2.1.

Animal Handling

Animal preparation and surgical procedures were performed under isoflurane anesthesia (5% induction, 2% to 3% maintenance, in $1\mathrm{L}/\min$ ${\mathrm{O}}_2$). Blood sampling was performed under 1.8% isoflurane anesthesia. All other experiments were performed under medetomidine sedation, which was induced by a subcutaneous bolus injection of $0.04\mathrm{mg}/\mathrm{kg}$ and followed by a continuous infusion of $0.05\mathrm{mg}/\mathrm{kg}/\mathrm{h}$. After medetomidine bolus, isoflurane was discontinued within 10 to 15 min and experiments were executed at least 40 min after bolus injection. Medetomidine sedation was supplemented by 0.4% isoflurane in eight pharmacological experiments. This attenuated the increase in respiration rate due to lactate infusion (Table S1 in the Supplemental Materials).

About 30 min prior to stereotaxic surgery for virus injection, fiber or electrode implantation, the animals received analgesia [Metacam ($1\mathrm{mg}/\mathrm{kg}$ s.c.) or Metamizol ($100\mathrm{mg}/\mathrm{kg}$ s.c.)].

16 animals were ventilated (MRI-1 Ventilator, CWE, Inc., Ardmore, Pennsylvania, United States). Three of them received the muscle relaxant Pancuronium ($2\mathrm{mg}/\mathrm{kg}$ bolus followed by continuous injection of $1.5\mathrm{mg}/\mathrm{kg}/\mathrm{h}$). The end-expiratory ${\mathrm{CO}}_2$ was continuously monitored in ventilated animals using a ${\mathrm{CO}}_2$ analyzer (Micro CapStar End-Tidal ${\mathrm{CO}}_2$ Analyzer, CWE, Inc., Ardmore, Pennsylvania, United States). Respiration rate of ventilated animals was set to 53 breaths per minute (bpm). The respiratory minute volume was between 95 and $127\mathrm{mL}/\min$. Respiration rate of spontaneously breathing animals was $(60\pm 15)\mathrm{bpm}$. Animals were placed on a temperature-controlled bed and fixated with bite and ear bars. Rectal temperature was held at $(36.6\pm 0.3)°\mathrm{C}$.

2.2.

Functional and Pharmacological Measurements

For functional measurements, electrical paw stimulation was used. For this purpose, electrodes were inserted into the forepaw contralateral to the implanted fiber/electrode. 1 ms pulses were applied at a frequency of 9 Hz and an amplitude of 1.5 mA according to a block paradigm (5-s stimulation, 25-s rest, and six repetitions). For LFP, recordings of one stimulation period were sufficient.

For pharmacological measurements, a 500 mM solution of l-lactate was injected via a lateral tail vein catheter. For phFRET, phMRI and MRS 2.5 or $2.6\mathrm{mmol}/\mathrm{kg}$ lactate were injected over 180 s (Table S1 in the Supplemental Materials). For blood analysis, a lower lactate concentration (250 mM) was used and $1.0\mathrm{mmol}/\mathrm{kg}$ lactate was injected in 150 s. This ensured that the maximum detectable lactate level of the detection device (Sec. 2.6) was not exceeded. As control, phFRET and phMRI measurements were performed by injecting $5\mathrm{mL}/\mathrm{kg}$ saline over 180 s.

2.3.

Fiber-Based FRET Fluorescence Recordings

2.4.

MR Measurements

All MR measurements were acquired on a 9.4 T Bruker Biospec 94/20 small animal scanner equipped with a $720\mathrm{mT}/\mathrm{m}$ gradient system (Bruker BioSpin GmbH, Ettlingen, Germany), a linearly polarized resonator, and a receive-only surface coil with 2 cm (MRI) or 1 cm (MRS) diameter. First, an anatomical image was acquired to identify the slice position for echo-planar imaging (EPI) or the voxel position for spectroscopy: 2D Turbo spin echo sequence (RARE), $\mathrm{TR}/{\mathrm{TE}}_{\mathrm{Eff}}=2000/50\mathrm{ms}$, RARE factor 8, Matrix $256\times 256$, field of view (FOV) $28\times 26{\mathrm{mm}}^2$, slice thickness 1.2 mm, nine contiguous slices. Subsequently, $B_0$ homogenization in the area of interest was achieved using the MAPSHIM Bruker routine.

