Thy1/GCaMP6 transgenic mice

Transgenic mice expressing GCaMP6 driven by the thymus cell antigen 1 (Thy1) promotor were acquired from Jackson Laboratories (JAX Strain: C57BL/6J-Tg(Thy1-GCaMP6f)GP5.5Dkim; stock: 024276)[17, 26]. In this model, GCaMP6 expression is widespread in the CNS (e.g. cortex, hippocampus, thalamus, and cerebellum) and localized predominantly to excitatory projection neurons [27]. Thy1/GCaMP6 transgenic genotypes were confirmed by PCR using the forward primer 5’-CATCAGTGCAGCAGAGCTTC-3’ and reverse primer 5’- CAGCGTATCCACATAGCGTA-3’.

Animal preparation

All procedures have been approved by the Washington University School of Medicine Animal Studies Committee. Seven GCaMP6 mice (12-16weeks of age, 28-36g) were used in all experiments in this study. Mice were sedated with isoflurane (3% induction, 1% maintenance, 0.5L/min) and placed in a stereotactic holder. The head was then shaved, and a midline incision made to expose the skull. Body temperature was maintained at 37°C using a temperature controlled heating pad. Following scalp retraction, pilot holes were drilled into but not penetrating the skull at approximately -1mm posterior to bregma, and +/- 5mm lateral to bregma (near barrel/auditory cortex). Small stainless steel self-tapping screws (BASI Inc., West Lafayette, IN, USA) were then inserted into these holes for collecting EEG recordings. Chronic cranial windows made of Plexiglas and with pre-tapped holes were fixed to the skull using dental cement (C&B-Metabond, Parkell Inc., Edgewood, NY, USA).

Ex vivo epifluorescence and confocal imaging

Mice were deeply anesthetized with FatalPlus^TM (Vortech Pharmaceuticals, Dearborn, MI, USA) and transcardially perfused with 0.01M PBS. The brains were removed and fixed in 4% paraformaldehyde for 24h and transferred to 30% sucrose in 0.2M PBS. After brains were saturated, they were snap-frozen on dry ice and coronal sections, 50μm thick, were made with a sliding microtome. Sections were stored in 0.2M PBS, 30% sucrose, and 30% ethylene glycol at −20°C. For viewing, cut sections were washed in PBS, mounted, and intrinsic GCaMP6 fluorescence was examined with epifluorescence microscopy (Nikon Eclipse 80i, Nikon Instruments Inc., Melville, NY, USA). For viewing using confocal microscopy, additional cut sections were washed in PBS, mounted, and coverslipped in DAPI containing mounting media (Vector Laboratories, Burlingame, CA, USA). Fluorescent images of DAPI labeled cells and intrinsic GCaMP6 fluorescence were then acquired with a Nikon A1-Rsi inverted confocal microscope. Excitation illumination was provided by 405nm and 488nm lasers for DAPI and GCaMP fluorescence, respectively (Nikon Eclipse 80i, Nikon Instruments Inc., Melville, NY, USA).

In vivo optical imaging system

Sequential illumination was provided by four LEDs: 470nm (measured peak λ = 454nm, LCS-0470-15-22, Mightex Systems, Pleasanton, CA, USA), 530nm (measured peak λ = 523nm, LCS-0530-15-22), 590nm (measured peak λ = 595nm, LCS-0590-10-22), and 625nm (measured peak λ = 640nm, LCS-0625-03-22). All LEDs are referred to by their measured wavelengths (i.e. 454nm, 523nm, 595nm, and 640nm) throughout the rest of the manuscript. The 454nm LED is used for GCaMP excitation, and the 523nm, 595nm, and 640nm LEDs are used for multispectral oximetric imaging. The 523nm LED was also used as an emission reference for GCaMP6 fluorescence in order to remove any confound of hemodynamics in the fluorescence signal (described below). Both the 454nm and 523nm LED light paths were made collinear by using a multi-wavelength beam combiner dichroic mirror (LCS-BC25-0505, Mightex Systems, Pleasanton, CA, USA). For image detection, we used a cooled, frame-transfer EMCCD camera (iXon 897, Andor Technologies, Belfast, Northern Ireland, United Kingdom) in combination with an 85mm f/1.4 camera lens (Rokinon, New York, NY, USA). The acquisition framerate was 16.8Hz per channel, with the overall framerate of the camera as ~67Hz. This framerate is well above respiration rates (approximately 3Hz) and fast enough to adequately characterize GCaMP6 activity. To increase frame rate as well as increase SNR, the CCD was binned at 4 x 4 pixels; this reduced the resolution of the output images from full-frame 512 x 512 pixels to 128 x 128 pixels but allowed for increased frame rate. Both the LEDs and the exposure of the CCD were synchronized and triggered via a DAQ (PCI-6733, National Instruments, Austin, TX, USA) using MATLAB (MathWorks, Natick, MA, USA). The field-of-view was adjusted to cover the majority of the convexity of the cerebral cortex with anterior-posterior coverage from the olfactory bulb to the superior colliculus, resulting in an area of approximately 1cm² with each pixel being approximately 78μm x 78μm. To minimize specular reflection from the skull, we used a series of linear polarizers in front of the LED sources and the CCD lens. The secured mouse was placed at the focal plane of the camera. The combined, collimated LED unit was placed approximately 8cm from the mouse skull, with a working distance of approximately 14cm as determined by the acquisition lens. A 515nm longpass filter (Semrock, Rochester, NY, USA) was placed in front of the CCD to filter out 470nm fluorescence excitation light and a 460/60nm bandpass filter (Semrock, Rochester, NY, USA) was used in front of the excitation source to further minimize leakage of fluorescence excitation light through the 515nm longpass filter. The pulse durations for the LEDs are 20ms, 5ms, 3ms, 1ms for 454nm, 523nm, 595nm, and 640nm, respectively. All imaging data from awake and anesthetized imaging sessions were acquired and stored in separate 5min imaging runs.

