Detection of laser-associated heating in the brain during simultaneous fMRI and optogenetic stimulation

Abstract

Purpose:

To calculate temperatures from T₂*-weighted images collected during optogenetic fMRI based on proton resonance frequency (PRF) shift thermometry, to monitor confounding heating effects and determine appropriate light parameters for optogenetic stimulation.

Methods:

fMRI is mainly based on long-echo time gradient-recalled echo acquisitions that are also suitable for measuring small temperature changes via the PRF shift. A motion- and respiration-robust processing pipeline was developed to calculate temperature changes based on the PRF shift directly from the T₂*-weighted images collected for fMRI with a 2-shot 2D GRE-EPI sequence at 9.4 T. Optogenetic fMRI protocols which differed in stimulation durations (3 s, 6 s and 9 s) within a total block duration of 30 s were applied in a squirrel monkey to validate the methods with blue and green light (20 Hz, 30 mW) delivery interleaved between periods. General linear modeling was performed on the resulting time series temperature maps to verify if light delivery with each protocol resulted in significant heating in the brain around the optical fiber.

Results:

The temperature standard deviation was 0.05 °C with the proposed imaging protocol and processing. Statistical analysis showed that the optogenetic stimulation protocol with a 3 s stimulation duration did not result in significant temperature rises. Significant temperature rises up to 0.13 °C (p < 0. 05) were observed with 6 s and 9 s stimulation durations for blue and green light.

Conclusion:

The proposed processing pipeline can be useful for the design of optogenetic stimulation protocols and for monitoring confounding heating effects.

Introduction

The recent development of optogenetic tools¹ has provided the ability to precisely stimulate, inhibit, or manipulate biochemical activity in specific cell types with light. The most widely used optogenetic tools are wavelength-specific opsins that are light-sensitive membrane-bound proteins. The opsins are activated by light (“opto”) and genetically encoded (“-genetics”). They respond only to light with a specific wavelength and enable precise activation or inhibition of genetically defined cell populations. Blood oxygenation level-dependent functional magnetic resonance imaging (BOLD fMRI) is a widely used neuroimaging technique to explore brain function noninvasively and has been applied to detect neuronal activity during optogenetic stimulation. Combining optogenetics and fMRI^2–4 (optogenetic fMRI) yields a powerful tool for neuron type-selective neuromodulation with simultaneous read-out of effects, both locally and over the whole brain. With the application of light, however, optogenetic tools bring the risk of local heat deposition in tissue, which may result in an increase of BOLD signal, i.e., a pseudo-BOLD response^4–6. Reported temperature changes in optogenetic-fMRI experiments range from 0.5–7 °C. Furthermore, many neuronal circuit processes are temperature-sensitive, which suggests that the light delivery may affect neural activity via tissue heating⁷. To avoid confounding effects induced by heating, light parameters during optogenetic stimulation need to be carefully considered.

MR thermometry⁸ is an imaging method that can noninvasively monitor temperatures based on temperature-sensitive MR parameters including proton density, T₁ and T₂ relaxation times, the diffusion coefficient, and the proton resonance frequency (PRF). The PRF-shift method of MR thermometry has been widely used to monitor temperature in vivo during thermal ablation and hyperthermia⁸. Maps of temperature changes can be measured using a gradient-recalled echo (GRE) sequence by measuring the phase difference between images before and during heating, which is proportional to the temperature-dependent PRF shift and the echo time (TE). Several PRF-shift temperature mapping techniques have been developed that are robust to motion and dynamic background magnetic field changes unrelated to heating for monitoring temperature during thermal therapies such as multi-baseline subtraction⁹, referenceless thermometry¹⁰, and a combination of the two¹¹.

