Odor-evoked layer-specific fMRI activities in the awake mouse olfactory bulb

Abstract

Awake rodent fMRI is increasingly common over the use of anesthesia since it permits behavioral paradigms and does not confound normal brain function or neurovascular coupling. It is well established that adequate acclimation to the loud fMRI environment and head fixation reduces stress in the rodents and allows for whole brain imaging with little contamination from motion. However, it is unknown whether high-resolution fMRI with increased susceptibility to motion and lower sensitivity can measure small, but spatially discrete, activations in awake mice. To examine this, we used contrast-enhanced cerebral blood volume-weighted (CBVw) fMRI in the mouse olfactory bulb for its enhanced sensitivity and neural specificity. We determined that activation patterns in the glomerular layer to four different odors were spatially distinct and were consistent with previously established histological patterns. In addition, odor-evoked laminar activations were greatest in superficial layers that decreased with laminar depth, similar to previous observations. Interestingly, the fMRI response strengths in the granule cell layer were greater in awake mice than our previous anesthetized rat studies, suggesting that feedback neural activities were intact with wakefulness. We finally determined that fMRI signal changes to repeated odor exposure (i.e., olfactory adaptation) attenuated relatively more in the feedback granule cell layer compared to the input glomerular layer, which is consistent with prior observations. We, therefore, conclude that high-resolution CBVw fMRI can measure odor-specific activation patterns and distinguish changes in laminar activity of head and body restrained awake mice.

1. Introduction

Advancements in high-resolution functional magnetic resonance imaging (fMRI) has enabled the noninvasive investigation of laminar microcircuit processing in both humans (Huber et al., 2017, 2019; Raimondo et al., 2021) and animals (Chen et al., 2013; Cho et al., 2022; Huber et al., 2017; Poplawsky et al., 2019a; Yu et al., 2014). However, while human subjects are awake, animals are typically anesthetized to minimize motion and reliably detect these layer-specific fMRI signals, which precludes animal behavior and suppresses neuronal and vascular responses. Thus, awake animal imaging has recently emerged as a reliable imaging technique to eliminate anesthesia and enhance human translatability (Cover et al., 2021; Gao et al., 2017). Such studies acclimated animals to being head fixed in a loud fMRI environment for short periods of time (<2 h) to reduce motion and mitigate stress (King et al., 2005; Lindhardt et al., 2022). Awake animal fMRI improved the detectability of hemodynamic responses in subcortical and secondary (i.e., downstream) brain regions (Dinh et al., 2021; Liang et al., 2015 as well as measuring behavior-associated neuronal activity and spontaneous resting-state activity (Cover et al., 2021; Ma et al., 2022). Although there are technical challenges associated with awake fMRI, it has proven to be highly reproducible without the confounds associated with anesthesia (Becerra et al., 2011; Ferris, 2022). However, it is unknown whether awake fMRI can be applied to the study of laminar microcircuit processing since high-resolution studies are sensitive to motion and require more trial averaging to overcome the reduced MRI signals at lower volumes.

To examine this, we took advantage of two known features of bulb activation that can only be differentiated at high spatial resolutions using contrast-enhanced, cerebral blood volume-weighted (CBVw) fMRI (Poplawsky and Kim, 2014; Poplawsky et al., 2015): glomerular odor mapping and layer-dependent olfactory adaptation. First, we measured odor-specific activation pattern maps of four different odors with a 100 × 100 × 300 μm³ resolution in awake mice. The unique patterns originate from glomeruli, which are ~100-μm diameter spherical structures in the glomerular layer (GL) that receive discrete synaptic input from peripheral olfactory sensory neurons that express a single olfactory receptor type. Different odors bind to different ensembles of receptors and, with 1000 – 2000 olfactory receptor types in rodents, creates odor-specific patterns of glomerular activations. These whole bulb “odor maps” have been extensively reported with terminal imaging techniques, such as c-fos immunostaining and 2-deoxyglucose (2DG) autoradiography, as well as with fMRI in anesthetized rodents (Liu et al., 2004; Muir et al., 2019; Sanganahalli et al., 2020; Schafer et al., 2006; Xu et al., 2003). However, mapping these glomerular activation patterns in the whole bulb has not been performed in awake animals with high-resolution fMRI. Here, we will determine whether awake CBVw fMRI is a reliable and sensitive technique to capture odor-specific activation patterns in the rodent olfactory bulb with minimal motion contamination. Second, we examined layer-dependent olfactory adaptation to repeated exposures. Olfactory adaptation serves as an intrinsic memory that reduces the perception of background odors to favor more novel stimuli and is mediated by peripheral OSNs, the olfactory bulb, piriform cortex, and other olfactory cortical regions (Wilson et al., 2004). Specifically, electrophysiological (Bolding and Franks, 2018; Chaudhury et al., 2010; Wilson, 1998) and fMRI responses (Schafer et al., 2005; Zhao et al., 2016, 2017) to repeated odor exposure in anesthetized rats attenuated more completely in piriform cortex compared to the bulb. Consistently, since the olfactory cortex provides feedback to the granule cell layer (GCL) of the bulb (Boyd et al., 2012; Markopoulos et al., 2012; Oswald and Urban, 2012; Otazu Gonzalo et al., 2015; Poplawsky et al., 2015), fMRI responses in deeper layers also attenuated more than in superficial layers (Zhao et al., 2016). However, natural layer-dependent mechanisms of neural adaptation may be obstructed since anesthesia dampens cortical feedback activity to reduce granule cell activity (Kato Hiroyuki et al., 2012) and fMRI signals in GCL. Therefore, we will also examine the effects of olfactory adaptation in awake mice on laminar CBVw fMRI responses, especially between the sensory input layer, GL, and cortical feedback input layer, GCL (Fig. 1A).

