Mapping of neurovascular and neurometabolic couplings by multimodal optical imaging

Abstract

Neurovascular coupling links calcium (Ca²⁺)-dependent neuronal activity to cerebral blood volume changes, whereas neurometabolic coupling describes alterations of neuronal activity and glucose uptake. While mesoscale optical imaging of neurovascular coupling is prevalent, neurometabolic coupling has been explored much less. We describe a multiplexed optical system with a closed cranial window setup for longitudinal studies in Thy1-jRGECO1a mice where neuronal activity is measured with Ca²⁺-dependent red fluorescence, glucose uptake with bolus injections of 2NBDG with green fluorescence, and cerebral blood volume (CBV) with near-infrared spectroscopy (NIRS). Genetically encoded calcium indicators (GECIs) provide strong fluorescent signals for assessing Ca²⁺-dependent neuronal activity. Thy1-jRGECO1a, a novel GECI with red fluorescence emission that penetrates deeper into tissue, allows for simultaneous imaging of metabolic activity using a green-fluorescent glucose analog, 2-(N-(7-nitrobenz-2-oxa-1,3-diazol-4-yl)amino)-2-deoxyglucose (2NBDG), which is taken up like glucose and then phosphorylated. Dual-fluorescent (red, green) and NIRS recordings confirm strong neurovascular coupling during hindpaw stimuli (Ca²⁺-CBV; P = 0.0033, r² = 0.91), whereas neurometabolic coupling (Ca²⁺-2NBDG; P < 0.001) was three times stronger during stimulation (r² = 0.75; slope = 0.6) compared to rest (r² = 0.49; slope = 0.23). In summary, multiplexed optical imaging can be used to reveal mechanisms of neurovascular and neurometabolic (un)couplings during ischemia, traumatic brain injury, aging, and Alzheimer’s disease.

Introduction

Optical imaging has enabled unique ways to explore brain activity in rodents (Lichtman and Denk 2011; Sepehri Rad et al. 2017), primarily using a relatively small cranial window. Recent advent of mesoscale optical imaging of the full dorsal cortex has revealed novel brain-wide dynamic interactions (Lin and Schnitzer 2016; Cardin et al. 2020). Mesoscale imaging with the newer, brighter fluorescent reporters enables scrutiny of both local and global neuronal interactions. Furthermore, new genetic engineering techniques like clustered regularly interspaced short palindromic repeats (CRISPR) provide novel options for fluorescent targeted cell lines in new rodent strains, albeit predominantly in mice. The use of fluorescent calcium (Ca²⁺) indicators has become a cornerstone technique in neuroscience for studying dynamic intracellular Ca²⁺ changes associated with neuronal activity. These indicators are designed to respond to the influx of Ca²⁺ by increasing their fluorescence intensity, thus providing a visual and quantifiable means of assessing cellular activity. Genetically encoded calcium indicators (GECIs) enable measurement of functional activity, since the brightness of the fluorescent signal increases manyfold as Ca²⁺ enters the cell during action potential generation and propagation (Akerboom et al. 2013). While most studies observe action potential generation and propagation of neuronal populations, aspects of astrocytic activity may also be measured (Sonoda et al. 2018; Lines et al. 2020; Lohr et al. 2021; Georgiou et al. 2022; Lines et al. 2022).

The conventional GECI is green fluorescent protein-based GCaMP6s or GCaMP6f under Thy1 neuronal promoter, which emits green light (Dana et al. 2014; Inoue 2021). However, a promising alternative is Thy1-jRGECO1a, which emits red light (Dana et al. 2018; Park et al. 2021). There are two advantages of red emitted light. First, it travels further in brain tissue due to reduced scattering, enabling the study of deeper cortical layers. Second, it frees up the green portion of the spectrum for other fluorescent agents of interest of which are available with green emission, to study a different modality alongside neuronal activity or to target an additional cell population.

Since the brain requires a continuous energy supply to sustain normal function with glucose as the major substrate, measurement of glucose consumption has always been of great interest for neuroscientific explorations (Sokoloff 1992). This is particularly true in disease states, where hypometabolism has been linked to various neurological disorders such as ischemia, stroke, traumatic brain injury, Alzheimer’s disease, and dementia (Zilberter and Zilberter 2017). The majority of energy expenditure supports demands of cell signaling (Yu et al. 2018; Yu et al. 2023), which includes restoring transmembrane ion gradients following neuronal firing and releasing/repackaging neurotransmitters, as well as “housekeeping” needs for biosynthesis (through the pentose-phosphate and hexosamine pathways) and energy storage (through the glycogen shunt) (Alkire 1998; Shulman et al. 2001; Hyder et al. 2013; Yellen 2018; Gonçalves et al. 2019). Without large glucose reserves, high energy demands are met by continuous glucose uptake from blood circulation. This process is maintained through highly regulated Ca²⁺-dependent signaling pathways between neurons, astrocytes, and the vasculature via neurovascular coupling, which relates changes in neuronal activity and cerebral blood flow (CBF) or volume (CBV) regulation. Similarly, the process relating neuronal activity and glucose metabolism—neurometabolic coupling—relates changes in neuronal activity to energy metabolism (Hyder et al. 2010; Shu et al. 2016).

Here we describe an optical imaging platform that allows parallel measurements of neuronal activity, CBV, and glucose uptake. Neuronal activity was measured with the red fluorescent Ca²⁺-signal in Thy1-jRGECO1a mice, whereas glucose uptake was assessed with the green fluorescent glucose analog 2-(N-(7-nitrobenz-2-oxa-1,3-diazol-4-yl)amino)-2-deoxyglucose (2NBDG), by using trace amounts of 2NBDG for repeated bolus injections to estimate glucose uptake multiple times during a single experimental session. In addition, using the upper part of the visible light spectrum, we used reflectometric near-infrared spectroscopy (NIRS) to extract CBV changes. Since this NIRS approach does not provide information about hemoglobin saturation and we did not measure CBF, we could not determine changes in oxidative metabolism. Regardless, we built a multiplexed optical system for simultaneous recordings of CBV (NIRS), neuronal activity (red fluorescence), and glucose uptake (green fluorescence) to determine neurovascular and neurometabolic couplings. Our final goal is to build a miniaturized optical system, which could be placed into a high-field small-bore magnetic resonance imaging (MRI) scanner so that functional MRI (fMRI) can be combined with optical imaging in mouse brain. We confirmed with this system the existence of strong neurovascular coupling during stimulation, and we observed stronger transcortical neurometabolic coupling during sensory stimulation compared to rest.

Materials and methods

Mice

Thy1-jRGECO1a, line GP8.20, indicator mice (n = 4, 2 male + 2 female) were obtained from Jackson Labs (JAX Stock No: 030525) and housed on a 12-h light/dark cycle. Food and water were available ad libitum. Mice were 4 months old, 20 to 28 g, at the time of imaging. jRGECO1a, a red fluorescent Ca²⁺ indicator, was developed in 2016 to improve compatibility with existing green fluorescent probes such as GCaMPs (Dana et al. 2016; Dana et al. 2018). The red-shifted excitation/emission spectra of jRGECO1a relative to GCaMP allowed us to almost simultaneously record glucose consumption (via 2NBDG; Yoshioka et al. 1996) and Ca²⁺ signaling by multiplexing (Figure S1). We implanted a closed cranial window and conducted the first experiment at least a week later. After each optical imaging experiment, each lasting 1.5 to 2 h, the animal recovered in a warm cage. The optical imaging experiment was repeated a maximum of 3 times per animal.

Closed cranial window

Surgical procedures and experiments were executed with approval from the Yale Institutional Animal Care and Use Committee and followed the NIH Guide for the Care and Use of Laboratory Animals. All mice were initially anesthetized with 3% isoflurane, followed by 1.5% during surgery. Additional analgesia administered consisted of meloxicam (2 mg/kg, Covetrus) for anti-inflammation, buprenorphine (0.1 mg/kg, Covetrus) for pain prevention, and bupivacaine (<8 mg/kg, Covetrus) below the scalp as a local anesthetic. A depilation cream (Nair) was applied to the scalp to remove fur and prevent incision contamination, and the skin was cleansed 3 times with iodine and ethanol to prevent postprocedure infection. The skull surface was cleared of tissue, and Neo-Predef (Zoetis, Parsipany, NJ, USA)—a veterinary antibiotic, analgesic, and anti-inflammatory powder—was applied along the edge of the incision. Bone above the parietal and frontal cortices was thinned until translucent using a dental drill (AB machinery) with 1.4 and 0.7 mm drill burrs. Once the cortical surface was visible, the thinned bone was cleaned using a brush, and a small amount (less than a drop) of cyanoacrylate glue (Loctite) was applied. After the glue cured, transparent dental cement C&B Metabond (Parkell Inc.) was applied, and a custom-printed head post was attached. The headposts included a resin frame with a waterjet-cut glass microscope slide in the center (James et al. 2023). The glass part covered the entire cortical surface.