Infrared cameras have been used to measure temperature changes near the optic fiber tip during light delivery for optogenetic stimulation⁴, but this requires extra equipment in the scanner if temperature is to be monitored simultaneously with fMRI, and it does not provide maps of temperature at depth. Fortunately, the most widely used sequence for BOLD fMRI is GRE echo planar imaging (EPI) with a long echo time (ideally TE~T₂*) which also makes it ideal for measuring temperature changes via the PRF shift⁸. Literature⁴ has shown that light delivery during optogenetic stimulation can cause temperature rises which suggests that PRF shifts may be observable using the GRE EPI sequences used in BOLD fMRI. At the same time, the temperature rises are likely to be an order of magnitude smaller than those encountered in thermal ablation and hyperthermia, and the relatively longer duration of fMRI scans as well as their longer TE compared to scans used to monitor ablations will make fMRI scans more susceptible to motion and background magnetic field changes due to sources such as respiration. In this work, we present a motion- and respiration-robust processing pipeline to calculate temperature changes with high precision based on the PRF shift from complex-valued T₂*-weighted GRE images collected in fMRI, and we perform statistical analysis on the resulting time series temperature maps to verify if light delivery with selected optogenetic stimulation parameters results in significant temperature rises during optogenetic fMRI. This method is valuable in three ways: (1) it provides a sensitive temperature monitoring tool that can be readily implemented to monitor confounding heating effects and to determine appropriate light parameters for optogenetic fMRI studies, without additional data collection; (2) it has high sensitivity and can detect small (<0.05 °C) temperature increases that are within the physiological range; (3) it could be applied to other neuromodulation studies in which heating is a potential confound, such as combined transcranial ultrasound stimulation and fMRI studies¹².

Methods

All procedures were conducted in accordance with NIH guidelines and with approval of the Institutional Animal Care and Use Committee (IACUC) at Vanderbilt University. The experiments were performed on two adult male squirrel monkeys with body weight in 700–812 grams range. Each animal was scanned 2–3 times.

Experiments

The squirrel monkeys were pre-anesthetized with ketamine and anesthesia was maintained with isoflurane (0.9–1.1%) delivered in a 70:30 N₂O/O₂ mixture during functional data acquisition. Figure 1a illustrates the experimental setup. Adeno-Associated Virus Types 5 & 9 (AAV5 & AAV9) with a commonly used opsin 9-CaMKIIa-ChR2 was transfected to neurons in the hand region of the secondary somatosensory cortex (S2) of the left hemisphere of the monkeys’ brains. ChR2 has a peak excitation wavelength of 470 nm. Subsequent neuron spiking recordings with intracranial optoelectrodes confirmed the successful expression of opsin at the targeted S2. An optical fiber (NA = 0.39, 200 μm, Thorlabs) was then inserted into S2 for the imaging studies.

Figure 1.

a: Schematic diagram of the simultaneous fMRI with optogenetic neuromodulation setup at 9.4 T. A laser fiber was implanted in S2 cortex where AAV5/9-CaMKIIa-ChR2 virus was transfected (left panel). Squirrel monkeys with optogenetically transfected cortical neurons were scanned simultaneously with alternating blue and green laser stimulation. The blue light activates neurons and evoked firing while the green light acted as a control and did not excite neurons (right panel). b: The stimulation block design with interleaved blue and green pulses with 10-ms pulse lengths and a 20 Hz pulse repetition frequency. The light stimulation was performed with the heating duration of 3 s, 6 s, and 9 s, respectively, within 30 second-long blocks.

fMRI and structural images were collected on a Varian DirectDriveTM/Agilent MR spectrometer horizontal 9.4-T magnet using a birdcage head coil, with an optical setup that delivered blue or green light during the functional scans. Functional images were acquired with a 2-shot T₂*-weighted 2D GRE-EPI sequence with TE/TR = 10 ms/1500 ms, in-plane resolution = 0.86 × 0.86 mm², slice thickness = 1 mm, slice gap = 0 mm, 24 slices and 290 volumes. The tip of the laser probe was placed in the center of one slice along the slice dimension. High-resolution T₂*-weighted structural images were acquired with in-plane resolution 0.107 × 0.107 mm² in the same sessions.