Electrical hindpaw stimulation

Electrical stimulation was generated using an isolated pulse stimulator (Model 2100, A-M Systems, Carlsborg, WA, USA) delivered to the left hindpaw of each mouse via micro vascular clips (Roboz Surgical Instrument Co., Gaithersburg, MD, USA). The somatosensory activation block paradigm consisted of an initial rest period of 5s followed by 4s of electrical stimulation at a frequency of 2Hz (pulse duration: 300μs, current: 0.5mA) and ending with 51s of rest. This 60s block was repeated five times during each 5min imaging session for subsequent block averaging.

Anesthetized imaging

For anesthetic imaging, mice were anesthetized with intraperitoneal injection of a ketamine/xylazine cocktail (86.9mg/kg ketamine, 13.4mg/kg xylazine). Anesthetic effect was verified by confirming that the animal was not responsive to a hind paw pinch. The animal was placed and kept on a solid state water circulating heating pad (T/Pump Classic, Stryker Co., Kalamazoo, MI, USA), maintained at 42°C, to minimize electronic noise in the EEG signal.

Awake imaging

We constructed a customized apparatus consisting of a felt pouch suspended by optical posts. This provided a dark, comfortable environment for the mouse throughout data acquisition. During the imaging session, the mouse was able to move freely while its head was secured, preventing the awake mouse from applying torque on their restrained head and optical window. Though no accelerometers or other behavioral measures were used to track motion within the pouch during imaging, after acclimation, mice are typically relaxed with infrequent limb motion. After recovery from this surgery (1 week), the mouse was acclimated to the apparatus by a training period consisting of two 20min sessions. Acclimation is indexed by a return to normal behavior (e.g., whisking, grooming, and walking with head restrained). Once acclimated, mice were mounted to the imaging system with a small bracket (using the window’s pre-tapped holes) with the body of the mouse placed inside the felt pouch. Imaging then proceeded as described above.

EEG recording and processing

Summed field potentials were concurrently recorded during GCaMP6 and OIS imaging using the 2 screws bilaterally implanted in the skull and subsequently digitized at 1kHz (Power Lab EEG Amplifier, AD Instruments, Dunedin, New Zealand). EEG frequency spectra were calculated by splitting data from each 5min session into thirty 10s epochs, applying a 10s Hann window to each, calculating the Fast Fourier Transform, then averaging the squared modulus of each epoch together.

Image processing

A representative frame of baseline light levels in a dark environment, calculated from a mean of dark images collected over 1min, was subtracted from the raw data. All pixel time traces were detrended to remove any variations in light levels due to photobleaching, LED current drift, and nonuniformity across the skull [28]. Reflectance changes in the 523nm, 595nm, and 640nm LED channels were used in combination to provide oximetric data using the Modified Beer-Lambert Law, described previously [2]. Images in each contrast were smoothed with a Gaussian filter (5x5 pixel box with a 1.3 pixel standard deviation).

The GCaMP6 fluorescent signal must be corrected for any contribution from the vascular compartment and varying concentrations of absorptive hemoglobin. Common correction methods to calculate relative fluorescence changes include using a reference wavelength for applying subtraction and ratiometric techniques. We implemented a ratiometric correction algorithm (Eq 1) to correct fluorescent emission for any absorption by hemoglobin and deoxyhemoglobin using the reflectance channels at the GCaMP6 emission wavelengths (523nm LED) as a reference.

