1. Neuroscience
Download icon

Mapping cortical mesoscopic networks of single spiking cortical or sub-cortical neurons

  1. Dongsheng Xiao
  2. Matthieu P Vanni
  3. Catalin C Mitelut
  4. Allen W Chan
  5. Jeffrey M LeDue
  6. Yicheng Xie
  7. Andrew CN Chen
  8. Nicholas V Swindale
  9. Timothy H Murphy Is a corresponding author
  1. Kinsmen Laboratory of Neurological Research, Canada
  2. Beijing Institute for Brain Disorders, Capital Medical University, China
  3. Djavad Mowafaghian Centre for Brain Health, University of British Columbia, Canada
  4. University of British Columbia, Canada
Tools and Resources
Cited
3
Views
2,256
Comments
0
Cite as: eLife 2017;6:e19976 doi: 10.7554/eLife.19976

Abstract

Understanding the basis of brain function requires knowledge of cortical operations over wide-spatial scales, but also within the context of single neurons. In vivo, wide-field GCaMP imaging and sub-cortical/cortical cellular electrophysiology were used in mice to investigate relationships between spontaneous single neuron spiking and mesoscopic cortical activity. We make use of a rich set of cortical activity motifs that are present in spontaneous activity in anesthetized and awake animals. A mesoscale spike-triggered averaging procedure allowed the identification of motifs that are preferentially linked to individual spiking neurons by employing genetically targeted indicators of neuronal activity. Thalamic neurons predicted and reported specific cycles of wide-scale cortical inhibition/excitation. In contrast, spike-triggered maps derived from single cortical neurons yielded spatio-temporal maps expected for regional cortical consensus function. This approach can define network relationships between any point source of neuronal spiking and mesoscale cortical maps.

https://doi.org/10.7554/eLife.19976.001

Introduction

Neural activity ranges from the microscale of synapses to macroscale brain-wide networks. Mesoscale networks occupy an intermediate space and are well studied in cortex forming the basis of sensory and motor maps (Bohland et al., 2009). These networks are largely defined by co-activation of neurons and have been evaluated with a variety of statistical approaches that capitalize on detecting synchrony. The study of large-scale networks (meso-to macroscale) has been mostly restricted to functional magnetic resonance imaging (fMRI), or magnetoencephalography that can capture whole-brain activity patterns (de Pasquale et al., 2010; Kahn et al., 2011; Logothetis et al., 2012), but lack high spatial and temporal resolution and sensitivity. To overcome these limitations, alternative approaches including mesoscopic intrinsic signal, voltage, glutamate, or calcium sensitive indicator imaging have been employed (Kleinfeld et al., 1994; Kenet et al., 2003; Ferezou et al., 2007; Chemla and Chavane, 2010; Chen et al., 2013b; Mohajerani et al., 2013; Stroh et al., 2013; Vanni and Murphy, 2014; Carandini et al., 2015; Chan et al., 2015; Madisen et al., 2015; Wekselblatt et al., 2016; Xie et al., 2016). New preparations using large-scale craniotomies (Kim et al., 2016b) and large format imaging systems (Tsai et al., 2015; Sofroniew et al., 2016) provide the ability to link mesoscale activity patterns to individual neurons. However, these measures are restricted to superficial layers of cortex and cannot assess linkages with sub-cortical structures. Developments in fiberoptic technology allow local optical functional assessment of brain activity in sub-cortical structures (Hamel et al., 2015; Kim et al., 2016a), but cannot simultaneously resolve cortex over large fields of view. Although the evolution of imaging has revealed new aspects of cortical processing in identified neurons (Harvey et al., 2012; Chen et al., 2013a, Chen et al., 2013b; Fu et al., 2014; Guo et al., 2014), the electrically recorded action potential is still a signal of prominence given its exquisite timing and ability to reflect the output of neuronal networks (Buzsáki, 2004).

We combine extracellular recordings of single units in the cortex, thalamus, and other sub-cortical sites with mesoscopic functional imaging in transgenic mice expressing the calcium indicator GCaMP (Zariwala et al., 2012; Vanni and Murphy, 2014; Silasi et al., 2016). While slower than protein-based or small molecule voltage sensors, GCaMP imaging offers a high signal-to-noise ratio and is associated with supra-threshold activity which allows a more direct comparison with spike activity. This work extends pioneering studies investigating the relationship between single neuron spiking and local neuronal population activity assessed by voltage-sensitive dye imaging. Spike-triggered averaging (STA) was used to identify the local activity profile related to the spiking activity of a single neuron within this population (Arieli et al., 1995), and it was further demonstrated that this activity profile could reveal the instantaneous spatial pattern of ongoing population activity related to a neuron’s optimal stimulus in visual cortex of anesthetized cat (Tsodyks et al., 1999). This current study extends these approaches and also exploits the main advantage of mesoscopic imaging allowing the simultaneous measurement of brain activity in multiple regions across most of cortex simultaneously and not only that of the local population of activity surrounding the recording site. This multiscale strategy has allowed us to define temporal relationships between the activity of single neurons at the microscopic scale and mesoscale cortical maps (Zingg et al., 2014; Madisen et al., 2015). Furthermore, we employ multisite, long shank, silicon probe recordings of single neuron activity that facilitates the assessment of long-distance activity relationships between remote subcortical single neuron activity and mesoscale cortical population activity. Spontaneous activity in awake and anesthetized mice was exploited as a source of diverse cortical network activity motifs (Mohajerani et al., 2010, 2013; Chan et al., 2015). Application of STA to cortical spontaneous activity linked single neurons to mesoscale networks. Single thalamic neuron spikes were found to functionally link to multiple primary sensorimotor maps, in contrast spiking cortical neurons were largely associated with consensus cortical maps. Thalamic neurons were found to both predict and report (firing before and after) specific cycles of wide-scale cortical inhibition/excitation, while cortical neuron firing was usually associated with excitation. These results are consistent with an active computational role of thalamus in sensory-motor processing (Theyel et al., 2010; Hooks et al., 2013; Petrus et al., 2014; Sheroziya and Timofeev, 2014; McCormick et al., 2015), as opposed to merely serving a relay function and is consistent with a diverse role of the thalamus in feedforward sensory processing. Thalamocortical transmission can dynamically and differentially recruit local cortical excitation and inhibition based on thalamic neuron firing patterns and where thalamocortical feedforward inhibition is a critical feature (Galarreta and Hestrin, 1998; Swadlow and Gusev, 2001; Gabernet et al., 2005; Cruikshank et al., 2007; Hu and Agmon, 2016). We expect that this spike-triggered cortical mapping technique, exploiting mesoscopic calcium imaging, can be extended to any brain location where electrodes can be placed to identify functionally linked cortical mesoscale networks.

Results

Linkage of individual spiking neurons to specific mesoscopic cortical maps

We exploit the wide field of view of mesoscale cortical imaging using GCaMP transgenic mice (Madisen et al., 2015) in combination with cellular electrophysiology recordings to derive cortical networks that reflect activity at targeted point sources of neuronal spiking throughout the brain. Cortical and sub-cortical neuron spiking activities were recorded electrically while simultaneously imaging cortical mesoscopic activity across a 9 × 9 mm bilateral window that encompassed multiple areas of the mouse dorsal cortex including somatosensory, motor, visual, retrosplenial, parietal association and cingulate areas (Figure 1A,B). Spectral decomposition of the mesoscopic spontaneous activity using GCaMP6 revealed the presence of information below 10 Hz that was distinct from non-specific green light reflectance (Figure 1—figure supplement 1). Given the slow Ca2+ binding and unbinding kinetics of GCaMP6, we expect imaging dynamics will be prolonged compared to actual spike records. In some cases, we employed deconvolution (Pnevmatikakis et al., 2016) to improve the time course of raw calcium signals (Figure 1—figure supplement 2). While deconvolution improved the temporal dynamics of the decay of the calcium signal, spike-triggered analysis was only marginally affected and it was not used throughout. Spiking signals were initially recorded in multiple brain areas using glass electrodes (n = 8 mice) to minimize obstruction of cortical imaging and reduce potential for damage from electrode placement. Subsequently, laminar probes (16 channel with 0.1 mm contact spacing) permitted the resolution of more spiking neurons simultaneously, and facilitated the recordings in multiple sub-cortical regions (n = 16 mice). Given the invasive nature and the long duration of recordings, initial data were obtained from urethane (n = 4) or isoflurane (n = 12) anesthetized adult mice, but were later optimized to include awake recordings (n = 12, see Materials and methods). The spike-triggered average maps (STMs) obtained under both these conditions were qualitatively similar, and this observation was consistent with previous work using VSD imaging (Mohajerani et al., 2013). To perform these assessments, we identified single neuron spikes from extracellular recordings using spike sorting methods based on clustering of principal components distributions of spike signals on sets of adjacent channels (Swindale and Spacek, 2014) (Figure 1C).

Figure 1 with 2 supplements see all
Experimental setup and multichannel electrode recordings and spike classification.

(A) Set-up for simultaneous wide-field calcium imaging and single unit recording using a glass pipette or laminar silicon probe. (Bi) Top view of wide-field transcranial window and (ii) cortical atlas adapted from the Allen Institute Brain Atlas. (C) Example of (i) cortical and (ii) subcortical pairs or spike recordings from separate channels showing the isolation in the two principal components axes. (D) The generation of a spike-triggered average map (STM) for unit located in barrel cortex. (Ei) STM generated from single neuron with 1158 spikes recorded in right barrel cortex. (ii) Red traces: Spontaneous calcium activity recorded from two different cortical areas (BCS1 and HLS1). Blue trace: spontaneous spiking activity recorded simultaneously from right BCS1. (iii) STMTD generated from average of calcium activity time-locked with each spike (red) and random spike (see Materials and methods, black, blue: subtraction of spike and random spike-evoked responses) in region-of-interest (ROI). These examples results were from mice under anesthesia. Source files for the generation of spike-triggered average map (STM) can be found at http://goo.gl/nHF29I. The folder ‘Matlab code and source data’ (see the sublink of https://github.com/catubc/sta_maps) contains calcium images (‘tif’ file), spike train and Matlab code used for the generation of STM shown in Figure 1E. The ‘tif’ file cannot not be viewed with a standard picture viewer, but must be viewed with a program, such as ‘ImageJ’. Spike times were exported as ‘txt’ file. Matlab code (named ‘STA_eLife.m’) was used for reading images and spike time files and generating STM.

https://doi.org/10.7554/eLife.19976.002

To investigate how single neuron spiking activity at a cortical/sub-cortical site was related to regional cortical activity, we calculated STMs from simultaneously acquired wide-field calcium imaging (Figure 1D; see also Materials and methods). For each individual spike, we considered cortical image frames from 3s before to 3s after the spike normalized as ΔF/F0 by subtracting and dividing the average calcium activity during the 3s preceding the spike. The static STM was then defined as the peak response (in units of ΔF/F0) calculated for each pixel within a time window of ±1 s. This peak method is a better reflection of correlated activity than the average over the ±1 s interval which would smooth out highly activated – but short duration activity increases. This method revealed that the activity recorded from a single right barrel cortex neuron yielded an STM showing strong and specific GCaMP signal in the barrel and motor cortices of both hemispheres (Figure 1Ei). STMs were thus calculated by averaging calcium activity during spiking activity (Figure 1Eii) and revealed the high spatial specificity of the mapping when compared with reference region (hind limb) or random spike averaging (see Materials and methods) (Figure 1Eiii).

We verified the calcium specificity of STMs (reflecting underlying neuronal activity) by imaging Thy-1 GFP-M mice (n = 6 mice) that lacked calcium-dependent neuronal fluorescent signals and failed to produce functional maps using the same procedures (Figure 2A,B). To investigate the sensitivity of the technique, the minimum number of spikes needed to make maps in GCaMP mice was measured by quantifying the similarities of pairs of STMs generated from a subset of increasing numbers of spikes. Stable STM maps were generally observed using 256 spikes (Figure 2C–E). A high stability of STMs was also confirmed by comparing the maps generated by splitting a unit's spikes into two halves, or into odd and even groups which yield similar STMs.

Sensitivity and specificity of STMs.

(A) Simultaneous calcium and spiking activity recording in GCaMP6f mouse and STM yielded from single unit recorded in barrel cortex. (B) Simultaneous GFP fluorescence and spiking activity recording in Thy-1 GFP-M mouse and STM yielded from single unit recorded in barrel cortex resulted in no clear regional map. (C) STMs generated from a subset of spikes (2–2048, on the left) randomly chosen in one experiment. Correlation coefficients (r-value on the right) between STMs were used to evaluate the consistency of mapping. In this example, STMs generated by more than 64 spikes generated a correlation >0.9 and revealed high similarity between the pairs of SPMs made using the same number of spikes. (D) Distribution of correlation values between pairs of STM for an increasing number of spikes. No significant change in r-value distribution was observed for 512 spikes in comparison to 256 spikes (Mann Whitney test, p=0.126, U = 948.5, 256 spikes group n = 58, r-value = 0.97 ± 0.01, mean ± SD; 512 spikes group n = 40, r-value = 0.98 ± 0.01). (E) STMs and profile of responses computed using spikes divided into halves or even-odd sets. These examples were performed under anesthesia.

https://doi.org/10.7554/eLife.19976.005

Thalamic neurons show more diverse STMs than cortical neurons

By combining wide-field calcium imaging and single unit recording, the mesoscopic network associated with any neuron of the brain can be mapped. After establishing the method, we then focused on thalamic and cortical recordings of spiking neurons and confirmed their anatomical location by labeling probes with Texas red-dextran or DiI to visualize tracks (Figure 3A, four experiments subcortical track and nucleus identification, see also Figure 2A,B). We observed that spiking cortical neurons were linked to consensus local and long-range cortical networks as defined by the stability of STMs between recordings (Figure 3B). We define consensus cortical networks as those that can be identified by assessment of correlated activity (seed pixel analysis) and reflect major mono-synaptic intra-cortical axonal projections (Mohajerani et al., 2013). Spiking barrel cortex neurons were consistently linked to regional GCaMP signal changes in the barrel and motor cortex, as well as showing signals in homotopic areas of both hemispheres, consistent with the previously observed pattern of long-distance connections (Ferezou et al., 2007; Mohajerani et al., 2013; Guo et al., 2014; Vanni and Murphy, 2014; Chan et al., 2015). To compare multiple neuron STMs, we displayed the ‘contour’ of each STM as the full-width-half-max value (Figure 3B,C-contours).

Figure 3 with 2 supplements see all
Topographic properties of cortical and thalamic STM.

(A) Electrode track for each recording (Blue channel: DAPI, yellow: DiI). (B) STM and overlay contours of neurons recorded in barrel cortex in the same animal. Each color represents one STM contour. (C) STM and overlay contours of neurons recorded in thalamus in the electrode track presented in panels A and B. Color bar in the right side indicated the depth of each recording site. (D) Diversity of overlap of STMs between neurons on neighboring laminar electrode channels. (i) Example of overlapping STMs (red area) between two cortical neurons recorded on adjacent channels. (ii) Example of overlapping STMs for neighboring pairs of neurons recorded subcortically showing differences across depth. (iii) Average neighboring cortical neuron map overlap (blue: 93%) and neighboring sub-cortical neuron overlap (78%) show significant differences (Mann Whitney test, p<0.0001, U = 617408.0, mean percentage overlap of cortical STM pairs = 92.77% ± 0.23%, mean ± SEM, n = 966; mean percentage overlap of sub-cortical STM pairs = 78.11% ± 0.61%, mean ± SEM, n = 1936). These results are from awake mice (except Mouse #1).

https://doi.org/10.7554/eLife.19976.006

We have also assessed other means of generating event triggered maps including Multi-Unit-Activity (MUA) and local field potential (LFP) frequency bands. Cortical and thalamic STMs computed from single unit activity or MUA strongly resembled one another and did not vary greatly according to laminar depth. LFP-triggered STMs for delta band activity were similar to STMs, while those associated with higher frequency bands showed more unique patterns that will be investigated in future work (Figure 3—figure supplement 1 and Figure 3—figure supplement 2). Delta band activity is where most mesoscale power functional imaging indicator power is located (Chan et al., 2015). In contrast, in the case of GCaMP6 higher frequency components are closer to hemodynamic and other noise sources making analysis more challenging.