To characterize heating versus stimulation durations at a fixed laser power level of 30 mW, data were collected in one monkey with three block design stimulation paradigms with a total period of 30 seconds and laser stimulation (heating) durations within that period of 3 (10%), 6 (20%), and 9 (30%) seconds with interstimulus intervals of 27, 24, and 21 seconds, respectively. The light pulses comprised trains of 10 ms-long 30 mW subpulses with a pulse repetition frequency of 20 Hz (duty cycle of 20%), yielding total energy depositions of 18 mJ (3-second stimulation), 36 mJ (6-second stimulation), and 54 mJ (9-second stimulation) as Figure 1b illustrates. The blue light at 473 nm and green light at 561 nm were delivered in an interleaved pattern between periods with the same heating duration in each run. Each light was repeated seven times, resulting in a total of 14 laser stimulation periods. The total duration of each imaging run was 12.5 minutes. A realistic simulation model of laser-induced tissue heating in optogenetic stimulation based on Monte Carlo modeling of light transport coupled to a discretized Pennes bioheat equation¹³ was further used to predict temperature rises induced by light stimulation, where the modeling was matched to the MR imaging and stimulation parameters including the spatial resolution of 0.86 mm, the temporal resolution of 3 s, and the stimulation durations of 3 s, 6 s, and 9 s. To further demonstrate heating detection at lower laser powers and in an additional animal, data were collected in both monkeys with 2 mW and 20 mW subpulses (Monkey 1) and 2 mW and 16 mW subpulses (Monkey 2), with 6-second stimulation durations and otherwise the same parameters as above. Finally, to illustrate BOLD or pseudo-BOLD activation at low and high laser powers for each laser color, data were collected in Monkey 1 with 1 mW and 30 mW subpulses, with 6-second stimulation durations and otherwise the same laser parameters as above, but with a 4-shot GRE-EPI sequence with TE/TR = 10.8/750 ms, in-plane resolution = 0.43 × 0.43 mm², slice thickness = 1 mm, slice gap = 0 mm. These images were statistically analyzed using the general linear model (GLM) with FEAT, part of FSL¹⁴ (FMRIB’s Software Library), with bandpass filtering of 0.01–0.25 Hz. Activated voxels were defined as those showing stimulus-related signal changes at a statistically significant level of p < 0.05.

Temperature Map Calculation and Processing

The workflow to calculate temperature changes from T₂*-weighted GRE fMRI images is illustrated in Figure 2. The raw imaging data were exported from the scanner and reconstructed to complex-valued images offline in MATLAB 2021a (Mathworks, Natick, MA, USA). The hybrid referenceless and multi-baseline subtraction thermometry method¹¹ was used to calculate temperature change maps for each time point’s image based on the PRF shift, while removing dynamic background phase variations due to motion, respiration, and cardiac pulsation. The hybrid method calculates temperature at each voxel j according to the model:

$$y_j=\left(\sum _{b=1}^{N_b}{x}_{b,j}{w}_b\right)e^{i\left({\left\{Ac\right\}}_j+{\theta }_j\right)}+{ϵ}_j$$

where ${\left\{x_b\right\}}_{b=1}^{N_b}$ are baseline complex-valued images, the w_b are baseline image weights, A is a matrix of low-order polynomial functions and c is a polynomial coefficient vector. θ is a phase shift induced by temperature changes, and ϵ is complex Gaussian noise. The phase shift θ can be converted to the temperature ΔT in units of Δ°C using the following equation:

$$\Delta T=-\frac{\theta }{\gamma \alpha B_{0}TE}$$

where γ is the gyromagnetic radio, α = −0.01 ppm/°C is the PRF change coefficient, B₀ is the magnetic field strength and TE is the echo time of GRE sequence. Hybrid thermometry was applied with third order background polynomials (9 basis functions in A). The baseline image library consisted of ten images (N_b = 10) measured before stimulation started. The hybrid algorithm used 25 iterations to calculate temperature maps.

Figure 2.

Temperature Map Calculation and Processing Pipeline. Raw images were reconstructed to a complex-valued time series which was converted to temperature maps using the hybrid multi-baseline thermometry method. Then GLM was applied using each laser stimulation protocol to generate parametric temperature maps and maps of temperature change z-statistics.

Conventional baseline subtraction thermometry (which does not compensate motion or background phase variations) was also applied to calculate temperature maps for comparison¹⁰. This method calculated temperature maps directly from the phase difference of images with and without heating:

$$\Delta T_{\mathit{\text{ref}}}=-\frac{{\phi }_t-{\phi }_{\mathit{\text{baseline}}}}{\gamma \alpha B_{0}TE}$$

where φ_t and φ_baseline are phase maps of current time point t and the baseline respectively.

Once the temperature maps were calculated in the first monkey, they were statistically analyzed to generate z-statistic maps identifying voxels with significant temperature changes and parametric temperature maps using general linear modeling with FEAT¹⁴. The temperature changes were modeled as stimulus patterns without convolution with other functions¹⁵. A statistical threshold was applied to identify voxels that exhibited significant temperature changes (p < 0.05, false discovery rate-corrected). Signals were extracted from these voxels and averaged for time course display as described below.