Chronic jugular vein catheter

On experimental day 1, mice were anesthetized with isoflurane (3%) and a preemptive dose of buprenorphine (0.1 mg/kg) was administered. Surgical sites were clipped free of fur, and 3 cycles of a two-stage (Betadine followed by isopropyl alcohol) scrub was used for each. A transverse incision was made over the trachea and the right jugular vein isolated. A sterile silastic catheter (Renasil size 10) was inserted, secured with a nonabsorbable suture (4-0 silk, s-1174 with 3/8-inch C-1 cutting needle) and filled with a saline solution containing heparin (30 U/ml) and closed with a suture. The catheter was tunneled subcutaneously to the back of the head and externalized through a midline skin incision between the scapulae. The catheter was fixed in the subcutaneous pouch above the scapula for ease of access during experimentation, and Neo-Predef was applied to the wound. Meloxicam (5 mg/kg) was administered at the completion of surgery and once every day for 3 d thereafter. Mice #1 and #2 were catheterized through jugular veins, but for mice #3 and #4, we applied tail vein injections for faster procedure.

Hindpaw stimulation

A pair of thin copper electrodes were inserted between the second and third and fourth and fifth digits of the left or right hind-paw (alternating each day) and a stimulus train of 10 Hz, 1% duty cycle, 0.5 or 1 mA amplitude was delivered using a constant current stimulus isolator (A365, World Precision Instruments, Sarasota, FL, USA) for short (30s) and long (455 s) stimulations. Figure S2 shows details of “resting” and “stimulation” protocols.

Optical imaging platform

2NBDG fluorescence (green), jRGECO1a Ca²⁺ fluorescence (red), and hemoglobin recordings (NIRS) were made on a custom-built optical imaging platform utilizing 4 separate light channels delivered by optical fibers (Fig. 1) two light-emitting diodes (LEDS) at λ_Ex1 and λ_Ex2 respectively: λ_Ex1 = <490 nm for 2NBDG (Thorlabs M470F3 LED, Newton, NJ, USA), λ_Ex2 = 560 nm for jRGECO1a (Fig. S1; Thorlabs M565F3 LED), and two lasers at λ_Ref1 and λ_Ref2 respectively: λ_Refl1 = 685 nm and λ_Refl2 = 808 nm for hemodynamic reflectance imaging (Thorlabs HL675MG and L808P010 laser diodes, with LDC201CU laser diode drivers). Since the laser light intensity is sensitive to the temperature of the diodes, the lasers were continuously on and two shutters (#87–208, C-Mount Motorized Shutter, Edmund Optics Inc., Barrington, NJ) in their respective light path controlled the chosen illuminations. λ_{Refl1 and} λ_Refl2 were chosen based on their differing absorbances of oxyhemoglobin (HbO) and deoxyhemoglobin (HbR), respectively: At 685 nm, HbR has 8-fold higher absorbance than HbO, whereas at 808 nm their absorbances are almost identical, ie isosbestic (Prahl 1999). This difference improves the accuracy of the calculation of their concentrations (Kocsis et al. 2006). λ_Ex1 was filtered using a bandpass filter (ET460/30x, Chroma Technology Corporation, Bellows Falls, VT) between 443 and 487 nm, and λ_Ex2 with a 560 ± 20 nm bandpass filter (Chroma ET560/40x). Emitted and reflected light was collected through a pair of achromatic doublet lenses (10 mm focal length, VIS-NIR coated, Edmund Optics, Inc, Barrington, NJ, USA), longpass filtered at 600 nm (Thorlabs FELH0600). In addition to obviating multi-bandpass emission filters or rotating filters, a 600 nm longpass filter confers multiple benefits in the current setup: (i) 2NBDG can be captured due to its wide emission band while blocking excitation light (Fig. S1), (ii) jRGECO1a emission can be recorded with reduced hemodynamic artifacts, while blocking excitation light, because hemoglobin absorbance drops after 590 nm (Fig. 1), and (iii) reflectance of both 685 nm and 808 nm laser light sources is captured (Svoboda and Block 1994; Dana et al. 2018). All channels were recorded at 640 × 480 pixels using a board camera (DMM 22BUC03-ML, The Imaging Source, Charlotte, NC, USA), with a 1/3 inch CMOS sensor (pixel size: 6 × 6 μm; color depth: 8-bit grayscale; MT9V024, ON Semiconductor, Phoenix, AZ, USA). This camera was selected because of its small size and possible adaptability to MRI, and its large pixel size (6 × 6 μm), producing relatively high sensitivity in a low-light environment. Its disadvantage is the 8-bit resolution and its readout noise pattern, which exists with every camera (see Figure S6 from Doran et al. 2024). The exposure time for each frame was 60 ms to collect sufficient light from the relatively weak emitted fluorescent light sources (Thy1-jRGECO1a and 2NBDG molecules) when long passed at 600 nm. One of the advantages of the long exposure time is that the laser speckle effect (Dunn et al. 2001) averages out, which lowers noise contribution on hemodynamic laser recordings. We minimized illumination crosstalk by adding a 20-ms delay between switching the LED light sources. The lasers were operated with a motorized shutter, with 40 ms between switching. Data was collected at 10 frames per second (2.5 fps for each channel) on a computer running MATLAB 2019a and multiplexed into a single multidimensional array. LEDs and shutters were switched in synchrony with the camera triggers by CED μ1401 and Spike2 software (Cambridge Electronic Design Ltd, Cambridge, UK). The time series from the cortex (in vivo) and from the reference area (outside the brain) show intensity variations, respectively (Fig. S3A). The definition of signal-to-noise ratio (SNR) depends on the nature of the experiment. In the case of a response to stimulation, the signal is the amplitude of the evoked response; the noise is the fluctuation amplitude during baseline. From another point of view, the fluctuation around the baseline contains physiological activity, physiological noise, and measurement noise. During resting-state activity in the case of an optical signal, the single signal amplitude is not relevant because it depends mainly on methodological aspects such as the illumination brightness of the optical recording, and it is dependent on the variation during the entire duration of the recording. Therefore, we defined the signal, as the fluctuation of the time series, characterized by the standard deviation (SD). The noise is similarly defined as the SD of the reflected signal measured outside of the living tissue (reference area), SNR = SD_signal/SD_noise. We estimated the SNR from different regions of the brain and from different resting-state recordings. The SNR for the green fluorescent signal was 5.22 ± 1.45, which includes all SNR variations from regions and recordings. Similarly, the SNRs for the red fluorescent, 685 nm, and 808 nm were 18.14 ± 6.384, 13.10 ± 3.12, 8.29 ± 0.50, respectively. The green fluorescent measurement without the 2NBDG injection shows the smallest SNR. The SNR of the 808 nm recording is the second smallest because the CMOS chip sensitivity decreases in this wavelength range. Spectral analysis confirms physiological signal changes, in the measured frequency range the power spectrum characteristics of the jRGECO Ca²⁺-signal was similar as mentioned in the literature under anesthesia (Vanni et al. 2017; Wang et al. 2024).

Fig. 1.

Multi-wavelength optical imaging setup. Schematic of the wide-field optical imaging setup with incident light sources. λ_Ex1 is produced by a blue LED shortpass filtered < 490 nm (shaded in light blue; first spectrum from top) and used for excitation of 2NBDG_. λ_Ex2 is produced by a broad-spectrum green/orange LED bandpass filtered at 560 ± 20 nm (shaded in orange; second spectrum from top) and used for excitation of jRGECO1a (spectra generated with FPBase; Lambert 2019). λ_Refl1 and λ_Refl2 are produced by 685 and 808 nm lasers, collected by a flexible fiber optic light guide and used for reflectance imaging of hemoglobin (see bottom two spectra). The four separate light channels were multiplexed, and the emitted or reflected light was collected using a custom-built objective, housing a pair of achromatic doublet lenses and a 600 nm longpass filter. Experimental animals were injected with 2NBDG via jugular or tail vein catheters. Hindpaw stimuli were delivered via microelectrodes. Parts of this figure were created with Biorender.com.