2.5.

LFP Measurements

For LFP recordings, two custom-made silver electrodes ($250\text{-}\mu \mathrm{m}$ diameter) were implanted in the brain of 10 animals. One electrode was placed in the S1Fl and the reference electrode was positioned in the cerebellum. Electrodes were glued to the skull using dental cement. LFP signals were recorded as the difference between the electric currents detected by the electrodes using a differential amplifier (DPA-2FX, NPI Electronics, Tamm, Germany) with an implemented amplifier and bandpass filter (HP: 3 Hz and LP: 30 Hz). Signals were acquired at 2 kHz using the same data acquisition device and LabView script as for the fluorescence recordings.

2.6.

Blood Analysis

Blood lactate levels were obtained in three animals before and after infusion of the l-lactate solution under isoflurane anesthesia. $100\text{-}\mu \mathrm{L}$ blood was sampled from the lateral tail vein, using heparinized glass capillaries and analyzed using a blood analyzer (Vet Scan i-STAT1, Abaxis, Union City, California, United States). Under medetomidine sedation, blood sampling was omitted due to peripheral vasoconstriction.

2.7.

Data Preprocessing

Analysis was done using MATLAB (Release 2018b, The MathWorks, Inc., Natick, Massachusetts, United States) if not explicitly specified otherwise. EPI datasets of GE EPI and CBV measurements were realigned using SPM 12 (Statistical Parametric Mapping, Functional Imaging Laboratory, Wellcome Trust Centre for Neuroimaging, London, United Kingdom). The first five scans of each dataset were discarded to avoid presteady-state artifacts. In each dataset, a region of interest (ROI) was positioned in S1FL with the implanted fiber within the ROI.

For fMRI blood oxygenation level-dependent (BOLD) data a voxelwise Mann–Whitney $U$-test⁴⁹ was performed within the chosen ROI to find significantly activated voxels. The $U$-test examined whether the signals of rest and stimulation periods differed significantly after Bonferroni correction ($p<0.05$). To take the delayed onset of the hemodynamic response into account, the stimulation paradigm was shifted by 2 s. Signal changes were classified as positive if the stimulation signal was larger than the signal of the rest periods. The sum of voxels with positive activation was averaged across stimulation trials. The resulting time courses were normalized to the baseline and linear signal drifts were removed using the function “detrend.”

Time courses of phMRI measurements were determined by summing up the signals of all voxels within the ROI. The resulting time courses were normalized to the baseline and downsampled to a temporal resolution of 4 s. Subsequently, phCBV time courses were exposed to a further correction: Empirically, the MR signal decreased $\sim 40\%$ during contrast injection. However, the signal drop ($D_I$) from each animal deviated slightly from this value. This deviation was corrected by multiplying each CBV signal by the factor $0.4/D_I$.

Fluorescence recordings of donor and acceptor signals were normalized to their baseline and downsampled to the temporal resolution of the MR measurements (1 s for fMRI and 4 s for phMRI). FRET time courses with a negative signal drift and functional data with an oscillating baseline were excluded from analysis.