STMs for single thalamic neurons indicated not only a functional link between GCaMP maps and their consensus cortical projection areas (Hunnicutt et al., 2014Oh et al., 2014; Zingg et al., 2014) but also showed more variability and complexity in behavior than for cortical units. Thalamic neurons were associated with both unilateral and bilateral hemispheric signals within multiple primary sensorimotor and higher order brain areas (Figure 3C). To quantify the difference of variability of STM between cortical and thalamic units, we compared the percentage overlap of static STMs (Figure 3D; red areas) for neighboring pairs of neurons (100 μm apart). We found cortical neurons in the same functional region exhibited substantial similarity, while subcortical neurons had more diversity even within the same sub-nucleus in thalamus (i.e. VPL, VPM, see Table 1 for nomenclature and compare 3 example in Figure 3C). The use of multichannel probes allowed us to obtain spiking profiles across cortical layers. However, there were only subtle topographic changes: the contours of static STMs were largely similar between and within superficial and deep layers, respectively (see the contour map in Figure 3B). To assess diversity in single neuron spiking derived cortical maps (STMs), we compared STMs derived from neighboring electrode contacts and found that sub-cortical derived maps were more varied (Figure 3Diii)

Table 1

Abbreviation used to define different cortical/sub-cortical areas.

https://doi.org/10.7554/eLife.19976.009

S1

Primary somatosensory area

S2

Supplemental or Secondary Somatosensory area

FL

Forelimb region of the Primary Somatosensory area (FLS1)

HL

Hindlimb region of the Primary Somatosensory area (HLS1)

BC

Barrel region of the Primary Somatosensory area (BCS1)

M1

Primary motor area

M2

Secondary motor area

MO

Mouth region of the Primary Somatosensory area

NO

Nose region of the Primary Somatosensory area

TR

Trunk region of the Primary Somatosensory area

UN

(Unassigned) region of the Primary Somatosensory area (S1)

AC

Anterior Cingulate area (ACC)

A

Anterior or Posterior Partial Association areas: PTLp or PTA

V1

Primary visual cortex

AL

AnteroLateral regions of the extrastriate visual areas

AM

AnteroMedial regions of the extrastriate visual areas

LM

LateralMedial regions of the extrastriate visual areas

PL

PosteroLateral regions of the extrastriate visual areas

LI

LateralIntermediate regions of the extrastriate visual areas

PM

PosteroMedial regions of the extrastriate visual areas

POR

Postrhinal regions of the extrastriate visual areas

RL

RostroLateral regions of the extrastriate visual areas

AU

Primary Auditory area

TEA

Temporal Association area

RS

Retrosplenial area

PTA

Parietal Association area

VPM

Ventral posteromedial nucleus of the thalamus

VPL

Ventral posterolateral nucleus of the thalamus

PO

Posterior complex of the thalamus

RT

Reticular nucleus of the thalamus

LGN

Lateral geniculate nucleus

CP

Caudoputamen

HPF

Hippocampal formation

Sub-cortical neurons are linked to cortical maps not predicted from consensus networks

To determine quantitatively how cortical/sub-cortical STMs were related to intra-cortical networks, static STMs were compared using cross-correlation with a cortex-wide library of seed pixel correlation maps (SPM) (Mohajerani et al., 2013; Vanni and Murphy, 2014; Chan et al., 2015) generated iteratively for all locations from the same recording of spontaneous activity (Figure 4A,B). To create SPMs, the cross-correlation coefficient r values between the temporal profiles of one selected pixel and all the others within the field of view were calculated (Mohajerani et al., 2013; Vanni and Murphy, 2014; Chan et al., 2015) (see Materials and methods). To evaluate the similarity between the static STM and SPMs, we calculated the correlation coefficient between pixels of both types of maps for all possible SPMs in the library. We then selected the SPM that resulted in best match: highest correlation between a given STM and the library of SPMs. The library of SPMs is expected to reflect cortical consensus activity motifs (areas undergoing temporally-correlated activity) and can be largely attributed to their underlying intra-cortical axonal projections (Mohajerani et al., 2013). Single cortical neuron-derived STMs were largely predicted by the pattern of cortical connectivity using SPM correlation mapping (Figure 4A) as correlations were relatively high between these SPMs and STMs (Figure 4C). In contrast, thalamic STMs were more complex and corresponded to more unique distributions of cortical patterns and had significantly lower correlations with the cortical consensus SPM library (Figure 4B,C). For example, VPM, VPL and CP neurons can functionally link to multiple cortical areas that are not predicted by SPM (SPM made by putting the seeds in either BCS1, HLS1 or RS areas). In other words, it is possible that the cortical STMs derived from single spiking sub-cortical neurons are the super-position of 2 or more cortical networks defined by SPMs. To support this theory, we show examples comparing subcortical STMs to pairs of SPMs made from two seed locations that can lead to these potentially more complex maps (Figure 4D). For example, the VPM STM could be constructed as a combination of a BCS1 and RS SPM (seed point a and b, respectively, Figure 4D).

STM compared with seed pixel correlation maps (SPM).

(A) Cortical STM (left) and the best fitting SPM (right) according to correlation coefficient (cc) values for different electrode placements (text to the left of panel). Similarity was calculated by measuring the r-value Pearson coefficient between each pair of map pixels (in title). Group data from 12 GCaMP6f mice are reported in panel C. (B) Sub-cortical STM (left) and the most similar SPM (right). Cells #2 and #4 were from GCaMP6s mice. Cells #3, 5–7, 11 were from GCaMP3 mice. Other cells were from GCaMP6f mice. These examples were performed under anesthesia. (C) Distribution of r-values (Mann Whitney test, p<0.0001, U = 5227, sub-cortical group n = 246 r-value=0.64 ± 0.18, mean SD; cortical group n = 168 r-value=0.85 ± 0.04, mean ± SD). (D) Examples of sub-cortical STMs compared with pairs of SPMs for seed indicated by ‘a’ and ‘b’.

https://doi.org/10.7554/eLife.19976.010

To better understand the underlying structural circuit basis of distinct thalamic STMs, we examined the Allen Mouse Brain Connectivity Atlas (Oh et al., 2014) as in our previous work (Figure 5) (Mohajerani et al., 2013). We processed the three-dimensional structural data and matched the composed anatomical 2D maps with our two-dimensional static STMs. As expected, we found that STMs of spiking cortical neurons correspond with underlying structural axonal projections (see BCS1 example in Figure 5, [Mohajerani et al., 2013]). However, sub-cortical STMs cannot be predicted by direct monosynaptic projections from sub-cortical to cortical areas. For example, the HPF has no strong direct structural link to RS area. Furthermore, CP was not directly linked to BCS1/HLS1 areas (Hunnicutt et al., 2014Oh et al., 2014), indicating that sub-cortical STMs reflect apparently polysynaptic links to cortex.

Examples of STMs and projection maps.

(A) Example of STMs from neurons recorded in BCS1, VPM, VPL, HPF and CP. (B) Example of projection maps (2D surface and 3D) reconstructed from Allen Brain Atlas with injection sites (Oh et al., 2014) in the same region as our recording. For example, for a spiking neuron recorded in BCS1 we show anterograde labeling of GFP emanating from an injection site in BCS1 that extends to motor cortex and is present across cortex in the 2D surface plot of cortex. This projection pattern for BCS1 matches the STM map quite well as in previous work (Mohajerani et al., 2013). In contrast, for sub-cortical injections of GFP tracer such as in HPF there was less overlap between STMs and projection maps perhaps indicating polysynaptic pathways. Website: 2015 Allen Institute for Brain Science. Allen Mouse Brain Connectivity Atlas [Internet]. Available from: http://connectivity.brain-map.org.

https://doi.org/10.7554/eLife.19976.011

Cortical and sub-cortical neuron firing is tuned to cortical network dynamics spanning millisecond to multi-second time scales

By analyzing the cortical GCaMP signal, time courses corresponding to single neuron firing (firing= time 0) we observed dynamic cortical activity states (Figures 6 and 7; Videos 14). We determined STM Temporal Dynamics (STMTD) by identifying a region-of-interest (ROI) and tracking the time course of the maximally activated/depressed cortical pixel of the region from 3s before to 3s after spiking. As all our extracellular recordings were in right barrel cortex and predominantly right sensory thalamus, we identified the left barrel cortex as a co-activated area of interest and tracked dynamics within this ROI for spiking neurons. As observed for static STMs, the cortical calcium dynamics associated with spiking cortical neurons were relatively homogeneous with an initial peak in activity within ~100 ms – 200 ms following spiking and a return to baseline (Figure 6A). However, some cortical cells (~20%) participated in multi-second depression dynamics (see distribution of profiles in Figures 7B,C and and 8A). In contrast, STMTDs generated by thalamic cells were more varied and were dominated by depression dynamics (~80% of cells) lasting up to 3 s (Figures 6B and 7B,C and Figure 8B).

Figure 6 with 6 supplements see all
Montages of cortical and thalamic spatio-temporal dynamics.

(A) Top: right hemisphere barrel cortex neuron spiking time montage stereotyped dynamics in left hemisphere barrel cortex region-of-interest (ROI). The maximally activated pixel in the ROI (red arrow) is tracked over time and reveals Spike-Triggered-Map Temporal Dynamics (STMTD) which rises quickly at spike time t = 0 and decays in 100–200 ms followed by 1–2 s cortical depression (red curve in right plot). Bottom: Additional examples of right hemisphere barrel cortex neuron spike-triggered montages show similar barrel-motor cortex activation pattern with peaks in cortical activation shortly following spiking and a return to baseline or prolonged depression. (B) Top: Same as in A, but for a thalamically recorded neuron (right hemisphere) which correlates strongly with motor cortex activation shortly after spiking. Bottom: Additional montage examples of thalamic neurons (also right hemisphere) reveal both the spatial diversity (i.e. different STMs) and temporal diversity (i.e. different STM dynamics).

https://doi.org/10.7554/eLife.19976.012
Figure 7 with 1 supplement see all
Classification of STMTD patterns.

(A) Example of two STMTDs from pixels within L-BCS1 (left barrel cortex) from a single mouse recording (both cortical and subcortical neurons STMTD pixel locations shown). Each recorded cell STMTD has a slightly different maximum pixel amplitude location, but all fall within the L-BCS1 region. (B) STMTD PCA distribution from all 428 cortical and subcortical neurons recorded from all mice separated using KMEANS (k = 3). (Ci) STMTD patterns (±SD) classifications from (B). The number of neurons from cortex and thalamus used for the average are presented in title. (ii) Distribution of STMTD classification between cortical (clear) and subcortical (hashed) neuron generated STMTDs.

https://doi.org/10.7554/eLife.19976.019
Figure 8 with 1 supplement see all
Spatiotemporal patterns of STMs.

(A) Top: Normalized STMTDs (as in Figure 7) from maximum pixels tracked in multiple ROIs (HLS1, FLS1, BCS1, RS, V1, M1, PTA and ACC, see Table 1) for 255 cortical cells. Each horizontal line represents a single neuron's STMTD in each of the eight ROIs considered normalized to that neuron's STMTD maximum or minimum activation. Bottom: average and standard deviation of STMTD within each ROI for all cells. (B) Same as A, but for all thalamic neuron generated STMTDs. The thalamic STMTDs are more diverse, less temporally precise, and contain longer depression epochs – revealing ROI specificity and cortical vs. subcortical differences. These results are from awake mice.

https://doi.org/10.7554/eLife.19976.022
Video 1
Cortical neuron-triggered bilateral mesoscale calcium activity.

Left: dorsal cortex calcium dynamics triggered from a right hemisphere barrel cortex recorded cell. Right, time-course reflecting dynamics in the left hemisphere barrel cortex (green) alongside a lower activated area (red). Dynamics in region-of-interest (green) exhibit high activation correlating with spiking time (t = 0 s) followed by a 2 s depression in fluorescence signal. Activity illustrated 3 s prior to, and 3 s following cell spiking.

https://doi.org/10.7554/eLife.19976.024
Video 2
Thalamic neuron #1 triggered bilateral mesoscale calcium activity.

As in Video 1 but from a right-hemisphere thalamic neuron. Region-of-interest (green) in the left hemisphere exhibits a transient pre-spiking depression followed by activation correlating with spike timing.

https://doi.org/10.7554/eLife.19976.025
Video 3
Thalamic neuron #2 triggered bilateral mesoscale calcium activity.

As in Video 2 but from another thalamic neuron revealing peak activation in a different region-of-interest.

https://doi.org/10.7554/eLife.19976.026
Video 4
Thalamic neuron #3 triggered bilateral mesoscale calcium activity.

As in Video 2. Depressed cortical calcium activity is present in the left hemisphere barrel cortex region-of-interest prior to spiking and persists for an additional 2 s.

https://doi.org/10.7554/eLife.19976.027

Our analysis indicates that averaging GCaMP cortical motifs from all spikes of a single cell produces converging STMs and STMTDs. However, thalamocortical synapses are known to be prone to synaptic depression and burst pattern firing may yield altered cortical responses (Castro-Alamancos, 1997; Gil et al., 1997). It is conceivable that averaging all spike activity may mask STMs from sub-groups of spikes that contribute different motifs during varying ongoing cortical dynamics. One way to divide the spikes into groups is to partition them into bursting versus tonic modes. We found that both STMs and STMTDs are similar across spiking modes to the all-spike average STMs in both cortical and thalamic cells examined (Figure 6—figure supplements 1, 2 and 3; see Materials and methods). We also tried grouping spikes according to motif similarity but did not find natural groupings or clusters. Manually partitioning the motifs into groups of four or more, and removing the spontaneous motif average, revealed that the sub-group STMs were the same as the overall average (Figure 6—figure supplements 4 and 5; see Materials and methods). These tests confirm that the averaging method reveals how single cells participate in largely stereotyped networks despite the variability of ongoing cortical activity.

While our analyses have employed spike-triggering averaging, spike-triggered co-variance (Aljadeff et al., 2016) or variance maps are another way to view the association between GCaMP activity and spiking as opposed to the mean. Spike-triggered variance maps revealed similar brain regions as defined using spike-triggered average (STA) mapping when examined as a difference to baseline (Figure 6—figure supplement 6). It is conceivably that during particular behavioral states or tasks that variance mapping may reveal markedly different maps than STA.

We next investigated whether the multi-second corticocortical and thalamocortical dynamics revealed by this novel functional approach can be categorized into distinct groups (Figure 7Figure 7—source data 1). Identification of the location of maximally activated cortical regions provided a means of comparing cortical and thalamic neurons (Figure 7A). Using Principal Component Analysis (PCA) we plotted the distribution of 428 STMTDs from both cortical and subcortical cells (Figure 7B). The time courses were then clustered in three patterns of STMTDs (using k-means algorithm) and averaged. Based on their temporal relationship to spikes, we termed them: pattern #1: spike-triggered-excitation; pattern #2 spike-triggered inhibition; pattern #3 inhibition-triggered spiking followed by inhibition (Figure 7Ci). The distribution of these three patterns were then compared between cortical and thalamic cells (Figure 7Cii).