Experimental protocol for glucose uptake measurements

To obtain optical measures of glucose uptake, we used the green-fluorescent glucose analog 2-(N-(7-nitrobenz-2-oxa-1,3-diazol-4-yl)amino)-2-deoxyglucose (2NBDG; Thermo Fisher Scientific, Inc.)—excitation/emission maxima at ~ 465/540 nm. 2NBDG was selected to measure glucose uptake because of its wide emission band, extending past the 600 nm cut-off of our long pass filter (Fig. 1, Fig. S1). 2NBDG enters the cell via glucose transporters (GLUTs) and/or sodium-glucose linked transporters (SGLTs) (Hamilton et al. 2021). 2NBDG has a higher affinity to GLUTs than glucose itself. In the cell, it is phosphorylated by hexokinase in the cytosol into the fluorescent metabolite 2NBDG-6-phosphate (2NBDG-6P). The presence of the fluorescent 7-nitrobenzofuran moiety on the C2-position prevents further metabolism, thus trapping the phosphorylated fluorophore in the cell until it decomposes into a nonfluorescent derivative (Speizer et al. 1985; Yoshioka et al. 1996). Previous studies have used dosages ranging from 3 mM (0.75 mM in blood) to 29 mM (2.75 mM in blood) (Tsytsarev et al. 2012; Yao et al. 2013). For the current experiments, trace amounts of 2NBDG were used (10 μl bolus of 15 mM (0.075 to 0.15 mM in blood) to infuse for 5 s at 2 μl/s, maximum 6 boluses/experiment by UMP3 UltraMicroPump & Micro4 controller, World Precision Instruments) to avoid competition with endogenous glucose. Each bolus was delivered at 90 s after the recording started both in rest and during stimulation, 30 s later after the stimulation began (Fig. S2). The recording of the first 2NBDG bolus injection was excluded from analysis as the volume of saline in the dead space of the catheter could not be determined, leaving 5 trials for analysis. We determined the effect of photobleaching on 2NBDG in vitro (Fig. S3B) with two concentrations (15 and 1.5 mM).

Animal monitoring during imaging

During the imaging session, anesthesia was maintained with 0.25% to 0.7% isoflurane (in a 30:70 ratio of oxygen:air) combined with 0.25 mg/kg/hr dexmedetomidine delivered via subcutaneous catheter. The combination of low concentration isoflurane + dexmedetomidine has been found to be more effective at preserving spontaneous neuronal activity, neurovascular coupling, and functional connectivity than higher concentrations of isoflurane alone (Pawela et al. 2009; Grandjean et al. 2014; Magnuson et al. 2014; van Alst et al. 2019; Steiner et al. 2021). Consequently, the stimulation responses are relatively small and frequently spatially distributed on the cortex (Maandag et al. 2007). Respiration rate and rectal temperature were continually monitored and recorded (Spike2, Cambridge Electronic Design Limited). Body temperature was maintained between 36.5 and 37.5 °C with a negative feedback heating pad (ATC1000, World Precision Instruments).

Data preprocessing

First, 100 frames of “dark,” or background, current were recorded using the same acquisition settings as in vivo recordings but without illumination. The dark current captures both the unwanted ambient illumination and the camera sensor’s intrinsic noise pattern (Ma et al. 2016; Valley et al. 2020). The average “dark current” image was subtracted from every frame in the recording. Following dark current subtraction, we compensated for uneven illumination, using the bias field correction method, which did not change the ratio between signal and fluctuation. Then all frames were spatially downscaled by 2 × 2 pixels (Fig. 2).

Fig. 2.

Multi-wavelength data analysis for signal preprocessing. Prior to each imaging session, an overview of the mouse cortex was collected under green illumination (see asterisk ‘*’ symbol) to define a brain mask and a vessel mask, as well as a “dark” image without illumination. Intensity fluctuations from neutral gray paper placed outside the cerebral region served as a reference light for laser regression. Following dark image subtraction, uneven illuminations were corrected (which step is optional), then all channels were spatially down-sampled by 2 × 2 pixels. ∆F/F₀ for 2NBDG and jRGECO1a recordings was calculated as (F_t − F₀)/F₀, where F₀ is the baseline signal average prior to stimulation and 2NBDG injection. The tissue absorption coefficient (∆μ_a) was calculated for 685 and 808 nm using the modified Beer–Lambert law. Relative CBV was estimated based on total Hb and the blood volume fraction of the brain cortex. A hemodynamic correction factor (HCx) was calculated from relative hemoglobin changes to correct for the attenuation due to absorption by blood.

A vascular/anatomical map was collected using λ_Ex2 (560 nm) to provide a reference image for defining brain-, vessel-, artifact-, and laser intensity reference masks. Brain masks were drawn manually as the outline of the visible cortex. Artifact masks were drawn manually by identifying regions within the brain mask not suitable for analysis due to, eg excessive bone, air bubbles, and other cranial window abnormalities. Regions within the artifact mask were excluded from all further analysis.

Processing reflectometric hemoglobin signals to estimate CBV

Calculating hemoglobin changes serves two purposes: (i) to record CBV changes in response to stimulation (CBV ~ ∆HbO + ∆HbR), and (ii) to correct the 2NBDG and Ca²⁺ signals for hemodynamic crosstalk. Due to the overlap between the absorbance spectra of HbO and HbR, and the excitation and emission spectra of the respective fluorophores, changes in [HbO] and [HbR] will contaminate the fluorescent signals—most noticeably during stimulation when CBV increases. Using the calculated values ∆[HbO] and ∆[HbR] from two well-separated wavelengths—λ_Refl1 and λ_Refl2—we can infer the degree of attenuation taking place at the excitation and emission wavelengths of 2NBDG and jRGECO1a, respectively, to correct them.

First, we regressed out light intensity fluctuations from the raw λ_Refl1 and λ_Refl2 signals using measurements of reflectance from two neutral gray paper fiducials placed in the diagonal corners of the recording field of view. We calculated the first principal component from the paper fiducial reflected regions, which explained 85% to 90% of the variability in the signals. This fluctuation was regressed out from each pixel over time. This procedure corrected for any laser intensity fluctuations such as those from diode temperature changes. Next, we subtracted the dark image and spatially down sampled the data twofold, as described for the other channels (Fig. 2).

We then estimated the change in tissue absorption coefficient (∆μ_a) at λ_Refl1 and λ_Refl2:

(1)

where PL_λ is the wavelength-specific pathlength of light through rodent brain tissue, and I_t and I₀ are the intensity of signals at time = t and baseline (average of 2 to 55 s), respectively. The values of estimated pathlengths were obtained from the literature (Ma et al. 2016). Assuming light scattering by the brain remains constant over time and that the pathlength at a specific wavelength can be estimated accurately, the modified Beer–Lambert law can be applied to calculate the change in absorption coefficient over time by scaling the relative light intensity signal change by the pathlength (Ma et al. 2016).

Using the wavelength-specific molar extinction coefficients (ε) of HbO and HbR (Prahl 1999) and ∆μ_a,t at λ_Refl1 and λ_Refl2, we calculated the changes in [HbO] and [HbR] over time (Kocsis et al. 2006):

(2)

(3)

The concentration change in total hemoglobin, ∆[tHb], was calculated as ∆[HbO] + ∆[HbR]. This calculation shows the hemoglobin fluctuations in micromolar units relative to the zero initial value. The relative change in tHb was taken to approximate changes in CBV, which is the metric that was used in our analysis. Hemoglobin has concentration units (micromolar = μM), while CBV has volume units (milliliter = ml). To convert hemoglobin concentrations (μM) to volume, we assumed that (i) the hemoglobin concentration did not change during the experiment; (ii) the murine hemoglobin concentration in blood was 16.5 g/dl (Yang et al. 1995); (iii) the resting baseline oxygen saturation in the murine brain cortex was 71.5% (Lyons et al. 2016); and (iv) the blood volume fraction is 5% in the cortex. These assumptions gave an Hb concentration of 2558 μM in blood, but considering that we measured it in the cortical volume, it was reduced to 127.9 μM in the cortex. Therefore, the relative change of CBV was calculated as ΔCBV/CBV_bl = tHb/127.9.