(1)

y(t) is the final corrected GCaMP6 time series for a given pixel, I^em refers to the detected fluorescent emission intensity. I^ref describes the measured reflectance changes at the emission wavelength. A single frame from the 628nm reflectance channel was loaded into Adobe Photoshop CC 2014 (Adobe Systems, San Jose, CA, USA) and all regions not corresponding to brain were manually painted white. The image was loaded back into MATLAB and used to create a binary brain mask. All subsequent analysis was performed on those pixels labeled as brain. Image sequences from each mouse (as well as the brain mask for each mouse) were affine-transformed to a common atlas space (based on the Paxinos mouse atlas) using the positions of bregma and lambda[29]; please see S3 Fig for cortical region parcels corresponding to the network seeds used in analysis and White and Bauer et al., for more information [2]. The time traces of all pixels defined as brain by the binary mask were averaged to create a global brain signal. For all functional connectivity and activation analyses, this global signal was regressed from every pixel’s time trace to remove global sources of variance, such as motion and cardiac pulsations, thus increasing correlation network specificity [2, 30, 31]. For all whole-brain delay analyses, the global brain signal was not regressed from the spontaneous data, as it served as the temporal point of reference for the cross-correlation calculations. All resting-state data were filtered over the canonical functional connectivity frequency range (0.009–0.08Hz) [1], an intermediate frequency band (0.08–0.4Hz), or the delta activity band (0.4–4.0Hz). Evoked activity was lowpass filtered at 6Hz to remove high-frequency noise.

Spectral analysis

To construct frequency spectra of spontaneous GCaMP6 activity, time series at each pixel were split into two 150s windows (per each 5min imaging session), windowed with 150s long Hann functions, and then the Fast Fourier Transform (FFT) was calculated. The squared moduli of these FFTs were then averaged across the brain to produce the final power spectra.

Seed-based functional connectivity

Using the Paxinos histological atlas as a reference [29], seed locations were chosen at coordinates corresponding to the left and right visual, motor, somatosensory, cingulate, parietal, auditory, and retrosplenial cortices as well as the right and left superior colliculi and olfactory bulbs. A 0.5mm diameter circle at each seed location was averaged to create a seed time trace. These seed traces were correlated against every other brain pixel using the Pearson correlation coefficient (r) to create functional connectivity maps. Functional connectivity maps were calculated for every 5min imaging session, Fisher z-transformed (using the inverse hyperbolic tangent function) to improve normality and stabilize the variance of the correlation coefficients, averaged across all imaging sessions and mice, then finally reverse transformed (using the hyperbolic tangent function) back to correlation coefficients. Matrices consisting of the correlation coefficient between each pair of network seed traces were constructed by plotting the pairwise correlation coefficients in each cell. Dice’s coefficient was adapted (Eq 2) to compare spatial similarity of variable length time and sampling window functional connectivity maps and maps created using group averaged. The adapted Dice’s coefficient was calculated using: (2)

Where s is the spatial similarity, A and B are the functional connectivity maps being compared, an i indexes each pixel within the maps.

Whole-brain delay analysis

The GCaMP6 time series from every pixel was cross-correlated with the global signal (described above) to reveal any phase shifts (quantified as time lags). The temporal shift that maximizes the correlation between both time series was plotted at each pixel to create a map of delay topography across the cortex.

Statistical analysis

To compare differences in functional connectivity network maps across the anesthetized and awake states, two-tailed Welch’s unequal variances t-tests were used to compare Fisher z-transformed correlation coefficients at every pixel. T-statistic maps were then created and subsequently thresholded using a Bonferroni-adjusted significance level of α = 3.1e-6, calculated by adjusting a conventional α = 0.05 significance level by the total number of pixel comparisons (128 pixels x 128 pixels, 16384 pixel comparisons in total). Two-tailed paired t-tests were used to compare differences at each pixel in the visual network functional connectivity maps in the mouse transitioning from anesthesia to wakefulness. Each 5min epoch was divided into ten 30s windows for all comparisons across epochs. Calculated t-statistic maps were thresholded using q = 0.001 after false discovery rate adjustment. Two-tailed Welch’s unequal variances t-tests were used to compare the full width at half maximum values for both the ipsilateral and contralateral horizontal line profiles from each seed-based functional connectivity map.

Concurrent imaging of GCaMP6 fluorescence and hemodynamics

In order to image both GCaMP6 activity and hemodynamics concurrently, we developed an optical imaging system with multiple illumination wavelengths that were temporally multiplexed. A filtered detection scheme provided both fluorescence and reflectance imaging. To image GCaMP6 fluorescence, we selected incident LEDs wavelengths that match both the excitation (peak lambda = 497nm) and emission (peak λ = 512nm) spectra of GCaMP6 (Fig 1A and 1B). To correct the raw fluorescence measurement for hemodynamic absorption changes, we incorporated a reference reflectance measure (green reflectance measurement using a LED with peak λ = 523nm) in a ratiometric correction technique (see methods for details). The green LED was also used along with yellow (peak λ = 595nm) and red (peak λ = 640nm) LEDs for concurrent calculation of differential changes of hemoglobin concentration (see Methods) (Fig 1C). The excitation light (peak λ = 454nm) was blocked between the lens and the detector using a 515nm longpass filter (Fig 1A–1C. Light was detected for all wavelengths using an electron multiplying charge coupled detector (EMCCD, iXon 897, Andor Technologies). Sequentially triggering each of the four LEDs yielded GCaMP6 and hemodynamic frame rate of16.8Hz.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Concurrent GCaMP6 fluorescence and optical intrinsic signal imaging system.