Fig. 1. Experimental overview of awake CBVw fMRI odor experiment using a 9.4 T Bruker scanner.

A) Laminar olfactory bulb circuitry highlighting the predominant cell type of the associated layers: ONL – olfactory nerve layer, GL – glomerular layer, EPL – external plexiform layer, MCL – mitral cell layer, and GCL – granule cell layer. B) Acclimation protocol used to condition awake mice to head fixation in the MRI scanner. C) A modified air-dilution olfactometer selectively delivered up to four unique odorants to the mouse in the scanner through pneumatic control valves. Odorants were amyl acetate (AA; 50 mL odorized air + 950 mL blank air or 5% air-dilution), nonanal (Nona; 15% air-dilution), 2-hydroxyacetophenone (2HA; 10% air-dilution), and limonene (Lim; 20% air-dilution). Subpanel created at BioRender.com D) Example contrast-enhanced gradient-echo EPI image of the olfactory bulb acquired during experimental imaging. E) Example average time-series data from the glomerular layer of the bulb during awake CBVw fMRI scans.

2. Methods

2.1. Animal preparation and surgery

All procedures and surgeries followed the National Institutes of Health guide for the care and use of Laboratory animals (NIH Publications No. 8023, revised 1978), and were approved by the University of Pittsburgh’s Institutional Animal Care and Use Committee. Seven adult B6129SF1/J male mice were used (Jackson Laboratory, Bar Harbor, ME) and kept on a 12:12-h light-dark cycle. Acrylic headplates were implanted for head-fixation. Briefly, mice were induced with 5% isoflurane anesthesia and maintained at 1.5 – 2% during surgery. All fur from the top of the head was removed and the scalp was sterilized with topical betadine and alcohol wipes. The scalp was removed, and the underlying skull exposed. A lightweight acrylic headplate (12 × 23 mm² , 1-mm thick, McMaster-Carr, Elmhurst, IL) was horizontally fixed to the skull with acrylic dental cement and Vetbond tissue adhesive. Mice were provided antibiotics and analgesics for three days following surgery. Mice were single housed for the remainder of the study. The mice recovered for two weeks prior to handling and acclimation procedures.

2.2. Head-Fixation and MRI acclimation

To minimize motion during the fMRI scans (Ferenczi et al., 2016) and to reduce stress (King et al., 2005; Stenroos et al., 2018; Yoshida et al., 2016), mice were acclimated to head fixation, body restraint, and the loud MRI environment for four weeks prior to scanning (Fig. 1B). Acrylic tubes with 28.575- or 31.75-mm inner diameters were used for body restraint. Mice were acclimated every other day for three days a week; and began with 30-min head fixation and body restraint that increased by 15 min per session until a maximum of 120 min was reached (i.e., 7 sessions). For the 5th acclimation session (i.e., 90-min acclimation), acoustic recordings of our gradient-echo echo planar imaging (GE-EPI) fMRI sequence were introduced through mounted speakers (Pioneer, model number: TS-A1676R, Tokyo, Japan) during the final 30-min of acclimation at 90 dB. The length of time and volume of the fMRI noise was systematically increased over the next five sessions until it was played for the entire 120 min at 120 dB (Miyati et al., 2001). Mice continued the 120 min of acclimation at 120 dB for three days per week for the duration of the experiment. No odor was introduced during benchtop acclimation. After a total of 12 benchtop acclimation sessions, mice underwent two 2-h fMRI sessions in the real scanner with odor exposure prior to the actual CBVw fMRI data collection to account for environmental differences in sound and exposure between the benchtop and scanner. No rewards were provided during the acclimation process.

2.3. MRI compatible cradle

An MRI compatible cradle and olfactometer was constructed for this study (Fig. 1C, Suppl. Fig. 1). The cradle was designed using SketchUp (Trimble, Westminster, CO) and 3D-printed using PLA (χ- χ_H20/PPM = 0.528;) and PMMA (χ- χ_H20/PPM = 0.528; −0.023) plastics (Wapler et al., 2014). The cradle consisted of four parts: 1) an animal bed that interfaced with Bruker’s rat cradle; 2) integrated Teflon tubing that delivered odors from an external olfactometer to the mouse nose and a vacuum port for odor removal; 3) plastic headplate holders; and 4) locking clamps to head-fix the mouse to the cradle. All 3D-printed components can be downloaded at https://github.com/neuroimlabpitt/Olfactory-fMRI.