Finally, we used the time-varying ∆[HbO] and ∆[HbR] concentrations, along with the known molar extinction coefficients of HbO and HbR and estimated pathlengths at the excitation and emission wavelengths of 2NBDG and jRGECO1a to compute two separate frame-by-frame hemodynamic correction factors, HCx, that were applied to our measured 2NBDG and jRGECO1a signals, respectively:

(4)

where ε_Ex and ε_Em are the molar extinction coefficients of HbO and HbR, for the excitation and emission wavelengths of the fluorophores, PL_Est is the estimated pathlength of the excitation and emission light after diffuse reflectance, scattering, and tissue absorption. Ma et al. (Ma et al. 2016) found that the optimal estimated pathlengths at the excitation and emission wavelengths of GCaMP6f were reduced 19% relative to those estimated using a Monte Carlo model, the same as we observed.

Processing fluorescent signals: green 2NBDG and red jRGECO1a

Following dark current subtraction and 2 × 2 pixel downsampling, the 2NBDG relative fluorescent intensity change (∆F/F₀) was calculated frame by frame with the baseline fluorescence calculated as the average intensity image from 2 to 55 s prior to stimulation and 2NBDG bolus. The measured green ∆F/F₀ was then adjusted using HCx as ∆F/F_0,corrected = ∆F/F_0,measured × HCx, yielding a hemodynamically corrected 2NBDG signal. Examples of hemodynamically corrected Ca²⁺ and 2NBDG signals are in Fig. S3C. Following dark current subtraction, 2 × 2 pixel downscaling, the jRGECO1a ∆F/F₀ was calculated frame by frame using the same baseline timing as for 2NBDG. The measured red ∆F/F₀ was then hemodynamically corrected as described above for 2NBDG.

Data analysis for glucose uptake with 2NBDG infusion

The glucose uptake analysis using 2NBDG as a proxy is complicated by the effects of CBF–related fluorescence increases and, as such, requires careful interpretation of available parameters. Our protocol for glucose uptake at rest was simple, as at the time of 2NBDG infusion, the brain was already at a steady “resting” state, whereas our protocol for glucose uptake during stimulation introduced the hindpaw stimuli 30 s before the 2NBDG infusion to allow a steady “stimulation” state to be reached (Fig. S2). The cerebral intravascular transport can be modeled with a lognormal distribution, where the descending part of the curve can be fitted very well with an exponential equation. Therefore, we determined five main parameters that can be extracted from 2NBDG dilution curves (detailed below) (Wietasch et al. 2000): peak fluorescence intensity, time to peak from bolus, exponential decay rates and half-lives, plateau fluorescence intensity, and area under the curve. Our dilution curve is not only the dispersion of the indicator in the blood but also its uptake by the tissue; therefore, we chose double exponential fits, to estimate both mechanisms. However, the bolus injection method allows only one glucose uptake estimation after every injection at steady state.

Estimation of neurovascular coupling upon stimulation

Since light anesthesia is prone to produce distributed and weak responses to stimulations (Maandag et al. 2007) relative to awake, we carefully analyzed the spatiotemporal responses to hindpaw stimulation. To account for the spontaneous, burst-like Ca²⁺ fluorescence activity, we applied the t-test to 1-s bins for the entire stimulation (vs. baseline) epochs (James et al. 2023). We generated t-statistic maps of cortical Ca²⁺ activation by calculating the t-statistic (P < 0.05) between each frame in the 30 s following the onset of stimulation (i.e. 30s × 2.5 fps = 75 frames), compared to the baseline average of 30 s before stimulation onset. All 75 single-frame t-maps were then averaged, and a Gaussian filter with a kernel size of 3 × 3 pixels was applied. We restricted the analysis to the primary somatosensory hindpaw area, where the individual pixels had P < 0.05 (t > 1.68 for 30-s stimulation). The same method was applied to generating CBV t-maps and time series. We further averaged the signals for every 6-s epochs to observe the evoked changes in relation to the spontaneous fluctuations. We chose 6-s epochs because the CBV-evoked response was 4 to 6 s delayed from the Ca²⁺-evoked response. We compared the extracted Ca²⁺ signal datapoints to the 6-s shifted CBV datapoints. Since the CBV response curves had higher fluctuations than Ca²⁺ responses (based on slight sampling volume differences of Ca²⁺ and CBV recordings, see Discussion), for the final analysis, we binned the Ca²⁺ datapoints and their respective CBV counterparts for parametrization of neurovascular coupling.

Cortical parcellation based on Ca²⁺ signal

We used the normalized cut method for spectral clustering-based image segmentation of the resting Ca²⁺ time courses via MATLAB’s built-in spectral cluster function (Shi and Malik 2000). To improve segmentation robustness, we first spatially down-sampled the 3D dataset (2 spatial dimensions and 1 temporal dimension) with 5 × 5 pixel averaging, yielding resting Ca²⁺ time series with a higher SNR. This reduced spatial resolution time course dataset was then masked to select only visible brain regions. For spectral clustering, graph edges were limited to neighboring and diagonal pixels such that segmented regions were spatially contiguous. Neighboring and diagonal pixel pairings were assigned weights based on the correlation between their respective resting Ca²⁺ time courses so that segmented clusters contained pixels with highly correlated time courses while pixels across the edge boundaries exhibited the lowest time course correlations. The algorithm was set up to cluster the brain into 200 parcels—100 per hemisphere. The resolution of the resulting parcellation map was upsampled to match the original image resolution to facilitate subsequent analysis.

We evaluated the trial-to-trial variability in 2NBDG patterns by comparing the two parcellated rest AUC maps from each animal on each day using the Structural Similarity Index (SSIM) function in MATLAB. The SSIM assesses image similarity based on the local means, standard deviations, and cross-covariance of the image luminance, contrast, and structure (Wang et al. 2004).

Estimation of neurometabolic couplings during rest and stimulation

While neuronal activity represented by the Ca²⁺ signal was continuous and dynamic because no agent was infused, the bolus injected 2NBDG allowed only an event-based analysis. We compared the AUC of the plateau phase of 2NBDG dilution curve to the average Ca²⁺ signal recorded during 2NBDG bolus. We analyzed two cortical modalities. Since Ca²⁺ activity drives 2NBDG uptake, we explored their relationship in different tiers. The Ca²⁺-map was divided into 8 regions according to the activity distribution: the highest tier contained Ca²⁺ activity above the 90th percentile of all pixels. Similarly, the following tiers were between the 80 and 90th percentile, then the 70 to 80th percentile, all the way down to the 20 to 30th percentile. The lowest tier below 20 percentile was excluded from the analysis. Based on this selection of regions, we calculated the data pairs as the regional average of Ca²⁺ activity and 2NBDG uptake. We used the average of 20 to 30 percentile tier as a cortical map baseline to calculate the relative changes through the cortex, ie along the selected tiers, to normalize both Ca²⁺ and glucose uptake datasets.

Statistical analysis

All statistical analyses were performed in MATLAB 2022, using built-in MATLAB functions.

Results

Wide-field imaging for neurovascular and neurometabolic couplings

We developed a novel combination of optical modalities to simultaneously measure in vivo glucose uptake, neuronal activity, and hemodynamics (Figs. 1, 2, S1, S2). While the excitation, emission, and reflectance spectra were spread over the visible spectrum into the near-infrared, we avoided spectral overlap. The spectral bands of 2NBDG and jRGECO1a display very limited overlap, enabling us to record glucose uptake and neuronal Ca²⁺ activity concurrently with little optical crosstalk. A 600 nm longpass emission filter permitted collection of emitted light from jRGECO1a and emitted light from 2NBDG, given its wide emission band and large signal change relative to jRGECO1a, while blocking excitation light for both fluorophores. It also enabled us to capture reflected light from 685 and 808 nm lasers to calculate concentrations of HbO and HbR, extract CBV, and correct the 2NBDG and Ca²⁺ fluorescence signals from the change of hemoglobin absorbance (Figs. 2 and S3). We found that the fluctuation of baseline Ca²⁺ fluorescent signals was around zero (ie 0.01 ± 0.012 based on group average and standard deviation of fluctuations) and increased during stimulation to 0.07 ± 0.03 (Fig. 3).