LED wavelengths must be selected to provide GCaMP6 excitation, distinguish between oxy- and deoxyhemoglobin absorption for oximetry, and enable correction of GCAMP6 fluorescence emission for correction of confounding hemoglobin absorption. (A) GCaMP6 dynamics are imaged using fluorescence measurements, wherein the fluorophore has a peak excitation at λ = 497nm and peak emission at λ = 512nm. A longpass filter at λ = 514nm is used to block fluorescence excitation light. (B) Spectra from the excitation LED (λ = 454nm, bandpass filtered at 460/60nm). A reference (λ = 523nm) LED is used for correcting for changes in optical properties due to the presence of absorptive hemoglobin. The transmission curve for the 514nm longpass filter at detection is shown for reference. (C) Schematic of the combined GCaMP and OIS imaging system. GCaMP excitation and reference LEDs share a collinear optical path using dichroic mirrors. Green (λ = 523nm), yellow (λ = 595nm), and red (λ = 640nm) are used together to determine differential concentration changes of hemoglobin. Detection is done using a cooled EMCCD with an 85mm f/1.4 lens.

https://doi.org/10.1371/journal.pone.0185759.g001

GCaMP6 fluorescence is detectable and temporally distinct from HbO₂

To confirm GCaMP6 fluorescence was present in superficial cortical layers, paraformaldehyde-fixed GCaMP6 brain slices were imaged using both epifluorescence (Fig 2Ai) and confocal (Fig 2Aii) microscopy. Individual neuronal somas (Fig 2Aii, green: GCaMP6, blue: DAPI stain) and their neurites are detectable throughout cortical layers I–VI (Fig 2A) and across the cortex (Fig 2Ai, green: GCaMP6).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Thy1/GCaMP6 mouse model validation.

Based on previous studies, Thy1/GCaMP6 fluorescence should provide a signal predominantly driven by cortical neuronal neurons and that is faster than hemodynamic measures. (A) Epifluorescence (i) and confocal (ii) images of 50um coronal slices collected from a representative Thy1/GCaMP6 mice. GCaMP6 expression (green) is located throughout the cortex (i) and localized to neuronal soma (ii, blue-DAPI stain) and can be seen throughout cortical layers I-VI. (B) Map of stimulus evoked responses to electrical stimulation of left hindpaw (thresholded at >50% peak response) in the anesthetized state. (C) Time series of evoked responses sampled using masks created from the right half of the 50% peak response images in part B. Green reflectance (green trace, top panel) is used to correct raw GCaMP6 emission (teal trace) for changes in absorption due to hemodynamics using a ratiometric approach (see Methods; the corrected trace is black). The corrected GCaMP6 trace (black) is sensitive to each stimulus presentation (black trace, lower panel), whereas HbO₂ (red) and HbR (blue) follow the slower, canonical hemodynamic response. Error bars are standard error of the mean. Data from 36 blocks across 7 mice are included in all evoked analysis.

https://doi.org/10.1371/journal.pone.0185759.g002

To evaluate the mapping fidelity of both the GCaMP6 and hemoglobin contrasts and the ratiometric correction strategy, we used a block design electrical hindpaw stimulation paradigm (each block consisted of 4s of stimulation at 2Hz, followed by 51s of rest; 36 blocks across n = 7 mice were included in analysis). Mice were under ketamine/xylazine anesthesia during electrical stimulation presentations to minimize discomfort and distress. Maps of the spatial extent of hindpaw cortex, contralateral to stimulation, for both GCaMP6 and HbO₂ show well localized responses (Fig 2B). Before ratiometric correction, each electrical pulse response is represented in the raw GCaMP6 signal (Fig 2C, dark green). However, during the stimulus, after initially rising, the raw fluorescence signal rapidly decreases, owing to increased absorption (Fig 2C, light green) from the increased blood volume and hemoglobin concentration (Fig 2C, red trace). After ratiometric correction for absorption, the GCaMP6 signal increases and stays elevated until returning to baseline approximate 6s post-stimulus. The evoked responses corresponding to HbO₂ (Fig 2C, red trace) and HbR (Fig 2C, blue trace) have an onset approximately 2s after the initial stimulus presentation. Also, the observed ratio of HbO₂ to HbR peak magnitudes is approximately 3:1, as expected from canonical functional responses [32]. These results collectively demonstrate the ability of the hardware and algorithms to concurrently image both calcium dynamics (via GCaMP6) and hemodynamics.