Results

Figure 3a shows calculated temperature maps through time from left to right from Monkey 1, using conventional single-baseline subtraction and hybrid multi-baseline thermometry for a 9 second-stimulation dataset. The white boxes indicate the S2 region surrounding the tip of the optical fiber. Compared with single-baseline subtraction thermometry, hybrid multi-baseline thermometry reduced confounding phase variations caused mainly by respiration that otherwise obscured true temperature changes caused by the laser. Temperature maps over one period of the 9 second-stimulation dataset are shown in Figure 3b. The maximum temperature at the center of hotspots increased ~0.25 °C from 0 s to 9 s and then decreased when the light stimulation stopped. The measured temperature standard deviations across all time points (prior to averaging across heating repetitions) and averaged over a 7 × 7-voxel ROI at the tip of optical fiber (where consistent BOLD signal changes in response to blue light stimulation were detected) were measured to be 0.15 °C using baseline subtraction and 0.05 °C using the hybrid method. The average temperature changes across time and space in the same ROI were measured to be 0.15 °C using baseline subtraction and 0.009 °C using the hybrid method.

Figure 3.

a: Temperature maps at selected time points from a 9-second stimulation dataset, processed with baseline subtraction thermometry vs. hybrid multi-baseline thermometry. White boxes indicate the position of the laser probe. Compared to single-baseline thermometry, hybrid multi-baseline thermometry removed confounding erroneous temperature changes caused mainly by respiration, making the laser’s hotspot easier to observe. b: Temperature maps with hybrid thermometry over one period of the 9-second stimulation dataset. White boxes show the position of the hotspot. The temperature peaked at 9 s and then decreased after stimulation was switched off.

The top of Figure 4a plots simulated temperature changes versus time with the three light pulse durations for blue and green light. The maximum temperature rises (blue light/green light) were 0.051/0.057 °C (3-second stimulation), 0.096/0.103°C (6-second stimulation) and 0.129/0.137°C (9-second stimulation). The temperature rises induced by green light were slightly higher than those induced by blue light. A two-voxel ROI of volume 1.48 mm³ with significant temperature rises close to the probe was determined from the 9 second-stimulation dataset by thresholding (z > 2) its z-statistic map (described below and FDR-corrected) within the zoomed region in Figure 3b and selecting the two voxels that were above the threshold and closest to the probe’s tip. The repetition-averaged temperatures in this ROI are plotted for all three stimulation durations in the bottom row of Figure 4a. The gray patches indicate when the light stimulation was on. The maximum averaged temperatures (blue light/green light) were 0.060/0.080 °C (3-second stimulation), 0.102/0.065 °C (6-second stimulation) and 0.135/0.132 °C (9-second stimulation). The bottom of Figure 4b plots temperature changes around the probe in an ROI determined the same way but extracted from both monkeys, with 6-second stimulation duration and reduced power levels. In both monkeys, the temperature rises were very low with 2 mW power but were detectable (~0.08 °C maxima) at 20 mW (Monkey 1) and 16 mW (Monkey 2) power levels.

Figure 4.

a: Top: Temperature time courses predicted by simulation. Bottom: Averaged time courses in a two-voxel (1.48 mm³) ROI adjacent to the laser probe. The gray patches indicate when the laser stimulation was on. Temperature peaks during laser illumination are increasingly apparent with increasing stimulation duration. The shaded area represents the standard deviation of measured temperature increases across epochs in a single run. b: Averaged temperature time courses extracted from two animals at lower power levels: 2 mW and 20 mW for Monkey 1 (top) and 2 mW and 16 mW for Monkey 2 (bottom), for a 6-second stimulation duration.

Figure 5a shows temperature change z-statistic maps and parametric temperature change maps derived from the three runs with different stimulation durations, overlaid on structural images, for each laser color. The white arrows indicate the position of the laser probe. The parametric temperature change maps represent the average amount by which the temperature increased due to the laser in each voxel. Similar to the ROI-averaged time courses, no significant temperature changes (no voxels with z > 2) were observed for the 3-second duration protocol for both blue and green light, while the 6-second and 9-second protocols show more significant temperature statistics and higher temperature rises. Figure 5b further shows multi-slice BOLD/pseudo-BOLD responses with 6-second stimulation durations with powers of 1 mW and 30 mW. The yellow circles indicate the position of the optical probe. BOLD responses were observed with 1 mW blue pulses but not with 1 mW green pulses. At 30 mW, activations with both blue (BOLD) and green (pseudo-BOLD) light pulses were observed.