Fig. 3.

Dynamic neurovascular coupling during hind paw stimulation. A) Example of Ca²⁺ activity and CBV changes during a short hind paw stimulation trial with a duration of 120 s. Hind paw stimulation (10 Hz, 1 mA) was delivered continuously between 30 and 60 s (gray area). The t-map shows the location of increased Ca²⁺ activity during stimulation at the hind paw regions. The t-map is overlaid on the enhanced anatomical image. The Ca²⁺ and CBV timeseries are the average of the significant pixels (t > 1.684, P < 0.05) of the hindpaw region based on Ca²⁺  t-map. Moving average filter was applied to the time courses. B) Example of Ca²⁺ and CBV changes during a long hindpaw stimulation trial with a duration of 515 s. Hindpaw stimulation time between 60 and 515 s shown in gray. The timeseries shows the spontaneous fluctuations in signal intensity in addition to the stimulus-related elevation, indicating high baseline activity due to the light anesthesia. C) The first 30 s of short and long stimulation responses of Ca²⁺ and CBV signals were plotted against each other, as an average of 6-s epochs (black dots, n = 34 trials from 7 experiments). Because of the 6-s delay of CBV signal response to Ca²⁺ response, we compared the signal values with 6-s shift. Clustered Ca²⁺ data points by 0.015 units and their CBV counterparts were plotted as means ± SD as circles and crosses. Linear regression fitted (ΔCBV/CBV = 0.184 ΔCa/Ca + 0.005, as a red dashed line), with P = 0.0033 significance, and the coefficient of determination was r² = 0.91.

Hemodynamic signals were extracted from the reflectometric 685 and 808 nm signals using NIRS principles (Elwell et al. 1994; Kocsis et al. 2006). The relative CBV changes during hindpaw stimulation were comparable with the literature values (Jones et al. 2002; Ma et al. 2016) (ie 10 to 15 μM increase in oxyhemoglobin and 3 to 4 μM decrease in deoxyhemoglobin relative to the initial value). At baseline, the CBV signal fluctuates around zero (0.007 ± 0.013) and, upon stimulation, the amplitude reached 0.042 ± 0.02. We did not analyze oxyhemoglobin and deoxyhemoglobin signals separately. The CBV reflectometric and Ca²⁺ fluorescent signals allowed for reproducible temporal (Fig. 3) and spatial (Fig. S4) patterns during short and long stimuli.

Our experimental protocol to assess neurovascular coupling (CBV and Ca²⁺ signals) did not require any infusion of agents, but determination of neurometabolic coupling (2NBDG and Ca²⁺ signals) required 2NBDG infusion (Figs. S2, S5, and Fig. 4). Thus, in essence, the neurovascular coupling was determined from dynamically evolving signals upon stimulation (Figs. 3 and S4), and the neurometabolic coupling was estimated under near steady-state conditions of rest and stimulation (Figs. 4, S5, and S6).

Fig. 4.

Key features of 2NBDG dilution curves. A) Typical 2NBDG ∆F/F dilution curve averaged from the entire cortex demonstrates relevant parameters that can be obtained to assess different elements of glucose metabolism in vivo. The time to peak (s) and peak value extracted shortly after injection are strongly related to CBF as this determines the speed with which 2NBDG is delivered and distributed in the brain. The exponential decay rate constants describe how rapidly fluorescence intensity decreases following the initial peak after injection. The decay is related to both CBF and metabolism, as it is influenced by the peak amplitude, blood velocity, and transmembrane transport, as well as retention. The plateau is the ∆F/F after 2NBDG has reached steady state at the end of a trial. It can be separated from the mainly CBF-associated initial signal increase and describes tissue uptake, retention, and phosphorylation of 2NBDG. Area under the curve (AUC) is the sum of ∆F/F from 95 to 515 s. It represents the total amount of 2NBDG uptake, retention, and clearance. B) Representative AUC map during the rest trial was calculated from 95 to 515 s. The left and right images are the vessel and AUC maps. The regions with higher relative AUC are associated with vasculature, eg the superior sagittal sinus along the midline.

Data acquisition during 2NBDG infusion was designed to examine relations between Ca²⁺ activity and 2NBDG uptake in rest and stimulated conditions within the same animal without saturation of glucose transporters. We achieved this by administering serial injections of trace amounts of 2NBDG (10 μl of 15 mM solution i.v. at 2 μl/s), which increases the blood 2NBDG concentration by 0.075 mM after every injection, with a maximum of 0.45 mM after 6 injections if there was no systemic clearance (Fig. S2), a highly unlikely scenario based on the decay of the 2NBDG following the bolus injection (Figs. 4 and S5). To prime and maximize 2NBDG uptake during stimulation trials, hindpaw stimulation was initiated 30 s prior to injecting 2NBDG, lasting until the end of the trial at 515 s. The fluctuation of 2NBDG fluorescent signal during rest was comparable to fluctuations of the Ca²⁺ signal (0.02 ± 0.04), whereas the fluctuation of green fluorescence signal increased to 0.38 ± 0.14 after 2NBDG bolus injection, which included both tissue and blood 2NBDG. The plateau phase of the 2NBDG fluorescent signal was 0.04 ± 0.06 (Fig. 4).

Dynamic neurovascular coupling upon stimulation

Ca²⁺ activity changes during hindpaw stimulation showed localized signal increases (Figs. 3 and S4). Peak jRGECO1a fluorescence increase during the first 30 s of hindpaw stimulation was ∆F/F₀ = 0.07 ± 0.03 for both short and long stimuli (Fig. 3A and B), in agreement with magnitudes reported in the literature (Park et al. 2021) and comparable to green fluorescent hindpaw stimulations (Lake et al. 2020). The amplitude of Ca²⁺ signal change was likely influenced not only by the electrical stimulation parameters (1 mA, 10 Hz) but also by the applied anesthesia (Maandag et al. 2007; Sanganahalli et al. 2008). In our case, the dexmedetomidine anesthesia provided weak suppression of baseline neuronal activity (Sanganahalli et al. 2008; Pawela et al. 2009), which indicated a relatively small evoked response, matching prior studies that have explored relations between evoked and baseline activities (Maandag et al. 2007). Rise in Ca²⁺ activity was followed by CBV increases, which peaked 26.5 s ± 6.9 s after stimulation and reached 0.042 ± 0.02 compared to baseline. We selected the first 30 s of short and long stimuli (n = 34 from 7 experiments) and plotted the relative Ca²⁺ and CBV signals against each other. Since the temporally filtered CBV signal is delayed to the Ca²⁺ signal by ~6 s, we compared the signal value pairs of 6-s average with 6-s shift of CBV. The Ca²⁺ and CBV signals were related linearly (Fig. 3C), calculated with 0.015 unit-wide Ca²⁺ intervals to decrease the effect of spontaneous fluctuations in both Ca²⁺ and CBV signals. The linear regression (ΔCBV/CBV = 0.184 ΔCa²⁺/Ca²⁺ + 0.005) showed a strong correlation between Ca²⁺ and CBV signal amplitudes (r² = 0.91, P = 0.0033). The activation maps generated from the Ca²⁺ and CBV signals were reproducible (Fig. S4).

Quantification of 2NBDG dilution curve

Each trial yielded a 2NBDG dilution curve showing a prominent rise in signal intensity within 1 s of the injection, peaking around 9 s, followed by a gradual decrease until the end of the trial. While there is a continuous recirculation of 2NBDG, the concentration in the blood was continuously decreasing, and its contribution to the signal became minimal after 2 to 3 min. Based on the dilution curves, we identified multiple extractable parameters to describe 2NBDG kinetics (Figs. 4 and S6). Since 2NBDG was delivered via the jugular/tail veins, the peak value (0.38 ± 0.14) and time to peak (8.6 ± 1.0 s) primarily signify the rate of vascular delivery, ie CBF increases during stimulation, but not the actual parenchymal uptake of the tracer (Fig. S6A and B). We fitted a 2-phase exponential model to the dilution curves to separate CBF- vs. metabolism-related effects [Equation (5)]: The early part of the decay is predominantly clearance due to both CBF and glucose uptake, whereas the latter part leading to the plateau is attributed to 2NBDG uptake and phosphorylation (Sheth et al. 2009). The maps for the first (“fast”) decay rate constant (−0.016 ± 0.005 1/s) and related half-life (Fig. S6C and E, and see Equation (6)) spatially overlap with the vasculature (eg superior sagittal sinus), while maps for the second (“slow”) component of decay (−8.82 * 10⁻⁵ ± 6.19 * 10⁻⁵ 1/s) and related half-life [Fig. S6D and F, and see Equation (6) in Discussion] overlap with parenchyma. The very small value of “slow” exponential decay (equivalent to very long half-life) indicates that the plateau phase (Fig. 4) of the 2NBDG dilution curve is relatively stable.