Functional connectivity mapping in GCaMP6 mice

Functional connectivity was quantified by calculating the correlation between the time traces in two distinct brain regions (Fig 3). To facilitate comparisons of fcMRI and the functional connectivity analysis using the concurrent GCaMP6/hemodynamic imaging system, we implemented a standard functional connectivity processing pipeline across three frequency bands: 0.009–0.08Hz (“infraslow”), 0.08–0.4Hz (“intermediate”), and 0.4–4.0Hz (“delta-band”) for all contrasts. Seed-based functional connectivity was then performed using spontaneous GCaMP6 and HbO₂ activity across frequency bands and awake and anesthetized states. In an illustrative example of spontaneous GCaMP6 data measured in wakefulness, the time series sampled from left motor and contralateral right motor cortices (Fig 3, black & red traces and circles, respectively) over the infraslow band are highly correlated (r = 0.95), whereas the correlation between the left motor seed and right cingulate cortices is low (r = 0.08). Functional connectivity maps were constructed by calculating the position-dependent correlation coefficient between this left motor seed and the spontaneous time series at every pixel across the field-of-view (Fig 3A, right; representative time series from all contrasts and frequency bands, S4 Fig). A recent wide-field in vivo optical imaging study of Thy1/GCaMP6 mice by Ma and colleagues showed that spontaneous GCaMP6f and hemodynamic activity from awake mice were most correlated when filtered with a 0.35Hz-wide frequency-domain window centered at 0.21Hz [33]. Based on this observation, resting-state data were also filtered over an intermediate frequency band (0.08–0.4Hz) to facilitate exploration of GCaMP6f and hemodynamic correlated activity. Fig 3B shows representative spontaneous seed traces from Motor and Cingulate networks from this 0.08–0.4Hz band. This same analysis was applied to the higher frequency delta band (Fig 3C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Performing functional connectivity analysis in the awake mouse brain.

Time series of spontaneous GCaMP6 activity filtered over the canonical functional connectivity band: (A) 0.009–0.08H, (B) 0.08–0.4Hz, and (C) the delta band: 0.4–4.0Hz. Corresponding functional connectivity maps constructed using a seed from left motor (Mot.) cortex (black circle) in 0.009–0.08Hz (top right), 0.08–0.4Hz (center right), and 0.4–4.0Hz (bottom right) spontaneous data are also shown. Black and red traces correspond to left and right motor cortices, respectively. The green trace is sampled from right cingulate cortex. Over both frequency bands, left and right motor cortex show highly correlated activity whereas right motor with left cingulate show reduced temporal correlation. At the pixelwise level, correlating the time series from the left motor (black) with every pixel in the brain produces a map showing homotopic functional somatosensory structure across hemispheres. Mot.: Motor, Cing.: Cingulate.

https://doi.org/10.1371/journal.pone.0185759.g003

Because GCaMP6 is able to probe activity at frequencies higher than traditionally associated with hemodynamic functional connectivity (i.e. <0.1Hz), we examined the effect of time window length on the stability of the functional connectivity patterns using the highest frequency GCaMP6 data examined in this study which are filtered over the delta frequency band (0.4–4.0Hz). To avoid confounds from the effects of anesthesia (e.g. ketamine-associated increase in cortical activity in this delta frequency range), awake data were used [21]. A spatial similarity analysis was employed to evaluate the convergence of network structures with increasing duration of resting-state data. For each window length (from 2.6s to 300s in duration) and for each seed location, we calculated the similarity between representative individual mouse data and a respective reference template calculated from longer data acquisitions with averaging across all mouse (i.e. group-level) maps (Fig 4A). The same analysis was carried out using the wakefulness data filtered over the infraslow frequency band (S5 Fig). In contrast to Fig 4A, the window length necessary to reach convergence is more network-dependent and most networks do not achieve a spatial similarity score of at least 0.5 compared to the group average reference with window lengths shorter than 100 seconds. Using window lengths as short as 30s for data from an individual mouse, the functional connectivity maps constructed with 7 canonical seeds exhibited distinct qualitative network-specific structure (Fig 4B). The similarity between functional connectivity maps based on 30s of data vs. the group is consistent for all 7 networks for all mice (Fig 4C). To evaluate the stability of the maps over time, the functional connectivity map for a seed placed in motor cortex was constructed for a sequence of overlapping 30s segments of data, shifted by increments of 2.6 seconds, for all mice and compared to the group-averaged map. The similarity score remains relatively consistent across the entire 5min session (Fig 4D). Collectively, these data imply that windows with a duration as short as 30s can produce stable functional connectivity maps when using GCaMP6 data. These results show that the robustness of GCaMP6 functional connectivity maps from these 30s windows is relatively independent of temporal sampling location, network seed, and mouse.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Stability and stationarity of spontaneous GCaMP6 activity in functional connectivity analysis.

(A) Functional connectivity maps were constructed using time windows of spontaneous 0.4–4.0Hz GCaMP6 (awake) data of increasing length from one mouse (Mouse 2). Windows were increased in length by ~2.6s up to ~300s. Spatial similarity (Dice coefficient) was calculated between each of these functional connectivity maps for all networks relative to their corresponding group level map. The functional connectivity maps calculated by using a seed in left motor cortex are highlighted as an example. Across all networks, similarity scores converge at 0.9 with a window length of ~30s. (B) Seed-based functional connectivity maps from Mouse 2 constructed using 30s of spontaneous GCaMP6 data. (C) For each of 7 networks, the same 30s window of spontaneous data for each network was compared to the corresponding group-level network functional connectivity map for each of seven mice (circles). The stability of similarity of 30s of data is present across network and across mice. (D) Using consecutive 30s windows (sampled every ~2.6s) from a single 300s imaging session from each mouse, the spatial similarity between all 30s motor functional connectivity maps and the group level motor functional connectivity map is shown to be relatively stationary. Error bars are standard error of the mean.