2.4. Olfactometer

Odor was delivered by a Knosys Olfactometer (Lutz, FL) that was modified for air-dilution odor delivery. Specifically, 100% medical air continuously flowed through the olfactometer at 1 L/min and, during odor exposure, a portion of that air flow was diverted to one of four pure odorants through Teflon tubing: amyl acetate (AA; 50 mL/min diverted or 5% air dilution), nonanal (Nona; 150 mL/min diverted or 15% air dilution), 2-hydroxyacetophenone (2HA; 100 mL/min diverted or 10% air dilution), and (+)-limonene (Lim; 200 mL/min diverted or 20% air dilution). Odor concentrations were selected in a preliminary study based on whether a single odor trial could reliably produce robust bulb activations (data not shown), in order to verify proper odor delivery during experiments. Check valves ~30 cm from the mouse nose allowed the Teflon tubing to be primed with odorant for fast odor switching. Odor delivery was TTL controlled by the fMRI sequence to precisely time-lock delivery to the fMRI TRs.

2.5. General MRI acquisition and procedures

MRI data were acquired using a 9.4 T/30-cm AVIII HD spectrometer (Bruker Biospin, Billerica, MA) with a 12-cm high-performance gradient set and ParaVision 6.0.1. An 86-mm quadrature volume coil was used for excitation, while a 10-mm surface coil centered over the olfactory bulb was used for signal reception. Prior to scans, the mouse tail vein was injected with monocrystalline iron oxide nanoparticles (MION; 25 mg/kg, Fereheme) under 1.5% isoflurane (<15 min duration) for contrast-enhanced CBVw fMRI (Poplawsky et al., 2019a). Mice were then head-fixed in the MRI compatible cradle and allowed to recover for 30 – 45 min before fMRI acquisition. A pressure pad was placed between the back of the mice and the body tube to record body motion during MRI. Since fMRI was performed in the olfactory bulb only, it was placed in the magnet isocenter and shimming was localized here to reduce field inhomogeneities from the nasal air sinus.

2.6. Anatomical imaging

We imaged the anterior commissure using a T2-weighted anatomical RARE sequence (2 s TR, 13.6 ms TE, 2 averages, RARE factor of 4, 11 × 11 mm² field-of-view (FOV), 128 × 128 matrix size, 21 slices) to standardize the fMRI slice alignment in the olfactory bulb across the fMRI imaging sessions. The center slice of the fMRI volume was then positioned 4.15 – 4.25 mm anteriorly into the bulb.

2.7. fMRI data acquisition

GE-EPI imaging parameters were as follows: 2 segments, 1 s TR (i.e., 2 s effective TR), 7.5 ms TE, 65° Ernst flip angle, 6.4 × 6.4 mm² FOV, 64 × 64 matrix (i.e., 100 × 100 μm² native in-plane resolution), 9 slices, 300-μm slice thickness, 178,571.4 kHz sampling bandwidth, no spatial saturation band, partial FT encoding to achieve an effective acceleration of 1.39, and the default navigator pulse with automatic ghost correction (Fig. 1D). An inter-volume delay of 806.48 ms captured the images within 193.52 ms of the TR to reduce motion artifacts. A single fMRI run consisted of a 120-s baseline (60 TRs), 64-s single odor exposure (32 TRs), and 120-s post-stimulus periods. A sequence of fMRI runs for the four odors (AA → Nona → 2HA → Lim; Fig. 1E) was repeated twice for each of two fMRI sequences in an interleaved order (GE-EPI → fast low-angle shot (FLASH) sequence), which resulted in a total of 16 fMRI runs for each scan session (Zong et al., 2014). However, the FLASH data were not examined in the current study, so only the 1st and 3rd odor exposures for each odor were recorded with GE-EPI.

2.8. fMRI data processing and analysis

All data were processed using the Analysis of Functional NeuroImages (AFNI) software (Cox, 1996; Cox and Hyde, 1997; Gold et al. 1998 and MATLAB (ver. 2019b, Natick, MA).

2.8.1. Individual analysis

All GE-EPI runs from an individual scan session were temporally concatenated and motion corrected to the first fMRI volume with 3dvolreg in AFNI. The resulting six motion parameter estimates (Δx, Δy, Δz, roll, pitch, yaw) were then used to calculate each volume’s Euclidean framewise displacement (Cover et al., 2021; Han et al., 2019; Power et al., 2014). Framewise displacement values exceeding 25 μm, or $\frac{1}{4}$ of a voxel, were censored from further analysis. β- and t-maps were calculated with 3dREMLfit in AFNI using a design matrix containing predictor variables for the four odors based on a previous CBVw impulse response function (Silva et al., 2007), as well as columns for the six motion parameters, linear drift, and constant (i.e., baseline) covariates.

2.8.2. Group analysis

Functional baseline images from all sessions were normalized to a single representative session using 3dAllineate with an lpa cost function and 3dQwarp; followed by linear interpolation to a higher in-plane resolution (128 × 128 matrix size; 50 × 50 μm²). These transformational values for each session were applied to the odor-evoked β- and t-maps, as well as the individual volumes in the fMRI time series. With all images in a common space, group z-maps with FDR correction were calculated with 3dMEMA and 3dFDR. To determine the spatial overlap of odor activations between the four odors (Fig. 3), we used a top percent threshold method where the 1%, 2.5%, 5%, and 10% most significant voxels within the bulb and a minimum voxel cluster of 10 were used.

