0.85. We excluded a small number of runs (N = 3 and N = 1 for two participants respectively), as their alignment could not be improved manually.
3D GRASE functional data analysis
Request a detailed protocolFunctional images were analysed using BrainVoyager (version 21.0, Brain Innovation, Maastricht, The Netherlands), custom MATLAB (The MATHWORKS Inc, Natick, MA, USA) code, and advanced normalisation tools (Avants et al., 2011) for registration of images. The first volume of each run was removed to allow for the magnetisation to reach a steady state. Head motion correction was performed using as reference the first image (10 volumes with TR = 6 s) acquired at the beginning of the functional runs. The higher contrast of this image facilitated the coregistration of the anatomical and functional images. After motion correction, temporal high-pass filtering was applied, using a GLM with Fourier basis set at three cycles per run. Preprocessed images were converted into Nifti files and an initial manual registration was performed between the first image and the anatomical image using the manual registration tool provided in ITK-Snap (http://www.itksnap.org/pmwiki/pmwiki.php, Yushkevich et al., 2006). The resulting transformation matrix was applied to coregister the anatomical image to the functional space and fine-tuned adjustments were provided by means of antsRegistration tools.
ROIs analysis
Request a detailed protocolWe used the data from the retinotopic mapping scan to identify ROIs. For each participant, we defined areas V1–V4 based on standard phase-encoding methods. Participants viewed rotating wedges that created travelling waves of neural activity (e.g. Engel et al., 1997). Due to limited coverage during acquisition, it was not possible to map area V4 in 1 of the 15 participants. Due to limited scanning time, it was not possible to perform an additional localiser scan to functionally identify posterior parietal cortex regions (comprising IPS1 and IPS2). We identified these regions based on a probabilistic atlas (https://hub.docker.com/r/nben/occipital_atlas/; Benson et al., 2014; Benson et al., 2012; Wang et al., 2015). This atlas defines IPS regions based on functional – rather than anatomical only – criteria based on previous work showing that these regions are involved in saccadic eye movements, spatial attention, and memory (e.g. Schluppeck et al., 2005; Sereno et al., 2001; Silver and Kastner, 2009; Wang et al., 2015). Finally, we defined hMT+ as a control ROI based on a functional atlas of visual cortex (Rosenke et al., 2020). For both the IPS regions and hMT+, we used the individual participant-based segmentation obtained with FreeSurfer and an anatomical probabilistic template to estimate the best location for the ROI (i.e. IPS). For each participant, we then visually inspected the ROI mapping to ensure good alignment and consistency across participants.
For each ROI and individual participant, we modelled BOLD signals using a block design GLM with two regressors, one per stimulus condition (adaptation = non-adaptation). We included estimated head motion parameters as nuisance regressors. The resulting t-statistical map was thresholded (t = 1.64, p=0.10) to select voxels within each ROI that responded more strongly to the stimulus conditions compared to fixation baseline (Figure 2B and C).
Voxel selection within each ROI was further refined by excluding voxels that were confounded by vasculature effects that are known to contribute to a superficial bias in the measured BOLD signal; that is, increased BOLD with increasing distance from white matter (see Results: Control analyses). In particular, it has been shown that the BOLD signal measured using GE-EPI, T2* weighted sequences is confounded by vasculature-related signals (Uğurbil et al., 2003; Uludağ et al., 2009; Yacoub et al., 2005) due to veins penetrating the grey matter and running through its thickness, as well as large pial veins situated along the surface of the grey matter (Duvernoy et al., 1981). This results in increased sensitivity (i.e. strong BOLD effect) but decreased spatial specificity of the measured signal.
Here, we took the following approach to reduce superficial bias due to vasculature contributions. First, following previous work (Olman et al., 2007), we computed tSNR for each voxel in each ROI (V1, V2, V3, V4, IPS1, and IPS2). We used this signal to identify voxels near large veins that are expected to have large variance and low intensity signal (mean tSNR across V1 smaller than 12.02 ± 2.02) due to the local concentration of deoxygenated haemoglobin resulting in a short T2* decay time. Second, it has been shown that high t-values on an fMRI statistical map are likely to arise from large pial veins (Kashyap et al., 2018; Polimeni et al., 2010). Therefore, voxels with t-score values above the 90th percentile (mean t-score across V1 larger than t = 15.47 ± 4.16) of the t-score distribution obtained by the GLM described above were removed from further analysis.
Further to account for possible differences in signal strength across cortical layers due to thermal and physiological noise, as well as signal gain (Goense et al., 2012; Havlicek and Uludağ, 2020), we (a) matched the number of voxels across cortical depths (i.e. to the layer with the lowest number of voxels) per participant and ROI, and (b) z-scored the time courses within cortical layer per ROI, controlling for differences in signal levels across cortical depths while preserving signal differences across conditions (after correction of vascular contributions, e.g. Lawrence et al., 2019b). Normalised fMRI responses for each condition (adaptation non-adaptation) were averaged across the stimulus presentation (excluding participant responses; 32–34 s after stimulus onset), blocks, and runs for each condition. For visual cortex ROIs, we focussed on the time window that captured the peak of the haemodynamic response to visual stimulus presentation (4–18 s after stimulus onset). To test for layer-specific differences between conditions (adaptation vs. non-adaptation) across cortical depths and brain regions, we conducted repeated measures ANOVAs with condition (adaptation non-adaptation), cortical depth (deeper, middle, superficial layers), and ROIs as factors. Post-hoc comparisons (Bonferroni corrected for multiple comparisons) were used to further test differences in fMRI adaptation following significant ANOVA effects.
Functional connectivity analysis
Request a detailed protocolWe followed standard analyses methods to compute functional connectivity across ROIs and cortical depths. We preprocessed the functional and anatomical data in SPM12.3 (v6906; http://www.fil.ion.ucl.ac.uk/spm/software/spm12/). We first performed brain extraction and normalisation to MNI space on the anatomical images (non-linear). The functional images were then corrected for distortions, slice-scan timing (i.e. to remove time shifts in slice acquisition), head motion (i.e. aligned each run to its single band reference image), coregistered all EPI runs to the first run (rigid body), coregistered the first EPI run to the anatomical image (rigid body), and normalised to MNI space (applying the deformation field of the anatomical images). Data were only resliced after MNI normalisation to minimise the number of interpolation steps.
Next, we used an ICA-based denoising procedure (Griffanti et al., 2014). We applied spatial smoothing (2 mm) and linear detrending, followed by spatial group ICA. The latter was performed using the Group ICA fMRI Toolbox (GIFT v3.0b) (http://mialab.mrn.org/software/gift/). Principal component analysis was applied for dimensionality reduction, first at the subject level, then at the group level. A fixed number (N = 35) of independent components was selected for the ICA estimation. The ICA estimation (Infomax) was run 20 times and the component stability was estimated using ICASSO (Himberg et al., 2004). Group information guided ICA back-reconstruction was used to reconstruct subject-specific components from the group ICA components (Du et al., 2016). The results were visually inspected to identify noise components according to published procedures (Griffanti et al., 2017). We labelled 12 of the 35 components as noise that captured signal from veins, arteries, cerebrospinal fluid pulsation, susceptibility, and multi-band artefacts.
To clean the fMRI signals from signals related to motion and the noise components, we followed the soft cleanup approach (Griffanti et al., 2014) on the BrainVoyager unsmoothed data in native space (see GE-EPI functional data analysis). That is, we first regressed out the motion parameters (translation, rotation, and their squares and derivatives; Friston et al., 1996) from each voxel and ICA component time course. Second, we estimated the contribution of every ICA component to each voxel’s time course (multiple regression). Finally, we subtracted the unique contribution of the noise components from each voxel’s time course to avoid removing any shared signal between neuronal and noise components.
