All the experimental procedures were approved by the state authority (Regierungspräsidium, Tübingen, Baden-Württemberg, Germany) and conducted in accordance with the guidelines. Nine male C57BL/6 mice (Charles River Laboratory) were employed in this project and two were excluded due to the headpost cracking during training. Animals were habituated individually at a 12 h−12 h light–dark cycle (light on from 8 a.m. to 8 p.m.) with food and water ad libitum, starting at age 10–15 weeks.

A 3D printed headpost and customized training boxes (Schwarz et al., 2010) have been designed for acclimating the animals (Supplementary Figure 1). Briefly, intermittent habitual procedures were applied to train animals for 5–8 weeks before fMRI experiments. The training procedure includes: (1) mice were quietly settled in a comfortable, natural posture in the training box; (2) mice were immobilized with a headpost in the training box; (3) mice were exposed to the loud scanning noise in the scanning bore of a 14.1T scanner; (4) the slow fluctuation of pupil dynamics was recorded; and (5) training was continued until less/no urination or defecation (fecal boli <10) during imaging/training. Only mice that have good performance (mostly quiet, little urination or defecation, slow pupil dynamics) will continue with the next phase of imaging.

Mice were first handled in the experimenter's gloved hands for approximately 5 min/section until they exhibited grooming behaviors. Then, they were fixed through a headpost to the training box. The training period was increased from 10–15 min/day to 60 min periodically (1–2 days of rest after each training). The increment protocol in the training time is as follows: 10–15 min/session at the very beginning (1 week), and then 20–30 min for the next training session (2–3 weeks), 30–40 min (3–4 weeks), 40–50 min (4–5 weeks), 50–60 min (5–8 weeks). Mice body posture and movements were monitored as an indicator of training efficacy. A video camera with infra light was used to continuously monitor pupil dynamics. One week after surgery, mice were first acclimated in the training box without head fixation (~2 weeks) in the scanner, and then mice were head-fixed through headpost and kept inside the MR scanner while the scanner was running with designated sequences. Between each training section, mice rested for 1–2 days, during which the training process was maintained for at least 5 weeks prior to the actual MRI experiment (Supplementary Video 1). Animals received sucrose water several times between each training section (Supplementary Video 2). After finishing the last training section, the animals were imaged in the 14.1T MRI scanner for 24 h.

Mice were first anesthetized by isoflurane (2% for induction; 1–1.3% for surgery in a mixture of 30% oxygen in the air). The depth of anesthesia was checked by the lack of pedal withdrawal to a firmly pinched hind toe or foot, or the lack of head movement to a firmly pinched ear. The temperature was maintained at 37°C using a controlled heating pad. A small amount of ophthalmic ointment was placed in each eye to protect it from drying. Then, mice were secured to a stereotaxic frame by two earpieces and a bite bar. The skull was shaved and disinfected. After exposing the skull with a surgical knife, the soft tissue and periosteum were carefully removed and cleared with Kerr Gel Etchant (37.5% phosphoric acid, OptiBondTM FL) for 15 s. Then, the skull surface for implantation was prepared following the procedure as suggested by the manufacturer. We applied dental cement (Charisma flow) to fix the headpost above the skull. The remaining incision area was sutured and disinfected with antiseptics. Painkillers and antibiotics were administered to relieve pain and inflammation immediately after finishing the surgery. Animals were maintained on a heated blanket until fully recovered from the anesthetic effect (15–30 min) and were placed back in their original cages. Post-surgical observations were documented for a minimum of five consecutive days following the surgery (twice a day for 5 days and then daily). Wound recovery was monitored daily in the first week after surgery to ensure that there was no infection and no sutures off (Supplementary Video 3).

To investigate BOLD responses under visual stimulation in awake mice. A customized LED light Matrix System (P160104 LED matrix, 470 nm) for visual stimulation (Supplementary Video 4) was triggered by the echo-planar imaging (EPI) sequence and was controlled with the master 9 system (Master-9, A.M.P.I). A block design paradigm of 4s-on-16s-off in 10 epochs for each imaging session was conducted. The light brightness (~6 μW) and flickering frequency (~5 Hz) were also controlled within the system. To monitor the chest movement, a pressure sensor (Graseby Sensor, Medicare Health & Living Ltd) was attached under the animals' abdomen, from which the motion effect was detected by the BIOPAC acquisition system (Biopac Systems Inc., USA).