This classification showed that ~80% cortical neurons were associated with a pure cortical excitation profile (pattern #1) and 20% triggered inhibition (pattern #2). In contrast, only 20% of thalamic neurons associated with post-spiking cortical excitation, while 80% of were associated with cortical inhibition patterns (45% with pattern #2% and 35% with pattern #3). Notably, pattern #3 was only identified with subcortical neuron spiking. Neither cortical cell depth, subcortical cell location (e.g. VPM vs VPL), nor cell-type classification (inhibitory and excitatory types [Connors and Gutnick, 1990; Nowak et al., 2005]) revealed any significant correlations between temporal dynamic clusters and cell-classification (not shown). We suggest future work with larger datasets and multiple cortical/subcortical areas could address the question of whether STMTD classification are a novel intrinsic single-cell property.

The profile of STMTDs across all mice and recordings were also presented by aligning their peak activation (red) or peak depression (blue) in various bilateral ROIs that span wide and varied regions of cortex (Figure 8). The analysis in Figure 8 provides a powerful means of visualizing group-data temporal relationships across ROIs for both cortical and sub-cortically derived spikes. We found that as observed for individual neuron STMTDs, aggregate cortical neuron spikes were generally followed by strong periods of persistent activation (Figure 8A), whereas sub-cortical neurons were linked to more diverse cortical activity profiles, in particular longer depression across multiple cortical ROIs (Figure 8B).

The role of movement and hemodynamics STMs

Our goal is to assess cortical functional connectivity based on coincidence between individual neuron spiking and ongoing spontaneous activity. However, we concede that in awake animals activity is rarely entirely spontaneous and that periods of volitional activity are present interspersed with relatively quiet intervals. Qualitative observations indicate limb twitches as well as tail and facial movements. In order to evaluate the impact of body movement on mapping, STAs were generated from spikes during periods of quietness and compared with STA from all spikes (Figure 8—figure supplement 1). Period of quietness and movement were identified by measuring behavior imaging collected simultaneously with the neurophysiological data (see Materials and methods, Figure 8—figure supplement 1A and Video 5). STA generated from periods of quietness did not differ from those generated with all spikes (see Materials and methods, Figure 8—figure supplement 1B). This suggests that, even when some movement could be observed during recordings, they minimally contribute to the mapping. This data also indicated that periods of high movement were relatively rare in awake head-fixed mice under the conditions we have employed and contributed negligibly to overall STA maps. Therefore, brain imaging activity obtained in awake states is mostly indicative of a quiet awake state and is not primarily movement-related activity.

Video 5
Evaluation of whisker movement.

Left: Average of absolute gradient within the region of interest (yellow box in Figure 8—figure supplement 1) between the frames 3400 and 5000. Right: Corresponding image frames displayed at real time.

https://doi.org/10.7554/eLife.19976.028

Other potential sources of error include apparent changes in GCaMP6 signal due to alterations in blood volume or oxygenation (Ma et al., 2016; Wekselblatt et al., 2016). Increases in blood volume are expected to decrease both excitation and emission light. While we have implemented a multi-wavelength correction strategy (see Materials and methods), the corrections in practice did not appreciably alter cortical or thalamic STA maps or dynamics altering peak % ΔF/Fby more than 1/10 of the maximum ΔF/F0. We provide examples of the strategy being employed (Figure 7—figure supplement 1A,B). In these examples, we employ a short blue reference signal that correlated positively with apparent blood volume artifacts that were revealed by parallel experiments using green reflected light imaging (Figure 7—figure supplement 1C,D). Consequently, the ratio of green over blue signal greatly reduced the blood volume hemodynamic response. To determine whether the short blue correction strategy was effective, we assessed data from GFP-m mice that exhibit green fluorescence signals that are not expected to be the calcium dependent as in GCaMP6 mice (Figure 7—figure supplement 1E). We used the ratio of F/F0 green fraction signal to blue reflected light signal F/F0 fraction to reduce nonspecific signals that were observed in GFP mice. Consistent with published work (Ma et al., 2016; Wekselblatt et al., 2016) blood volume artifacts were greatly reduced. This same strategy was then applied to GCaMP fluorescence data. Although the approach was effective at removing smaller non-specific signals in GFP-m mice, in practice, using GCaMP mice, where much larger activity-dependent signals were present, we revealed only a relatively small apparent contribution of blood volume to cortical and subcortical STA fluorescent signals as in other work (Vanni and Murphy, 2014; Murphy et al., 2016; Silasi et al., 2016) (Figure 7—figure supplement 1F,G). Furthermore, thalamic STA maps and STA dynamic plots still indicated cases where thalamic spiking was associated with cortical inhibition. For these reasons, we have attempted to correct other STA maps using this multi-wavelength procedure.

Discussion

Using STM of wide-field spontaneous calcium imaging data, we have characterized mesoscopic cortical maps defined by the spiking activity of individual cortical/sub-cortical neurons. Our results demonstrate that STMs can reveal functional cortical architecture related to the activity of individual cortical and subcortical-cortical neurons. Cortically recorded STMs reflect the cortical state when a neuron spikes in connected areas. We observed that the STMs of individual sub-cortical neurons had more variation than maps attributed to spiking cortical neurons. For example, sub-cortical STM patterns of neighboring neurons were more diverse than cortical neurons, and were less likely to match intra-cortical consensus activity patterns defined using SPM. Sub-cortical-neuron-derived STMs revealed multiple areas of activation and multimodal kinetic behavior, while intra-cortical spiking neuron networks were simpler in structure and kinetics. Furthermore, spiking sub-cortical neurons reflected diverse cortical multi-phasic excitation/inhibition timing patterns that were reflected in dynamic STMTDs. In contrast, most spiking cortical neurons were linked to a single phase of cortical excitation.

Event-triggered brain mesoscale mapping

Previously, STA of local field potentials has been used to investigate of how single neurons in visual cortex were linked to on-going state-dependent activity (Nauhaus et al., 2009); however, this work only examined such correlations locally within visual cortex and did not assess regional connectivity using imaging or investigate differences with individual sub-cortical neurons. Other similar applications where single neuron spiking was recorded and related to spontaneous activity using calcium imaging have been restricted to in vitro brain slices (Aaron and Yuste, 2006)). Our study extends previous in vivo work that assessed spike-triggered events using voltage-sensitive dye imaging (Arieli et al., 1995; Tsodyks et al., 1999) to encompass a more expansive spatial scale, electrode arrays, awake recordings, and selective genetically encoded indicators of activity. While being important seminal findings (Arieli et al., 1995; Tsodyks et al., 1999), previous STA work was largely confined to the visual system, performed under anesthesia and was unable to define how collections of brain areas interact. The approach is probably most analogous to event-triggered MRI imaging from the standpoint of larger spatial scale (Logothetis et al., 2012). Interestingly, in this study, it was observed that during hippocampal ripple states that cortex exhibited net positive bold responses and thalamus net negative BOLD responses. This anti-correlated relationship is consistent with some of our observations of thalamic spiking activity corresponding with cortical temporal dynamics exhibiting slow depression of calcium signals and may point to a larger coordinated network involving other brain structures. However, as we did not record simultaneously from hippocampus we cannot definitively comment on whether these observations necessarily correspond to hippocampal ripple events. Furthermore, MRI signals lack temporal resolution and can be more difficult to relate to neuronal activity than GCaMP signals that are isolated within excitatory neurons of GCaMP6f mice using specific promoters (Chen et al., 2013b; Madisen et al., 2015). Unique to our approach is the power to assess the functional connectivity and temporal dynamics between specific sub-cortical neurons and areas of cortex not predicted by previous knowledge such as linkages between thalamic neurons and cortical state as defined by GCaMP signal dynamics. We anticipate that the approach can be further refined as tools such as more selective cre-dependent CGaMP6f transgenic mice allow for the expression of calcium indicators in particular neuron types (Madisen et al., 2015). Furthermore, two-photon microscopy could be used to provide information about behavior of individual cells within the context of larger maps (Chen et al., 2013a; Guo et al., 2014; Okun et al., 2015). Because of the high sensitivity of the indicator and the possibility of measuring the activity of tens of single-units using multiple electrode channels simultaneously, a large number of functional connections can be mapped in only a few minutes of recording. Although we have only used a single electrode shank, we anticipate future applications using higher density electrodes and multiple shanks to collect spikes from more neurons simultaneously.

Relationship between cortical and sub-cortical spiking derived maps

Our results indicate that the neocortex is divided into discrete subdivisions where individual spiking cortical neurons generally belong to spatial-temporal maps that follow a consensus function that can be defined using correlation as in previous work (Mohajerani et al., 2013; Chan et al., 2015). In contrast, we show that single thalamic neurons tend to fire when cortex is in more kinetically diverse states which is dominated by inhibition. The more diverse dynamics between thalamic neurons and cortical mesoscopic networks indicate that sub-cortical thalamic neurons play an instructive role with respect to cortical state, particularly with respect to feedforward cortical inhibition (Stroh et al., 2013; Urbain et al., 2015), whereas cortical neurons may serve as relay endpoints or amplifiers (Douglas et al., 1995). A better understanding of these dynamics may yield insight into how disorders, such as epilepsy, and dementia, emerge when interactions between brain areas are disrupted (Paz et al., 2013; Busche et al., 2015; McCormick et al., 2015). The diversity in sub-cortical spiking derived maps may also reflect differing receptive field properties in thalamus and cortex based on varying types of functional convergence described previously (Miller et al., 2001). Indeed, in the somatosensory whisker barrel system, evidence for ‘ensemble convergence’ has been described where input from the thalamus can extend outside of the boundaries of the corresponding cortical receptive field (Simons and Carvell, 1989; Linden and Schreiner, 2003). The larger diversity of maps derived from the spiking of different thalamic neurons may be expected because of the smaller size of thalamic nuclei compared to the cortex and the recording of thalamic neurons from more-varied structures. Another potential source of variance may arise from the diversity of thalamocortical impulses that can be comprised of patterns of activity ranging from tonic, 'relay' transmission consisting of high regular rates of firing to burst-like activity where firing rates are low and interspersed with high-frequency events (Steriade and Llinas, 1988; McCormick and Feeser, 1990; Sherman and Guillery, 1996). Thalamic bursting can powerfully activate neocortical circuits and has been suggested to serve a ‘wake-up’ signal to sensory cortices (Sherman and Guillery, 1996; Swadlow and Gusev, 2001). When we segregated our recordings into various firing configurations, we did not observe profound differences in STMs or their temporal dynamics indicating that the averaging methodology is not constrained by a particular firing pattern. However, interpreting these results is caveated by the mesoscale resolution and calcium dynamics present in the recorded data.

Applications of spike-triggered mapping

Mapping the functional connectivity around identified spiking neurons is important for understanding brain function and finding therapeutic targets for brain stimulation or brain machine interfaces. Identification of networks linked to individual neurons may help reveal the mechanism of brain machine interfaces where key signals are often attributed to only a small number of neurons (Stanley et al., 1999; Serruya et al., 2002; Taylor et al., 2002; Guggenmos et al., 2013). Other applications include understanding of how small groups of epileptic neurons (Paz et al., 2013) are coupled to brain networks leading to seizure propagation. Given that reciprocal connections between mesoscale structures are widespread, the cortical maps associated with a spiking neuron in a sub-cortical structure such as the sub-thalamic nucleus may provide clues as to how cortical activity can be manipulated to affect a sub-cortical target. This hypothesis can be tested by recording sub-cortically using electrode arrays, while stimulating regions of cortex that show coincident STMs using Channelrhodopsin-2 or other opsin-activity sensor pairs (Lim et al., 2012; Rickgauer et al., 2014; Zou et al., 2014; Abdelfattah et al., 2016; Kim et al., 2016a).

Kinetics of spike-triggered mapping

By computing spike-triggered calcium imaging averages, we reduce the contribution of neurons which fire out of phase. Analysis of STMTDs indicate slower time to peak (~100 ms) than postsynaptic potentials evoked by a single synaptic connection (time to peak ~20 ms) (Bruno and Sakmann, 2006). Slower dynamics are expected given the kinetics of GCaMP6 (Chen et al., 2013b). We have used deconvolution (Pnevmatikakis et al., 2016) to take into consideration the slower kinetics of GCaMP6 and compensate for it. Using this approach, we observed a significant acceleration of raw data but very modest effects on STMTD indicating that STMTD may already be accelerated relative to GCaMP6 kinetics by the statistical nature of spike/Ca2+ transient temporal convergence. It is also possible that slower dynamics of STMTD reflect sequences of spiking activity propagating through specific polysynaptic circuits. This speculation was supported by the similar time range of STMTD and cue-triggered recall of learned temporal sequences (Xu et al., 2012). STMTD kinetics can be improved using faster sensors such as organic voltage-sensitive dyes (Shoham et al., 1999; Mohajerani et al., 2013), or genetically encoded voltage (Carandini et al., 2015; Gong et al., 2015; Abdelfattah et al., 2016) or glutamate sensors (Xie et al., 2016).

Extension to behaviorally driven activity

We acknowledge that the same approach can be extended to making STMs during specific behaviors. During specific behaviors we do not expect major shifts in area map boundaries found during spontaneous activity, as we believe these are largely determined by projection anatomy (Mohajerani et al., 2013; O'Connor et al., 2013Oh et al., 2014; Zingg et al., 2014) and in the case of sub-cortical neuron maps (HPF for example) poly-synaptic, hard-wired connections. During behavior we expect more nuanced changes in the weighting, timing, and frequency-dependence of STM networks during an active task. It is possible that specific behaviors will reveal the superposition of multiple cortical motifs associated with progression through the task. STM mapping of cortex would be particularly interesting the context of rhythmic whisking-related centers within the medulla and thalamus and their linkage to cortical maps within barrel-motor areas (Moore et al., 2013; Deschênes et al., 2016; Sreenivasan and Petersen, 2016).

Conclusion

We conclude that the analysis of single neuron spiking activity can reliably reflect mesoscale activity transitions within mouse cortex. STMs together with connectomic information (Hunnicutt et al., 2014Oh et al., 2014; Zingg et al., 2014; DeNardo et al., 2015) may help to bridge the gap between single neuron function and larger networks. Our data have already revealed that thalamic neurons interact with cortex during specific state transitions that not reflected by typical consensus cortical neuron behavior. We have exploited spontaneous activity as a means of sampling active cortical networks. The presence of such obligate long-range relationships in even spontaneous activity may suggest new opportunities and routes by which brain stimulation and inhibition can be applied to affect synaptically connected areas.

Materials and methods

Animals

Animal protocols (A13–0336 and A14–0266) were approved by the University of British Columbia Animal Care Committee and conformed to the Canadian Council on Animal Care and Use guidelines and animals were housed in a vivarium on a 12 hr day light cycle (7 AM lights on). Most experiments were performed toward the end of the mouse light cycle. Transgenic GCaMP6f mice (males, 2–4 months of age, weighing 20–30 g; n = 16), were produced by crossing Emx1-cre (B6.129S2-Emx1tm1(cre)Krj/J, Jax #005628), CaMK2-tTA (B6.Cg-Tg(Camk2a-tTA)1Mmay/DboJ, Jax #007004) and TITL-GCaMP6f (Ai93; B6;129S6-Igs7tm93.1(tetO-GCaMP6f)Hze/J, Jax #024103) strain (Madisen et al., 2015). Transgenic GCaMP6s mice (n = 3) were produced by crossing Emx1-cre (B6.129S2-Emx1tm1(cre)Krj/J, Jax #005628), CaMK2-tTA (B6.Cg-Tg(Camk2a-tTA)1Mmay/DboJ, Jax #007004) and TITL-GCaMP6s (Ai94;B6.Cg-Igs7tm94.1(tetO-GCaMP6s)Hze/J, Jax #024104) strain. Transgenic GCaMP3 mice (n = 8) were produced by crossing Emx1-cre and R26-GCaMP3 (Ai38; B6;129S-Gt(ROSA)26Sortm38(CAG-GCaMP3)Hze/J, Jax #014538) strain (Zariwala et al., 2012; Vanni and Murphy, 2014). The presence of GCaMP expression was determined by genotyping each animal before each surgical procedure with PCR amplification. These crossings are expected to produce a stable expression of the three calcium indicator variants (GCaMP3, GCaMP6s and GCaMP6f [Chen et al., 2013b]) specifically within all excitatory neurons across all layers of the cortex (Vanni and Murphy, 2014). Control experiments, assessing the specificity of STM, were performed in Thy1-GFP-M mice (n = 6; Jax #007788). No method of randomization was used since all mice belonged to the same sample group. Samples sizes were chosen based on previous studies using similar approaches (Mohajerani et al., 2013; Vanni and Murphy, 2014; Chan et al., 2015). Given the use of automated acquisition and analysis procedures we did not employ blinding.