Further, following recent work (Cole et al., 2019), we deconvolved the denoised time courses using FIR functions. In particular, we fitted 23 regressors per condition that covered the duration of each task block, including the response period and fixation block, to capture the whole haemodynamic response. This method allows us to accurately model and remove the cross-block mean response for each condition (adaptation and non-adaptation) to account for potential task-timing confounds that have been shown inflate the strength of the computed task-based functional connectivity. Within the GLM, the data were high-pass filtered at 0.01 Hz and treated for serial autocorrelations using the FAST autoregressive model (Corbin et al., 2018; Olszowy et al., 2019). For each ROI and cortical depth, we then computed the first eigenvariate across all voxels within the region to derive a single representative time course per cortical depth and ROI for connectivity analysis. We computed functional connectivity as the Pearson correlation between the eigenvariate time courses across ROIs and cortical depths. Finally, we performed repeated measures ANOVAs with connection (V1–V2, V1–V3, V1–V4, V1–IPS1, V1–IPS2), pathway (feedforward feedback), and condition (adaptation non-adaptation) as factors to test for differences in the functional connectivity values (after Fisher z-transform) between conditions. Post-hoc comparisons (Bonferroni corrected for multiple comparisons) were used to further test differences in functional connectivity following significant ANOVA effects.
Data availability
Source data have been provided for Figures 3, 4, and 5. Data can also be found on the Cambridge Data repository.
-
Cambridge Data repositoryFine-scale computations for adaptive processing in the human brain.https://doi.org/10.17863/CAM.60330
References
-
Comparison of BOLD and CBV using 3D EPI and 3D GRASE for cortical layer functional MRI at 7 TMagnetic Resonance in Medicine 84:3128–3145.https://doi.org/10.1002/mrm.28347
-
Perceptual adaptation: motion parallels orientationTrends in Cognitive Sciences 6:136–143.https://doi.org/10.1016/S1364-6613(00)01856-8
-
Accurate modeling of temporal correlations in rapidly sampled fMRI time seriesHuman Brain Mapping 39:3884–3897.https://doi.org/10.1002/hbm.24218
-
Confidence intervals in within-subject designs: a simpler solution to Loftus and Masson's methodTutorials in Quantitative Methods for Psychology 1:42–45.https://doi.org/10.20982/tqmp.01.1.p042
-
How do expectations shape perception?Trends in Cognitive Sciences 22:764–779.https://doi.org/10.1016/j.tics.2018.06.002
-
Recurrent neuronal circuits in the neocortexCurrent Biology 17:R496–R500.https://doi.org/10.1016/j.cub.2007.04.024
-
Artifact removal in the context of group ICA: A comparison of single-subject and group approachesHuman Brain Mapping 37:1005–1025.https://doi.org/10.1002/hbm.23086
-
Microvascular BOLD contribution at 4 and 7 T in the human brain: gradient-echo and spin-echo fMRI with suppression of blood effectsMagnetic Resonance in Medicine 49:1019–1027.https://doi.org/10.1002/mrm.10472
-
Cortical blood vessels of the human brainBrain Research Bulletin 7:519–579.https://doi.org/10.1016/0361-9230(81)90007-1
-
Changes in "top-down" connectivity underlie repetition suppression in the ventral visual pathwayJournal of Neuroscience 31:5635–5642.https://doi.org/10.1523/JNEUROSCI.5013-10.2011
-
Orientation-tuned FMRI adaptation in human visual cortexJournal of Neurophysiology 94:4188–4195.https://doi.org/10.1152/jn.00378.2005
-
A theory of cortical responsesPhilosophical Transactions of the Royal Society B: Biological Sciences 360:815–836.https://doi.org/10.1098/rstb.2005.1622
-
The mismatch negativity: a review of underlying mechanismsClinical Neurophysiology 120:453–463.https://doi.org/10.1016/j.clinph.2008.11.029
-
Adaptation, after-effect and contrast in the perception of tilted lines. I. quantitative studiesJournal of Experimental Psychology 20:453–467.https://doi.org/10.1037/h0059826
-
High-resolution fMRI of macaque V1Magnetic Resonance Imaging 25:740–747.https://doi.org/10.1016/j.mri.2007.02.013
-
fMRI at high spatial resolution: implications for BOLD-ModelsFrontiers in Computational Neuroscience 10:66.https://doi.org/10.3389/fncom.2016.00066
-
Repetition and the brain: neural models of stimulus-specific effectsTrends in Cognitive Sciences 10:14–23.https://doi.org/10.1016/j.tics.2005.11.006
-
BookAdaptive Network Connectivity in the Human BrainOrganization for Human Brain Mapping.
-
Sub-millimeter T2 weighted fMRI at 7 T: comparison of 3D-GRASE and 2D SE-EPIFrontiers in Neuroscience 9:163.https://doi.org/10.3389/fnins.2015.00163
-
Variable flip angle 3D-GRASE for high resolution fMRI at 7 teslaMagnetic Resonance in Medicine 76:897–904.https://doi.org/10.1002/mrm.25979
-
Visual adaptation: physiology, mechanisms, and functional benefitsJournal of Neurophysiology 97:3155–3164.https://doi.org/10.1152/jn.00086.2007
-
Adaptation: from single cells to BOLD signalsTrends in Neurosciences 29:250–256.https://doi.org/10.1016/j.tins.2006.02.008
-
Orientation-selective adaptation to first- and second-order patterns in human visual cortexJournal of Neurophysiology 95:862–881.https://doi.org/10.1152/jn.00668.2005
-
Anatomy of hierarchy: feedforward and feedback pathways in macaque visual cortexJournal of Comparative Neurology 522:225–259.https://doi.org/10.1002/cne.23458
-
Cortical depth profiles of luminance contrast responses in human V1 and V2 using 7 T fMRIHuman Brain Mapping 39:2812–2827.https://doi.org/10.1002/hbm.24042
-
Using high spatial resolution fMRI to understand representation in the auditory networkProgress in Neurobiology 1:101887.https://doi.org/10.1016/j.pneurobio.2020.101887
-
Contextual feedback to superficial layers of V1Current Biology 25:2690–2695.https://doi.org/10.1016/j.cub.2015.08.057
-
Accurate autocorrelation modeling substantially improves fMRI reliabilityNature Communications 10:1220.https://doi.org/10.1038/s41467-019-09230-w
-
Topographic organization for delayed saccades in human posterior parietal cortexJournal of Neurophysiology 94:1372–1384.https://doi.org/10.1152/jn.01290.2004
-
Neural elements for predictive codingFrontiers in Psychology 7:1792.https://doi.org/10.3389/fpsyg.2016.01792
-
Topographic maps in human frontal and parietal cortexTrends in Cognitive Sciences 13:488–495.https://doi.org/10.1016/j.tics.2009.08.005
-
Moving sensory adaptation beyond suppressive effects in single neuronsCurrent Biology 24:R1012–R1022.https://doi.org/10.1016/j.cub.2014.09.001
-
Neural repetition suppression reflects fulfilled perceptual expectationsNature Neuroscience 11:1004–1006.https://doi.org/10.1038/nn.2163
-
Expectation in perceptual decision making: neural and computational mechanismsNature Reviews Neuroscience 15:745–756.https://doi.org/10.1038/nrn3838
-
How accurate is magnetic resonance imaging of brain function?Trends in Neurosciences 26:108–114.https://doi.org/10.1016/S0166-2236(02)00039-5
-
Anatomically motivated modeling of cortical laminaeNeuroImage 93 Pt 2:210–220.https://doi.org/10.1016/j.neuroimage.2013.03.078
-
Probabilistic maps of visual topography in human cortexCerebral Cortex 25:3911–3931.https://doi.org/10.1093/cercor/bhu277
Decision letter
-
Ming MengReviewing Editor; South China Normal University, China
-
Chris I BakerSenior Editor; National Institute of Mental Health, National Institutes of Health, United States
In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.