It is important to mention that the 2NBDG uptake may vary due to the animal’s changes in physiology, and the brain state may vary trial-to-trial (Fig. S4) since the low-level anesthesia is close to the awake state (Musall et al. 2019). The real estimation of glucose uptake can be achieved by kinetic modeling, which has two technical challenges. First, the fluorescent properties of 2NBDG and derivatives are the same in the blood and in the tissues; therefore, the signal origin cannot be defined. Second, it requires the identification of the arterial input function, which can be measured from the supplying artery, either from the carotid (Fig. S5A) or from the local cortical supply (Fig. S5B). The measurement from the brain vessels, eg a branch of the middle cerebral artery as input and the neighboring cerebral vein as output, shows that the temporal dynamic is beyond our frame rate capability.

Neurometabolic coupling during rest and stimulation

The brain uses glucose as the predominant source of fuel (Siesjo 1978). Positron Emission Tomography (PET) studies show high glucose consumption in the resting human cerebral cortex (Mortensen et al. 2018). We examined the uptake uniformity of 2NBDG, a glucose analog, through the cortical surface (Fig. 5) at rest in relation to regional Ca²⁺ activity. We segmented the cortex into functional parcels (100 parcels on each hemisphere) using the resting-state Ca²⁺ fluctuations and examined their 2NBDG uptake. We eliminated the areas where failures of the preparation did not allow a clear view of the cortex. The Ca²⁺ signal clusters provided very consistent AUC patterns for 2NBDG through trials on the same day and other days. The average SSIM of AUC patterns for 2NBDG fluorescence was 0.72 ± 0.03 (Fig. S7), indicating reproducible resting 2NBDG patterns across trials. Closer examination showed, however, that the AUC of 2NBDG was consistently higher in large vessels; thus, we compared Ca²⁺ and 2NBDG signals only in parenchymal areas, excluding areas with visible vessels from the analysis.

Fig. 5.

Parcellation of cortical surface based on resting Ca²⁺ signals. We developed a hyperspectral correlation clustering algorithm for neighborhood clustering of Ca²⁺ signals with a predefined number of clusters. The parcellations of two trials (A and B) shown with light colored boundaries above the cortical vessels. The AUCs of averaged 2NBDG dilution curves are calculated for every parcel and presented for the two resting trials. Examples of averaged Ca²⁺ and 2NBDG time courses for high and low AUC values are shown on the right. The AUC values of clusters are divided by the AUC value of the whole cortex, helping to visualize the regional heterogeneity of glucose uptake. The AUC patterns of the two trials are highly similar, indicating high reproducibility between trials. For each bracket on the right, the top and bottom traces are Ca²⁺ and 2NBDG signals, respectively.

First, we separated the large vessels and parenchymal/tissue masks (Fig. 6A) using the anatomical image. Only the tissue mask was used in the analysis of the Ca²⁺ and 2NBDG maps (Fig. 6B). These maps were created from the hemodynamically corrected Ca²⁺ and 2NBDG recordings by calculating the AUC of the last 100 s epochs (415 to 515 s), more than 5 min after the injection of the 2NBDG bolus. The reason for selecting this epoch was 2-fold: (i) the selected time was during the plateau phase of the 2NBDG dilution curve, which has minimal hemodynamic influence, so the signal is shifted toward glucose uptake (Sheth et al. 2009), and (ii) the long epoch duration minimizes the temporal fluctuation effect while retaining regional heterogeneity of neuronal activity (Fig. S6). Based on the sorted Ca²⁺-signal percentiles, we selected 8 regions (Fig. 6C) and we calculated the regional average of Ca²⁺ and 2NBDG signals. Since the fluorescent imaging does not have an absolute reference point from experiment to experiment, we normalized both Ca²⁺and 2NBDG signals to the lowest Ca²⁺ activity region. These values represent the spatial heterogeneity of the mouse brain activity for Ca²⁺ and 2NBDG signals. We compared the spatial Ca²⁺ and 2NBDG changes at rest (n = 12 from 7 experiments) and during continuous hindpaw stimulation (n = 12 from 7 experiments) (Fig. 6D). The individual linear regressions of Ca²⁺ and 2NBDG signals were statistically significant during both rest (P < 0.001) and stimulation (P < 0.001) trials. The interaction term was also statistically significant (t = 10.91, P < 0.0001), indicating that the slopes are significantly different between rest and stimulation conditions. Specifically, the coefficients of determination (r²) were 0.49 for the resting state and 0.75 during stimulation. The slopes were 0.23 for rest and 0.60 for stimulation, while the intercepts were 0.07 and 0.02, respectively. This indicates that during stimulation, the glucose uptake is ~160% (or 2.6-fold) more responsive to changes in Ca²⁺ signaling compared to the resting state. This enhanced response was observed not only in the primary somatosensory region but across the entire cortex. These data match metabolic responses in the somatosensory cortex under light anesthesia (Maandag et al. 2007).

Fig. 6.

Static neurometabolic coupling during rest and hind paw stimulation. The spatial distribution of plateau phase measured 2NBDG uptake is compared to the equivalent temporally averaged neuronal activities (Ca²⁺ signal). The Ca²⁺ intensity−based ROIs (8 tiers altogether) defined the average neuronal and glucose uptake values for rest (n = 12 from 7 experiments) and hindpaw stimulation (n = 12 from 7 experiments) 2NBDG infusion trials. A) The intensity field−corrected enhanced vessel map used to create the tissue mask, without large vessel contribution in the analysis (white regions). B) Neuronal activity and glucose uptake maps are shown with 3 examples: rest and stimulation from mouse #3 and stimulation from mouse #4. C) The highest neuronal activity regions (last two tiers) are shown together. D) The average values of the upper 7 tiers were normalized to the lowest tier for neuronal activity and glucose uptake. These values represent the spatial heterogeneity of the mouse brain activity. We compared the spatial Ca²⁺ and glucose changes at rest (black open dots) and during hind paw stimulation (red filled dots). Linear regression was statistically significant in both cases (P < 0.001) with r² = 0.49 (dashed line) and 0.75 (solid line) for rest and stimulation, respectively. The slopes and intercepts are 0.23, 0.07, and 0.60, 0.02, respectively, indicating ~3 times larger glucose uptake during stimulation than in rest.

Discussion

Wide-field optical imaging setup

In the last decade, several multi-modal imaging systems have been developed for fluorescence imaging and hemodynamic/metabolic measurements. Recently, Wang et al. described a system with 3 wavelengths and 2 cameras for concurrent measurement of neuronal activity (with jRGECO1a), hemoglobin concentration, and mitochondrial oxidation with flavoprotein autofluorescence (Wang et al. 2024). The advantage of this system is the parallel usage of reflectometric and fluorescent signals with high-speed recordings. A single camera system with 2 fluorescence and 2 reflectance channels (like our design) was described using jRGECO1a for neuronal measurements and enhanced green fluorescent protein-based acetylcholine probe for cholinergic signal recording, where 525 and 625 nm reflectometric signals were used to calculate hemoglobin concentrations (Doran et al. 2024). While this system is like our design, it was not designed for miniaturization for use in the restricted space of MRI scanner. Integrating our system into a high-field MRI scanner further enhances its potential, enabling comprehensive insights into brain function by combining high-resolution anatomical data with dynamic functional and metabolic information (Lake et al. 2020).

Usually, fluorescent recordings illuminate the surface and collect images along the same light path, ie epifluorescence, establishing an even illumination (Park et al. 2011; Ma et al. 2016; Sanganahalli et al. 2016; Lake et al. 2020) by applying a dichroic beam splitter, which reflects excitation wavelengths while transmitting emitted wavelengths. Since we collected two fluorescent signals and two reflectometric signals, requiring both transmission and reflectance, we applied direct surface illumination. We used a longpass filter, which allowed the recording of all four target wavelengths (Figs. 1 and S1).