Customized and copper-sealed MRI-compatible cameras (RS-OV7949-1818, Conrad Electronic SE) were applied for monitoring head movements and acquiring pupillometry videos during imaging (32 bites/pixel, 60 frames/s, and resolution: 1920 × 1080). The camera was held on a home-built adjustable holder (Supplementary Figure 2). A movable infrared LED light (850 nm) was positioned beside the eyes of mice for pupil recording.

MRI images were acquired using a Bruker Avance III System (Bruker BioSpin, Ettlingen, Germany) with a 14.1-T superconducting magnet with a 12 cm diameter gradient providing 100 G/cm with a 150 μs rise time. A custom-made transceiver surface coil (8 mm in diameter) covering the whole brain of mice was applied for imaging. Functional scans were acquired using a 2D echo-planar imaging (EPI) sequence, with the following parameters: TR = 1 s, TE = 8 ms, FOV = 12 × 12 mm, 40 × 40 acquisition matrix, 24 slices, 0.3 × 0.3 × 0.5 mm³ resolution, 210 TRs total time. The anatomical RARE (rapid acquisition with relaxation enhancement sequence) images were acquired using the same geometry as functional scans with the following parameters: TR = 2.5 s, TE = 7 ms, FOV = 12 × 12 mm, 96 × 96 acquisition matrix, 24 slices, 0.125 mm² in-plane resolution, 0.5 mm slice thickness, 6 × RARE factor, 4 repetitions. The magnetic field homogeneity was corrected using the FASTMAP scout. A well-trained mouse was used to adjust the magnetic field homogeneity, which was also used as a reference for imaging other mice. The duration for multi-order shimming and positioning adjustment is ~17 min.

All fMRI data were analyzed using the AFNI software package [Analysis of Functional NeruoImage, https://afni.nimh.nih.gov/(R. W. Cox, 1996)). The preprocessing steps include (1) motion correction (3dvolreg), (2) co-registration to the anatomical images (align_epi_anat.py), (3) segmentation (3dSkullStrip, using a brain mask), (4) spatial smoothing (3dmerge, using a Gaussian kernel of FWHM = 0.2mm), (5) image normalization, and (6) voxel-wise linear regression analysis to calculate BOLD statistical maps (3dDeconvolve). This final step applied the hemodynamic response function (HRF) model: BLOCK (d, p), which is convolved with a square wave of duration “d” and made to have a peak amplitude of “p” (the amplitude of the basis function and usually set to 1).

The activated ROIs in functional maps were extracted and detrended for baseline shift. The baseline signal of EPI images was normalized to 100 for later multiple trial statistical analysis. To characterize the signal changes, custom-written MATLAB (MathWorks, Natick, USA) scripts were used to average the BOLD signal time series from each epoch and convert them to percentage change based on the signal intensity of 6 TRs prior to stimulation. The anatomical MRI images were registered to a template across animals for later image processing by applying the 3dAllineate function.

For pupil data preprocessing, all the pupillometry videos were cropped and processed using the DeepLabCut toolbox (Mathis et al., 2018). A mouse pupil network was built in this project to extract the pupil size from each video (Supplementary Video 4). The network was generated by using 5,670 frames from 27 videos (5 mice). Through this specially built network, four boundary dots with traced coordinates (x_i,y_i), i = 1, …, 4, were used to calculate the pupil diameter (22):

The resting-state pupil data were collected and applied for statistical analysis to monitor the physiological states. One-way ANOVA was applied to estimate the animals' pupil data in 1^st, 5th, and 8^th weeks of training.