Acceptance summary:
Sensory adaptation reflects short-term brain plasticity that optimizes the efficiency of information processing. The present study uses cutting-edge ultra-high field fMRI to examine cortical layer-specific neural basis for adaptation. Valuable new data with sub-millimeter resolution are provided to advance our understanding of mechanisms supporting adaptive processing in human visual cortex.
Decision letter after peer review:
Thank you for submitting your article "Fine-scale computations for adaptive processing in the human brain" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Chris Baker as the Senior Editor. The reviewers have opted to remain anonymous.
The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.
As the editors have judged that your manuscript is of interest, but as described below that additional experiments are required before it is published, we would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). First, because many researchers have temporarily lost access to the labs, we will give authors as much time as they need to submit revised manuscripts. We are also offering, if you choose, to post the manuscript to bioRxiv (if it is not already there) along with this decision letter and a formal designation that the manuscript is "in revision at eLife". Please let us know if you would like to pursue this option. (If your work is more suitable for medRxiv, you will need to post the preprint yourself, as the mechanisms for us to do so are still in development.)
Summary:
This study investigates the neural mechanisms of adaptation in human visual cortex with ultra-high-field fMRI. The stimuli were gratings that either had the same orientation repeatedly presented (adaptation) or gratings of different orientation repeatedly presented (non-adaptation). Attention was maintained at fixation throughout with a rsvp task. A primary claim is that adaption is stronger in superficial depths in visual areas V1 through V4, but is not modulated with depth in IPS1 and IPS2. Functional connectivity analyses are used to assess the relative strength of feedforward and feedback connections between the regions studied during adaptation, indicating enhanced feedback connectivity from IPS to V1 and enhanced feedforward connectivity from V1 to V2, V3, and V4 during adaptation. The study combines cutting-edge imaging techniques with clear experimental design and careful analysis to make an important and valuable contribution to our understanding of mechanisms supporting adaptive processing in human visual cortex.
Essential revisions:
Assuming that a direct physiological measure (e.g., spikes) for the present study at this time is difficult, some additional psychophysical experiments would be needed to successfully address reviewer #3's first main concern. Adding some psychophysical experiments would also help to address comments from reviewer #2 better than just adding analyses and tempering claims.
Reviewer #1:
1) The study overall did not add much new evidence that may advance our understandings of feedforward and feedback processes in the visual adaptation, and the reported results were to some extent predictable from previous studies (see review, Lawrence et al., 2019; Self et al., 2019).
2) The authors claimed higher functional connectivity between V1 deeper layers and IPS, however, the reported analysis showed that the difference in adaptation and non-adaptation conditions between V1 deeper layers and IPS1 just reached the significance (p =.049), while that in V1 deeper layers and IPS2 was not significant (p =.281). I feel that these results were not strong enough for the authors make the solid conclusion on the significant difference between adaptive conditions. Also, I noticed that the IPS region was defined using anatomical templates, did the authors included functional localization scan for the IPS regions? In addition, the above analyses were between V1 deeper layers and the overall IPS1 and IPS2 respectively. According to the model (see Figure 1 in the manuscript), the feedback connection was between V1 deeper layers and IPS deeper layer, so the functional connectivity analyses should be conducted between the deeper layers of V1 and deeper layers of IPS1/IPS2. Given that the manuscript has reported the significant different neural responses in different layers of IPS1 and IPS2, so I suggest that the authors do additional analyses on the functional connectivity between V1 deeper layer and IPS1/IPS2 deeper layers, and the results would provide more specific and stronger evidence on the feedback connections in visual adaption.
Reviewer #2:
My concerns fall into two broad categories: 1) that the statistical tests employed in this work don't sufficiently support the authors' claims, and 2) that, even when properly analyzed, the data presented here are insufficient evidence for the circuit-level conclusions presented throughout the manuscript. The authors could improve the evidence and present interpretations that are more closely tied to the data.
Conceptual
1) The language used does not adequately clarify the differences (and potential lack of alignment) between cortical depths and cortical layers. For example, the claim that "UHF imaging affords the sub-millimetre resolution necessary to examine fMRI signals across cortical layers" is not strictly true. Even at the small voxel size used here, cortical curvature, variable thickness, and partial volume effects all make it extremely challenging to map cortical depth to physiologically-distinct cortical layers. This distinction is critical given the motivation of this work as dissecting feedforward and feedback projections.
Suggestion: avoid the term "layer" and instead use "depth", and clarify in the manuscript that inferences about cortical layers, and thus, alignment to existing anatomical models of feedforward and feedback projections is limited.
2) It's not immediately clear how one should combine the functional connectivity and adaptation results into a single framework for thinking about mechanisms of adaptation. Is it possible, for example, that other regions (besides IPS1, IPS2, V2, V3, and V4) feedback to V1 and contribute to suppression? Is the strength of the IPS feedback during the adaptation condition predictive of the amount of adaptive suppression? Conducting these additional analyses would strengthen the authors' claim that top-down feedbacks from the IPS contribute to the adaptive processing reported in visual cortex.
Suggestion: (i) Compare the degree of feedforward and feedback connectivity between conditions for a control region, e.g., hMT+, to demonstrate that IPS is uniquely (or especially) involved in these computations. (ii) Demonstrate that the amount of feedback from IPS correlates with the amount of adaptation in deeper cortical depths in V1 across individual subjects, and (iii) Elucidate it this feedback from IPS that is correlated with adaptation is specific to V1 or occurs to other visual areas that show adaptation (V2, V3, V4).
Data Analyses and Validation of Results
1) The reporting of ANOVA results of differences in adaptation across ROIs and depths throughout the paper is confusing. It is unclear whether a) separate ANOVAs were run to test each effect (main or interaction), or b) that the results of the ANOVAs are misreported. For example, consider the ANOVA reported which tests the effects of ROI, depth, and condition on z-scored BOLD responses. The degrees of freedom (3, 39), indicates that there are four levels to the factor of interest (four ROIs), and 40 total measurements. Where does 40 come from? If there are 15 subjects and each contributes 12 data points (4 ROIs x 3 depths), I would expect that the variance being analyzed is that of 15 x 12 = 180 data points. Then, the next result presented is the main effect of condition, which has a reported (1, 13) degrees of freedom, suggesting that a different model was fit.
Suggestion: Run a single ANOVA and report main effects and interaction terms from that analysis.
2) The authors claim that adaptation is stronger in superficial V1 than in other cortical depths in the Introduction but write "visual cortex" instead of "V1" or "primary visual cortex" elsewhere. It is unclear when the authors are claiming that the result applies to V1 and when it applies to V1 through V4 in aggregate. This is problematic for three reasons. First, the claim being made should be clear, and the Introduction and Discussion should reflect the scope (V1 or V1-V4) intended. Second, if the authors claim is that suppression is stronger in superficial V1, a post-hoc test with appropriate multiple comparisons correction is needed. The tests are insufficient in that they don't compare superficial to middle depths directly, and in that they aggregate across all four ROIs instead of testing V1 separately. Third, if the authors claim is that suppression is stronger in superficial V1 through V4, then the direct comparison of superficial to middle depths is still lacking. Furthermore, the framing of the rest of the paper which compares V1 connectivity to IPS and V2-V4 and discusses V1 in the Introduction and Discussion needs to be justified if V1 is no different from V2, V3, and V4.
Suggestion: Make the claim being made clearer, then compute the appropriate statistical tests to support that claim. If needed, revise the Introduction and Discussion to explain special attention paid to each ROI.
3) In relation to point (2) above, they claim that adaptation is layer specific rests on the results of post-hoc tests. However, it is unclear (i) which ROIs are tested, and (ii) why they are comparing superficial and deep as well as middle and deep but not superficial and middle. Further the Materials and methods indicate that pairwise t-tests were used, if so multiple comparisons correction is needed to validate the result.