${}^1\mathrm{H}$ NMR spectra were analyzed using LCModel⁵⁰ and a dedicated simulated basis-set of metabolite spectra based on the density matrix formalism using a spin system simulation software implemented in MATLAB. Relative metabolite concentrations were derived using total Creatine ($[\mathrm{Cr}+\mathrm{PCr}]=8\mathrm{mmol}/\mathrm{L}$) as an internal reference of concentration. The macromolecule and lipid background signals were parameterized as proposed by Lopez-Kolkovsky et al.⁵¹ An additional constraint was set to ensure that the background remained stable throughout experiments: per animal only those spectra were examined, for which the total background (sum of all macromolecules and lipids) did not deviate by $>10\%$ from the background of the first spectrum in the time series. Spectra exhibiting Cramer–Rao lower bounds $>30\%$ for lactate were excluded from analysis.

Spectra were used to calculate the relative increase of cortical lactate level caused by i.v. lactate injection for each animal. Lactate concentrations measured before injection were averaged. A rise in lactate level can be expected during the lactate injection and in the first minutes after the injection.²⁰ It was assumed that the maximum lactate concentration of the three spectra recorded during and after injection corresponds to the rise of the lactate level caused by the injection. This maximum lactate concentration was determined for each animal and normalized to the preinjection concentration.

2.8.

Calculation of FRET-Signals Expected in Twitch-2B-Expressing Animals

Our method for estimating the expected FRET signal of Twitch-2B-expressing animals during sensory stimulation is based on two assumptions: first, the expected FRET signal corresponds to the convolution of the IRF of Twitch-2B and the neuronal response to the applied sensory stimulation. Second, the measured LFP response reflects the neuronal response. Accordingly, the convolution of the measured LFP response and the IRF of Twitch-2B was calculated to determine the expected fFRET signal of Twitch-2B-expressing animals during one cycle of the sensory stimulation [Fig. 1(e)].

For this purpose, the IRF of Twitch-2B was taken from the study of Thestrup et al.³⁹ with a temporal resolution of 0.5 s. This time resolution was chosen because it was close to the resolution of the fMRI data (1 s) and the time to peak (ttp) of the IRF of Twitch-2B could still be resolved. The same temporal resolution was used to extract the measured LFP response to sensory stimulation: first, LFP responses to sensory stimulation were averaged across 10 animals. Subsequently, the envelope of the averaged LFP response was obtained with the desired time resolution of 0.5 s. Finally, the convolution of the envelope of the LFP response and the IRF of Twitch-2B was calculated.

2.9.

Correction of the Hemodynamic Artifact

Ma et al.³⁸ have shown that hemodynamic artifacts in the measured fluorescence signal $F_m$ can be modeled by an absorption term as shown in Eq. (1). It consists of the true fluorescence signal ($F$), the change in the absorption coefficients of the excitation ($\mathrm{\Delta }{\mu }_{\mathrm{ex}}$) and the emitted light ($\mathrm{\Delta }{\mu }_{\mathrm{em}}$), as well as the light traveling path length $x$. For fiber-based recordings, all possible light paths can be combined since the light coming from all emitting cells is measured

We extended the absorption model by introducing a proportionality constant $a_1$ between changes in an absorption coefficient $\mathrm{\Delta }\mu$ and changes in the longitudinal relaxations rate $\mathrm{\Delta }{R}_2^{*}$

This correlation is justified, considering that the relaxation rate $R_2^{*}$ is also dependent on hemodynamics. For small changes in relaxivity, the relative MR signal $S_r$ depends on the echo time TE and the change in relaxation rate $\mathrm{\Delta }{R}_2^{*}$^52,53

Summarizing Eqs. (2) and (3) show that changes in an absorption coefficient $\mathrm{\Delta }\mu$ are proportional to the relative MR signal $S_r$. The absorption coefficient $\mathrm{\Delta }\mu$ can be described by a proportionality constant $a_2$, which depends on the echo time TE

Combining all fixed constants ($\sum x,a_{2,\mathrm{ex}}$, $a_{2,\mathrm{em}}$) of Eq. (1) into one absorption parameter $b$, yields the following equation:

In FRET measurements, this equation applies to the signals of donor and acceptor, each with its own proportionality constant ($b_d$, $b_a$).