2.8.3. Laminar mask categorization

We created a laminar mask that categorized each voxel in the bulb to one of the six bulb layers using the normalized and averaged group maps. Acquisition of high-resolution anatomical images with sufficient contrast to delineate the bulb layers, as we performed in anesthetized rats (Poplawsky et al., 2015), was not feasible with the 2-h limitation for awake scanning. Since the largest odor-evoked responses were previously observed in GL, we used the superficial activations from the four odor-evoked group 3dMEMA t-maps (see above) to manually delineate the outer GL boundaries. Using this reference line, the outer ONL boundaries were calculated with a 50-μm image dilation using imdilate and the outer EPL boundaries by a 200-μm image erosion using imerode in Matlab. Next, we used the deep regions of signal hyperintensities on the group CBVw baseline images, indicative of lower blood volumes, to delineate the outer MCL boundaries (Poplawsky and Kim, 2014; Poplawsky et al., 2019b). With this reference line, the outer GCL boundaries were calculated by a 50-μm image erosion. Finally, the bulb core was manually drawn as a 50-μm strip at the center of GCL. This process was repeated for both the left and right bulb hemispheres.

2.8.4. Layer-dependent analysis

We used normalized fMRI signal change (ΔS/S₀) maps from each of the fMRI sessions that were in the same image space as our laminar mask; and calculated the mean signal changes from all voxels in each layer (i.e., no threshold applied). The average laminar fMRI signal change was then calculated across imaging sessions. For relative laminar changes between odors, these values were further normalized by the mean signal change of all six regions-of-interest (ROIs) (Fig. 5A).

To ascertain the relative strengths of peripheral input to GL and cortical feedback to GCL (Boyd et al., 2012), we calculated the GL-to-GCL ratio. For the layer-dependent fMRI responses, the ratio of odor-evoked signal changes in each layer (i.e., GL/GCL) was calculated for each fMRI session before the mean ratio across sessions was determined. For relative laminar differences due to repeated odor exposures, GL/GCL was calculated for the 1st and 3rd odor exposures separately, followed by their ratio differences for each session (3rd – 1st exposure ratio), and the mean ratio difference across sessions.

2.8.5. Laminar time series analysis

We used normalized fMRI volumes for each TR and processed the time series data on a voxel-wise basis. Specifically, we linearly detrended the data and calculated the percent signal change ([S_t – S₀] / S₀; S_t is the fMRI signal in time, S₀ is the mean baseline fMRI signal from time 0 – 120 s and 280 – 304 s) for each odor run (i.e., 304 s total). One of two procedures was then followed: 1) for the mean laminar time series, we averaged the two repeated runs for each of the four odors (Fig. 5B) or 2) for the fMRI signal attenuation to repeated odor exposures, the 1st and 3rd exposures for each odor were lowpass filtered (2nd-order Butterworth filter with a normalized cutoff frequency of 0.3; butter in Matlab) before the difference was calculated (3rd – 1st) (Fig. 6A). For both procedures, we finally averaged across the fMRI sessions. Missing values in time due to motion censoring were ignored and the number of sessions for each time point used in the SEM calculation was adjusted accordingly. We further calculated the temporal signal-to-noise ratio (tSNR) from fMRI time series that were linearly detrended only by dividing the mean of the 0 – 120 s baseline period by its standard deviation. The tSNR was calculated for all voxels in the bulb and averaged for each individual odor run.

2.8.6. Flat map generation and center-of-mass calculation

Odor-evoked flat maps are commonly used to visualize three-dimensional bulb glomerular activity in two dimensions (Sanganahalli et al., 2020). Responses in GL from individual t-maps were extracted using the predefined laminar mask. We approximated the thickness of GL to be two native fMRI voxels or 200 μm (Franklin and Paxinos, 2007) (see, laminar mask categorization). Therefore, to linearize GL with this thickness, we averaged the t-values within a moving 250 × 250 μm² square to ensure all voxels in GL were included in the analysis while voxels outside of GL were excluded. This moving average began at the ventral-most point of GL and progressed circularly around GL in the lateral direction. The GL flattening process was repeated for each slice, bulb hemisphere, and fMRI session to create flat maps of GL (Fig. 4A). Flat maps for each odor were averaged across bulb hemispheres since left and right bulb flat maps were qualitatively similar (Suppl. Fig. 4).

To quantify the uniqueness of the spatial activation patterns, we calculated the weighted center-of-masses (CoMs). Only the top 5% of the flat map t-values were used for the CoM calculations. ROIs were drawn around the two distinct clusters of activations in the lateral and medial regions of the bulb for each odor (Fig. 4B), and the CoM for each ROI was calculated. To characterize intra-mouse reproducibility of GL activation patterns, we calculated the average Euclidean distance of CoM values between different scan sessions for each mouse. For mice that had more than two scan sessions, the centermost CoM value was used as the reference point.

2.9. Statistical analysis

Statistical analysis was run in GraphPad Prism Version 9.4.1 for univariate analysis and IBM SPSS Version 28.0.1.1 for multivariate analysis. All datasets were tested for normality using a Kolmogrov-Smirnov test to determine if parametric or non-parametric techniques were used. For multivariate parametric analyses, we used a one-way MANOVA with Sidak’s post hoc tests for multiple comparisons. For univariate parametric techniques we used a one- or two-way ANOVA with a Geisser-Greenhouse correction and Šidák’s or Dunnett’s post hoc tests for multiple comparisons. A paired t-test was used for simple pairwise comparisons. Data points greater than three standard deviations from the mean were considered outliers and removed from further analysis. For non-parametric techniques, we utilized a Kruskal-Wallis Test and Dunnett’s test for multiple comparisons, with a Wilcoxon Rank sum with corrected p-value for post hoc analysis. A Fisher transformation was utilized to calculate the t- and p-values from the Pearson’s correlation coefficient, r. All data are expressed as mean ± SEM, unless otherwise noted.