Suggestion: The authors should conduct all tests and clarify the reporting of the results. If multiple-comparisons correction is not currently being used, the authors should employ either Tukey's Honest Significant Difference, Bonferroni correction, or an alternative correction. If the tests are already corrected, that fact should be reflected in the Materials and methods section.
4) The inclusion of GRASE results strengthens the work substantially. However, the claim that the same adaptation patterns are observed are not supported numerically. In fact, the results presented in Figure 3—figure supplement 2 suggest that adaptation is not stronger in superficial than middle or deep depths, and that the correlation between adaptation indices across scan sequences (panel B) are moderate; at most, the adaptation indices from one scan sequence predicts 22% of the variance in the adaptation indices from the other scan sequence.
Suggestion: Report statistics for the 3D GRASE results.
Reviewer #3:
The motivation and framing is to characterize "circuit properties of adaptation" but I have two issues with this. First, is the inference that neural adaptation is occurring. Yes, the signal is smaller when stimuli are repeated vs. when they are not, and this is *consistent* with adaptation. But, the origin(s) of fMRI repetition effects is controversial. Without a secondary measure – e.g., psychophysical evidence of adaptation (e.g., reduced sensitivity, tilt aftereffect, etc) in the adapted vs. non-adapted conditions or a direct physiological measure (e.g., spikes) – all we can say is that the fMRI signal is reduced during repetition. What I think this paper does a great job of doing is the set up for an experiment that specifically examines adaptation – for example, is there a specific layer-response that best predicts psychophysical differences in adaptation?
The second issue is the inferences that are made between depth and feedback/feedforward processing. Take, for example, a measured difference in superficial layers. I don't understand how it is possible to know whether a change in the BOLD signal in superficial layers is due to neurons in these layers being affected by within-area circuits (e.g., from known connections between middle-layers to superficial layers) or due to feedback-mediated effects (as feedback affects both superficial and deep layers). In light of this, it's difficult to parse the first paragraph of the Discussion. "First, visual adaptation is implemented by recurrent processing of signals in visual cortex, as indicated by fMRI adaptation (i.e. BOLD decrease due to stimulus repetition) across layers with stronger effects in superficial than middle and deeper layers." What does "recurrent processing" mean here? If it means "feedback" shouldn't deeper layers have stronger effects? Does it mean with-area processing (middle/input -> superficial/output). Overall, I find the report of layer-specific responses interesting but I cannot infer the level of "circuit properties" the authors' wish to ascribe to the effects.
Along similar lines, as a way to refresh my memory of layers/connections in early visual cortex, I looked at this recent paper: "Anatomy and Physiology of Macaque Visual Cortical Areas V1, V2, and V5/MT: Bases for Biologically Realistic Models". It is difficult to map the relatively simple characterization presented in the current paper with the real complexity described in the Vanni et al. paper. As just one example, from Vanni et al., "FF connections from V1 to V5 arise from layers 4B (both blobs and interblobs) and 6 and target primarily L4 and less so L3 of V5." "FB projections from V5 to V1 terminate predominantly in layers 4B and 6 (Maunsell and Van Essen 1983b; Ungerleider and Desimone 1986b; Shipp et al. 1989), that is, the source layers of the V1-to-V5 FF projection." There is very little consistency between this characterization and what is depicted in Figure 1A (though, I understand these quotes are specifically about MT-V1 connections).
Reviewer #2:
1) "while feedback occipito-parietal connectivity" should be "while feedback was enhanced for occipito-parietal connectivity".
2) Introduction second sentence: This claim is lacking citations
3) Results final paragraph: why isn't IPS considered visual cortex? It's definition in the Wang et al. atlas is based on topographic representation of visual space.
Reviewer #3:
Introduction paragraph 1 and paragraph 2: plethora. Maybe don't use plethora twice.
Results: It seems like all ROIs should be in Figure 3A and remove Supplementary figure 3. All ROIs are in Figure 3B and all are included in the ANOVA. It would just be easier to refer to a figure that included all ROIs if ROIs are discussed in the ANOVA.
"Post-hoc comparisons showed significantly decreased fMRI responses for adaptation across cortical layers (deeper: t(14)=-3.244, p=0.006; middle: 126 t(14)=-3.920, p=0.002; superficial: t(14)=-4.134, p=0.001)." What are the t-tests comparing? Is that just in V1?
Error bars: Unfortunately, the error bars make it look as if there are no effects. Visually, looking at Figure 3A, I'd say the responses are identical across layers. And, I'd conclude the same about all ROIs in Figure 3B. In fact, as I write this, it's difficult to reconcile the p-values in the text and what is presented in the figures. I'm guessing the issue is the repeated-measures nature of the analysis in which case between-subject error bars are misleading. You might consider:
https://doi.org/10.7554/eLife.57637.sa1Author response
Essential revisions:
Assuming that a direct physiological measure (e.g., spikes) for the present study at this time is difficult, some additional psychophysical experiments would be needed to successfully address reviewer #3's first main concern. Adding some psychophysical experiments would also help to address comments from reviewer #2 better than just adding analyses and tempering claims.
We thank the reviewers and the Editor for their constructive feedback. We agree that neurophysiology experiments are beyond the remit of this study that focuses on understanding the human brain computations that underlie adaptive processing. Following the reviewers’ suggestion, we have conducted a behavioural study employing a classic tilt after effect paradigm that has been used extensively to study orientation adaptation. Our results show that perceptual adaptation as measured by this paradigm correlates significantly with fMRI adaptation in superficial layers of visual areas, suggesting that the layer-specific fMRI adaptation we observed relates to adaptive behaviour.
Reviewer #1:
1) The study overall did not add much new evidence that may advance our understandings of feedforward and feedback processes in the visual adaptation, and the reported results were to some extent predictable from previous studies (see review, Lawrence et al., 2019; Self et al., 2019).
We have revised the Introduction and Discussion sections to clarify the advances provided by our study in understanding the feedforward vs. feedback processes involved in visual adaptation using UHF laminar fMRI.
First, previous studies have used laminar fMRI to discern feedforward vs. feedback processes involved in a range of cognitive functions (e.g., attention, memory, visual motion, multisensory processing, tonotopic maps, visual imagery and illusions, to name just a few; De Martino et al., 2015; Fracasso, Petridou and Dumoulin, 2016; Lawrence, Norris and de Lange, 2019; Scheeringa et al., 2016). However, to the best of our knowledge, no previous study had investigated the contributions of these processes to sensory adaptation and plasticity due to stimulus repetition.
Second, previous studies focused on fMRI adaptation as a tool for interrogating selectivity at the level of large-scale neural populations for a given stimulus dimension. This is typically measured by recording fMRI responses to a test stimulus following the repeated presentation of stimuli that have similar or different dimensions to the test stimulus (e.g. Engel, 2005; Fang, Murray and He, 2007; Fang et al., 2005). In contrast, our study interrogates the mechanisms underlying adaptive processing, as a signature of short-term sensory plasticity, by measuring fMRI responses during stimulus repetition rather than responses to a test stimulus following adaptation. Employing this paradigm in combination with UHF laminar fMRI allows us to interrogate fMRI responses during stimulus repetition at a submillimetre resolution to gain insights into the circuit processes (feedforward vs. feedback) underlying adaptive processing in the human brain.
After receiving the reviews on our manuscript, a new study was published (Ge et al., 2020), using laminar fMRI to investigate the mechanisms underlying visual adaptation in the context of the Flash Grab aftereffect. The paradigm involves adaptation to wedged discs rotating clockwise and anti-clockwise with vertical bars presented on the disc’s wedge boundary at the time of rotation reversal. This study reports stronger fMRI responses in superficial V1 layers, as measured by the difference in mean BOLD response between clockwise and anti-clockwise adaptors conditions. At first glance, this result is similar to our finding showing laminar-specificity of fMRI adaptation in V1. Yet, our study extends beyond this finding to provide further insights into the functional circuit processes underlying visual adaptation. In comparison to Ge et al., 2020, our study advances our understanding in the following respects:
1) We interrogated laminar fMRI specificity for adaptation in regions beyond V1: a) across visual cortex (V1 and higher visual areas) involved in sensory processing, b) posterior parietal cortex known to be involved in expectation of familiar stimuli.