Surgery

Mice were anesthetized with isoflurane (1.5–2%) for induction and during surgery and a reduced maintenance concentration of isoflurane (0.5–1.0%) or urethane was used later during anesthetized data collection. In some cases, animals were allowed to wake up following isoflurane anesthesia for awake imaging (see section ‘Multimodal recording in awake mice’). Throughout surgery and imaging, body temperature was maintained at 37°C using a heating pad with a feedback thermistor. For cortical experiments, mice were placed on a metal plate that was mounted on a macroscope. The skull was fastened to a stainless steel head-plate and was connected with tubing to a water pump, which circulated temperature controlled 37°C water to ensure physiological temperature. A 9 × 9 mm bilateral craniotomy (bregma 3.5 to −5.5 mm, lateral −4.5 to 4.5 mm) covering multiple cortical areas was made as described previously (Mohajerani et al., 2013). For sub-cortical experiments, mice were placed in a stereotaxic apparatus and an incision was made in the midline to expose the skull as in cortical experiments. A burr hole was then unilaterally drilled (usually in the right hemisphere) above the thalamic area (stereotaxic coordinates considering a 45° angle (Figure 1A): between 1.7 ± 0.3 mm posterior to bregma and 1.6 ± 0.4 mm lateral to midline. We estimated angular tilt relative to a perpendicular penetration to the cortical surface of less than 5° (Hunnicutt et al., 2014). In order to minimize movement artifact (due to breathing and heartbeat), the exposed skull was fastened to a stainless steel head-plate with cyanoacrylate glue and dental cement. In cases where the laminar probe was inserted (as opposed to a glass electrode), a craniotomy was only made for the probe insertion site and cortical GCaMP imaging was performed through intact bone.

In vivo single unit recording

For initial glass pipette recordings, the pipette was advanced into the targets (HLS1, FLS1, BCS1, V1, M1, ACC and RS) on the cortical surface (within the right hemisphere) at an angle of 30° from the horizontal using a motorized micromanipulator (MP-225, Sutter Instrument Company). Signals were recorded through a silver wire placed inside the micropipette and were amplified (MultiClamp 700A or Axopatch 200B, Molecular Devices) and digitized at 12.5 kHz (Digidata 1322A, 16-bit Data Acquisition System). A reference electrode, teflon-coated, chlorided silver wire (0.125 mm) was placed on the right edge of the craniotomy. For multichannel 16-channel laminar electrode (NeuroNexus, A16–10 mm-100-177) recordings, the electrode was directed toward the center of the burr hole using a motorized micromanipulator (MP-225, Sutter Instrument Company). The electrode was inserted into the right hemisphere with a 45° angle from the lateral surface of the cortex to avoid contact with the region where wide-field imaging was performed. The electrode was first held in cortex (BCS1) for cortical recordings and then advanced into the sub-cortical areas. The multichannel signal was amplified using 16-channel data acquisition system (20 kHz, USB-ME16-FAI-System, Multi-Channel Systems) and recorded for at least 5 min for each recording site. To minimize tissue damage, all experiments were performed by only a single insertion of the laminar electrode, and most trajectories were from barrel cortex to thalamus.

Calcium imaging

Images of the cortical surface were recorded through a pair of front-to-front video lenses (50 mm, 1.4 f:30 mm, 2 f) coupled to a 1M60 Pantera CCD camera (Dalsa) (Vanni and Murphy, 2014). To visualize the cortex, the surface of the brain was illuminated with green light (but not during image acquisition). Calcium indicators were excited with blue-light-emitting diodes (Luxeon, 470 nm) with bandpass filters (467–499 nm). Emission fluorescence was filtered using a 510–550 nm bandpass filter or collected in a multi-band mode as described below. For single wavelength green epifluorescence, we collected 12-bit images at varying time resolution (20–100 ms; i.e., 10–50 Hz) using XCAP imaging software. In order to reduce file size and minimize the power of excitation light used, we typically bin camera pixels (8 × 8) thus producing a resolution of 68 µm/pixel. These imaging parameters have been used previously for voltage-sensitive dye imaging (Mohajerani et al., 2013) as well as anesthetized GCaMP3 imaging of spontaneous activity in mouse cortex (Vanni and Murphy, 2014) and awake GCaMP6 imaging in mouse cortex with chronic window (Silasi et al., 2016).

In some experiments (as indicated), we employed a multi-wavelength strategy to correct for potential green epifluorescence signals that were associated with non-calcium dependent events. We employed a variant of the elegant spectral correction strategy described by others (Ma et al., 2016; Wekselblatt et al., 2016) that monitor changes in green reflected light near the isobestic point of hemoglobin. This strategy was inspired by previous work using blue excitation/reflected light(Sirotin, 2010). We assume hemoglobin is the primary absorber in brain tissue in vivo and changes in blood volume or oxygenation affect both excitation and emission of light used for wide-field imaging (Ma et al., 2016). Our strategy makes use of short blue wavelength reference light that is also near a hemoglobin isosbestic point. While others have used a strobed LED presentation with a subset of frames providing a green reflected light reference image (Ma et al., 2016; Wekselblatt et al., 2016), we took advantage of an RGB camera sensor to allow simultaneous acquisition of a shorter wavelength blue ~447 nm signal that correlates strongly with green reflected light signals. This strategy provides a short blue light reference without the need for strobing which can limit time resolution and potentially entrain some neuronal rhythms (Iaccarino et al., 2016) and is more technically demanding from a hardware synchronization standpoint. Our strategy employs the Raspberry Picam’s RGB sensor (Waveshare Electronics RPi Camera F) to independently resolve signals attributed to blood volume changes as blue reflected light, while simultaneously collecting green epi fluorescence (GCaMP6). We used a Chroma 69013m multi-band filter 10 mm diameter mounted just over the image sensor allowing blue, green, and red signals to be simultaneously obtained in separate channels of the cameras RGB sensor with less than 10% cross talk between channels. We employ 2 Luxeon LEDs: (1) Royal-Blue (447.5 nm) LUXEON Rebel ES LED with added Brightline Semrock 438/24 nm filter to provide a short blue wavelength reflected light signal that is expected to report blood volume changes; (2) a blue 473 nm Luxeon Rebel ES LED for excitation of GCaMP6 with a Chroma 480 nm/30 nm excitation filter. In preliminary analysis, we found that the short blue signal correlated positively with apparent blood volume artifacts that were revealed by parallel experiments using green reflected light imaging (r = 0.93, see Figure 7—figure supplement 1C). Given that the short-blue reflected light signals provided a surrogate indicator of green reflected light (they are highly positively correlated) we used this in a ratiometric correction strategy. While the shorter blue wavelength light will scatter more than a green strobed reflected light signal used by others (Ma et al., 2016; Wekselblatt et al., 2016), our analysis of green reflected light and short blue reflected light, indicating that two were highly correlated, suggests that the major artifacts we observe are associated with large blood volume changes in superficial cortical layers.

Multimodal recording in awake mice

To initiate wakefulness isoflurane and oxygen were stopped and the anesthesia mask was removed. Calcium imaging data were obtained over the following 1 hr. The body temperature of mice was maintained with a heating pad. Awake calcium imaging of spontaneous activity was performed in the absence of visual and auditory stimulation. A behavioral monitoring camera was used to confirm that the mice were indeed awake and relatively unstressed as grooming and whisking were occasionally observed. An analgesic, buprenorphine, was injected (0.075 mg per kg body weight intraperitoneally) 24 hr before awake calcium recordings. We used a second Dalsa 1M60 camera (150 Hz) or Raspberry Picam’s RGB sensor (60 Hz) to capture body and whisker movements under infrared illumination.

While relatively few large body movements were observed during awake imaging sessions, their impact on mapping was evaluated by generating STA from period of quietness. To identify regions of movement or quietness, the standard deviation of luminance fluctuation was calculated for each pixel. This approach showed that most of the movements were localized on the facial (whisker and jaw) and forepaw regions. A region of interest was manually drawn for each frame and the sum of absolute value of the gradient was calculated by subtracting each frame by their previous within this region. This gradient profile within the region of interest was temporally smoothed at 0.1 Hz, and the median and standard deviation were calculated (σ). Periods of quietness were identified as having a gradient lower than [median+σ/10], while periods of movement were higher than [median+σ]. To more selectively identify periods of quietness isolated from any movement, an exclusion window of 10 s was applied at the beginning and the end of each period of quietness and only periods of more than 10 s were kept. STA was then generated only from spikes for periods of quietness and compared with STA of all spikes.

Sensory stimulation

Sensory stimuli were used to confirm sensory cortical and sub-cortical areas using forelimb, hindlimb, whisker and visual stimulation. To stimulate the forelimbs and hindlimbs, thin acupuncture needles (0.14 mm) were inserted into the paws, and a 0.2–1 mA, 1 ms electrical pulse was delivered. To stimulate a single whisker (C2), the whisker was attached to a piezoelectric device (Q220-A4-203YB, Piezo Systems, Inc., Woburn, MA) and given a single 1 ms tap using a square pulse. The whisker was moved at most 90 μm in an anterior-to-posterior direction, which corresponds to a 2.6° angle of deflection. A 1 ms pulse of combined green and blue light was delivered as visual stimulation. Averages of sensory stimulation were calculated from 20 to 40 trials of stimulation with an inter-stimulus interval of 10s.

Single unit activity analysis

Raw extracellular traces were imported into Spike2 (Cambridge Electronic Design, Cambridge, UK) or SpikeSorter software (Swindale and Spacek, 2014) for spike sorting and analysis. Briefly, data were high pass-filtered at 1 kHz, and single spikes were detected using a threshold of 4.5 times the median of the standard deviation over 0.675. Sorting was carried out by an automated method previously described (Swindale and Spacek, 2014) and followed by manual visual inspection of units. For analysis, we only used only units with a peak-to-peak extracellular amplitude of at least 40 μV, a minimum of 200 spikes, and a calcium cortical response (STMTD, see next section) of at least 0.5% to improve single unit sorting isolation, STM map stability and signal-to-noise ratios for STMTD clustering (see main text), respectively.

Calcium imaging analysis

TIF files of raw fluorescence were imported and processed using Matlab (Mathworks, Inc. Natick, MA) or Python2.7 custom codes. For each spike, we averaged the cortical fluorescence of every pixel during the period preceding the spike (−3 to 0 s: baseline). ΔF/F0 was then performed by subtracting and dividing the baseline to each frame for each spike within a time window of ±3 s. We tested additional methods for computing ΔF/F0 including computing baseline from the average of the entire recording or band pass filtering the data (0.1 Hz to 6.0 Hz) but the results were similar. STA (spike-triggered average) sequences were created by averaging the ΔF/F0 responses for each trial and were compared with random spike STA. STA maps (STM) were then defined as the maximum response calculated for each pixel within a time window of ±1 s.

STM temporal dynamics (STMTDs) were defined as the time course of activity of the maximally activated pixel in a region of interest. The resulting clusters were separated using k-means clustering algorithm. Putative cell classification into inhibitory and excitatory cell types was based on the full-width-half-max of each unit's positive and negative phases (Connors and Gutnick, 1990; Pape and McCormick, 1995).

To calculate seed pixel correlation maps (SPM), the contribution of global and illumination fluctuations was subtracted from the signal of each pixel (Vanni and Murphy, 2014) and the spontaneous activity recording sequences were temporally pass-band filtered (0.3–3 Hz). Then, cross-correlation coefficient r values between the temporal profiles of one selected pixel and all the others were calculated (White et al., 2011; Mohajerani et al., 2013; Vanni and Murphy, 2014). Similarities between STM and SPM maps were performed by measuring the r-value Pearson correlation coefficient between each pair of pixels.

To compare the STM with anatomical database, brain stacks of 140 slices were downloaded from the Allen Mouse Brain Connectivity Atlas providing AAV-virus tracing database (http://connectivity.brain-map.org/, [Oh et al., 2014]). For each slice, the first dorsal 300 µm of brain fluorescence in the Z-axis were summed to generate partial maximum Z-projection maps similarly to previous studies (Mohajerani et al., 2013).

Deconvolution

We temporally deconvolved our calcium imaging pixel-by-pixel using the method presented by Pnevmatikakis et al. (2016) and code provided by the authors on Github (https://github.com/epnev/ca_source_extraction). Briefly, the method uses an autoregressive approach to estimate the calcium transient as an impulse response from the data itself (Pnevmatikakis et al., 2016). Making use of this, the time course of calcium transients is then deconvolved using a computationally efficient non-negative, sparse, constrained deconvolution.

Cell spiking mode determination

We implemented methodology previously described for defining main spiking modes (Sherman et al., 2006) (Figure 6.5, pg 236). Briefly, the method requires determination of the distribution of each spike’s inter-spike-interval (ISI) between the previous (x-axis) and following (y-axis) spike (Figure 6—figure supplement 1). The distribution is then plotted using logarithmic scales and naturally arising clusters are grouped or clustered. Spike groups occurring in approximately each quadrant of the plot indicate different spiking modes: first spikes in a burst (bottom right), spiking occurring during a burst (bottom left), last spikes in a burst (top left), and tonic spikes (top right). The vast majority of cortical cells we recorded in barrel cortex did not exhibit multiple classes of spiking modes (2 examples provided where some natural clustering is present: Figure 6—figure supplement 1A and B), and only a few of thalamic cells we recorded showed clear bursting modes (two-examples provided Figure 6—figure supplement 1C and D) while also passing our minimum thresholds (see Materials and methods: Single unit activity analysis).

Single-spike motif sub-network analysis

We sought to determine whether sub-groups of spike STMs from a single cell could cluster and yield average STMs that were different than the all-spike average STM. Our approach was to group single spike motifs by similarity in a high-dimensional space (e.g. 64 × 64 pixels=4096 dimensions) and removed spontaneous motifs averages that were similar to our groupings to reveal the underlying single cell STM. The first step was to compute distributions of the cortical STMs of all spikes (which have high variability; Figure 6—figure supplement 4A) in a high dimensional space. The lack of obvious clusters indicated that there were no sub-groups of STMs present in the data which are removed by the averaging procedure. We proceeded to group the STM distributions by similarity into four (or more) partitions to reveal active sub-networks present during single spiking (Figure 6—figure supplement 4B). While inter-spike-interval (ISI) distribution were largely similar for the sub-grouped networks (Figure 6—figure supplement 4C) the resulting four sub-networks had substantial diversity indicating that (on averae) spiking occurred during different types of active cortical networks with only one of these sub-networks resembling our all-spike average STM (Figure 6—figure supplement 4D; four sub-network STMs and sum at the bottom). We next identified spontaneous STMs – that is STMs occurring without spiking – that were most similar to our sub-networks to subtract their contribution and reveal the cell's component STM in the sub-network STMs. We thus grouped spontaneously occurring motifs (Figure 6—figure supplement 4E) into sub-networks similar to spike triggered networks by re-using the spike generated sub-network centres (Figure 6—figure supplement 4B,F). This ensured that the spontaneous sub-networks would be similar to the spike-triggered sub-networks. The resulting spontaneous sub-networks are similar – but not identical – to the cell spike triggered sub-networks (Figure 6—figure supplement 4H; note that the sum is mostly noise as expected when summing over all activity). Importantly, when subtracting the spontaneously active sub-network STMs (Figure 6—figure supplement 4H) from the spike-triggered sub-network STMs (Figure 6—figure supplement 4D), we recovered STMs which represented mainly single cell spiking components and were largely similar to the overall average STM. We provide additional examples of this STM partitioning using other cortical and thalamic cells and also an example where we partitioned the STM distribution into 12 sub-networks but still recovered the all-spike average STM from each of the partitioned networks (Figure 6—figure supplement 5).