Figure 5.

a: Maps of temperature change z-statistics and quantitative parametric temperature change maps of single runs (each run contains seven laser stimulation on and off blocks) obtained from general linear modeling. The white arrows point to the positions of the laser probes. Detectable temperature rises were seen for 6-second and 9-second laser stimulation durations with both blue and green lasers. b: Multi-slice maps of BOLD (blue) and pseudo-BOLD (green) responses to laser stimulation with two laser colors and power levels (blue and green, 1 mW amd 30 mW). The yellow circles indicate the position of the optical probe.

Discussion

Optogenetic stimulation in combination with fMRI provides a powerful method for neuron type-selective neuromodulation with simultaneous readout of effects, both locally and over the whole brain. The heat resulting from light delivery during optogenetic fMRI may produce pseudo-BOLD responses or affect neuronal activity in the brain. The ability to monitor heating induced by light is therefore desired to help determine the parameters of light delivery for optogenetics and remove cofounding effects induced by heating. The results shown in Figure 5b emphasize this need for temperature measurement, since they show that BOLD or pseudo-BOLD responses can result from applying laser light to brain tissue even without the presence of corresponding opsins, and that this effect depends on the laser power level. Understanding the causes of these responses (heating versus possible hemodynamic effects²¹) requires the ability to measure temperature changes in situ during optogenetic-fMRI experiments. In this work, we demonstrated how T₂*-weighted GRE images collected in fMRI can be used to simultaneously image brain function while monitoring small temperature changes on the order of tenths of a degree Celsius caused by the stimulating light delivery, but a motion- and respiration-robust temperature map processing is required. The maximum temperature difference that can be resolved is 24 °C with the current TE setting (10 ms) at 9.4 T without phase wrapping. This approach has advantages over other temperature monitoring approaches such as infrared cameras because it does not require additional hardware but instead uses the intrinsic temperature sensitivity of GRE image phase. This also means that unlike infrared cameras, temperature changes can be imaged deep in brain tissue. The method can be easily implemented in animal and human fMRI studies and could be especially important for optogenetic studies of larger animals including nonhuman primates in which a relatively large volume of tissue needs to be stimulated. It could further be deployed in non-invasive focused ultrasound neuromodulation studies where unwanted confounding or unsafe tissue heating is also a concern.

The precision of PRF-shift thermometry is usually quantified as the standard deviation of temperature measurements; in this study, the measured temperature standard deviation across time and averaged over a spatial ROI at the tip of optical fiber was measured to be 0.05 °C. This is much smaller than typical ~1 °C standard deviations reported for temperature mapping during ablation¹⁶: for example, Weidensteiner et al¹⁷ measured a temporal temperature standard deviation of 2.3°C change in human liver at 1.5T, while Marx et al¹⁸ found a precision of 0.64 °C with single-echo GRE using a protocol for transcranial FUS (multi-echo approaches are typically used now and achieve better precision than this) at 3T and down to 0.34 °C with protocols described in that paper. The improved temperature precision in the present study resulted from a combination of the high temporal and spatial SNR at 9.4 T, the hybrid multibaseline and referenceless subtraction processing, and a 2–3x-longer TE than would be used to monitor ablation. More specifically, while temperature precision is theoretically maximized at TE ~ T₂*⁸ which is ~60 ms at 3T¹⁹, a much shorter TE~10 ms is typically used at 3T¹⁸ to monitor ablation, to avoid phase wraps at high temperatures and minimize scan time. Adjusting for the PRF shift’s dependence on field strength, at 9.4T this would correspond to a temperature imaging TE < 3 ms for ablation monitoring. In this study, we used a much longer TE of 10 ms to obtain BOLD contrast, which is more than 3x longer than the expected ablation monitoring TE and is closer to the theoretical optimal TE for temperature precision. The repetition-averaged temperature time courses in Figure 4 and the z-statistic map and parametric temperature change maps in Figure 5a further benefited from GLM processing which effectively averaged the temperature change signal across multiple heating repetitions. The standard deviation of temperature measurements²⁰ is inversely proportional to the imaging SNR which increases with voxel size, matrix size and the sampling time. In this study, we found that the signal amplitude in heated voxels decreased approximately 2% when lasers were switched on, which would not have a significant effect on temperature precision. There are some limitations when employing PRF shift thermometry to detect laser-associated heating, one of which is that the inserted optical fiber will cause local tissue damage and loss of signal around its tip. In this work the temperature changes directly at the tip of the optical fiber could not be reliably measured because of such signal loss.

Conclusion

In conclusion, the GRE-EPI data collected during optogenetic fMRI can be processed to calculate temperature maps with high precision, for use in monitoring confounding heating effects and designing optogenetic stimulation protocols.