3. Results

3.1. Motion characterization of awake scans

Scanning awake mice increases motion contamination of the functional data (Fig. 2A; Suppl. Video 1 for example scan). We first characterized the motion parameters to evaluate the efficacy of our adapted acclimation approach and custom-made MRI compatible cradle for awake head-fixation. Overall, the motion characteristics of our awake mouse fMRI protocol was comparable to previous values reported in literature (Chen et al., 2020; Cover et al., 2021; Han et al., 2019; Tsurugizawa et al., 2020). Rigid body motion was determined using AFNI’s 3dvolreg’s motion parameter estimates. Due to the motion estimates having a gamma distribution, median and median absolute deviation (MAD) values from all runs were reported. For all 5-min runs, the olfactory bulb of the awake mice moved 0.10 ± 0.07° roll, 0.11 ± 0.08° pitch, 0.19 ± 0.15° yaw, 8.3 ± 6.84 μm dX (left-right axis), 22.3 ± 36.23 μm dY (dorsal-ventral axis), and 8.5 ± 8.12 μm dZ (rostral-caudal axis; Fig. 2C) of the 100 × 100 × 300 μm³ voxel. Motion estimates were not statistically different between odors (Supplementary Table 1; Kruskal-Wallis Test, Bonferroni corrected p-value for multiple comparisons with p<0.0125 considered significant).

Fig. 2. Motion characteristics of awake mice during olfactory fMRI scans.

A) Example traces of framewise displacement and censoring of high motion data points (>25 μm) with associated body motion measured with a pneumatic respiratory pillow sensor during an fMRI scan. Additionally, an example censored cerebral blood volume (CBV) hemodynamic response function. B) Total frequency distribution (histogram count) of framewise displacement values during all scans. Median value of distribution prior to thresholding was 13.5 μm, which decreased to 11.26 μm after censoring. Inset image shows zoomed in bin count of motion and censored values. C) AFNI’s 3dVolReg motion parameter estimates broken down into their median and median absolute deviation (MAD) rigid body components and first derivative components. D) Total percent of TR’s censored per GE-EPI scan. On average 5.26% ± 4.4% (8 ± 7 TR’s out of 152) were censored. E) Temporal breakdown of when censored events occurred during the length of the scan for all scans (black line), and for each odor (AA – green; Nona – blue; 2HA – red; Lim – yellow).

Interestingly, the largest motion was in the dY direction. This could potentially be due to: 1) sporadic motion throughout time, 2) slow drifting, and/or 3) sustained shifts in the baseline brain position. To determine if the motion parameter estimates were due to sporadic motions, we calculated the change in motion estimates over time. The awake mouse moved 0.056 ± 0.035° roll/dt, 0.073 ± 0.046° pitch/dt, 0.10 ± 0.063° yaw/dt, 3.9 ± 2.5 μm dX/dt (left-right axis), 2.0 ± 1.3 μm dY/dt (dorsal-ventral axis), 1.7 ± 1.1 μm dZ/dt (rostral-caudal axis; Fig. 2C; median ± MAD). This suggests that sporadic motion was not the main source of the larger head motion in the dY direction. We therefore examined slow drifting and sustained shifts. Qualitatively, 43% of scans had a sudden and sustained shift in head position (24.28 μm ± 24.6 μm; median ± MAD), 10% had a slow drift in the head position (14.8 μm ± 14.8 μm; median ± MAD), and 14% had attributes of both (31.7 μm ± 24.5 μm; median ± MAD), while the remaining 32% had a standard distribution of dY values (13.8 μm ± 15.7 μm; average ± STD). Motion parameters values and first-time derivatives were not dependent upon odor type (Supplementary Table 2; Kruskal-Wallis Test, Bonferroni corrected p-value for multiple comparisons with p<0.0125 considered significant). Our results suggest that larger dY motion was predominantly due to both slow drifting and sudden, but enduring, shifts in head position.

Framewise displacement (FD) values, the Euclidean distance between two successive scan volumes, has been regularly used in resting-state and rodent imaging to index and censor volumes with high degrees of motion. To limit contamination of large motions on the fMRI activation maps, we censored timepoints in which FD exceeded a threshold of 25 μm (i.e., $\frac{1}{4}$ of the in-plane resolution) (Cover et al., 2021; Han et al., 2019; Power et al., 2014). The FD for 5-min awake runs was 13.49 ± 2.74 μm (median ± MAD) before censoring that reduced to 11.26 ± 1.20 μm after censoring, which represents a 16.5% decrease in motion (Fig. 2B). FD values were not dependent upon odor type (Suppl. Fig. 2; Kruskal-Wallis Test before censoring, χ²(3147) = 2.09, p = 0.55; one-way ANOVA after censoring, F(3147) = 1.05, p = 0.37). Additionally, a FD threshold of 25 μm resulted in an average censoring of 5.26 ± 4.4% or 8 ± 7 out of 152 scan volumes being omitted (Fig. 2D). The average number of omitted volumes per scan was not dependent on odor type (One-way ANOVA, F(3159) = 0.8, p = 0.49).