2) Combining functional connectivity analyses with laminar fMRI, we provide clearer insights in interpreting the laminar fMRI specificity by testing for connectivity that relates to feedforward (i.e. between superficial and middle layers across regions) vs. feedback (i.e. across deeper layers) processing. These analyses demonstrate the following novel findings related to the fine-scale circuit involved in adaptive processing: a) feedforward processing within visual cortex from V1 to higher visual areas b) recurrent processing as indicated by short-range feedback from V2 to V1, c) long-range feedback from IPS to V1.
3) We demonstrate that laminar specificity of fMRI adaptation in visual cortex relates to perceptual bias in orientation discrimination due to adaptation. Using a tilt aftereffect paradigm, we measured perceptual adaptation on the same individuals who participated in the fMRI study and demonstrated a significant correlation between fMRI adaptation in superficial layers of visual areas and perceptual bias, suggesting that processing in superficial visual cortex layers supports perceptual adaptation. These results strengthen the evidence we provide through a range of controls that the laminar fMRI specificity we report is unlikely to be due to vasculature-related confounds.
We now discuss these points and the Ge et al., 2020, study in the revised manuscript.
2) The authors claimed higher functional connectivity between V1 deeper layers and IPS, however, the reported analysis showed that the difference in adaptation and non-adaptation conditions between V1 deeper layers and IPS1 just reached the significance (p =.049), while that in V1 deeper layers and IPS2 was not significant (p =.281). I feel that these results were not strong enough for the authors make the solid conclusion on the significant difference between adaptive conditions.
We thank the reviewer for raising this point. We checked the localisation of the IPS ROIs and improved the coverage for 4 participants. Performing the connectivity analysis with these modified ROIs showed a clearer result of enhanced feedback connectivity between V1 deeper layers and IPS1 for adaptation compared to non-adaptation (p=0.009). There was no significant difference between conditions for connectivity between V1 deeper layers and IPS2, suggesting specificity of feedback from posterior IPS to V1.
Also, I noticed that the IPS region was defined using anatomical templates, did the authors included functional localization scan for the IPS regions?
To localise visual areas, we used a standard retinotopic mapping protocol using a checkerboard stimulus. Previous studies have reported that mapping regions in the posterior parietal cortex requires participants to engage in saccadic eye movements, spatial attention, or memory tasks (see Wang et al., 2015; Silver and Kastner, 2009; Schluppeck, Glimcher and Heeger, 2005; Sereno, Pitzalis and Martinez, 2001). Due to time constraints on the fMRI scans, we were not able to include a specific IPS functional localiser. Instead, we used the Atlas provided by Benson, based on the work from Wang et al., 2015, to identify subregions of the intraparietal sulcus. IPS ROIs in this atlas are based on a functional –rather than anatomical– definition following memory-guided saccade mapping. We have now clarified this in the revised manuscript.
In addition, the above analyses were between V1 deeper layers and the overall IPS1 and IPS2 respectively. According to the model (see Figure 1 in the manuscript), the feedback connection was between V1 deeper layers and IPS deeper layer, so the functional connectivity analyses should be conducted between the deeper layers of V1 and deeper layers of IPS1/IPS2. Given that the manuscript has reported the significant different neural responses in different layers of IPS1 and IPS2, so I suggest that the authors do additional analyses on the functional connectivity between V1 deeper layer and IPS1/IPS2 deeper layers, and the results would provide more specific and stronger evidence on the feedback connections in visual adaption.
We thank the reviewer for this suggestion. We averaged the signal across layers in IPS1 and IPS2, as we did not observe laminar-specificity in IPS (i.e. there was no significant differences in fMRI-adaptation across IPS layers). Following the reviewer’s suggestion, we tested for feedback connectivity between deeper V1 and IPS1 or IPS2 layers. Correlating fMRI adaptation index (i.e. difference in z-scored BOLD signal between adaptation and nonadaptation) between deeper layers of V1 and IPS1 showed a significant correlation (r=0.715, p=0.003), suggesting that feedback from deeper IPS1 layers contributes to adaptation in deeper V1 layers. The same analysis between deeper layers of IPS2 and V1 did not show a significant correlation (r=0.427, p=0.113), suggesting specificity of feedback from IPS1 to V1.
Reviewer #2:
My concerns fall into two broad categories: 1) that the statistical tests employed in this work don't sufficiently support the authors' claims.
We have substantially revised our statistical tests, adding more information on the ANOVA models and clarifying the main and post-hoc tests performed, as well as the related corrections for multiple comparisons (Bonferroni correction). Further, we added a substantial section of control analyses. In addition, we have described the analyses across areas in the visual cortex, rather than simply V1, and have clarified both significant and non-significant tests.
2) Even when properly analyzed, the data presented here are insufficient evidence for the circuit-level conclusions presented throughout the manuscript. The authors could improve the evidence and present interpretations that are more closely tied to the data.
Following the reviewers’ suggestions, we have provided additional analyses and tests of functional connectivity, strengthening the evidence for the finer scale connectivity supporting adaptive processing. In addition, we have conducted a behavioural experiment, providing evidence that the fMRI laminar-specificity we observed (i.e. stronger fMRI adaptation in superficial V1 layers) relates to perceptual bias.
Conceptual
1) The language used does not adequately clarify the differences (and potential lack of alignment) between cortical depths and cortical layers. For example, the claim that "UHF imaging affords the sub-millimetre resolution necessary to examine fMRI signals across cortical layers" is not strictly true. Even at the small voxel size used here, cortical curvature, variable thickness, and partial volume effects all make it extremely challenging to map cortical depth to physiologically-distinct cortical layers. This distinction is critical given the motivation of this work as dissecting feedforward and feedback projections.
Suggestion: avoid the term "layer" and instead use "depth", and clarify in the manuscript that inferences about cortical layers, and thus, alignment to existing anatomical models of feedforward and feedback projections is limited.
We followed the terminology used in recent UHF imaging studies that often use different terms (e.g., cortical layers, cortical depth, mesoscopic scale, laminae, …) interchangeably (see, for example: https://layerfmri.com/2019/02/21/terminology/). However, we agree with the reviewer that UHF imaging does not support one-to-one mapping between MRI-defined cortical depths and cyto-architectonically defined cortical layers. As we describe in the manuscript, voxels within the grey matter/white matter and grey matter/CSF borders were assigned to three different subregions corresponding to superficial, middle and deeper layers. Following the reviewer’s suggestion, we have clarified this limitation of UHF imaging and have replaced “layers” with “cortical depths” where possible.
2) It's not immediately clear how one should combine the functional connectivity and adaptation results into a single framework for thinking about mechanisms of adaptation. Is it possible, for example, that other regions (besides IPS1, IPS2, V2, V3, and V4) feedback to V1 and contribute to suppression? Is the strength of the IPS feedback during the adaptation condition predictive of the amount of adaptive suppression? Conducting these additional analyses would strengthen the authors' claim that top-down feedbacks from the IPS contribute to the adaptive processing reported in visual cortex.
Suggestion:
i) Compare the degree of feedforward and feedback connectivity between conditions for a control region, e.g., hMT+, to demonstrate that IPS is uniquely (or especially) involved in these computations.