Using four different wavelengths means that there are sampling volume differences in the recordings. Generally, shorter wavelengths have more superficial penetration depths in brain tissue. Comparing the optical pathlengths, we concluded that the two fluorescent wavelengths carry the information roughly from the same sampling volume at ~0.3 mm depth from the cortex (Ma et al. 2016). The reflectometric signals, however, have information contribution from the whole cortical depth (~1 mm), but even the NIR signals are superficially weighted, as was demonstrated earlier (Eke et al. 1997). When we calculated the hemodynamic corrections (Figs 2 and S3C), we included wavelength-dependent optical pathlengths into the calculation to partially compensate for volume sampling differences [Equation (4)].

Data analysis pipeline for multiplexed optical data

The multiplexed raw data were separated to four channels according to their illumination wavelengths and preprocessed to remove noise contributions (Fig. 2). In the first part of preprocessing for each channel, we regressed out the reference light intensity fluctuations, removed the ambient light, corrected for the uneven illumination, and increased the SNR by averaging every 2 × 2 pixels into a lower resolution image. Uneven illumination correction is necessary only with the structural images to get a better description of vessels. The second part of the preprocessing included calculating hemoglobin concentration changes from the near-infrared channels (Kocsis et al. 2006). In the mammalian head, three major chromophores with time-varying concentrations absorb near-infrared light: HbO, HbR, and the cytochrome oxidase aa3 in the mitochondria (Elwell et al. 1994). Since the latter has an order of magnitude smaller concentration than hemoglobin, its effect can be neglected from the analysis (Quaresima et al. 1998). Using the modified Beer–Lambert law, which includes the scattering term of brain tissue, we calculated the [HbO] and [HbR] changes (Kocsis et al. 2006), which in turn allowed estimation of regional CBV changes and calculation of the hemodynamic correction factor for the fluorescent signals.

Hemodynamic correction is necessary for correct estimation of fluorescent signal changes, because the fluctuation of blood volume itself acts as an absorption filter for the emitted fluorescent signals. Different techniques have been developed to correct for the hemodynamic effect on the Ca²⁺ signal, such as ultraviolet light excitation without Ca²⁺ response (Lake et al. 2020), signal separation from the Ca²⁺ signal using a high number of signal components (Valley et al. 2020), and the calculation of hemodynamic correction factor based on the Beer–Lambert equation (Ma et al. 2016) as in this study. All techniques are imperfect, so it is important to emphasize that the hemodynamic correction offers only acceptable, not perfect, solutions.

Dynamic neurovascular coupling upon stimulation

Neurovascular coupling is influenced by many factors, such as brain baseline activity, anesthesia, stimulation frequency, cortical–subcortical localization, and baseline vasodilation/vasoconstriction balance (Sanganahalli et al. 2008; Herman et al. 2009; Sanganahalli et al. 2009; Herman et al. 2013). The baseline level of brain activity influences the elicited response the most: In deep anesthesia, the evoked response is larger and more localized; in anesthesia, the response is smaller and more delocalized (Maandag et al. 2007). Because of the many factors influencing amplitudes of neuronal and hemodynamic responses and consequently neurovascular coupling, comparison across studies is not straightforward. Park et al. (Park et al. 2021) found that the localized hemodynamic time-to-peak response was 5.58 ± 0.17 s with 1.61 ± 0.11 s latency and the Ca²⁺ signal time to peak response was 35 ± 2.1 ms for a 3 mA current single-stimulus. These differences in Ca²⁺ and CBV responses could be due to differences in stimulation paradigm, anesthesia (α-chloralose), temporal resolution (80 Hz with 10-ms exposure), or jRGECO1a expression by viral injection.

We selected very light anesthesia induced by dexmedetomidine (Pawela et al. 2009), balanced with low concentration (0.5% to 0.75%) of isoflurane to compensate for the vasoconstrictor effect of dexmedetomidine, which allows strong spontaneous resting activity, but weaker and more diffuse evoked responses to somatosensory stimuli. We selected this anesthetic method to be closer to the awake state of the brain; therefore, our observations can directly extrapolate to general wakefulness. The short (30 s) and the long (455 s) stimulations (Fig. 3A and B) elicited the Ca²⁺ signal as well as CBV responses from the hindpaw area. In our analysis, we focused on the contralateral hindpaw region (Figs. 3 and S3); however, smaller responses were also observed ipsilaterally.

Neurovascular coupling can be defined spatially or temporally. We chose the latter method, so we calculated the time series of Ca²⁺ and CBV responses. As expected, CBV responses were delayed by 4 to 6 s, so we compared Ca²⁺ response at t₀ s and CBV response at t₀ + 6 s (ie 6-s shift) (Fig. 3C) to analyze the neurovascular coupling. Because of light anesthesia, the large spontaneous regional activity interfered with the hindpaw stimulation responses, causing fluctuations in the signal (Fig. 3A and B). We compensated for this large fluctuation in two steps. First, we used a 6-s average for both modalities and compared the first 30 s duration of the response (altogether 34 data sets from 7 experiments). Second, the Ca²⁺ data points were clustered by 0.015 units of fractional changes and the mean and standard deviation of the clusters were calculated together with their CBV counterparts (Fig. 3C). The linear fit of these datapoints demonstrate a strong neurovascular coupling between Ca²⁺ and CBV, where the Ca²⁺ change relates to more than 90% of the CBV change. The ratio of ΔCBV/ΔCa²⁺ (neurovascular coupling) was 18.4%. However, some caution is needed before generalizing these findings. The ΔCBV was calculated from hemoglobin concentration changes, which are relative to the beginning of the measurement (Elwell et al. 1994). The cortical CBV was based on a blood volume fraction of 5% and a normal hematocrit level (Yang et al. 1995). Clearly, a change in these parameters would influence the ΔCBV amplitude.

Quantification of 2NBDG dilution curve

The direct measurement of 2NBDG uptake is challenging, since the bound 2NBDG6P and free intravascular and intracellular 2NBDG have the same fluorescence property. One way to confirm the uptake is the continuous accumulation with high vascular concentration of 2NBDG (Tsytsarev et al. 2012), which compromises cell metabolism. We demonstrated the 2NBDG uptake by separation of intravascular and intracellular compartments with post-experimental saline perfusion, which left the intracellular 2NBDG signal intact (Fig. S5C).

The measured fluorescence intensity of 2NBDG is expected to be the sum of a fast decay (vascular clearance) and a slow decay (intracellular decomposition of 2NBDG-6P) and the effect of photobleaching occurring simultaneously. We measured that the effect of photobleaching is negligible (−0.045%/min) and therefore excluded it from further analysis (Fig. S3B). The exponential decay rate of 2NBDG was determined by fitting a pixelwise two-phase exponential model to the dilution curve between the peak time and the end of the trial at 515 s, as given by

(5)

where coefficients β₁ and β₂ describe the fast and slow decay rate constants (in s⁻¹) of the dilution curve, and a and c are the starting values of ∆F/F₀. Each phase of the exponential decay has an associated half-life, determined by

(6)

The plateau fluorescence intensity is the average ∆F/F₀ in the last 10 s of the dilution curve when the 2NBDG has reached steady state. As it is separated from the fluorescence intensity increases of the early phase of the 2NBDG bolus, this late phase better represents the amount of brain glucose uptake and retention (as 2NBDG, 2NBDG-6P, or even stored in glycogen; Louzao et al. 2008; Zhu et al. 2020), although recirculating 2NBDG also contributes to the signal (Sheth et al. 2009).

The AUC of the dilution curve is the sum from 95 to 515 s. The AUC encompasses the entire dilution curve and thus represents the total amount of 2NBDG supply, uptake, retention, and clearance. Due to the undesired confounding effects of hemodynamics on the other parameters, only the last 100-s epoch of the curve was selected for further analysis, where the hemodynamic influences were the smallest.