2.9.1.

Fitting procedure

To determine the proportionality constants, the absorption term ${\mathrm{e}}^{-b·S_r(t)}$ was fitted to the measured fluorescence signals by adjusting the absorption parameter $b$. For fitting of fFRET time courses, the simultaneously recorded fMRI signal was inserted as MR signal $S_r$ [Fig. 2(a)]. For fitting of phFRET-signals, time courses of the averaged CBV signals were used [Fig. 2(b)].

Fig. 2

Principle of MRI-based correction method: to each measured fluorescence time course an absorption term was fitted via optimizing the absorption parameter $b$. For fitting fFRET data, the simultaneously measured fMRI signal was used. For phFRET recordings, the averaged CBV signal was used. Correction of the artifact was performed by dividing the fluorescence signal by the resulting absorption term. Correction procedure for (a) an exemplary fFRET fluorescence and (b) an exemplary phFRET time course.

The fit was optimized by minimizing the sum of squared residuals using the function “fmincon.” The optimization procedure was repeated for 100 starting points to find the global minimum. Starting points were generated with the function “multistart.” The fitting procedure showed the best results when omitting time points with the baseline signal. Therefore, for fFRET, only active time points were considered in the calculation of the least-squares deviation (stimulation start until 2 s after the end of stimulation). For phFRET, only time points during injections were used.

In the first step, the fluorescence signal of the acceptor fluorophore was fitted. Limits for the parameter $b_a$ were set to $[-\mathrm{10,10}]$. Afterward donor signals were fitted. Limits of the parameter $b_d$ were set according to the parameter $b_a$ (fFRET: $\mathrm{abs}(1.5\times b_a)\times [-\mathrm{1,1}]$; phFRET: $\mathrm{abs}(1.1\times b_a)\times [-\mathrm{1,1}]$). Limiting the parameter $b_d$ prevented parts of the actual fluorescence signal were removed by artifact correction.

2.10.

Data Representation

Throughout the text, results are stated as mean ± standard deviation, or minima and maxima are specified. Figures with time courses show averages and standard deviation. In the case of wide variability of time courses, individual signal courses are shown. If not specified otherwise, individual data clusters are displayed as scatterplots and boxplots. In boxplots the central mark represents the median, boxes include the 25th and 75th percentiles, whiskers extend to the last data points within the $1.5\times$ interquartile, and outliers are indicated as crosses.

Before testing for statistical differences, data were subjected to a Kolmogorov–Smirnov test for normal distribution. Since none of our data were normally distributed, a nonparametric test ($U$-test) was used for further statistical testing.

3.1.

fMRI-FRET

To assess whether fiber-based in vivo detection of calcium and lactate levels is feasible, we performed simultaneous fMRI-FRET measurements with rats expressing either Twitch-2B or Laconic as sensors, respectively. We performed functional measurements with electrical paw stimulation. fMRI activation maps and time courses ($n=16$) showed a distinct response to stimulation with a ttp of $(4.6\pm 1.0)\mathrm{s}$ and an onset around 1 s after the start of each stimulation train [Figs. 3(a) and 3(b)]. Fluorescence time courses ($n=10$) of Twitch-2B-expressing animals showed signal changes starting directly with stimulation onset [Fig. 3(c)]. Acceptor signals increased while donor signals decreased. After $\sim 1\mathrm{s}$, a signal decrease of both fluorophores was observed. FRET signals of uncorrected fluorescence data started to rise directly with stimulation onset and peaked at ttp of $(3.8\pm 0.9)\mathrm{s}$ [Fig. 3(b)]. For artifact correction, the absorption term ${\mathrm{e}}^{-b·S_r}$ was fitted to the fluorescence signals. For each measurement, the simultaneously recorded fMRI signal was inserted for $S_r$. Application of this correction removed major components of the signal decrease [Fig. 3(d)] and the remaining FRET signal had a ttp of $(2.7\pm 1.4)\mathrm{s}$ [Fig. 3(b)]. To compare these data with the theoretically expected signal in artifact-free fluorescence recordings of Twitch-2B-expressing animals, we calculated a reference by convolving the neuronal response with the known IRF of the Twitch-2B sensor. LFP measurements were used to identify the neuronal response to stimulation. The resulting time course showed a ttp of 2.0 s and started to rise directly with stimulation onset. Comparing this time course to our measured data, it became evident that prior to correction, the measured FRET signal deviated from the expected signal curve [Fig. 3(c)]. After applying the correction, FRET-signal time courses agreed well with the expected time course [Fig. 3(d)].