Code availability

A mixture of custom Python and Matlab code was used for analysis (https://github.com/catubc/sta_maps).

Histology

Prior to each recording, pipettes were filled with dye (Texas red-dextran) or the rear of a laminar electrode shank (side opposite the recording sites) was painted with fluorescent 1, 1-dioctadecyl-3,3,3,3-tetramethylindocarbocyanine perchlorate (DiI, ~10% in dimethylfuran, Molecular Probes, Eugene, OR). Dye-labeled pipettes and electrodes were not used until the dimethylfuran solvent had evaporated. At the end of each experiment, animals were killed with an intraperitoneal injection of pentobarbital (24 mg). Mice were transcardially perfused with PBS followed by chilled 4% PFA in PBS. Coronal brain sections (50 µm thickness) were cut on a vibratome (Leica VT1000S). Images of diI labeling with counter-stained DAPI were acquired using confocal microscopy (Zeiss LSM510) to reveal the electrode track and help identify the approximate subcortical location of recorded single units.

Statistics

Data were analyzed using GraphPad Prism six and custom written software in MATLAB and Python2.7. Mann-Whitney non-parametric tests were used to compare correlation coefficients between STMs, percentage overlap between STM pairs, and the correlation coefficients between STMs and SPMs. **** denote p<0.0001. Collection of data and analysis were not performed blind to the nature of the experiment, and there was no randomization of animal groups. STMs included in analyses were generated from a minimum of 200 spikes. Only STMTDs exhibiting fluorescence exceeding 0.5% △F/F0 were used for analysis. Sample sizes were not pre-determined but are consistent with previous experiments using similar methodology (Mohajerani et al., 2013; Vanni and Murphy, 2014; Chan et al., 2015).

References

  1. 1
  2. 2
  3. 3
  4. 4
    Coherent spatiotemporal patterns of ongoing activity revealed by real-time optical imaging coupled with single-unit recording in the cat visual cortex
    1. A Arieli
    2. D Shoham
    3. R Hildesheim
    4. A Grinvald
    (1995)
    Journal of Neurophysiology 73:2072–2093.
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
    Dynamics of propagating waves in the olfactory network of a terrestrial mollusk: an electrical and optical study
    1. D Kleinfeld
    2. KR Delaney
    3. MS Fee
    4. JA Flores
    5. DW Tank
    6. A Gelperin
    (1994)
    Journal of Neurophysiology 72:1402–1419.
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
  61. 61
  62. 62
  63. 63
    Exploring the Thalamus and Its Role in Cortical Function
    1. SM Sherman
    2. RW Guillery
    3. SM Sherman
    (2006)
    Cambridge, Massachusetts: MIT Press.
  64. 64
    Functional organization of thalamocortical relays
    1. SM Sherman
    2. RW Guillery
    (1996)
    Journal of Neurophysiology 76:1367–1395.
  65. 65
  66. 66
  67. 67
  68. 68
    Thalamocortical response transformation in the rat vibrissa/barrel system
    1. DJ Simons
    2. GE Carvell
    (1989)
    Journal of Neurophysiology 61:311–330.
  69. 69
  70. 70
  71. 71
  72. 72
    Reconstruction of natural scenes from ensemble responses in the lateral geniculate nucleus
    1. GB Stanley
    2. FF Li
    3. Y Dan
    (1999)
    The Journal of Neuroscience 19:8036–8042.
  73. 73
    The functional states of the thalamus and the associated neuronal interplay
    1. M Steriade
    2. RR Llinás
    (1988)
    Physiological Reviews 68:649–742.
  74. 74
  75. 75
  76. 76
  77. 77
  78. 78
  79. 79
  80. 80
  81. 81
  82. 82
  83. 83
  84. 84
  85. 85
  86. 86
  87. 87
  88. 88
  89. 89

Decision letter

  1. David Kleinfeld
    Reviewing Editor; University of California, San Diego, United States

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Mapping cortical mesoscopic networks of single spiking cortical or sub-cortical neurons" for consideration by eLife. Your article has been reviewed by two peer reviewers, and the evaluation has been overseen by David Kleinfeld as Reviewing Editor and Timothy Behrens 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.

Summary

The authors combined single neuron electrical recordings of spiking output with wide field imaging of the calcium transients in cortex. This allows them to form a transfer function of the spatiotemporal patterns in cortex in response to spiking from a single neuron in with cortex or thalamus. These patterns may extend across multiple cortical areas, consistent with long-range connections. Further, they are more robust when derived with respect to cortical versus thalamic spikes. These data provide key, additional evidence for a highly restricted number of patterns generated by cortex, even from "spontaneous" activity, compared with the number of neurons in a region of interest.

Essential revisions:

Both reviewers as well as the Reviewing editor find this to be a carefully executed set of experiments and praise to quality of the imaging over an unprecedented nearly 1 cm by 1 cm field of view. The reviewers raise excellent issues and I would like the author to respond to all comments. Regarding "essential" points, one reoccurring theme that will require additional analysis concerns the averaging. First, it would be very useful to see variance maps as well as average or spike triggered maps (STMs) as an extension of Figures 3 and 4. The two may not be scaled versions of each other if the calcium responses are nonlinear. For that reason, the authors may further consider an analysis of the covariance of the data, as opposed to only the mean, as would reveal properties of "complex" versus "simple" visual cells. Perhaps this issue is at bay with the thalamic data. An example of how to do this was reviewed in Aljadef et al. (Neuron 2016), where the calcium images would replace the visual stimuli in the analysis.

The reviewers raise three additional points that must be considered and preferably implemented. The first is the important issue of removing potential blood flow contributions, as was done by Chris Niell in a 2016 Journal of Neurophysiology manuscript. This should, ideally, be applied to the present data if at all possible. The second concerns the nature of the STM calculated from temporally isolated spikes versus spikes that are part of a burst. Given the known depression of thalamo-cortical synapses, this could likely contribute the variability of the STM data using thalamic spikes. To address this issue, the thalamic bursts can be isolated from the spike records and, as one possibility, the STM can be computed with respect to only the first spike. Lastly, the "spontaneous activity" may currently include sensory-evoked activity from body motion. As the authors recorded body motion, it would be germane to check if "spontaneous activity" in fact results from self motion, as this changes the interpretation of "spontaneous activity' from internally generated to reafferent driven.

Reviewer #1:

Spontaneous activity recorded with various methods (e.g., fMRI, intrinsic signal, voltage, and Ca2+ imaging) in the brain is useful to predict anatomical and functional connectivity in the brain. For example, seed pixel correlation maps (SPMs) can show the cross-correlation map of the spontaneous activity for a defined seed pixel in the images. In this study, Xiao et al. combined GCaMP imaging of an entire cortex with single neuron electrophysiological recordings, and used the spike-triggered averaging approach to determine the mesoscopic network motif linked to the recorded neurons. The spike trigger averaging approach has already been reported by Grinvald lab (Arieli et al., 1995), and the current study does not provide a new methodological framework. However, combined with state-of-the-art GCaMP imaging, Xiao et al. nicely demonstrates the power of this approach by finding subcortical to cortical network motifs, which have been difficult with optical imaging alone.

Overall, this study is technically sound. I suggest the following points to improve the quality and/or the presentation of the work.

Major issues:

1) Results section. Spike-trigger averaging of optical signals was originally reported by Grinvald lab and authors are simply adopting this approach to the large-scale GCaMP imaging. Authors need to introduce the original work in the Introduction and/or Results section.

2) Figure 1. Authors are using single-spike events to calculate the STM. However, in GCaMP imaging, single spikes can produce only marginal response changes (ΔF/F). In addition, a single spike in a single neuron is unlikely to effectively activate all the post-synaptic neurons. Thus, it is possible that multiple-spike events (or bursts) can produce a more reliable STM. The authors could test this possibility by reanalyzing data and the results could be shown as in Figure 2C, D.

3) Figure 2C. Authors demonstrate that a conserved network motif emerges by averaging the spike-triggered maps. However, this is based on an assumption that a single spike is always linked to a single type of network motif. Is it possible that multiple motifs are linked to a single neuron and authors' approach is averaging out multiple types of motifs? Authors could, for example, try classification of the spike-triggered maps.

4) Figure 3. Given the diversity of cell types (and projection patterns) within a defined brain region, it is possible that STM is diverse even within a defined cortical area. Did authors find such cases? (In fact, STMTD is diverse in Figures 7 and 8.) It is also possible that neurons in different cortical layers show distinct STMs even within the same area. Throughout this manuscript, box plots or beeswarm plots may be more appropriate if they show non-normal distributions with rare examples.

Reviewer #3:

Neuroscience research linking ongoing activity in brain-wide networks to that of single neurons is sorely lacking. fMRI studies have revealed interesting default brain networks that correspond to different behavioral states. The current study analyzes such patterns using widefield calcium imaging, selectively in excitatory neurons throughout dorsal cortex – arguably a useful complementary approach to the aforementioned human work given that calcium signals are more tightly linked to spiking activity. In addition, previous work using hippocampal electrophysiology and fMRI in non-human primates (Logothetis et al., Nature 2012) showed that hippocampal ripple activity during states other than active waking was linked to widespread cortical activation as well as to suppressed activity in subcortical regions (including thalamus), but few studies have taken a similar approach using widefield calcium imaging. The current study uses impressive tools, the combination of which is somewhat novel, to link spiking activity from single cortical and subcortical neurons with widespread GCaMP6 calcium activity during anesthesia and quiet waking. Further, only few studies have performed imaging with such a large field of view (9 mm x 9 mm). In general, the manuscript is well written. The successful collection of these datasets is exciting, and the combination of dense imaging, electrophysiology and behavioral information provides an extremely rich and unique dataset. However, the analyses performed do not effectively extract much of this richness, and do not, in the current form, reveal clearly interpretable insights of broad interest gained by this approach (particularly given the caveats involved in averaging across many neurons). Further, several interpretations of the data are confusing, and some key references require further consideration. These issues are discussed in detail below:

1) The authors show that seed pixel correlation maps (SPMs) are better matched to cortical single-unit spike-triggered maps (STMs) than to thalamic STMs, and that thalamic STMs are generally more diverse than cortical (specifically barrel cortex) STMs. However, as suggested by the authors in the Discussion, this is likely due, at least in part, to greater diversity across thalamic recordings from neurons in a larger number of smaller thalamic nuclei (vs. more consistently from S1 barrel region of cortex, albeit across layers) vs. recordings in a single cortical area with a large magnification factor. Further, cortical neurons within S1 are highly correlated, likely more so that nearby neurons in thalamus, further ensuring strong correlation between the cortical spiking activity and the SPM maps (since the seed cortical pixel is likely highly correlated with the cortical single-unit recording), more so that would be likely be possible with most thalamic units). The other suggested explanations for these findings made by the authors in the Discussion are far more speculative, and many would be difficult to test. Further, little effort is made to relate specific thalamic recordings from specific areas to any specific cortical patterns (e.g. did lemniscal thalamic recordings have more similar STMs to SPMs vs non-lemniscal thalamic recordings?). Given that the correlation between cortical neurons and their neighbors is likely very different than the correlation between single thalamic neurons and their neighbors, it might be helpful to compute STMs for multiunit activity in cortex and in thalamus for a given electrode contact, and then subtract a scaled version of this STM from the single unit STM – that may put the cortical and thalamic analyses on more even footing (all the more true given data in Figure 3Diii). Similar analyses could be done for high-passed LFP or CSD recordings, allowing a better sense of the relationship between a single neuron's spike with network activity, vs. the correlated spiking of that neuron and many nearby neurons. This analysis may also reveal more spatially localized patterns (e.g. for barrel cortex recordings, it might reveal relatively more barrel-specific response profiles). Similarly, for cortical recordings, one could subtract STMs in superficial recordings from STMs with simultaneously recorded neurons in deep layers, thus removing the 'columnar' component of the correlation, and allowing evaluation of residual, layer-specific contributions.

2) Figure 8 is interesting in terms of differences between thalamic and cortical recordings, given the consistent net and prolonged suppression with thalamic STMs and net excitation with cortical STMs. While the authors mention Logothetis 2012 Nature, they don't mention that this Logothetis et al. paper (their Figure 3) shows hippocampal ripple-triggered excitation of most of cortex, and ripple-triggered suppression of most subcortical regions including thalamus. This could very well explain the effects seen in the current Figure 8, of net suppression in thalamic STMs (at most delays) and net activation for cortical STMs, as it provides evidence that cortical and thalamic neurons are likely anti-correlated during much of spontaneous activity (e.g. during ripple events). It would be worthwhile following this up – if most spontaneous cortical spiking is ripple-related and involved distributed, correlated cortical patterns, this could also partially explain why cortical STMs look more like SPMs than thalamic SPMs. This is a place where the true level and time course of activity would likely be different if STM analysis methods were used that didn't involve subtraction of GCaMP6 activity in the 3-s window prior to spike onset (see point 3c below).

3) Several analysis choices make the findings difficult to interpret: (a) there is no correction for vascular effects despite analyses spanning several seconds surrounding the spike: it is surprising that no vascular responses are seen in STMs in the GFP control, please discuss. (b) the authors suggest using deconvolution of signals as a future direction, but this would seem to be highly useful to use in the present analyses in separating GCaMp6 dynamics from neural dynamics, given the emphasis, in part, on temporal resolution (relative to fMRI). Both points (a) and (b) are effectively addressed in a study that is in some ways quite similar, by Wekselblatt et al., J Neurophys June 2016 -- a paper that is not cited by the authors. In that paper, Wekselblatt et al. nicely show the use of stimulus-evoked widefield GCaMP6 mapping across an intermediate sized cranial window in mice, after deconvolution and removal of vascular signals. (c) Another issue that is difficult to interpret is the normalization of GCaMP6 signal to a window of time prior to spike onset. Clearly this is throwing out a lot of useful information, and may distort much of the information that remains. An alternative approach might be to use a running median subtraction at each pixel, perhaps with removal of any slow drifts over long timescales, which would more effectively reveal what GCaMP6 activity is leading vs. lagging the spiking activity. (d) No discuss of the nature of the spiking activity is presented, yet burst firing (and thus, counting spikes with short ISI as similar to those with long ISI) may also distort the STMs. (e) The cortical delineation of areas is crude, and most secondary areas are not well delineated, making it difficult to assess whether any specific spatial resolution afforded by the technique actually yields insights at the level of specific higher cortical areas. (f): While it is interesting and useful to see that STM patterns become similar after averaging >128 or 256 trials (Figure 2D), most of the information in the dataset is lost in this averaging process. Do subsets of spike events share a common spike-triggered pattern, albeit one that is not present on the majority of spike events, and which might be washed out by this averaging?