We next investigated when the motion occurred during the 5-min run (Fig. 2E). Interestingly, censored timepoints exceeded a two standard deviation threshold at two different instances: 1) the start of the scan for a duration of one volume (2 s) and 2) following odor onset for three volumes (6 s). We also found that AA and Lim evoked more censored motions at odor onset compared to 2HA and Nona (Fig. 2E). We next examined whether more motion occurred with increased time in the scanner (~2 h in length). We found no correlation between the time in the scanner and the number of censored timepoints, (Pearson’s r = −0.05, p = 0.36; Suppl. Fig. 3a). Even after removing outlier data predominately from a single mouse, the Pearson’s correlation was not significant (r = 0.09, p = 0.14; Suppl. Fig. 3b).

Lastly, we determined whether head motion correlated with recorded body motion. To accomplish this, we adopted a pneumatic pillow sensor traditionally used as a respiratory monitor to act as a nonin-vasive motion sensor. The pillow sensor was placed between the mouse’s back and body tube to allow for real-time feedback of body motion during scans (Fig. 2A). Time point censoring corresponded with body motion when there was simultaneous supra-threshold FD values and pillow data that exceeded two standard deviations above baseline, respectively. Overall, 33 ± 27% (median ± MAD) of censored points from all 5-min runs had concomitant body motion, while 13 ± 22% of censored points did not. Interestingly, 33 ± 25% of large body movements did not lead to a censored timepoint during scans. These values were not dependent upon odor type (Kruskal-Wallis Test for censored points explained by body motion; χ²(3156) = 3.74, p = 0.29; Kruskal-Wallis Test for censored points not explained by body motion χ²(3156) = 1.77, p = 0.62; and Kruskal-Wallis Test for large body motion that did not lead to a censored time point, χ²(3156) = 1.7, p = 0.64).

3.2. Odor-Evoked fMRI activation maps

Odor stimulation produces discrete spatial patterns of activation in the bulb that are specific to the odorant. Therefore, the mouse bulb is an ideal model to test whether neighboring fMRI activations can be resolved in the awake state at high resolutions (100 × 100 × 300 μm³). Two pairs of odors were chosen for having spatially distinct (10% 2HA, 15% Nona) and spatially similar (5% AA, 20% Lim) activation maps. The four odors were sequentially delivered in a block design experiment (4-min blank air, 64-s odorant) and CBVw fMRI was used for enhanced sensitivity and neural specificity. Group activation maps were calculated using 3dMEMA with FDR correction (n = 19 sessions, 7 mice, 2–3 sessions/mouse). Greater than 99.9% of voxels were positive (i.e., CBV increase) at a q < 0.001 threshold and a minimum cluster of 10 active voxels (data not shown) for all odors, so negative activations were ignored for the remainder of the analyses. However, at the same threshold, each odor significantly activated different volumes of the bulb. Specifically, AA activated 82.8% and 85.8% of the whole bulb and GL only, respectively, Lim activated 74.5% and 77.9%, Nona activated 29.7% and 32.4%, and 2HA activated 9.0% and 13.3%, respectively. Differences can be attributed to the glomerular specificity of the odorant molecules, odorant sensitivity, and relative odor concentrations. Therefore, to compare odor-specific fMRI activation maps between odors, the number of activated voxels was fixed for each odor using top-percentile statistical thresholds. Top 5% activation maps to odor stimulation (Fig. 3A) were qualitatively consistent with odor-evoked 2-DG maps reported by other groups: lateral and ventromedial activations for AA and Lim (Johnson et al., 1998, 2002, 2009), dorsolateral and dorsomedial activations for 2HA (Smear et al., 2013), and ventromedial and ventrolateral activations for Nona (Johnson et al., 2009). To further show the separation of the activation maps, the four odor maps and their spatial overlap (∩) were plotted together (Fig. 3B). At the top 5% threshold, the dorsal-activating 2HA and ventral-activating Nona had highly distinct spatial patterns with a low degree of overlap (0.3%), while AA and Lim were still mostly separable despite the larger overlap (8.1%) (Fig. 3C). Few voxels had overlapping activation for three odors (2.5%) and there was no voxel that was activated by all four odors at the top 5% threshold. The map separability improved with stricter thresholds, as seen by the increasing single-odor shares and decreasing overlap shares, which is consistent with the activation maps being odor-specific (Fig. 3D). Specifically, the voxels that were exclusively activated by a single odor accounted for 63.9% (top 10% threshold), 79.8% (top 5%), 88.1% (top 2.5%), and 93.3% (top 1%) of the total number of activated voxels.

Fig. 3. Odor-Specific Group Maps.