Fig. 3

Averaged time courses of fMRI-FRET with standard deviation, exemplary activation map, and ttp: (a) averaged fMRI signal and exemplary activation map. Time courses showed a clear response to each stimulation period (black bars). (b) Overlay of scatter plots and boxplots of ttp’s of fMRI and FRET signals. Note the temporal resolution of 1 s. (c) Time courses of the donor (blue), acceptor (green), and the FRET signal (gray) of Twitch-2B-expressing animals before and (d) after correction. The expected time course of the FRET signal (red) was normalized to the maxima of the actual FRET signal. (e) Fluorescence and FRET-signal time courses of Laconic-expressing animals before and (f) after correction of hemodynamic artifacts.

Fluorescence measurements ($n=6$) of donor and acceptor fluorophores in Laconic-expressing animals showed a signal decrease starting around 1 s after the start of each stimulation train, resembling the signal decrease observed for Twitch-2B-expressing animals [Fig. 3(e)]. The amplitude of signals from donor fluorophores was $(31\pm 11)\%$ larger compared with signals from acceptor fluorophores. Accordingly, a decrease was evident in FRET signals of fluorescence data. Ttp of FRET signals was $(4.2\pm 0.6)\mathrm{s}$ [Fig. 3(b)], similar to ttp observed in fMRI measurements. Application of the MR-based correction algorithm removed the signal drops in fluorescence time courses of Laconic expressing animals [Fig. 3(f)]. The averaged time course of corrected FRET signals showed no pronounced signal change triggered by stimulation. However, a $U$-test revealed significant differences between stimulation and the first 5 s of rest periods, for the averaged corrected FRET signals ($p=0.006$, Bonferroni corrected). This difference was not found for the uncorrected data ($p=0.090$, Bonferroni corrected).

3.2.

phMRI + phFRET

To better assess the sensitivity of FRET-based lactate detection, we performed phMRI and phFRET measurements during lactate injection in rats expressing Laconic or Twitch-2B. Signal changes in $n=11$ phEPI measurements upon lactate injection were low ($<1\%$) and showed large variability, and no clear signal trend was evident [Fig. 4(a)]. Therefore, we deemed the phEPI data unsuitable for MR-based correction of hemodynamic artifacts in fluorescence measurements. In contrast, phCBV measurements showed a substantial signal increase of around 6% and 4% for both lactate ($n=4$) and saline ($n=5$) injections, respectively [Fig. 4(b)]. Therefore, phCBV measurements were used for MR-based correction of the fluorescence recordings, and the average of the phCBV signals was inserted into the absorption term ${\mathrm{e}}^{-b·S_r}$ for the correction procedure.

Fig. 4

Averaged time courses of phMRI and phFRET measurements for lacate and saline injections. Individual signal time courses are shown for data sets with strongly varying time courses. (a) phEPI signals (red) showed large variations and no clear trend for injection of lactate (black bars). (b) phCBV signal increased during the injection of both lactate (light blue) and saline (dark blue). (c) Time courses of the donor (blue), acceptor (green), and the FRET signal (gray) in Twitch-2B-expressing animals during lactate injection prior to and (d) after artifact correction (right). (e) Fluorescence and FRET-signal time courses of Laconic-expressing animals during lactate injection before and (f) after correction.