https://doi.org/10.1371/journal.pone.0185759.g004

Intercontrast functional connectivity between delay-shifted GCaMP6 and HbO₂ recapitulates hemodynamic functional connectivity structure in the anesthetized state

Though the origins of the GCaMP6 and HbO₂ signals are distinct, they are associated via neurovascular coupling [34, 35]. Accordingly, we examined the potential interaction between 0.08–0.4Hz GCaMP6 and HbO₂ dynamics. This intermediate frequency band was selected given the maximal correlation recently observed between spontaneous GCaMP6 and HbT at approximately 0.21Hz [33]. Data from the anesthetized state were selected to directly build upon previously-reported literature on functional connectivity using OIS hemodynamic contrasts [2–6, 36–38]. Further, the delay times derived from spontaneous data could be compared to the delay time from forepaw stimulation. By calculating the lagged-correlation between the GCaMP6 and HbO₂ evoked responses to electrical hindpaw stimulation (Fig 2C), we measured the temporal delay of HbO₂ relative to GCaMP6 to be ~0.65s (Fig 5A and 5B), which is comparable in magnitude to the previously reported delay of maximal correlation calculated between spontaneous GCaMP6 and HbT (~0.86s) in this frequency band [33]. Due to the lack of full characterization of the relationship and neurophysiology of neurovascular coupling in evoked versus spontaneous activity conditions, we also carried out this analysis by delay-shifting spontaneous data using delays determined from seed-trace pairs of resting-state data (Fig 5C and 5D). For reference, we evaluated the same seed locations previously used for hemodynamic functional connectivity mapping in spontaneous data acquired from mice under anesthesia [2, 3, 6](S1 Fig). Within the intermediate frequency band, GCaMP6 produced intact functional connectivity patterns (Fig 5D, middle) similar to concurrent HbO₂ maps (Fig 5D, top; S2 Table), as well as to maps made using deoxyhemoglobin (HbR) (Fig 5D, bottom; S2 Table). Next, to evaluate the potential relationship between spontaneous GCaMP6 activity and HbO₂ and HbR dynamics, we temporally shifted spontaneous GCaMP6 traces for the seven seeds by the computed spontaneous delay between HbO₂ (Fig 5D, top, bottom row) and HbR (Fig 5D, bottom, top row) and then calculated correlation maps between the GCaMP6 seeds and the shifted HbO₂ and HbR data (Fig 5D, middle). The resulting maps show distinct patterns spatially similar to the within-GCaMP6 functional connectivity maps (Fig 5D middle, S3 Table). The same spontaneous delay-shift analysis was carried out on infraslow data (S6 Fig, S3 Table), producing functional connectivity maps with ill-defined network boundaries. To further investigate the relationship and neurophysiology of neurovascular coupling in evoked conditions, we also carried out the same delay-shifting analysis using the fixed calculated ~0.65s shift (S7 Fig); the results generally recapitulate those reported here (Fig 5, S3 Table). Moreover, intercontrast functional connectivity analysis between GCaMP6 and HbR produced functional connectivity maps with largely indeterminate network-specific spatial structure (Fig 5, S6 and S7 Figs, S3 Table). These results imply that both signals are synchronized, with GCaMP6 potentially containing predictive information of downstream hemodynamics.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Intercontrast functional connectivity mapping in the anesthetized state.

Based on the temporal delay between evoked GCaMP6 and HbO₂/HbR responses combined with similar observed 0.08–0.4Hz functional connectivity structure, delay-shifted GCaMP6 likely shares functional network information with HbO₂ and HbR. (A) GCaMP6 (black), HbO₂ (red), and HbR (blue) evoked responses to 2Hz 0.5mA electrical stimulation of left hind paw (see Fig 2B and 2C). Data from a sampling window spanning from approximately t = 0s to t = 7.5s were included in subsequent cross-correlational delay analysis. (B) Cross-correlation from the GCaMP6 and HbO₂ time courses in (A), and the GCaMP6 and HbR time courses in (A), both show a delay of 0.65s between responses. (C) An example of spontaneous GCaMP6, HbO₂, and HbR time traces from the right motor cortex seed over the 0.08–0.4Hz band. HbO₂ and HbR data was shifted based on the lag with the maximum correlation calculated between the spontaneous hemoglobin data and spontaneous GCaMP6 data per seed (as shown by arrows). (D) Top and Bottom: Seed-based group-averaged functional connectivity mapping in filtered (0.08–0.4Hz) anesthetized data using HbO₂ (top) or HBr (bottom) and cross-correlation curves (with mean delays of maximum correlation in black text with standard errors of the mean in parenthesis) between each HbO₂ network seed trace and corresponding GCaMP6 seed trace. Middle: Seed-based GCaMP6 functional connectivity maps and corresponding intercontrast seed-based functional connectivity maps for GCaMP6 vs temporally-shifted HbO₂ (top) and GCaMP6 vs temporally-shifted HbR (bottom). Correlation between spontaneous HbO₂ and GCaMP6 contrasts (filtered between 0.08–0.4Hz) are better optimized by removing the intercontrast time shift between these contrasts compared to the analogous correlation between spontaneous HbR and GCaMP6 contrasts.