We thank the reviewer for this suggestion. Our choice of ROI was: a) motivated by our aim to test the fine scale computations that support adaptive processing in sensory areas and IPS that is known to be involved in expectation of familiar stimuli, b) limited by the coverage afforded when acquiring data at sub millimetre resolution. Following the reviewer’s suggestion, we tested hMT+ as a control ROI. Briefly, we used a probabilistic atlas (Rosenke et al., 2020) to obtain a functionally defined ROI for hMT+ for each participant. We manually adjusted the segmentation for this ROI and further assigned grey matter voxels within this ROI in three groups, generating cortical layers as described in more detail in the manuscript (see Materials and methods, MRI data analysis: Segmentation and cortical depth sampling). First, we extracted z-scored BOLD responses for each condition and cortical depth. A repeated measures ANOVA with cortical depth (deeper, middle, superficial) and condition (adaptation, non-adaptation) as factors showed no significant effect of cortical depth (F(2,28)=2.898, p=0.101), nor condition (F(1,14)=3.874, p=0.069), nor a significant interaction between these two factors (F(2,28)=2.392, p=0.138). Second, we performed laminar functional connectivity analysis as described in the revised manuscript (see Materials and methods, Functional Connectivity analysis) between V1 and hMT+. A repeated measures ANOVA with pathway (feedforward, feedback) and condition (adaptation, non-adaptation) as factors showed a main effect of condition (F(1,14)=5.828, p=0.03), but no significant effect of pathway nor an interaction between condition and pathway (F(1,14)=2.597, p=0.129; F(1,14)=0.475, p=0.502, respectively). These analyses suggest specificity of functional connectivity between V1 and IPS, i.e. fMRI adaptation in V1 deeper layers was due to feedback from IPS rather than other visual areas (e.g. hMT+).
ii) Demonstrate that the amount of feedback from IPS correlates with the amount of adaptation in deeper cortical depths in V1 across individual subjects.
We thank the reviewer for this suggestion. Our results showed stronger feedback connectivity for adaptation (i.e. connectivity across deeper layers) between: a) V1 and V2 (paired t-test comparing feedback connectivity between adaptation and non-adaptation: t(14)=2.223, p=0.043; no significant differences for V1-V3 connectivity: t(14)=0.703, p=0.494; nor V1V4 connectivity: t(13)=0.813, p=0.431), b) V1 and IPS (t(14)=3.014, p=0.009). Following the reviewer’s suggestion, we tested whether this feedback connectivity correlates with adaptation in deeper V1 layers. In particular, we correlated the fMRI adaptation index (i.e. difference in z-scored BOLD signal between adaptation and non-adaptation) between deeper layers of V1 and IPS1. We observed a significant correlation (r=0.715, p=0.003), suggesting that feedback from deeper IPS1 layers contributes to adaptation in deeper V1 layers. We have now included this analysis in the revised manuscript.
iii) Elucidate it this feedback from IPS that is correlated with adaptation is specific to V1 or occurs to other visual areas that show adaptation (V2, V3, V4).
We thank the reviewer for this suggestion and have now included additional statistics in the revised manuscript. In particular, feedback (i.e. correlation across deeper layers) from IPS1 was specific to V1. There was no significant correlation across deeper layers between a) IPS1 and V2, V3, or V4; that is, repeated measures ANOVA with pathway (feedforward, feedback), condition (adaptation, non-adaptation) showed no significant interactions between these factors; IPS1-V2: F(1,14)=0.241, p=0.631; IPS1-V3: F(1,14)=0.039, p=0.846; IPS1-V4: F(1,13)=2.387, p=0.146), b), IPS2 and V1, V2, V3, or V4; that is, repeated measures ANOVA with pathway (feedforward, feedback), condition (adaptation, non-adaptation) showed no significant interactions between these factors; IPS2V1: F(1,14)= 0.714, p=0.412; IPS2-V2: F(1,14)=0.386, p=0.545; IPS2-V3: F(1,14)=0.49, p=0.495; IPS2-V4: F(1,13)=0.264, p=0.616).
Data Analyses and Validation of Results
1) The reporting of ANOVA results of differences in adaptation across ROIs and depths throughout the paper is confusing. It is unclear whether a) separate ANOVAs were run to test each effect (main or interaction), or b) that the results of the ANOVAs are misreported. For example, consider the ANOVA reported which tests the effects of ROI, depth, and condition on z-scored BOLD responses. The degrees of freedom (3, 39), indicates that there are four levels to the factor of interest (four ROIs), and 40 total measurements. Where does 40 come from? If there are 15 subjects and each contributes 12 data points (4 ROIs x 3 depths), I would expect that the variance being analyzed is that of 15 x 12 = 180 data points. Then, the next result presented is the main effect of condition, which has a reported (1, 13) degrees of freedom, suggesting that a different model was fit.
Suggestion: Run a single ANOVA and report main effects and interaction terms from that analysis.
We apologise for the confusion and we have now revised the presentation of the results, as well as clarified the ANOVA models in the Materials and methods and Results sections. Please note that the definition of area V4 was not possible for one of the participants due to limited coverage. As a result, the ANOVA model had missing data (n=14 rather than n=15 for V4) and three main factors, ROI (V1, V2, V3, V4), cortical depth (deeper, middle, superficial), and condition (adaptation, nonadaptation).
That is degrees of freedom were calculated as follows: ROI with [(4-1),[(4-1)*(14-
1)]]=(3,39) degrees of freedom; cortical depth with [(3-1),[(3-1)*(14-1)]]=(2,26) degrees of freedom; condition with [(2-1),[(2-1)*(14-1)]]=(1,13) degrees of freedom. Accordingly, the interactions between factors were: ROI x cortical depth [(4-1)*(3-1), (4-1)*(3-1)*(141)]=(6,78); ROI x condition [(4-1)*(2-1), (4-1)*(2-1)*(14-1)]=(3,39); cortical depth x condition [(3-1)*(2-1), (3-1)*(2-1)*(14-1)]=(2,26); ROI x cortical depth x condition [(41)*(3-1)*(2-1), (4-1)*(3-1)*(2-1)*(14-1)]=(6,78).
2) The authors claim that adaptation is stronger in superficial V1 than in other cortical depths in the Introduction but write "visual cortex" instead of "V1" or "primary visual cortex" elsewhere. It is unclear when the authors are claiming that the result applies to V1 and when it applies to V1 through V4 in aggregate. This is problematic for three reasons. First, the claim being made should be clear, and the Introduction and Discussion should reflect the scope (V1 or V1-V4) intended. Second, if the authors claim is that suppression is stronger in superficial V1, a post-hoc test with appropriate multiple comparisons correction is needed. The tests are insufficient in that they don't compare superficial to middle depths directly, and in that they aggregate across all four ROIs instead of testing V1 separately. Third, if the authors claim is that suppression is stronger in superficial V1 through V4, then the direct comparison of superficial to middle depths is still lacking. Furthermore, the framing of the rest of the paper which compares V1 connectivity to IPS and V2-V4 and discusses V1 in the Introduction and Discussion needs to be justified if V1 is no different from V2, V3, and V4.
Suggestion: Make the claim being made clearer, then compute the appropriate statistical tests to support that claim. If needed, revise the Introduction and Discussion to explain special attention paid to each ROI.
We thank the reviewer for these suggestions and have revised the manuscript accordingly.
First, we have clarified when we refer to V1 (primary visual cortex) vs. extrastriate visual areas (V2, V3, V4). We have revised the Introduction and Discussion to include all visual areas (not only V1) and IPS.
Second, we have added post-hoc comparisons (Bonferroni corrected for multiple comparisons) showing significantly stronger fMRI adaptation (i.e. fMRI responses for nonadaptation minus adaptation) across visual areas in superficial compared to deeper layers (V1: t(14)=-2.556, p=0.023; V3: t(14)=-2.580, p=0.022; V4: t(13)=-2.091, p=0.012) and middle compared to deeper layers (V1: t(14)=-2.429, p=0.029; V2: t(14)=-2.524, p=0.024; V3: t(14)=-3.528, p=0.003; V4: t(13)=-2.519, p=0.026). No significant differences were observed between fMRI adaptation in superficial and middle layers across visual areas (V1: t(14)=-2.093, p=0.055; V2: t(14)=0.331, p=0.746; V3: t(14)=0.942, p=0.362; V4: t(13)=2.096, p=0.056), nor between superficial and deeper layers of V2 (t(14)=-0.881, p=0.393).