Fluorescence time courses of $n=5$ Twitch-2B-expressing animals showed a signal increase during lactate injection [Fig. 4(c)], with roughly similar magnitudes for donor and acceptor. Only for two animals, the corresponding FRET signal showed an SNR higher than three [Fig. 5(a)] and the FRET-signal change was considered. After application of the correction, fluorescence signals and FRET signal remained constant during lactate injection [Figs. 4(d) and 5(a)]. In Twitch-2B-expressing animals, fluorescence signals also increased during saline injection ($\mathrm{n}=2$), which was completely removed by the correction (Fig. S3 in the Supplemental Materials).

Fig. 5

FRET-signal change and increase in lactate level caused by lactate injection. Data points of individual animals are connected by lines. Individual change of the FRET signal in (a) $n=5$ Twitch-2B-expressing and (b) $n=8$ Laconic-expressing animals upon lactate injection, before (left) and after (right) application of the correction algorithm. Only values with $\mathrm{SNR}\ge 3$ (black) were considered as FRET-signal change. Values with $\mathrm{SNR}<3$ (gray) were classified as no signal change. (c) Lactate level in blood plasma before (left) and after injection (right) of lactate ($1\mathrm{mmol}/\mathrm{kg}$) in $n=3$ animals. (d) Relative increase in cortical lactate level caused by lactate injection ($2.5\mathrm{mmol}/\mathrm{kg}$) as determined by MRS in $n=5$ animals.

In Laconic-expressing animals ($n=6$), prior to correction, both fluorescence signals and FRET-signal time courses showed an increase during lactate injection [Fig. 4(e)]. After correction, the acceptor signal remained stable, whereas the donor signal increased, resulting in a net increase of the FRET signal [Fig. 4(f)]. However, changes in FRET signals showed a large variability between individual Laconic-expressing animals [Fig. 5(b)]. Five-time courses showed an increase in the FRET signal ($\mathrm{SNR}\ge 3$) before applying the MR-based correction. The signal change was between 0.5% and 1.7%. This increase persisted after correction, albeit with slightly reduced values between 0.2% and 1.6%. One of three-time courses that showed no signal change before correction showed a signal increase of 0.6% after correction, whereas the FRET signals of the other two measurements remained unchanged.

3.3.

Blood Analysis and ${}^1\mathrm{H}$ MRS

To verify that lactate injection leads to a sufficient increase in lactate levels in blood and brain, blood analysis and ${}^1\mathrm{H}$ MRS experiments were performed, respectively. Blood analysis showed a doubling of the lactate blood plasma level from $(2.9\pm 0.5)\mathrm{mmol}/\mathrm{L}$ to $(6.1\pm 0.5)\mathrm{mmol}/\mathrm{L}$, caused by lactate injection [Fig. 5(c)]. One outlier was found in preinjection values of the ${}^1\mathrm{H}$ MRS data using the interquartile range method. Spectra of this animal were excluded from further analysis. Cortical lactate level of remaining animals ($n=5$) increased by $(64\pm 13)\%$ during i.v. lactate injection [Fig. 5(d)].

3.4.

Absorption Parameters

Absorption parameters $b_d$ and $b_a$ were calculated by fitting the absorption term $e^{-b·S_r}$ to the measured fluorescence separately for donor and acceptor signals, respectively. The calculated parameters ranged from 0.2 to 6.2 for fMRI-FRET measurements [Fig. 6(a)] and from $-1.3$ to $-0.1$ [Fig. 6(b)] for phCBV and phFRET measurements with lactate injection. Overall, the absorption parameters showed a large variability with standard deviations ranging from 0.5 to 0.8. Although not significant ($U$-test, $p>0.05$, Bonferroni corrected), absolute parameters for measurements with Twitch-2B-expressing animals tended to be larger than those from Laconic-expressing animals. We, therefore, conclude that the absorption strength also depends on the used FRET sensor.