https://doi.org/10.1371/journal.pone.0185759.g005

At higher frequencies, GCaMP6 shows a striking state-dependence in correlation map structure

To examine a potential dependence of GCaMP6 activity and functional connectivity structure on state (anesthesia vs. awake), we evaluated seed based functional connectivity maps of GCaMP6 in the higher 0.4–4.0Hz frequency band (Fig 6) given the established sleep- and anesthesia-associated phenomena that occur within it [20, 21, 39]. Under ketamine/xylazine anesthesia (Fig 6A, top row), GCaMP6 functional connectivity shows an anterior-posterior separation, with large regions of high magnitude correlations and anti-correlations across the cortex. In contrast, functional connectivity correlation structure in the awake state (Fig 6A, second row) exhibits higher spatial specificity for each network of interest. To quantify the degree of detail and localization, we thresholded using pixels with positive correlations >0.2 of the maximum correlation value in each image and then summed the remaining pixels to generate an estimate of area (here referred to as correlation area) and averaged these across all seven seeds. The mean correlation area in the awake state was 39.2% (std +/- 15.9%) smaller than that of the corresponding anesthesia maps. The difference images constructed between the group level anesthetized and awake functional connectivity maps (Fig 6A, third row) show approximately the same overall topography in cingulate, motor, somatosensory, retrosplenial, and visual functional connections. This result suggests that the awake maps have finer detail than the anesthesia maps. Further support for these state-dependent differences is seen in the pixelwise t-statistic maps (Welch’s t-test) which again are similar to the top row of Fig 6A and reinforces the structure observed in the difference images (Fig 6A, third row). Finally, horizontal line profiles through the center of each seed for both the awake (red) and anesthetized (blue) functional connectivity maps show higher focality (measured using reductions in ipsilateral and contralateral full width at half maximum, table in S1 Table.) in cingulate, motor, somatosensory, retrosplenial, and visual networks in the awake state. Visualized in matrix form (Fig 6B), the pairwise functional network connections under anesthesia (Fig 6B, top) show clustering of positive correlations in anterior networks (cingulate, motor, and somatosensory) and negative correlations in posterior networks (retrosplenial, parietal, visual, and auditory), whereas awake animals (Fig 6B, bottom) exhibit stereotypical high correlation strength only in the off-diagonals. Note the functional connectivity structure over the delta band during wakefulness is consistent across all mice (S2 Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Functional connectivity mapping in GCaMP6 mice.

(A) Seed-based functional connectivity maps for seven canonical functional networks in anesthetized and awake GCaMP6 mice (N = 7) at delta-band frequencies. Under anesthesia (top), delta activity (0.4–4.0Hz) drives a strongly correlated/anticorrelated network structure between anterior and posterior brain regions that is not observed in awake animals (second row). This high magnitude correlation feature is preserved in the difference images (third row) and after performing pixelwise t-tests and thresholding the t-statistic maps at a Bonferroni adjusted α = 3.1e-6 (fourth row). Together, these results show that the topography of functional connectivity maps in the anesthetized state is less sensitive to seed-based network-specific features compared to functional connectivity topography during wakefulness. Finally, horizontal line profiles through the center of each seed show higher ipsilateral and contralateral focality in the awake (red) functional connectivity maps compared to anesthetized (blue) maps (fifth row and table in S1 Table). (B) Seed-based pairwise functional connectivity matrices. Regional correlation coefficients between each seed reveal heterogeneous connectivity structure within each contrast across different states. Clusters of networks are present in the anesthetized matrix (top), including between cingulate, motor, and somatosensory networks that are diminished in the awake mouse (bottom).

https://doi.org/10.1371/journal.pone.0185759.g006

Strong cortical delta activity is associated with anterior-posterior spontaneous GCaMP6 dynamics

Compared to the functional connectivity structure of GCaMP6 during wakefulness, functional connectivity maps under anesthesia largely contain strong anti-correlations between anterior and posterior brain structures (Fig 6A). These anti-correlations might be indicative of coherent propagating waves of spontaneous GCaMP6 activity over the anterior-posterior axis. To determine if our model and system are sensitive to these dynamics, we examined the frequency content of global spontaneous GCaMP6 activity in a representative mouse under ketamine/xylazine anesthesia and while awake (Fig 7A). A phenomena commonly reported in the sleeping and anesthetized brain is the presence of prominent propagating slow-wave delta activity [40]. The GCaMP6 spectra of mice under anesthesia (Fig 7A, black trace) reveal a peak centered at ~1.5Hz that is absent in wakefulness (Fig 7A, gray trace), which is consistent with previous electrophysiological studies reporting delta-band activity [21]. Further, the state-dependence of delta activity exhibited in GCaMP6 dynamics is recapitulated in the simultaneously-acquired global field potential EEG data (anesthetized: dark green, awake: light green).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Spontaneous activity in anesthetized and awake mice reveals regional temporal delay structure.