4) Description of the choice of epochs included in "spontaneous activity" is unclear. The authors do not seem to use body-tracking data that they collect and describe in Methods in order to segregate epochs of true spontaneous activity. The "spontaneous activity" may currently include sensory-evoked activity (i.e., internally generated spontaneous whisking), even though these behaviors were observed, as stated in "Methods: Multimodal recording in awake mice."

5) Figure 7 in interesting given the PCA patterns show differential connection with thalamic or cortical spiking, but no obvious effort was made to understand which cortical cells or thalamic cells fall into each pattern category (other than spike width analysis). What do spike autocorrelations look like? Sensory receptive fields? Layer in cortex? Location within specific thalamic nuclei? Without any such insights linking the findings to known entities, the value of this figure is somewhat low.

https://doi.org/10.7554/eLife.19976.034

Author response

Essential revisions:

Both reviewers as well as the Reviewing editor find this to be a carefully executed set of experiments and praise to quality of the imaging over an unprecedented nearly 1 cm by 1 cm field of view. The reviewers raise excellent issues and I would like the author to respond to all comments.

We are pleased by the positive response of the reviewers and we have been able to address their collective and individual concerns as we describe below. As part of our revised manuscript, we now incorporate 11 new supplemental figures and 1 new video. We now hope the paper is ready for publication at eLife.

In summary, in response to the points highlighted as Essential Revisions: 1) we have addressed alternatives to averaging GCaMP signals and produced both variance and spike triggered co-variance maps, as well as devising a partitioning method to examine heterogeneity in single trials (see Figure 6—figure supplement 15). In the end, these approaches confirm the validity of an averaging approach; 2) Potential for changes of blood volume/oxygenation to alter fluorescence signals. We provide new data using a reference wavelength to confirm minimal effects of blood volume contamination of spike triggered average GCaMP signals (Figure 7—figure supplement 1); 3) We report that spike-triggered average maps are similar for bursting versus non-bursting neurons (Figure 6—figure supplement 1, 2, 3); and 4) We now show that periods of body movement have little impact on spike triggered average maps Figure 8—figure supplement 1).

Although not mentioned in the reviews we also take this opportunity to address verbal comments made by Matteo Carandini (at SFN during his lecture) about the Ai93 mouse and his observation of seizures. We are surprised by this observation as we have not seen activity consistent with the behavioral consequences of a seizure in our colony of these mice. We show long records of LFP activity indicating relatively normal up down states (Author response image 1). We also show examples of spike waves that were generated after topical cortical application of a convulsant drug (picrotoxin). These results indicate that we can indeed detect seizures.

Author response image 1
Temporal and spectral signatures of spontaneous and epileptic events measured from the local field potential (LFP).

(A) Top (black trace), 100s segment of LFP recording of spontaneous activity from superficial layer of cortex. Bottom, spectrogram (Morlet-wavelet scalograms) of the LFP trace. (B) Top (black trace), 170s segment of LFP recording after application of Picrotoxin (Sigma) on the top of cortical surface during the same experiment. (i) Top (black trace), 40s segment of LFP recording, epileptic events initiated with reversed phase and gradually increased amplitude (red arrows). Bottom, spectrogram of the LFP trace. (ii) Top (black trace), 40s segment of LFP recording showed repetitive high amplitude (>1 mV) epileptic discharge. Bottom, spectrogram showed high frequency power during the epileptic events but suppressed background activity.

https://doi.org/10.7554/eLife.19976.029

Regarding "essential" points, one reoccurring theme that will require additional analysis concerns the averaging. First, it would be very useful to see variance maps as well as average or spike triggered maps (STMs) as an extension of Figures 3 and 4. The two may not be scaled versions of each other if the calcium responses are nonlinear. For that reason, the authors may further consider an analysis of the covariance of the data, as opposed to only the mean, as would reveal properties of "complex" versus "simple" visual cells. Perhaps this issue is at bay with the thalamic data. An example of how to do this was reviewed in Aljadef et al. (Neuron 2016), where the calcium images would replace the visual stimuli in the analysis.

We thank the editor and reviewers for suggesting these extensions to the analysis. As suggested, we calculated spike triggered variance maps in an analogous way to the spike triggered maps and show an example of their temporal dynamics. While the spike triggered variance maps were of interest, spatially, they largely reveal familiar brain regions as defined using spike-triggered average (STA) mapping (when examined as a difference to baseline, see Author response image 2). Considering the time course, there is a subtle, but global, shift to higher variance, but the variance is still very small (△F/F0from -0.2% to 0.2%) compared to the STA response.

Author response image 2
Spike-triggered variance map.
https://doi.org/10.7554/eLife.19976.030

A. Spatiotemporal dynamic of spike triggered average map of a cortical neuron. The time window is from -3s to 3s. B.Spatiotemporal dynamic of spike triggered variance map of the same neuron. The amplitude of the variance is small (△F/F0 (%) from -0.2 to 0.2).

The reviewer also makes a suggestion of examining co-variances making use of recently published methods (Aljadeff et al., 2016). We now include the spike-triggered covariance (STC) analysis as supplemental figures (Author response image 3 and 4). To do this we made use of the code provided in the supplemental information in Aljadeff et al. 2016 and used our calcium imaging in place of the visual stimuli, as suggested, following the analysis steps presented in this excellent resource. To make the computation feasible (in terms of computer time) it was necessary to down sample the imaging data to 32x32. In order to compare the resulting STA from the Aljadeff et al. 2016 with the spike triggered maps we previously calculated, we considered 6 time points. With these reduced dimensions it took 1-3 days to process a neuron depending on the length of the recording.

Author response image 3
Spike Triggered Covariance Analysis of BCS1 Neuron.

(A) Each calcium image was recorded each Δt = 33.33 ms (frame rate = 30 Hz), and the spikes recorded in BCS1 neuron were binned within the same interval. In this example, we recorded 15250 frames and 4625 spikes simultaneously. (B) We constructed the covariance matrix of the spatiotemporal calcium images, followed by eigenvector analysis of the covariance matrix and plotted its spectrum (black). We compared this spectrum to that expected theoretically for the same-sized random matrix (red). (C) PCA of calcium images covariance matrix. The eigenvectors of the calcium images covariance matrix that correspond to the largest eigenvalues (mode 1 to 12) are seen to contain sequence of spatiotemporal patterns. (D) Spatiotemporal dynamic of STA map for cell #1 and cell#2. Calcium images were binned from 128*128 to 32*32 square pixel array with mask to select brain region. (E) Distribution of underlying calcium image matrix projected on Spike triggered average (STA) for cell #1 and cell #2. (F) The input/output nonlinearity of cell #1 and cell#2, input project on STA (S*sta).

https://doi.org/10.7554/eLife.19976.031
Author response image 4
Spike Triggered Covariance Analysis of sub-cortical Neuron.

(A) a1. STA of this sub-cortical neuron shows unique spatiotemporal pattern. a2. Distribution of underlying calcium image matrix projected on STA of this sub-cortical neuron. a3. The input/output nonlinearity of this neuron, input project on STA (S*sta). (B) b1. Spike triggered covariance mode 1 (STC1) of this neuron. b2. Distribution of underlying calcium image matrix projected on STC1. b3. The input/output nonlinearity of this neuron, input project on STC1 (S*stc1). (C) c1. Spike triggered covariance mode 2 (STC2) of this neuron. c2. Distribution of underlying calcium image matrix projected on STC2. c3. The input/output nonlinearity of this neuron, input project on STC2 (S*stc2). (D) The significance of each candidate STC feature, were determined by comparing the corresponding eigenvalue (red and black) to the null distribution (gray shaded area). We used 1,000 repetitions of the calculation for randomized spike trains, corresponding to a confidence level of 0.05. (E) Three dimensional plot of the nonlinearity in the space, spanned by the STA and other two orthogonalized STC feature, with surfaces for firing rates (FR) equal to 20% and 50% of the max.

https://doi.org/10.7554/eLife.19976.032

Author response image 3 (panel A) illustrates a montage of subsequent frames of calcium imaging with the recorded spikes as in Figure 3a of Aljadeff et al. 2016. Following Aljadeff et al. 2016 Figure 3, Author response image 3 (panel B) shows the spectrum (eigenvalues) of the covariance matrix of the “stimulus” generated from our calcium imaging and Author response image 3 (panel C) shows the spatial mode shapes. As expected, and in contrast to white noise, these have spatial structure and some familiar anatomical/functional regions are apparent. In Author response image 3 (panels E & F) we show the underlying stimulus distributions and input/output nonlinearity curves for two cells (as in Aljadeff et al. 2016 Figure 4C and D). The distributions can be expected to be Gaussian for a large n (central limit theorem, as pointed out in Aljadeff et al. 2016) and we would expect them to be more Gaussian-like if we were able to use our data at the recorded resolution (128x128). Author response image 3 (panel D) shows the resulting STA and it compares well to our previously calculated spike triggered maps.

Continuing the analysis we next tested for significance of the STC features as in Aljadeff et al. 2016 Figure 5. We also used 1000 repetitions of random shifts of the spike train to generate null distributions. We first tested at a significance level of 0.01. At this significance level we were unable to find any significant STC feature for either cortical or subcortical neurons. The SNR of optical recordings may contribute to this so we re-tested at a more relaxed significance level of 0.05. At this level none of the n=4 cortical neurons tested had significant STC features. However, for 5 subcortical neurons of a total of n=5 tested we found significant STC features. Author response image 4 illustrates this for a subcortical neuron similar to Aljadeff et al. 2016 Figure 5.

The analysis indicates that single cortical cells were most likely associated with a single dominant map, whereas single sub-cortical cells are not only associated with dominant STA map, but also more likely to be associated with multiple STC maps. While initial STC analysis indicated that BCS1 neurons were associated with more simple maps than subcortical neurons, GCaMP images may have had lower signal to noise ratio than some of the examples in Aljadeff et al. 2016 and we have considerably less spikes making it harder to draw firm conclusions from group analysis. Therefore, we show this analysis as a reviewer/editor-only example and have not included it as supplemental or regular figure.

The reviewers raise three additional points that must be considered and preferably implemented. The first is the important issue of removing potential blood flow contributions, as was done by Chris Niell in a 2016 Journal of Neurophysiology manuscript. This should, ideally, be applied to the present data if at all possible.

The reviewers mention that our data should be corrected for blood volume changes which can potentially alter conclusions regarding spike-triggered mapping.

All of the data in the original manuscript was collected with a single green epi-fluorescence wavelength and not dual wavelength imaging needed for a reflected light reference image. In the original paper we gave examples of GFP mice that would be subject to similar artifacts yet were unable to produced spike-triggered maps (Figure 2B). As suggested by the reviewer these animals do show small slow transients that are typically less than 0.1% and importantly do not produce clear STA maps with known anatomical motifs.

To satisfy the reviewers concerns we have now done additional experiments using a multi-wavelength approach. We have now implemented a very similar correction to that performed by the Niell lab manuscript (Wekselblatt et al., 2016) and is also similar to work performed by the Hillman lab (Ma et al., 2016). In summary, the Niell lab implemented a strobed light presentation using alternating blue and green lights to capture GCaMP epi-fluorescence (blue illumination) and green reflectance (dim green illumination) in alternating frames. We have used a similar approach but monitor blue reflected light as our reference signal for fluorescence changes due to hemodynamics. We feel that there is some loss of time resolution using the strobed approach and it also depends very strongly on camera timing signals. There is also the issue of the flashing lights potentially stimulating/entraining the animal. Accordingly, we have developed a very similar strategy using a color RGB camera (Picam) (Murphy et al., 2016), which allows simultaneous acquisition of a short blue light reflected signal (447 nm LED and 438/24 nm filter near an isosbestic point for hemoglobin), and a green epi-fluorescence signal (GCaMP). Using the short blue reflected light signal, we now show that 438 nm reflected light is strongly correlated with 532 nm green reflected light in a control experiment (r = 0.93; Figure 7—figure supplement 1). This analysis indicates that short blue reflected light signals can be used as a surrogate for a green reflected signal. Accordingly, we now use the short blue reflected light signal in a division strategy very similar to previous approaches (Sirotin and Das, 2010; Ma et al., 2016; Wekselblatt et al., 2016). The division was performed after both frame-by-frame signals were converted a ΔF/F0 and employed a green/blue weight of 1. We have tried weightings than >1.0 for the short blue signal, but found that some aspects of the kinetics were over-corrected. This new analysis dampens non-specific fluctuations associated with blood volume changes, which are aggravated in the awake state. We now show using validation with green light reflection, as well as GFP mice and that our strategy results in a reduction in baseline noise.

However, after implementing this corrective strategy, we find little change in spike-triggered map activity consistent with control investigations of functional connectivity or task-related connectivity in GFP animals in experiments done previously by our lab (Vanni and Murphy, 2014; Murphy et al., 2016; Silasi et al., 2016). Furthermore, notable features, such as some areas showing apparent cortical inhibition (reductions in calcium activity) were also preserved in the new corrected analysis. We now include this new data as a supplemental figure (Figure 7—figure supplement 1) and outline the procedure in the Materials and methods and Results section. Given that the correction does not have a large impact and failed to change the appearance of the maps we have kept the original figures as is (they cannot be corrected given they reflect a single wavelength) and now point readers to the new confirmatory experiments and previous papers using the GFP-mouse controls (Vanni and Murphy, 2014; Murphy et al., 2016; Silasi et al., 2016).

See amended text from Materials and methods:

“In some experiments (as indicated), we employed a multi-wavelength strategy to correct for potential green epifluorescence signals that were associated with non-calcium dependent events. […]While the shorter blue wavelength light will scatter more than a green strobed reflected light signal used by others (Ma et al., 2016; Wekselblatt et al., 2016), our analysis of green reflected light and short blue reflected light, indicating that two were highly correlated, suggests that the major artifacts we observe are associated with large blood volume changes in superficial cortical layers.”

See amended text from Results section:

“Other potential sources of error include apparent changes in GCaMP6 signal due to alterations in blood volume or oxygenation (Ma et al., 2016; Wekselblatt et al., 2016). […]Furthermore thalamic STA maps and STA dynamic plots still indicated cases where thalamic spiking was associated with cortical inhibition. For these reasons we have attempted to correct other STA maps using this multi-wavelength procedure.”

The second concerns the nature of the STM calculated from temporally isolated spikes versus spikes that are part of a burst. Given the known depression of thalamo-cortical synapses, this could likely contribute the variability of the STM data using thalamic spikes. To address this issue, the thalamic bursts can be isolated from the spike records and, as one possibility, the STM can be computed with respect to only the first spike.

It is conceivable that averaging over many spikes could mask spatial heterogeneity observed in sub-groups of spikes. Therefore, we performed additional analyses where we classified spike triggered STMs and STMTDs into sub-groups based on spiking modes (bursting vs. tonic) or motif similarity in a high-dimensional space – with the goal of determining whether during different types of active cortical dynamics our single spike-sorted neurons contribute different types of STMs or temporal dynamics (Figure 6—figure supplement 15).