Third, we have now added statistics showing no significant changes in functional connectivity between extrastriate and intraparietal areas and included additional discussion on the connectivity results.
3) In relation to point (2) above, they claim that adaptation is layer specific rests on the results of post-hoc tests. However, it is unclear (i) which ROIs are tested, and (ii) why they are comparing superficial and deep as well as middle and deep but not superficial and middle. Further the Materials and methods indicate that pairwise t-tests were used, if so multiple comparisons correction is needed to validate the result.
Suggestion: The authors should conduct all tests and clarify the reporting of the results. If multiple-comparisons correction is not currently being used, the authors should employ either Tukey's Honest Significant Difference, Bonferroni correction, or an alternative correction. If the tests are already corrected, that fact should be reflected in the Materials and methods section.
We thank the reviewer for this suggestion. We had performed Bonferroni correction for multiple comparisons. We have now clarified this in the revised Materials and methods section. Further, we have clarified the factors compared for each test in the revised Results section. We have also added the non-significant results for comparisons between superficial and middle layers.
4) The inclusion of GRASE results strengthens the work substantially. However, the claim that the same adaptation patterns are observed are not supported numerically. In fact, the results presented in Figure 3—figure supplement 2 suggest that adaptation is not stronger in superficial than middle or deep depths, and that the correlation between adaptation indices across scan sequences (panel B) are moderate; at most, the adaptation indices from one scan sequence predicts 22% of the variance in the adaptation indices from the other scan sequence.
Suggestion: Report statistics for the 3D GRASE results.
As we report in the manuscript, the subset of participants that took part in both sessions (GEEPI and 3D GRASE acquisitions) was small (N=4), as the 3D GRASE acquisitions were intended as a control for vasculature-related confounds. This small sample size does not support further statistical analyses; yet, we believe it to be a valuable control dataset. We tested the same participants with both GE-EPI and 3D GRASE to assess reproducibility of our results independent of the MR sequence used, as these sequences are known to be affected by vasculature-confounds at different degrees. To assess reproducibility, we correlated the fMRI adaptation index measured between sequences, across scanning sessions.
As reported in Figure 3—figure supplement 2, we observed substantial (r> 0.45) and statistically significant correlations for middle and superficial layers, suggesting that the laminar specificity of the GE-EPI results was unlikely to be significantly confounded by vasculature related confounds. We have now clarified in the revised manuscript our approach in using the GRASE measurements as a test of reproducibility.
Reviewer #3:
The motivation and framing is to characterize "circuit properties of adaptation" but I have two issues with this. First, is the inference that neural adaptation is occurring. Yes, the signal is smaller when stimuli are repeated vs. when they are not, and this is *consistent* with adaptation. But, the origin(s) of fMRI repetition effects is controversial. Without a secondary measure – e.g., psychophysical evidence of adaptation (e.g., reduced sensitivity, tilt aftereffect, etc) in the adapted vs. non-adapted conditions or a direct physiological measure (e.g., spikes) – all we can say is that the fMRI signal is reduced during repetition. What I think this paper does a great job of doing is the set up for an experiment that specifically examines adaptation – for example, is there a specific layer-response that best predicts psychophysical differences in adaptation?
We agree with the reviewer that the neural mechanisms underlying fMRI adaptation (i.e. decreased BOLD response due to stimulus repetition) remain controversial. Our study aimed to investigate the feedforward vs. feedback mechanisms that contribute to fMRI adaptation by testing the laminar-specificity of fMRI adaptation. Neurophysiological recordings are beyond the remit of our study; yet our results showing laminar specificity have the potential to inspire future neurophysiological studies to test feedforward vs. feedback mechanisms of neural adaptation across cortical depths.
Our experimental design aimed to ensure that the comparison of BOLD signals across conditions was not confounded by differences in task performance. To this end, participants engaged in an attentional control task rather than in a behavioural task that measures perceptual adaptation. Following the reviewer’s suggestion, we conducted an additional behavioural experiment on the same participants who participated in the fMRI experiment. We used a classic tilt-aftereffect task to measure perceptual adaptation to orientation. In brief, participants were presented with blocks of gratings of either the same (adaptation block), or different orientations (non-adaptation). Following the presentation of 21 gratings, a test stimulus with orientation near vertical was shown, and participants were asked to indicate whether the test was oriented clockwise or anti-clockwise compared to vertical. During stimulus presentation, participants were asked to perform the same RSVP task as in the fMRI experiment to ensure that attention was maintained at fixation. Performance in the tilt after effect task showed a perceptual bias, that is, a significant shift in the distribution of responses away from the adapted orientation for the adaptation compared to non-adaptation condition (t(11)=-3.197, p=0.0085).
Correlating perceptual adaptation index and fMRI adaptation showed a significant correlation for superficial V1 layers (r=0.640, p=0.046) but not middle (r=0.562, p=0.091) or deeper V1 layers (r=0.563, p=0.09). Further, correlating perceptual adaptation index with mean fMRI adaptation across visual areas (V1, V2, V3, V4) showed similar results; that is significant correlation for superficial (r=0.706, p=0.039) but not middle (r=0.562, p=0.091), nor deeper (r=0.567, p=0.087) layers. These results suggest that the laminar-specificity of fMRI adaptation in visual areas relates to a behaviour measure of perceptual adaption (i.e. perceptual bias away from the adapted orientation).
The second issue is the inferences that are made between depth and feedback/feedforward processing. Take, for example, a measured difference in superficial layers. I don't understand how it is possible to know whether a change in the BOLD signal in superficial layers is due to neurons in these layers being affected by within-area circuits (e.g., from known connections between middle-layers to superficial layers) or due to feedback-mediated effects (as feedback affects both superficial and deep layers). In light of this, it's difficult to parse the first paragraph of the Discussion. "First, visual adaptation is implemented by recurrent processing of signals in visual cortex, as indicated by fMRI adaptation (i.e. BOLD decrease due to stimulus repetition) across layers with stronger effects in superficial than middle and deeper layers." What does "recurrent processing" mean here? If it means "feedback" shouldn't deeper layers have stronger effects? Does it mean with-area processing (middle/input -> superficial/output). Overall, I find the report of layer-specific responses interesting but I cannot infer the level of "circuit properties" the authors' wish to ascribe to the effects.
Along similar lines, as a way to refresh my memory of layers/connections in early visual cortex, I looked at this recent paper: "Anatomy and Physiology of Macaque Visual Cortical Areas V1, V2, and V5/MT: Bases for Biologically Realistic Models". It is difficult to map the relatively simple characterization presented in the current paper with the real complexity described in the Vanni et al. paper. As just one example, from Vanni et al., "FF connections from V1 to V5 arise from layers 4B (both blobs and interblobs) and 6 and target primarily L4 and less so L3 of V5." "FB projections from V5 to V1 terminate predominantly in layers 4B and 6 (Maunsell and Van Essen 1983b; Ungerleider and Desimone 1986b; Shipp et al. 1989), that is, the source layers of the V1-to-V5 FF projection." There is very little consistency between this characterization and what is depicted in Figure 1A (though, I understand these quotes are specifically about MT-V1 connections).
We thank the reviewer for this comment. We agree that stronger fMRI adaptation in superficial layers could be due to a) output from middle layers where significant fMRI adaptation was also observed, suggesting recurrent processing within V1, b) feedback from higher visual areas or IPS. Although we did observe significant fMRI adaptation in deeper layers, previous work has shown that synaptic input to superficial layers may result due to increase in feedback signals carried by neurons that have dendrites projecting to the superficial layers and their cell bodies in deeper layers (Larkum, 2013). Our functional connectivity analysis sheds more lights into these mechanisms. We observed significant connectivity across deeper layers between V1 and V2, suggesting a local circuit based on recurrent processing within these regions and local feedback between them. Further, we observed significant connectivity across deeper layers between V1 and IPS, suggesting long-range connectivity between occipital cortex involved in sensory processing and parietal regions involved in expectation.