Intravenous injection of 2NBDG does not allow instantaneous glucose uptake measurement. Large bolus injection of high dose 2NBDG during epileptic seizure allows 2NBDG to accumulate for an extended period (Tsytsarev et al. 2012). However, with a large bolus, we expect 2NBDG to compete with glucose. Considering that 2NBDG phosphorylation inhibits glycolysis, a high concentration of 2NBDG could compromise our experimental goals. In contrast, small boluses using trace amounts of 2NBDG allow repeated measurements of glucose uptake in a single experimental session, making this approach more suitable for our study. The injection, through jugular or tail veins, arrives at the heart’s right chamber and ventricle (already diluted), then to the lung (further dilution) during ~ 30 to 40 heart cycles at 400 bpm, returns to the heart, and finally to the brain through the carotid arteries. We did not observe any differences in dilution curves either with jugular or tail vein injection, since both injection sites deliver the contrast agent to the right chamber via the vena cava superior or inferior. Thus, detectable amounts of 2NBDG arrive in the brain within 6 to 8 s after the beginning of the bolus (Fig. 4A; Fig. S6A and B).

The 2NBDG dilution curve is characterized by a fast increase and an exponential decrease, which can be parametrized by time-to-peak, plateau, and decay time constants (Fig. S6). A disadvantage of 2NBDG is that its fluorescence property prevents delineation of blood or tissue borne 2NBDG signals, even after phosphorylation to 2NBDG-6P. Additionally, our spatial resolution cannot separate the 2NBDG signal’s origin between tissue and capillary (Murata et al. 2014). Since the plateau phase at the end of the recording is more influenced by tissue glucose uptake than blood flow, we used the AUC of the last section of the recording to characterize the local glucose uptake (Fig. 4).

Other groups have attempted to circumvent the confound of large peak values and short delivery times in highly vascularized regions by delivery-correcting the measured signal (Frees et al. 2014). During a 75-min imaging session of 2NBDG in tumor tissue, Frees et al. quantify glucose uptake by dividing the peak 2NBDG fluorescence by the time to reach peak fluorescence and found increased uptake in metastatic vs. nonmetastatic tumors. In our analysis, correcting for the peak fluorescence and time-to-peak did not yield different results from standard AUC or plateau calculations. This may be due to the shorter imaging times used (8.5 min in ours vs. 75 min in theirs), the effectiveness of our vascular masking, or the specific tissues being studied (healthy brain vs. mammary tumors).

The plateau AUC maps showed large spatial heterogeneity (Fig. 5), which indicates that the accumulation of glucose is nonuniform across the cortex. We investigated whether this inhomogeneity is random or repeated across trials. Segmenting the brain with a functional parcellation scheme as opposed to atlas-based anatomical parcellation better accommodates individual variations in functional organization from subject to subject (Ren and Komiyama 2021), making it suitable for intra-animal comparisons as in the current study. Functional parcellation of resting-state Ca²⁺ activity was recently found to better represent individual differences in functional Ca²⁺ dynamics compared to static anatomical atlases (Barson et al. 2020). While the similarity between trials is high within a session, individual animals show very different patterns between experimental days (Fig. S7). This indicates that each animal’s neuronal activity pattern depends on daily activity, even if the same anesthesia is applied. While this segmentation to provide functional maps is promising, it requires more data than are available in this study for detailed analysis of Ca²⁺ and 2NBDG signals.

Neurometabolic coupling during rest and stimulation

PET techniques for measuring glucose and oxygen consumption are available for the whole human/animal brain (Hyder et al. 2016). Similarly, calibrated fMRI in animal/human brain allows the measurement of oxygen consumption (Hoge and Pike 2001) or direct glucose metabolism with ¹³C-MRS (Rothman et al. 1999). However, neither technique can directly measure neuronal activity. In this study, we describe a unique opportunity to measure hemodynamics, glucose uptake, and neuronal activity together in the animal brain (Fig. 6).

The fluorescent Ca²⁺ signal used here is a reporter of neuronal activity. The Ca²⁺ signal is an integral part of the electrical conductivity of neurons, but it is expressed only in those neurons that produce the jRGECO1a protein, namely, Thy-1 excitatory neurons (Dana et al. 2018). This includes pyramidal cells, the main excitatory cell type in layers III and V in the cortex (Dehghani et al. 2016).

The 2NBDG signal has vascular and tissue components, which include its presence in extracellular and intracellular compartments. Moreover, the phosphorylated form (2NBDG-6P) also contributes to the fluorescent signal. Since wide-field optical recording does not allow determination of the spatial origin of the fluorescent signal, this approach is qualitative because the absolute 2NBDG value was not measured in the brain. Using a tiered analysis of the Ca²⁺ activity to probe the 2NBDG signal, we were able to assess the neurometabolic couplings at rest and stimulated conditions. The neurometabolic coupling depends on the daily variation of neuronal and metabolic activities, but the neurometabolic coupling itself changes only if the two components change independently. Our method to calculate the neurometabolic coupling was based on the regional activity changes, and the regional activity was defined experiment to experiment, but in every timepoint measurement required a single bolus of 2NBDG. Therefore, we do not assume that the variability of 2NBDG pattern modifies the neurometabolic couplings, but we expect some variation between two different brain states. Our results show that glucose uptake is almost 3 times higher during stimulation than during rest (Fig. 6). This indicates a general increase of glucose uptake during stimulation, not only in the hindpaw area but also cortex-wide, beyond the hindpaw region. One explanation for this distributed 2NBDG signal is the high level of ongoing neuronal activity (Fig. S8) (Maandag et al. 2007), similar to the open/close eye human PET measurements that reveal brain-wide increase in glucose consumption, not only in the visual cortex (Riedl et al. 2014; Thompson et al. 2016).

Dexmedetomidine-induced systemic hyperglycemia can modify not only the neuronal activity but also the available glucose in the blood compared to the injected 2NBDG. Since we did not monitor the blood glucose concentration during the experiment, we cannot disprove the effect of hyperglycemia. We assumed that the same level of anesthesia always caused the same hyperglycemic effect, and therefore, the rest and stimulated neurometabolic couplings are influenced by this factor equally. We tried to mitigate this effect by using light anesthesia. The only method to keep the animal continuously euglycemic is to apply a glucose clamp, which involves continuous infusion of glucose and variable insulin infusion (Herzog et al. 2013). It also requires very frequent blood glucose measurements, which is very challenging with the mouse.

Future directions

In this study, we demonstrated that it is possible to simultaneously measure neurovascular and neurometabolic couplings. Our 2NBDG injection technique does not allow us to measure the instantaneous glucose uptake, and hence, we only described spatial neurometabolic coupling. With additional improvements of 2NBDG delivery, we could measure dynamic couplings. Another possibility is to get absolute 2NBDG uptake by measuring the inflow and outflow of the 2NBDG signal and apply kinetic modeling. However, this would require significant optical probe miniaturization around major blood vessels entering/exiting the brain. Additionally, since we primarily designed and built the system to be MRI compatible, we could implement the system inside a high-field MRI scanner, allowing us to use calibrated fMRI for oxygen consumption. Even without these advances, the current multiplexed optical imaging system can be used to reveal mechanisms of neurovascular and neurometabolic (un)coupling(s) during brain disorders and/or injury.

Author contributions

P.H., S.S., S.K.M., B.G.S., J.V.V., and FH designed research and concepts. P.H., S.S., A.A., and F.H. wrote and vetted the codes for analysis of the data. P.H., S.S., and F.H. analyzed the data. All authors wrote the manuscript. Peter Herman (Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Software, Visualization, Writing—original draft, Writing—review & editing), Simon Sanggaard (Conceptualization, Formal analysis, Investigation, Methodology, Software, Visualization, Writing—original draft, Writing—review & editing), Shaun D. James (Investigation, Writing—review & editing), Adil Akif (Software, Writing—review & editing), Sandeep Kumar Mishra (Conceptualization, Writing—review & editing), Basavaraju Sanganahalli (Conceptualization, Writing—review & editing), Justus Verhagen (Conceptualization, Methodology, Writing—review & editing), Hal Blumenfeld (Writing—review & editing), and Fahmeed Hyder (Conceptualization, Funding acquisition, Methodology, Resources, Software, Writing—review & editing).

Funding

This work was supported by National Institute of Health grants (grant numbers R01-MH067528, R01 NS100106, R01-AG084681, R21- AG085366).

Conflict of interest statement: F.H. is the founder of InnovaCyclics LLC. He has an equity interest in InnovaCyclics LLC, which is a biotechnology company creating next-generation MRI contrast agents. All other authors declare no conflict of interest.

Data availability

The preprocessed data (>100GB) underlying this article will be shared on reasonable request to the corresponding author.