We searched for STM sub-groups by dividing all spikes from a single cell into bursting versus tonic modes. We implemented methodology previously described for defining the main spiking modes of a single cell (Sherman and Guillery, 2006; Figure 6.5, pg 236). Briefly, the method requires computing the distribution of each spike’s inter-spike-interval (ISI) between the previous (x-axis) and following (y-axis) spike (Figure 6—figure supplement 1). The resulting pre- and post-spike 2D distribution is then plotted using logarithmic scales and naturally arising clusters are grouped or clustered. Spike groups occurring in approximately each quadrant of the plot indicate different spiking modes: first spikes in a burst (bottom right), spiking occurring during a burst (bottom left), last spikes in a burst (top left), and tonic spikes (top right). The vast majority of cortical cells we recorded in barrel cortex did not exhibit multiple classes of spiking modes, however we provide 2 examples where some natural clustering is present (Figure 6—figure supplement 1(Ai) and (Bi)). The spike sub-groups representing different spiking modes yielded very similar STMs and motifs (Figure 6—figure supplement 1(Aii) and (Bii)) and similar STMTDs in the area of highest activation, i.e. left barrel cortex (Figure 6—figure supplement 1 (Aiii) and (Biii)). We also provide examples of two identified thalamic cells which showed more apparent spiking modes (Figure 6—figure supplement 1(C) and (D)). The thalamic cells we identified showed better bursting versus tonic clusters (Figure 6—figure supplement 1 (Ci) and (Di)) but much like cortical cells their STM and STMTDs were largely stable across the clustered spiking modes. The 3 cortical and 4 thalamic cells provided in Figure 6 did not exhibit clear spiking modes. For those cells we implemented a simpler methodology and compared all-spike STMs against STMs generated by spikes preceded by at least 500ms of silence as indicative of bursting activity following a period of quiescence (Figure 6—figure supplements 2 and 3). Using both of these methodologies to divide spiking times into different modes we found that both STMs and STMTDs across spiking modes are similar to the all-spike average STMs in both cortical and thalamic cells examined.

We additionally sought to determine whether single-spike motifs were similar to each other and formed natural clusters in a high-dimensional representation space. There were no naturally arising clusters suggesting that single spikes occur during many phases of ongoing, i.e. spontaneous, cortical activity. Accordingly, we manually partitioned the motifs into several (4 to 12) groups by similarity and after removing the spontaneous component we were able to recover the all-spike average STMs (Figure 6—figure supplement 4 and 5; see Methods). This single-spike motif sub-grouping suggests that: (i) cells fire during many different ongoing cortical states; and (ii) that despite the high degree of variability single cell spikes appear to correlate with (or possible contribute to) similar overall STM patterns. Given that single spike motifs are very diverse (Figure 6—figure supplement 4(A)) our approach was to project single-spike STMs into a high-dimensional space, compute their distributions and search for clusters or manually partition the resulting distributions. There were no natural groups in the resulting high-dimensional distribution so we partition the data into 4 (or more) sub-groups (Figure 6—figure supplement 4(B), see also Figure 6—figure supplement 5). The resulting 4 “sub-networks” represent (on average) qualitatively different types of ongoing cortical activity during which our cell fires action potentials. While these partitions were manually created, we were interested in determining whether they could generate STMs that resembled all-spike average STMs and whether this was independent of the number of partitions. The 4 partitions chosen generated somewhat different STMs, only one of which strongly resembled our all-spike average STM (Figure 6—figure supplement 4(D); all spike average is bottom STM). The inter-spike-interval (ISI) distributions during activation of these sub-networks is also similar across all partitions further suggesting that this partitioning method did not identify highly bursting or tonic periods of activity (Figure 6—figure supplement 4(C)). Because we had enforced a manual STM-space partition, our resulting sub-network STM averages contained an offset that essentially represented ongoing “spontaneous” cortical activity particular to each STM-space partition. Accordingly, we sought to remove this spontaneous component by carrying out the same partitioning method – but this time on spontaneous activity (i.e. not using single spike triggers). The distribution of spontaneous STMs (i.e. ongoing STMs not related to our cell spiking) also have substantial variability (Figure 6—figure supplement 4(E)). We next projected all spontaneous motifs into an STM-space but partitioned this space based on the sub-network centres derived previously (Figure 6—figure supplement 4(B) and (D)). This guaranteed that the 4 STMs derived from spontaneous activity would be the closest to our spike-triggered STMs. The spontaneous activity motifs showed a very strong ISI distributions peak at ~33ms (i.e. single-frame time) (Figure 6—figure supplement 4(G)) indicating that STMs that were neighbouring in time (i.e. separated by single frames) were substantially more likely to be in the same region of STM-space and to be grouped together. This is expected as transitions meso-scale cortical dynamics usually last several frames at our 30Hz sampling rate. The sum of the 4 spontaneous STMs yield an STM with an approximately 0% ΔF/F0 value – which is expected when averaging all spontaneous motifs (or frames) during a recording (Figure 6—figure supplement 4(H) bottom STM). The last step is to remove the spontaneously activity STMs (Figure 6—figure supplement 4(H)) from the single cell spike-triggered sub-networks (Figure 6—figure supplement 4(D)). The results reveal that the 4 partition STMs reduce to the all spike average STMs. This analysis supports the averaging method of computing STMs: despite firing during different types of cortical dynamics, single cells contribute to – and participate in – ongoing dynamics in a stereotyped way that is represented by the all-spike STM average.

We further applied this partitioning method on additional cells: two cortical and two thalamic cells with similar results (Figure 6—figure supplement 5). We also tested a 12 partition approach (Figure 6—figure supplement 5(C)) and showed that when subtracting spontaneously activity STMs (Figure 6—figure supplement 5(C)(ii)) from the spike triggered STM (Figure 6—figure supplement 5(C)(i)) we obtain STMs that are very similar to the all-spike averages (Figure 6—figure supplement 5(C)(iii)). This generally confirms that average based STMs are representative of a single cell’s contribution to ongoing cortical activity.

See amended text from Materials and methods:

“Cell Spiking Mode Determination. We implemented methodology previously described for defining main spiking modes (Sherman et al., 2006) (Figure 6.5, pg 236). […]We provide additional examples of this STM partitioning using other cortical and thalamic cells and also an example where we partitioned the STM distribution into 12 sub-networks but still recovered the all-spike average STM from each of the partitioned networks (Figure 6—figure supplement 5).”

See amended text from Results section:

“Our analysis indicates that averaging GCaMP cortical motifs from all spikes of a single cell produces converging STMs and STMTDs. […]These tests confirm that the averaging method reveals how single cells participate in largely stereotyped networks despite the variability of ongoing cortical activity.”

Lastly, the "spontaneous activity" may currently include sensory-evoked activity from body motion. As the authors recorded body motion, it would be germane to check if "spontaneous activity" in fact results from self motion, as this changes the interpretation of "spontaneous activity' from internally generated to reafferent driven.

Mice when habituated to our head-fixed recording apparatus do not exhibit extensive active body movements. However, the reviewer’s point is well-taken and we explored the potential contribution of active movements, however minute, to our recordings of spontaneous data. In new analysis, we now use simultaneously acquired video imaging (webcam) data to segment quiet periods of animal activity for spike-triggered averaging. Despite this procedure, we found little difference in spike-triggered mapping when movement periods were removed from awake mouse spontaneous activity data (Figure 8—figure supplement 1 and Video 5).

See amended text from Materials and methods:

“While relatively few large body movements were observed during awake imaging sessions, their impact on mapping was evaluated by generating STA from period of quietness. […] STA was then generated only from spikes for periods of quietness and compared with STA of all spikes.”

Amended section from Results section:

“Our goal is to assess cortical functional connectivity based on coincidence between individual neuron spiking and ongoing spontaneous activity. […] Therefore, brain imaging activity obtained in awake states is mostly indicative of a quiet awake state and is not primarily movement-related activity.”

Reviewer #1:

[…] Major issues:

1) Results section. Spike-trigger averaging of optical signals was originally reported by Grinvald lab and authors are simply adopting this approach to the large-scale GCaMP imaging. Authors need to introduce the original work in the Introduction and/or Results section.

Although we have referenced these papers, including within the Introduction, we agree that we could do more to highlight their seminal contributions that include the use of early application of spike-triggered averaging wide field imaging methods and conclusions. We have now revised our Introduction section to better acknowledge this pioneering work.

The amended Introduction now states:

This work extends pioneering studies investigating the relationship between single neuron spiking and local neuronal population activity assessed by voltage-sensitive dye imaging. […] Furthermore, we employ multisite, long shank, silicon probe recordings of single neuron activity that facilitates the assessment of long-distance activity relationships between remote subcortical single neuron activity and mesoscale cortical population activity.

We also do note differences between our GCaMP wide-field mouse approach and the previous work in visual cortex. Notable differences include the neuronal specificity of the GCaMP imaging signals, as well relatively wider field of view and additional information concerning sub-cortical spiking neurons.

2) Figure 1. Authors are using single-spike events to calculate the STM. However, in GCaMP imaging, single spikes can produce only marginal response changes (ΔF/F). In addition, a single spike in a single neuron is unlikely to effectively activate all the post-synaptic neurons. Thus, it is possible that multiple-spike events (or bursts) can produce a more reliable STM. The authors could test this possibility by reanalyzing data and the results could be shown as in Figure 2C, D.

The reviewer raises an important point regarding the potential nuanced relationship between the pattern of cell firing and the efficacy in producing a reliable STM. The reviewer suggests that in the case of GCaMP imaging, individual spikes may not go readily detected. We reference papers which show that GCaMP6 can detect individual spikes in vivo (Chen et al., 2013). However, we feel we must clarify that our intention is not to imply a causal relationship between the unitary contributions of a single cell’s spiking activity to mesoscale cortical calcium activity or that our method requires the faithful detection of individual spikes. Instead, we interpret the single cell spiking activity as being correlated with the activation of these large cortical networks and likely acting in synchrony with other cells. However, the issue of averaged cell firing pattern diversity and its impact on STMs is important and explored in greater detail in our response to the Essential Revisions Comment #3.

3) Figure 2C. Authors demonstrate that a conserved network motif emerges by averaging the spike-triggered maps. However, this is based on an assumption that a single spike is always linked to a single type of network motif. Is it possible that multiple motifs are linked to a single neuron and authors' approach is averaging out multiple types of motifs? Authors could, for example, try classification of the spike-triggered maps.

The reviewer raises an excellent point and it is not our intention to imply that a single spike is always linked to a single type of network motif but rather that the average response yields a pattern consistent with these consensus maps. However, one of our principle conclusions is that numerous, diverse sub-networks of cortical activity patterns exist and can be linked with the activity of a single cell as illustrated in Figure 6—figure supplements 1 to 5). The discussion of averaging and heterogeneity and our new analyses are described in detail in our response to the Essential Revisions Comment #3.

4) Figure 3. Given the diversity of cell types (and projection patterns) within a defined brain region, it is possible that STM is diverse even within a defined cortical area. Did authors find such cases? (In fact, STMTD is diverse in Figure 7 and 8.) It is also possible that neurons in different cortical layers show distinct STMs even within the same area. Throughout this manuscript, box plots or beeswarm plots may be more appropriate if they show non-normal distributions with rare examples.

As mentioned, given the strong agreement between seed pixel maps and single neuron spike-triggered maps in cortex, we do not expect a lot of averaged-STM diversity. In the original submitted manuscript we provided “contour” plot analysis and quantification of individual STM maps across laminar depth (Figure 3A and B). We found that they were largely super-imposed and somewhat invariant across cortical depth. In the revised manuscript, we better highlight these findings with new analyses where we illustrate the STMs generated by single unit activity, multi-unit activity, and LFP across laminar depths from cortical and subcortical recordings (Figure 3—figure supplement 1 and Figure 3—figure supplement 2).

Another consideration is that we insert the electrode using an angular tilt (45°) to achieve an approximately vertical penetration to the curved cortical surface such that the electrode is roughly recorded in the same functional cortical column. In the same functional column, the single neuron related functional maps may be more likely similar, even there are diverse cell types. We have also replaced the bar graph in Figure 4C with a box plot.

Reviewer #3:

[…] 1) The authors show that seed pixel correlation maps (SPMs) are better matched to cortical single-unit spike-triggered maps (STMs) than to thalamic STMs, and that thalamic STMs are generally more diverse than cortical (specifically barrel cortex) STMs. However, as suggested by the authors in the Discussion, this is likely due, at least in part, to greater diversity across thalamic recordings from neurons in a larger number of smaller thalamic nuclei (vs. more consistently from S1 barrel region of cortex, albeit across layers) vs. recordings in a single cortical area with a large magnification factor. Further, cortical neurons within S1 are highly correlated, likely more so that nearby neurons in thalamus, further ensuring strong correlation between the cortical spiking activity and the SPM maps (since the seed cortical pixel is likely highly correlated with the cortical single-unit recording), more so that would be likely be possible with most thalamic units). The other suggested explanations for these findings made by the authors in the Discussion are far more speculative, and many would be difficult to test. Further, little effort is made to relate specific thalamic recordings from specific areas to any specific cortical patterns (e.g. did lemniscal thalamic recordings have more similar STMs to SPMs vs non-lemniscal thalamic recordings?). Given that the correlation between cortical neurons and their neighbors is likely very different than the correlation between single thalamic neurons and their neighbors, it might be helpful to compute STMs for multiunit activity in cortex and in thalamus for a given electrode contact, and then subtract a scaled version of this STM from the single unit STM – that may put the cortical and thalamic analyses on more even footing (all the more true given data in Figure 3Diii). Similar analyses could be done for high-passed LFP or CSD recordings, allowing a better sense of the relationship between a single neuron's spike with network activity, vs. the correlated spiking of that neuron and many nearby neurons. This analysis may also reveal more spatially localized patterns (e.g. for barrel cortex recordings, it might reveal relatively more barrel-specific response profiles). Similarly, for cortical recordings, one could subtract STMs in superficial recordings from STMs with simultaneously recorded neurons in deep layers, thus removing the 'columnar' component of the correlation, and allowing evaluation of residual, layer-specific contributions.

Reviewer 3 is generally positive about the work and appreciates the linkage between wide-scale cortical activity and studies which have examined event-related activation using fMRI imaging.

The reviewer makes a number of good points, mostly about better examining the relationship between individual neurons and the ensemble brain activity.

1) The reviewer wonders about our interpretation of findings which compare diversity in cortical spike-triggered GCaMP maps for spike trains produced by cortical versus sub-cortical neuron groups. They acknowledge our discussion points, including diversity being attributed to the physical layout of the thalamus versus cortex and the ability to record from potentially more diverse neurons in the thalamus given the relatively smaller structure over which things are mapped. Accordingly, they suggest experiments to place subcortical thalamic neurons on more even footing with cortical neurons. These include making a spike-triggered single cortical neuron map in which multi-unit activity or LFP-triggered activity has been subtracted away. This would potentially allow the contributions of individual cortical neurons to be more easily seen.

This is a great suggestion and we have already done preliminary studies using single-channel MUA spiking (Figure 3—figure supplement 1 and 2). We also computed band-passed LFP triggered STMs for Delta (0.1-4Hz), Theta (4-8Hz) and Gamma (25-100Hz). LFP triggered STMs were computed using 60 second periods by multiplying the average LFP amplitude in each imaging time frame. Thus, imaging frames where the average LFP amplitude was large and positive contributed substantially to the STM, whereas frames where the average LFP values were closer to zero did not (negative LFP values were clipped). The findings for both cortical and subcortical recordings are that MUA triggered STMs are similar to single-unit STMs in cortex while LFP triggered STMs – representing mostly synaptic and some spiking activity, were substantially different from single cell and MUA triggered STMs and across different LFP frequency bands. However, it is important to note that LFP amplitude (or power-) -triggered maps have very low ΔF/F values, for example most have peak ΔF/F values < 0.05% (whereas single-unit or MUA peaks are between 1%-5%). Nonetheless, cortical LFP triggered maps do reveal anatomically discrete maps (Figure 3—figure supplement 1). However, while there appear to be some slight differences across layers, the differences do not appear to be substantial or systematic. It is likely that as LFP contains mostly synaptic and some spiking activity from within a region with radius of at least 250μm (Buzsaki et al., 2012) the STMs may represent average activity from larger neighbouring regions thus blurring the overall effect of local LFP activity.