We agree with the reviewer that the scale of UHF imaging is limited compared to anatomical or neurophysiology studies; as a result, it is not possible to achieve one-to-one mapping between fMRI cortical depths and cyto-architectonically defined. Our analyses of cortical depth and functional connectivity follows the framework presented in Figure 1, as adopted by several previous UHF imaging studies investigating feedforward vs. feedback processing using a range of tasks and stimuli (e.g. perceptual illusions, finger tapping, word reading, attention; Huber et al., 2017; Kok et al, 2016; Moerel et al, 2020; Sharoh et al., 2019). Future studies using similar paradigms (i.e. stimuli, tasks) across animals and humans would be necessary to investigate functional correspondence across anatomical scales. We have clarified and discussed this point further in the revised manuscript.
Reviewer #2:
1) "while feedback occipito-parietal connectivity" should be "while feedback was enhanced for occipito-parietal connectivity".
We have revised the text; See also response to point 1 of reviewer #1.
2) Introduction second sentence: This claim is lacking citations
We have included the following references: Clifford, 2002; Kohn, 2007
3) Results final paragraph: why isn't IPS considered visual cortex? It's definition in the Wang et al. atlas is based on topographic representation of visual space.
The IPS regions we tested are located in the posterior parietal cortex. The IPS definition in Wang et al., 2015, is based on a memory-guided saccade task. This is consistent with previous work (Silver and Kastner, 2009) that reports topographic representation in posterior parietal cortex using memory-guided saccades and spatial attention tasks rather than the standard visual stimulation protocol (flickering checkerboard) used for mapping visual retinotopic areas. In the revised manuscript, we clarify that our choice of IPS as a region of interest was motivated by the role of IPS in processing a) visual information, b) expectation due to stimulus familiarity.
Reviewer #3:
Introduction paragraph 1 and paragraph 2: plethora. Maybe don't use plethora twice.
Thank you-we have revised the text.
Results: It seems like all ROIs should be in Figure 3A and remove Supplementary figure 3. All ROIs are in Figure 3B and all are included in the ANOVA. It would just be easier to refer to a figure that included all ROIs if ROIs are discussed in the ANOVA.
We thank the reviewer for this suggestion. We have now modified Figure 3 to include results from all ROIs. Figure 3—figure supplement 1 and 2 shows data from the behavioural task and GRASE data.
"Post-hoc comparisons showed significantly decreased fMRI responses for adaptation across cortical layers (deeper: t(14)=-3.244, p=0.006; middle: 126 t(14)=-3.920, p=0.002; superficial: t(14)=-4.134, p=0.001)." What are the t-tests comparing? Is that just in V1?
These tests related to V1 data. We have now revised the text to include post-hoc comparisons (Bonferroni corrected) related to the two-way ANOVA on fMRI adaptation index with ROI (V1, V2, V3, and V4), and cortical depth (deeper, middle, superficial) factors.
Error bars: Unfortunately, the error bars make it look as if there are no effects. Visually, looking at Figure 3A, I'd say the responses are identical across layers. And, I'd conclude the same about all ROIs in Figure 3B. In fact, as I write this, it's difficult to reconcile the p-values in the text and what is presented in the figures. I'm guessing the issue is the repeated-measures nature of the analysis in which case between-subject error bars are misleading. You might consider:
We thank the reviewer for pointing this out and for the resourceful information. We have now implemented the suggested method for computing error bars. This has helped with consistency between figures and statistics.
https://doi.org/10.7554/eLife.57637.sa2Article and author information
Author details
Funding
Biotechnology and Biological Sciences Research Council (H012508)
- Zoe Kourtzi
Biotechnology and Biological Sciences Research Council (BB/P021255/1)
- Zoe Kourtzi
Wellcome Trust (205067/Z/16/Z)
- Zoe Kourtzi
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
Acknowledgements
We would like to thank Christopher Wiggins and Esther Steijvers (Scannexus) for technical support, and Peter Kok (University College London), Denis Schluppeck (University of Nottingham), Federico De Martino (University of Maastricht), Laurentius Huber (University of Maastricht), and Cheryl Olman (University of Minnesota) for the expert and insightful comments to the manuscript. We would also like to thank Adrian Ng, Valentyna Chernova, and Cher Zhou for help with the analysis. This work was supported by grants to ZK from the Biotechnology and Biological Sciences Research Council (H012508 and BB/P021255/1) and was funded in part by the Wellcome Trust (205067/Z/16/Z). For the purpose of Open Access, the author has applied a CC BY public copyright licence to any Author Accepted Manuscript version arising from this submission.
Ethics
Human subjects: Participants gave written informed consent. The study was approved by the local Ethical Committee of the Faculty of Psychology and Neuroscience at Maastricht University and the University of Cambridge Ethics Committee (ethics number PRE2017.057).
Senior Editor
- Chris I Baker, National Institute of Mental Health, National Institutes of Health, United States
Reviewing Editor
- Ming Meng, South China Normal University, China
Publication history
- Received: April 7, 2020
- Accepted: November 9, 2020
- Accepted Manuscript published: November 10, 2020 (version 1)
- Version of Record published: November 25, 2020 (version 2)
Copyright
© 2020, Zamboni et al.
This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.
Metrics
-
- 1,180
- Page views
-
- 177
- Downloads
-
- 5
- Citations
Article citation count generated by polling the highest count across the following sources: PubMed Central, Crossref, Scopus.
Download links
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Further reading
-
- Neuroscience
Comparing connectomes can help explain how neural connectivity is related to genetics, disease, development, learning, and behavior. However, making statistical inferences about the significance and nature of differences between two networks is an open problem, and such analysis has not been extensively applied to nanoscale connectomes. Here, we investigate this problem via a case study on the bilateral symmetry of a larval Drosophila brain connectome. We translate notions of'bilateral symmetry' to generative models of the network structure of the left and right hemispheres, allowing us to test and refine our understanding of symmetry. We find significant differences in connection probabilities both across the entire left and right networks and between specific cell types. By rescaling connection probabilities or removing certain edges based on weight, we also present adjusted definitions of bilateral symmetry exhibited by this connectome. This work shows how statistical inferences from networks can inform the study of connectomes, facilitating future comparisons of neural structures.
-
- Neuroscience
Temporal synchrony of signals arriving from different neurons or brain regions is essential for proper neural processing. Nevertheless, it is not well understood how such synchrony is achieved and maintained in a complex network of time-delayed neural interactions. Myelin plasticity, accomplished by oligodendrocytes (OLs), has been suggested as an efficient mechanism for controlling timing in brain communications through adaptive changes of axonal conduction velocity and consequently conduction time delays, or latencies; however, local rules and feedback mechanisms that OLs use to achieve synchronization are not known. We propose a mathematical model of oligodendrocyte-mediated myelin plasticity (OMP) in which OLs play an active role in providing such feedback. This is achieved without using arrival times at the synapse or modulatory signaling from astrocytes; instead, it relies on the presence of global and transient OL responses to local action potentials in the axons they myelinate. While inspired by OL morphology, we provide the theoretical underpinnings that motivated the model and explore its performance for a wide range of its parameters. Our results indicate that when the characteristic time of OL’s transient intracellular responses to neural spikes is between 10 and 40 ms and the firing rates in individual axons are relatively low (⪅ 10 Hz), the OMP model efficiently synchronizes correlated and time-locked signals while latencies in axons carrying independent signals are unaffected. This suggests a novel form of selective synchronization in the CNS in which oligodendrocytes play an active role by modulating the conduction delays of correlated spike trains as they traverse to their targets.