High-resolution (100 × 100 × 300 μm³) CBVw fMRI images in the awake mouse olfactory bulb were acquired during 5% amyl acetate (AA), 10% 2-hydroxyacetophenone (2HA), 15% nonanal (Nona), or 20% limonene (Lim) odor stimulations (64-s on, 4-min off). (A) Group z-maps were calculated by 3dMEMA with FDR correction (n = 19 sessions, 7 mice, 2–3 sessions/mouse) and a top 5% activation threshold (AA: q < 5.5 × 10⁻⁹; 2HA: q < 4.9 × 10⁻⁴; NA: q < 7.6 × 10⁻⁵; Lim: q < 8.9 × 10⁻⁸) was applied with a minimum voxel cluster of 10. No negative z-scores were measured at these thresholds. Dynamic color-bar ranges were used to highlight localized hotspots (dark red) relative to each odor. Ascending slice numbers (see AA panel) on the grayscale underlays (EPI baseline images) progress from anterior to posterior. (B) 4-odor overlay map where activations for each odor in A (top 5% threshold) and their intersections (∩) with other odors are color-coded. (C) Proportion of each odor combination relative to all activations (same maps as B). (D) Proportions for other top% thresholds. 1-mm scale bars. D: dorsal, V: ventral, L: left, R: right.

3.3. Flat map representation of fMRI activation maps

To better visualize and compare the spatial activation patterns from odor stimulations, we generated two-dimensional flat maps of fMRI activation from non-thresholded data in GL (Fig. 4; n = 17). Two scan sessions were omitted due the loss of one slice during group map normalization. All flat maps showed similar activation patterns between the left and right bulbs for each odor, so the two hemispheres were averaged (Suppl. Fig. 4). Our CBVw fMRI flat maps had similar activation maps as previously reported by other groups. Specifically, 2HA activated narrow dorsolateral regions of the bulb that were relatively more anterior than the dorsomedial activations. This pattern corresponded to the locations of the M72 glomeruli in the bulb, which are strongly activated by 2HA (Smear et al., 2013). In contrast to the 2HA dorsal activations, NA activated ventromedial and ventrolateral regions of the bulb and was similar to 2DG activation patterns (Johnson et al., 2009). Lastly, AA and Lim similarly activated lateral regions of the bulb that were anterior to the other ventromedial activations, which corresponded to 2DG activation maps for both odors (Johnson et al., 1998, 2002, 2009). Flat maps of GL and GCL were also similar for all four odors (Suppl. Fig. 5), which is consistent with the functional (Johnson et al., 1999) and anatomical (Willhite et al., 2006) columnar organization of the bulb across all layers. This further supports that natural feedback from higher order olfactory brain regions was preserved in awake mice.

Fig. 4. Olfactory bulb odorant flap maps and discrete spatial activation patterns.

A) Overview of flat map generation from the glomerular layer (yellow) of the bulb. B) For each odor (AA – amyl acetate, Nona – nonanal, 2HA - 2-hydroxyacetophenone, Lim – limonene), the individual session t-maps were flattened and averaged (n = 17 sessions). Regions-of-interest (ROIs) 1 and 2 were outlined (dotted white line). Vertical scale is in-plane voxel number (50 μm/voxel); horizontal scale is slice number (300 μm/slice). C) Due to the symmetry of flat map profiles in the bulb, images were split into two discrete locations capturing the medial and lateral aspects for analysis labeled ROI1 (lateral bulb) and ROI2 (medial bulb). Weighted center-of-mass calculations (mean ± SD; n = 17) were performed on the top 5% of t-values. D) Intra-mouse Euclidean distance (mean ± SE; n = 7 mice) was calculated between CoM locations to capture the variability of activation patterns.

To better assess finer differences in the odor-evoked fMRI flat maps, we calculated the weighted CoMs (Fig. 4C) to assess the variability of odor mapping between mice. Qualitative observations of the averaged odor-evoked flat maps revealed two distinct regions of activation in the dorsolateral and ventromedial ROIs (Fig. 4B) for each odor. Therefore, we limited our CoM calculations to these regions. A one-way, repeated measures MANOVA with Sidak’s multiple comparisons post-hoc testing was used to assess whether the CoMs for each odor were spatially independent. We determined that all four odors in both ROIs had significantly separable and reproducible CoMs across sessions (p<0.001, n = 17) (Supplementary Tables 3-6). Namely, CoMs for AA and Lim were statistically discrete with the Lim activation being slightly more ventral than AA for both ROIs.

We then further characterized the reproducibility of odor-evoked flat maps acquired on different days for each mouse (n = 7 mice) to assess the variability of odor mapping between scan days. Since each mouse was longitudinally scanned for 2 – 3 sessions on different days, we compared the average Euclidean distance between CoM locations for each ROI across different days (Fig. 4D). The average intra-mouse difference in CoM locations for all odors was more reproducible for the dorsolateral ROI1 (175.3 ± 128.2 μm) compared to the ventromedial ROI2 (339.9 ± 218.6 μm). 2HA had the smallest difference in CoM locations for both ROI1 and ROI2 compared with other odors. Most odors had similar in-plane and out-of-plane differences in CoM locations for both bulbs, except for AA-ROI1 and 2HA/NA-ROI2 that had greater in-plane differences (Suppl. Fig 4d). Together, these results support that awake rodent CBVw fMRI is sensitive, odor-specific, and reproducible within ~200 – 300 μm (i.e., ~2 – 3 in-plane voxels or ~1 slice thickness). Differences in CoM location were not likely due to changes in motion since we did not observe any differences in framewise displacement between scan days. Interestingly, motion during 2HA runs was nearly different across scan days (p = 0.08, paired t-test, n = 17), but it had the smallest intramouse CoM variability of ~100 μm and ~200 μm for ROI-1 and ROI-2, respectively. While we cannot definitively claim that misregistration error did not contribute to the variability in intramouse CoM location, such error would be reflected in this small ~200–300 μm reproducibility range.