In response to the reviewer’s suggestions that we subtract “a scaled version” of MUA based STM from single-unit STM, unfortunately the results appear largely as very noisy maps (i.e. maps with <1% ΔF/F values that have no ROI specificity). This is due to the similarities between single-unit and MUA STMs before subtraction. However, we agree overall that this general direction could be pursued further in future work to determine whether the limit of the GCaMP temporal and spatial resolution allows for more nuanced single-unit STMs to be obtained by removing common baselines.

See amended text from Results section:

“We have also assessed other means of generating event triggered maps including MUA and LFP frequency bands. […] In contrast, in the case of GCaMP6 higher frequency components are closer to hemodynamic and other noise sources making analysis more challenging.”

2) Figure 8 is interesting in terms of differences between thalamic and cortical recordings, given the consistent net and prolonged suppression with thalamic STMs and net excitation with cortical STMs. While the authors mention Logothetis 2012 Nature, they don't mention that this Logothetis et al. paper (their Figure 3) shows hippocampal ripple-triggered excitation of most of cortex, and ripple-triggered suppression of most subcortical regions including thalamus. This could very well explain the effects seen in the current Figure 8, of net suppression in thalamic STMs (at most delays) and net activation for cortical STMs, as it provides evidence that cortical and thalamic neurons are likely anticorrelated during much of spontaneous activity (e.g. during ripple events). It would be worthwhile following this up – if most spontaneous cortical spiking is ripple-related and involved distributed, correlated cortical patterns, this could also partially explain why cortical STMs look more like SPMs than thalamic SPMs. This is a place where the true level and timecourse of activity would likely be different if STM analysis methods were used that didn't involve subtraction of GCaMP6 activity in the 3-s window prior to spike onset (see point 3c below).

We thank the reviewer for highlighting these findings from Logothetis et al. As we did not record simultaneously from the hippocampus we cannot definitely comment on the relationship between thalamic spiking activity and the observation of net cortical inhibition with respect to hippocampal ripples in our experiments. However, the data presented in Logothetis et al. are remarkable and consistent with this interpretation and we now more clearly discuss our observations in this interesting context in our revised Discussion.

See amended text in Discussion section:

“The approach is probably most analogous to event-triggered MRI imaging from the standpoint of larger spatial scale (Logothetis et al., 2012). […] However, as we did not record simultaneously from hippocampus we cannot definitively comment on whether these observations necessarily correspond to hippocampal ripple events.”

With respect to the specific comment of “the true level and time course” of activity, we have addressed this in greater detail in regard to the reviewer suggestion of implementing a running median subtraction in comment (3C).

3) Several analysis choices make the findings difficult to interpret: (a) there is no correction for vascular effects despite analyses spanning several seconds surrounding the spike: it is surprising that no vascular responses are seen in STMs in the GFP control, please discuss. (b) the authors suggest using deconvolution of signals as a future direction, but this would seem to be highly useful to use in the present analyses in separating GCaMp6 dynamics from neural dynamics, given the emphasis, in part, on temporal resolution (relative to fMRI). Both points (a) and (b) are effectively addressed in a study that is in some ways quite similar, by Wekselblatt et al., J Neurophys June 2016 -- a paper that is not cited by the authors. In that paper, Wekselblatt et al. nicely show the use of stimulus-evoked widefield GCaMP6 mapping across an intermediate sized cranial window in mice, after deconvolution and removal of vascular signals.

The reviewer notes that there is no correction for potential vascular effects despite the signals spanning several seconds around the spike and it is surprising that we do not see vascular responses in the spike-triggered maps from GFP controls. We have more thoroughly addressed this point in our response to the Essential Revisions (comment 2) and now report new data using a normalization procedure in a new supplemental figure (Figure 7—figure supplement 1). In our characterization of this strategy, we have also done additional GFP mice. In no case do we see a clear reduction in baseline signal. Perhaps this is because spike-triggered maps are relatively discreet events and are not associated with global changes in activity which could potentially recruit hemodynamic responses. Another issue the reviewer brings up is the normalization of GCaMP signals to an average time window. We should note that in the spike-triggered average movies and images of temporal dynamics that we do indeed show imaged which reflect relative changes in ΔF over F during the preceding time. We have also attempted to implement the median subtraction approach.

The reviewer suggests a temporal deconvolution for signals to improve time resolution. We have now performed such an analysis and report these results (Figure 1—figure supplement 2). While sharpening the kinetics for raw data the procedure does not significantly change maps and it was not applied to all data.

(c) Another issue that is difficult to interpret is the normalization of GCaMP6 signal to a window of time prior to spike onset. Clearly this is throwing out a lot of useful information, and may distort much of the information that remains. An alternative approach might be to use a running median subtraction at each pixel, perhaps with removal of any slow drifts over long timescales, which would more effectively reveal what GCaMP6 activity is leading vs. lagging the spiking activity.

The reviewer suggests different methods for averaging calcium activity frames, in particular a “running median” method. We previously employed two different methods including a “global signal regression” method in which the baseline fluorescence is the mean of the entire ~5 minute recording stack. We also employed a low-band pass filtering method where the entire image stack was filtered at 0.1Hz-6Hz. The STMs provided by these methods appeared to be largely similar to the sliding window methodology likely because averaging over 3sec, 5 minutes or low-band pass filtering yield similar components in the calcium signal.

Slow drifts were very occasionally observed during recordings. However, in the initial manuscript, STA generated from random spike times, not carrying neuronal activity but still affected by slow drifts in the same manner, were used for correction (by subtraction, see Author response image 5). The use of running median gives the same results on slow drift but in contrast to random spike STA, a median window has to be determine and if inappropriately set, could remove important components such as depression. For this reason this normalization strategy, while otherwise elegant, was avoided.

Author response image 5
Running median and random spike normalization.

(A) STA from real spikes (1st line), random spikes (2nd line) and real spikes after running median filtering (3rd line). 4th and 5th lines are subtractions of STA by random spikes and running median respectively. (B) STA dynamic within BCS1 for real spikes (black curve), random spikes (thin red) and the subtraction (Thick red). (C) Same as B but for running median. (D,E) While drift was not often observed it was artificially introduced by reducing the DC of spontaneous activity by 50% in 9min. Both subtractions (random spikes and running median) were able to remove the slow drift contribution. However, post-spike depression (black arrow) at ∼1s was only preserved by using random spikes correction.

https://doi.org/10.7554/eLife.19976.033

(d) No discuss of the nature of the spiking activity is presented, yet burst firing (and thus, counting spikes with short ISI as similar to those with long ISI) may also distort the STMs.

We have now addressed the issue of the potential relationship of burst firing and STMs in our response to the Essential Revisions Comment #3.

(e) The cortical delineation of areas is crude, and most secondary areas are not well delineated, making it difficult to assess whether any specific spatial resolution afforded by the technique actually yields insights at the level of specific higher cortical areas.

The reviewer would like to see better delineation of secondary cortical areas. Unfortunately, using a bilateral imaging window implant, we do not have optical access to most secondary somatosensory areas. We have replaced the figure panel in Figure 1B with an enlarged and more clearly defined cortical reference atlas and updated table of regions-of-interest and abbreviations used to better elucidate our imaging perspective.

Regarding spatial resolution, we have revised the Methods section to include acknowledgement of the spatial resolution of our imaging and references to our previous work from which these measurements are described in detail.

See the amended text in Materials and methods:

“In order to reduce file size and minimize the power of excitation light used, we typically bin camera pixels (8 × 8) thus producing a resolution of 68 µm/pixel. These imaging parameters have been used previously for voltage sensitive dye imaging (Mohajerani et al., 2013) as well as anesthetized GCaMP3 imaging of spontaneous activity in mouse cortex (Vanni and Murphy, 2014) and awake GCaMP6 imaging in mouse cortex with chronic window (Silasi et al., 2016).”

(f): While it is interesting and useful to see that STM patterns become similar after averaging >128 or 256 trials (Figure 2D), most of the information in the dataset is lost in this averaging process. Do subsets of spike events share a common spike-triggered pattern, albeit one that is not present on the majority of spike events, and which might be washed out by this averaging?

We have now addressed the issue of heterogeneity of STMs and the use of averaging in our response to the Essential Revisions Comment #3.

4) Description of the choice of epochs included in "spontaneous activity" is unclear. The authors do not seem to use body-tracking data that they collect and describe in Methods in order to segregate epochs of true spontaneous activity. The "spontaneous activity" may currently include sensory-evoked activity (i.e., internally generated spontaneous whisking), even though these behaviors were observed, as stated in "Methods: Multimodal recording in awake mice."

We have now addressed the issue of the potential contribution of active behaviours in our response to the Essential Revisions Comment #4.

5) Figure 7 in interesting given the PCA patterns show differential connection with thalamic or cortical spiking, but no obvious effort was made to understand which cortical cells or thalamic cells fall into each pattern category (other than spike width analysis). What do spike autocorrelations look like? Sensory receptive fields? Layer in cortex? Location within specific thalamic nuclei? Without any such insights linking the findings to known entities, the value of this figure is somewhat low.

Our results presented in Figure 7 indicate that single cortical and thalamic cells participate in ongoing contralateral barrel cortex activating/depressing dynamics that fall into 2 or 3 discrete groups, respectively. The reviewer’s question of whether this grouping correlates with other single cell properties or whether it may be a novel intrinsic single cell property is very interesting but we were not able to identify correlations given our available datasets. Our cell-type, i.e. inhibitory versus excitatory, classification did not show correlation between PCA clusters and cell type. Additionally, we did not carry out stimulus paradigms to allow us to compute receptive fields for our barrel cortex cells. However, we did not find a correlation between cortical layers and PCA clusters nor between VPM and VPL nuclei and specific PCA clusters. Lastly, we did not pursue spike autocorrelation analysis as it would require classification or grouping of autocorrelation results which would go beyond the scope of our work. We anticipate that further work with substantially larger numbers of single cells may better classify activation/depression dynamics and allow for the use of multiple ROIs (e.g. barrel, motor, retrosplenial). We agree that the alternative, i.e. that single cells may have intrinsic properties that bias them to participating in networks acting on substantially different temporal scales, is certainly a topic worth investigating further.

See amended text in Results section:

“This classification showed that ~80% cortical neurons were associated with a pure cortical excitation profile (pattern #1) and 20% triggered inhibition (pattern #2). […] We suggest future work with larger datasets and multiple cortical/subcortical areas could address the question of whether STMTD classification are a novel intrinsic single-cell property.”

https://doi.org/10.7554/eLife.19976.035

Article and author information

Author details

  1. Dongsheng Xiao

    1. Department of Psychiatry, Kinsmen Laboratory of Neurological Research, Vancouver, Canada
    2. Beijing Institute for Brain Disorders, Capital Medical University, Beijing, China
    3. Djavad Mowafaghian Centre for Brain Health, University of British Columbia, Vancouver, Canada
    Contribution
    DX, Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing—review and editing
    Contributed equally with
    Matthieu P Vanni and Catalin C Mitelut
    Competing interests
    The authors declare that no competing interests exist.
  2. Matthieu P Vanni

    1. Department of Psychiatry, Kinsmen Laboratory of Neurological Research, Vancouver, Canada
    2. Djavad Mowafaghian Centre for Brain Health, University of British Columbia, Vancouver, Canada
    Contribution
    MPV, Conceptualization, Software (Matlab), Formal analysis, Investigation, Methodology, Writing—review and editing
    Contributed equally with
    Dongsheng Xiao and Catalin C Mitelut
    Competing interests
    The authors declare that no competing interests exist.
  3. Catalin C Mitelut

    1. Department of Psychiatry, Kinsmen Laboratory of Neurological Research, Vancouver, Canada
    2. Djavad Mowafaghian Centre for Brain Health, University of British Columbia, Vancouver, Canada
    Contribution
    CCM, Conceptualization, Software (Python), Formal analysis, Investigation, Methodology, Writing—review and editing
    Contributed equally with
    Dongsheng Xiao and Matthieu P Vanni
    Competing interests
    The authors declare that no competing interests exist.
  4. Allen W Chan

    1. Department of Psychiatry, Kinsmen Laboratory of Neurological Research, Vancouver, Canada
    2. Djavad Mowafaghian Centre for Brain Health, University of British Columbia, Vancouver, Canada
    Contribution
    AWC, Conceptualization, Software, Formal analysis, Investigation, Methodology, Writing—review and editing
    Competing interests
    The authors declare that no competing interests exist.
  5. Jeffrey M LeDue

    1. Department of Psychiatry, Kinsmen Laboratory of Neurological Research, Vancouver, Canada
    2. Djavad Mowafaghian Centre for Brain Health, University of British Columbia, Vancouver, Canada
    Contribution
    JML, Conceptualization, Formal analysis, Methodology, Writing—review and editing
    Competing interests
    The authors declare that no competing interests exist.
  6. Yicheng Xie

    1. Department of Psychiatry, Kinsmen Laboratory of Neurological Research, Vancouver, Canada
    2. Djavad Mowafaghian Centre for Brain Health, University of British Columbia, Vancouver, Canada
    Contribution
    YX, Conceptualization, Investigation
    Competing interests
    The authors declare that no competing interests exist.
  7. Andrew CN Chen

    Beijing Institute for Brain Disorders, Capital Medical University, Beijing, China
    Contribution
    ACNC, Conceptualization, Supervision, Methodology
    Competing interests
    The authors declare that no competing interests exist.
  8. Nicholas V Swindale

    Department of Ophthalmology and Visual Sciences, University of British Columbia, Vancouver, Canada
    Contribution
    NVS, Conceptualization, Resources, Software, Methodology, Writing—review and editing
    Competing interests
    The authors declare that no competing interests exist.
  9. Timothy H Murphy

    1. Department of Psychiatry, Kinsmen Laboratory of Neurological Research, Vancouver, Canada
    2. Djavad Mowafaghian Centre for Brain Health, University of British Columbia, Vancouver, Canada
    Contribution
    THM, Conceptualization, Supervision, Funding acquisition, Validation, Investigation, Methodology, Writing—original draft, Project administration, Writing—review and editing
    For correspondence
    thmurphy@mail.ubc.ca
    Competing interests
    The authors declare that no competing interests exist.
    ORCID icon 0000-0002-0093-4490

Funding

International Alliance of Translational Neuroscience

  • Dongsheng Xiao

Brain Canada, Canadian Neurophotonics Platform

  • Jeffrey M LeDue
  • Timothy H Murphy

Canadian Institutes of Health Research (MOP-12675)

  • Timothy H Murphy

Canadian Institutes of Health Research (FDN-143209)

  • Timothy H Murphy

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

This work was supported by Canadian Institutes of Health Research (CIHR) Operating Grant MOP-12675 and Foundation Grant FDN-143209 to THM, THM is supported by the Brain Canada Neurophotonics Platform, an International Alliance of Translational Neuroscience (IATN) program to DX. We thank Pumin Wang, Cindy Jiang for surgical assistance; Jamie D Boyd and Federico Bolanos for technical assistance and Alexander McGirr for helpful discussions and comments on the manuscript. We thank Hongkui Zeng and Allen Brain Institute for providing transgenic mice.

Ethics

Animal experimentation: Animal protocols (A13-0336 and A14-0266) were approved by the University of British Columbia Animal Care Committee and conformed to the Canadian Council on Animal Care and Use guidelines.

Reviewing Editor

  1. David Kleinfeld, Reviewing Editor, University of California, San Diego, United States

Publication history

  1. Received: July 24, 2016
  2. Accepted: February 2, 2017
  3. Accepted Manuscript published: February 4, 2017 (version 1)
  4. Version of Record published: February 27, 2017 (version 2)

Copyright

© 2017, Xiao 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

  • 2,256
    Page views
  • 649
    Downloads
  • 3
    Citations

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

Comments

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)