(A) Power spectra from the whole brain from a representative mouse. The presence of a peak at ~1.5Hz in both the anesthetized GCaMP6 (black) and EEG (dark green) corresponds to increased delta band (0.4–4.0Hz) activity observed in the brain under anesthesia and deep, slow-wave sleep (see Introduction). Note that this peak is absent when the mouse is awake (GCaMP6, transparent black, EEG, transparent green). Maximum cross-correlation between the whole brain signal and each pixel’s timeseries were calculated across the FOV. Cortex-wide delay topography may underlie the striking correlated/anti-correlated structure observed in the anesthetized functional connectivity maps due to the effect of phase shifts on the magnitude and sign of zero-lag correlations. Asterisks represent significant (p<0.0001) comparisons across state following two-sample Welch’s t-tests. Error bars are standard error of the mean. (B) Group-wise delay maps in anesthetized (top) and awake (bottom) mice (N = 7). The map from anesthetized mice show a temporal separation along the anterior-posterior axis of the brain, with anterior somatomotor regions leading the whole brain signal by < = 0.06s and posterior visuoparietal areas lagging the whole brain signal by < = 0.06s. This time shift is in agreement with the striking correlated/anti-correlated structure in the delta band functional connectivity maps (see Fig 4). This structure is not present in awake mice. Group-average maps are masked using pixels identified to be significant (p<0.001) following one-sample t-testing. (C) Top: Delay maps from consecutive 300s epochs from one mouse as it transitions out of anesthesia to awake reveals a loss of anterior-posterior delay structure across time. Middle section: functional connectivity maps for a left visual seed from each epoch (top row), difference images between each successive epoch and E1 (middle row), and pixelwise t-statistic maps after performing paired t-tests between E1 and each other epoch (bottom row). The boundary of the area of maximum delay in the anterior, cingulate network in the delay map in E1 (top section of C) is overlaid on the E6 t-statistic map for reference. These data show that the correlation magnitude of the parietal region differs across state and that there is agreement across delay and functional connectivity analyses in the implicated cingulate network. All analysis was performed after dividing each 300s epoch into ten 30s windows. Bottom: Spectrogram over the delta frequency band (0.4–4.0Hz) from each epoch showing the loss of delta-band activity (peak at ~1.5Hz) with time across epochs.

https://doi.org/10.1371/journal.pone.0185759.g007

In order to better understand the observed state-dependent GCaMP6 functional connectivity structure, we examined the delay topography of the GCaMP6 signal by calculating the pixelwise lag (via cross-correlation) relative to the whole-brain signal (Fig 7B and 7C). At the group level (Fig 7B), the statewise contrast observed in Fig 6 is prominent, with somatomotor regions tending to lead the brain by up to 0.08s and visuoparietal areas lagging by a comparable interval. This temporal separation across the cortex could underlie the observed anterior/posterior structure found in the functional connectivity maps of GCaMP6 activity in anesthetized animals (Fig 6).

Finally, we evaluated spontaneous GCaMP6 activity in a mouse while it transitioned from anesthesia to wakefulness across six consecutive 5min epochs (Fig 7C). In the first epoch (E1), the mouse shows an anterior-posterior delay structure (Fig 7C, top), with cingulate regions leading the whole brain signal by ~0.08s and visual/parietal regions lagging by ~0.08s. As the mouse wakens (determined by the attenuation of delta-band power, peak at ~1.5Hz, Fig 7C spectrogram), this topography deteriorates (Fig 7C, top, transition from E2 to E3). The corresponding functional connectivity map constructed using a left visual seed is shown for each epoch (Fig 7C, second row). To improve SNR when creating average functional connectivity maps (Fig 7C, second row) and enable statistical testing using paired t-tests (Fig 7C, fourth row), each 5min epoch was divided into ten 30s windows. This window length was selected based on the observed convergence in spatial similarity of functional connectivity maps constructed using at least 30s of spontaneous data compared to the group average reference (Fig 4A). During wakefulness (Fig 7C, columns E5 & E6), the correlation maps show smaller, more detailed regions of reduced correlation magnitude, both ipsilateral and contralateral to the seed, as compared to the maps during anesthesia. Successive difference maps calculated between E1 and each subsequent epoch (E2-E6) show that connectivity in the parietal cortex preferentially changes during the transition from anesthesia (E1) to wakefulness (E6). There is also a subtle but visible difference in cingulate cortex between E1 and the later epochs. t-statistic maps (thresholded at q<0.001 after false discovery rate adjustment), constructed from pixelwise paired t-tests between E1 and all subsequent epochs, recapitulate these findings (Fig 7C, fourth row). The anterior, cingulate region leads the brain during anesthesia (E1, Fig 7C, top), which is consistent within the difference t-statistic maps of functional connectivity during the awake epochs (E5 and E6). The boundary of maximum delay value from this cingulate network (E1, Fig 7C, top) is overlaid on the t-statistic map in column E6, showing the agreement across both analysis strategies.