The schematics of the fMRI-pupillometry platform and awake mouse training scheme are shown in Figure 1A. Different from the pervious “mock scanner”-based restraining system, the animals were exposed to the real scanner noise fixed in a home-designed restraining system (Supplementary Figure 1). There were no anesthetics applied during the entire training and imaging sections. After 5–8 weeks of training, mice were adapted to the restraining system and showed an inclination to keep an “imaging posture” even without head fixation compared with the animals trained in the short time period (Supplementary Videos 5–8). The chest movements of mice with longer training periods showed less motion, indicating a quantitative measurement of well-trained mice during fMRI (Supplementary Figure 3).

An infrared camera with copper shielding and a blue LED light matrix was implemented to record the pupil diameter and to deliver visual stimulation, respectively (Figure 1B, Supplementary Video 4, and Supplementary Figure 2). The flickering light stimulus (0–6 μW, 0–10 Hz) was delivered during imaging sessions. Whole-brain fMRI datasets were acquired in both evoked and resting states (Figure 1C).

Head motion assessment was calculated and compared after different periods of training sessions (1st, 5th, and 8th weeks). Based on the registration process of 3D fMRI datasets acquired at different training phases, six motion parameters (three translation displacements: anterior–posterior (AP), right–left (RL), inferior–superior (IS), and three rotational displacements of roll, pitch, and yaw) were processed (Figure 2). The rotation and movement time courses were gathered from six mice. The average time courses of translation and rotation displacements corresponding to the 1st, 5th and 8th weeks are shown in Figure 2A (A–P, 0.04 ± 0.03 vs. 0.01 ± 0.01 vs. 0.003 ± 0.002 mm; I–S, 0.04 ± 0.03 vs. 0.01 ± 0.01 vs. 0.01 ± 0.009 mm; R–L, 0.003 ± 0.002 vs. 0.002 ± 0.002 vs. 0.002 ± 0.001 mm; Pitch, 0.27 ± 0.22 vs. 0.11 ± 0.09 vs. 0.14 ± 0.08 degree; Yaw, 0.08 ± 0.05 vs. 0.05 ± 0.04 vs. 0.02 ± 0.01 degree; Roll, 0.05 ± 0.05 vs. 0.03 ± 0.02 vs. 0.03 ± 0.02 degree; mean ± SD). The variance of those motion parameters from each individual trial were extracted and compared (Figure 2B, one-way ANOVA, A–P, F (2, 47) = 3.26, P = 0.047; I–S, F (2, 47) = 6.94, P = 0.002; R–L, F (2, 47) = 2.36, P = 0.106; Pitch, F (2, 47) = 8.63, P = 0.0006; Yaw, F (2, 47) = 5.64, P = 0.006; Roll, F(2, 47) = 2.6, P = 0.085).The estimated framewise displacement during intermittent acclimation from different weeks after different weeks of training is shown in Supplementary Figure 4. Head motion was reduced along all planes following the training protocol.

Under isoluminance, the pupil dynamic changes were recorded at different phases of the training protocol when exposed to scanner noise during scanning. Figure 3 shows the statistical comparisons of raw pupil data in the 1st, 5th, and 8th weeks of training (3 mice, one-way ANOVA, F (2, 27) = 5.92, 1st vs. 8th = 0.0078, Bonferroni post-hoc test). These results indicate that the animal pupil dynamics dilated at the very beginning of training and subsequently reduced at later training time periods, indicating a more relaxed state of mice during fMRI. Meanwhile, the resting state pupil dynamics were plotted as a function of time, as well as the light exposure-induced pupil constriction (Supplementary Video 4), indicating the feasibility of awake mouse fMRI with real-time pupillometry.

To validate the awake mouse fMRI-pupillometry platform, BOLD fMRI signals were acquired under visual simulation. Three main brain regions along the visual pathway showed positive BOLD activation (p < 0.001): the visual cortex (VC), superior colliculus (SC), and dorsal lateral geniculate nucleus (LGd) (Figure 4A). Averaged time courses of BOLD signal changes of the three regions from different animals (VC, SC, and LGd) showed a maximum of 0.72, 2.5, and 0.5%, respectively (Figure 4B). The representative raw RARE and EPI images are shown in Supplementary Figure 5. Similar to previous fMRI studies(Niranjan et al., 2016; Desjardins et al., 2019; Tsurugizawa et al., 2020; Dinh et al., 2021), these responses were detected reproducibly across different mice.