Fig. 6

Absorption parameters $b_d$ and $b_a$ for signals of donor (blue) and acceptor (green) fluorophores for (a) fMRI-FRET and (b) phCBV and phFRET measurements during lactate injection. A $U$-test was used to test for significant differences.

4.1.

Performance of the MR-Based Correction Algorithm

In vivo calcium recordings using Twitch-2B verified that fiber-based FRET measurements are possible and showed that hemodynamic artifact correction is necessary. FRET-based calcium measurements are well suited to examine hemodynamic artifacts. Cortical calcium dynamics are well-understood and closely linked to neuronal activity.^25,26,54,55 Since the IRF of Twitch-2B is known,³⁹ we were able to calculate the expected calcium response to electrical paw stimulation using LFP recordings. Regarding phFRET recordings, we did not expect any calcium release-eliciting effect for saline or lactate injections. Calcium signals of uncorrected FRET data showed a significant deviation from the expected signals for both functional and pharmacological measurements. The high agreement of the corrected FRET signals with expected signals demonstrated the suitability of the presented MR-based correction method.

4.2.

Fiber-Based Lactate Recordings are Feasible In Vivo

We detected changes in lactate level with fiber-based FRET measurements using Laconic, for both sensory stimulation and lactate injections, when the correction algorithm was applied. Uncorrected fluorescence signals clearly matched the BOLD fMRI time courses, which was particularly evident from similar values of onset and ttp. We, therefore, conclude that mainly a hemodynamic artifact was detected. Using the correction algorithm these artifacts were successfully corrected, revealing a small but significant change in lactate signal upon electrical paw stimulation. Previously, lactate changes during electrical brain stimulation or brain needle prick were investigated using microdialysis. Caesar et al.¹² reported that a change in lactate level was detectable only after 30-s stimulation. In contrast, other studies showed that a shorter duration of stimulation was sufficient to cause a slow lactate rise lasting several minutes, and starting 5 s to 1 min after the stimulation onset.^11,13 We did not detect a further lactate increase during resting periods. Hu and Wilson¹¹ additionally reported an initial dip in the lactate level that started immediately with stimulation. This initial dip coincides with the lactate change detected in our study.

Furthermore, our fiber-based measurements demonstrated a lactate increase of up to 1.6% after i.v. injection. The rise in lactate level was verified by MRS and blood analysis. MRS measurements showed that the cortical lactate level increased around 60% to 70% during injection of lactate. Based on blood analysis, it was estimated that the blood lactate level was increased fivefold after injecting the lactate volume used for MRS ($2.5\mathrm{mmol}/\mathrm{kg}$). However, blood vessels represent only around 2% to 3% of the cortical tissue.^56,57 Accordingly, only 15% to 20% of the lactate increase detected by MRS can originate from the increase in blood lactate level in vessels. We, therefore, attribute part of the detected signal change during lactate injection to an increase in intra- and extracellular lactate levels. Histological validation found sensor expression exclusively in neurons, suggesting that signal changes represented an increase in neuronal lactate level. Mächler et al.²⁰ found an increase in neuronal lactate level upon i.v. lactate injection in mice using two-photon microscopy. According to that study, an increase in neuronal lactate level of around 4% can be expected for i.v. injection of $2.5\mathrm{mmol}/\mathrm{kg}$ lactate [Fig. 7(a)]. However, a substantially lower signal is expected for measurements using fiber-based detection since it is not possible to examine the signal from individual cells. Instead, a bulk signal is detected from all FRET sensor-expressing cells as well as background signal from the neuropil and auto-fluorescence from tissue.

Fig. 7

(a) Mächler et al.²⁰ showed the increase in FRET signal for Laconic-expressing mice (blue points, average ± standard deviation). According to a linear fit (dotted line), an increase in FRET signal of around 4% can be expected for injection of 2.5 mmol/kg lactate. (b) Absorption coefficients of Hb (black) and HbO (red) are wavelength-dependent.⁵⁸ For the blue and green light emitted by the donor (blue double-headed arrow and dashed lines) and acceptor (green double-headed arrow and dashed lines) fluorophores of Twitch-2B and Laconic, the absorption coefficient of HbO is greater than that of Hb. At the isosbestic points 479 and 526 nm (circles) absorption coefficients of Hb and HbO are identical. Each point lies within the emission spectra of one of the fluorophores^44,46 (blue and green background; arbitrary units). If narrow bandpass filters were used, centered at the isosbestic wavelengths, oxygenation change-related artifacts could be minimized. Full absorption and emission spectra of Twitch-2B and Laconic are given in Figs. S2(a) and S2(b) in the Supplemental Materials.

4.3.

Hemodynamic Artifacts in FRET Recordings

Hemodynamic changes lead to artifacts in fluorescence recordings since hemoglobin in blood absorbs light. Furthermore, hemodynamic changes modulate MR contrast. However, the generic term “hemodynamics changes” comprises various effects such as changes in blood oxygenation, changes in blood volume as well as blood dilution. Here, we observed that in functional recordings, the effect of oxygenation dominated hemodynamic artifacts (Sec. 4.3.1), whereas in pharmacological measurements oxygenation remained constant and the combination of dilution and volume change was prominent (Sec. 4.3.2).

4.4.

Artifact Correction Is Necessary

We found large variations in absorption parameters and signal amplitude of lactate between animals. These variations may originate from the fiber location relative to pial and penetrating vessels within the cortex^65,66: if the implanted fiber is close to the surface of the cortex or one of the diving vessels, the hemodynamic artifact will be larger than for a fiber remote from the closest blood vessel. Furthermore, our results show that the strength of the absorption also depends on the FRET sensors used. Therefore, artifact correction is required separately for each animal and no general correction function can be calculated.

Confocal fluorescence imaging often uses reflectance measurements to correct the fluorescence signal.^23,37,38 However, reflectance and fluorescence light need to have the same wavelength for effective correction. Therefore, this method is not applicable to fiber-based recordings since a bulk signal is measured, making it impossible to distinguish between reflectance and fluorescence light. A recent preprinted study presented a correction method for fiber-based recordings, where fluorescent signals from several fluorescence dyes were recorded simultaneously using fiber and a spectrometer.³³ However, fiber-based detection usually does not employ spectrometers, requiring substantial modifications to perform corrections based on reflectance measurements. The MR-based method presented in this study provides a feasible alternative, for combined fluorescence-MRI measurements.^25–32,34 It is also readily applicable with separate MRI measurements, as shown here for phCBV. Furthermore, our correction method can be applied to recordings of any fluorescence sensor. A potential limitation of our proposed method is the required downsampling of fluorescence data to the temporal resolution of MR scans, which may cause a loss of kinetic information. However, this was not the case for our study since the fluorescence signals had slow dynamics in all experiments. The half decay time of Twitch-2B is 2.8 s.⁶⁷ Further, the response of cerebral lactate to electrical stimulation is on the timescale of seconds,^14,15 and finally, a very slow but constant signal change is expected for phFRET upon a 3-min injection. For other fluorescence recordings with fast kinetics (e.g., calcium imaging with GCaMP6f⁶⁸), line scanning MRI could be used for correction, enabling temporal resolutions of 50 ms.^69,70

Interestingly, hemodynamic artifacts are often not reported in fiber-based studies, although their impact on fluorescence measurements has already been demonstrated in the 1990s.⁷¹ Their magnitude and importance may vary with the choice of detection wavelength and amplitude of fluorescence signal relative to the artifact. Our measurements clearly show that the absorption properties of blood lead to prominent hemodynamic artifacts in fiber-based FRET measurements and require proper correction.