3.4. Layer-dependent fMRI responses

We previously observed odor-evoked CBVw fMRI signals that were greatest in the superficial GL that decreased with laminar depth in anesthetized rats (Poplawsky et al., 2015), which were similar to a laminar profile of 2-DG metabolic signal in awake rats (Sharp et al., 1977). In the current study, the mean fMRI signal change of all voxels in each layer (i.e., no threshold applied, see Suppl. Fig. 6 for laminar definitions) was normalized by the mean signal change across all six layers for each odor to account for global fMRI strength differences between odors. A similar trend was observed in awake mice where the greatest fMRI changes were observed in the superficial ONL and GL and decreased with laminar depth for all four odors (Fig. 5A). Signals in GL and GCL were also compared to further examine the impact of wakefulness on the presumed laminar input and feedback strengths, respectively. The normalized percent signal changes for the four odors (Fig. 5A) were not statistically different within GL (F(3,72) = 0.109, p = 0.954) or GCL (F(3,72) = 0.456, p = 0.714) using a one-way ANOVA, so the layer-dependent data were pooled across odors. The mean normalized signal change was 1.478 ± 0.639 in GL and 0.606 ± 0.285 in GCL; corresponding to a GL-to-GCL ratio of 2.496 ± 0.622 (± SD, n = 75, 1 outlier removed).

Fig. 5. Layer-specific CBVw fMRI responses.

(A) The normalized fMRI signal change (ΔS/S₀) was calculated across layers to show the relative laminar distribution of odor-evoked activations for each odor (n = 19, no threshold applied). The value of 1.0 represents the average laminar response for each odor. (B) For time series data, the percent signal change from baseline (0 – 120 s and 280 – 304 s) was calculated for each session and averaged across sessions (n = 19). Horizontal bars indicate time of odor exposure (i.e., 120 – 184 s). (Inset) The responses in GL and GCL were normalized to their maxima, respectively, compared to the CBVw hemodynamic response function (black curve). All graphs are mean ± SEM.

Next, we calculated the layer-dependent time series (Fig. 5B) by averaging all voxels in each layer (i.e., no threshold applied) to compare the temporal dynamics of fMRI responses between layers. Similar to the laminar profile in A, the greatest odor-evoked signal changes were in superficial layers that decreased with laminar depth for all four odors. The mean time series for the whole bulb (i.e., independent of layer, data not shown) reached 80% of their maximum 6 – 8 s after odor onset for all four odors and maintained this level until 6 – 8 s after odor offset for AA, Nona, and Lim; while 2HA decreased below this level 24 s after odor onset. The odor-evoked fMRI signal changes in GL during these peak ranges were: 11.09 ± 0.708% for AA (mean ± SEM, n = 19), 6.721 ± 0.479% for Lim (n = 19), 5.377 ± 0.726% for Nona (n = 19), and 3.660 ± 0.205 for 2HA (n = 19). Since each odor significantly activated different relative GL volumes (see above), we repeated this calculation using the same peak ranges but included only significantly activated voxels in GL (q < 0.001, minimum voxel cluster of 10): 11.32 ± 0.717% for AA, 7.329 ± 0.504% for Lim, 8.363 ± 0.999% for Nona, and 5.020 ± 0.304% for 2HA (data not shown). As expected, the fMRI signal changes increased when only including significant voxels, but the relative strength of activation between the four odors was similar. The fMRI responses to all four odors also did not attenuate much over the course of the 64-s odor stimulation, suggesting that odorant concentrations were not sufficiently high to cause rapid peripheral adaptation (Lecoq et al., 2009; Zhao et al., 2016, 2017). Temporal response shapes were similar between GL and GCL when normalized by their respective maxima for all four odors (Fig. 5B, insets), indicating that CBVw responses were not spatially varying. Although the CBVw hemodynamic response function used in 3dREMLfit was adapted from forepaw stimulations in anesthetized rats (Silva et al., 2007), it was mostly consistent with the awake functional time series except for a faster time-to-peak in awake mice. A longer return to baseline was also noted, but this may be due to the slower clearance of residual odor.

3.5. Layer-dependent olfactory adaptation

We next examined the effects of adaptation to repeated odor exposures on the bulb laminar responses. Only sessions with two exposures to the same odor were included in the analysis, and one outlier was removed (see arrowhead in Fig. 6C, Nona). First, to examine time-dependent adaptation effects, we calculated the temporal response differences between the 3rd and 1st odor exposures for each session before averaging (Fig. 6A). Negative time series were apparent in all layers because mean responses to the 1st exposure were larger than the 3rd. Although the temporal responses were qualitatively similar between layers for each odor, the time-to-minimum and its duration were different between odors. Specifically, the mean difference time series for the whole bulb (i.e., independent of layer, data not shown) maintained 80% of their minimum (see dashed gray lines in Fig. 6A) 32 – 82 s following AA onset (i.e., 18 s after offset of the 64-s stimulation), 14 – 26 s after 2HA onset, 14 – 52 s after Nona onset, and 24 – 48 s after Lim onset. These time intervals of “maximum adaptation” were then used in subsequent calculations.

Fig. 6. Adaptation of laminar fMRI responses to repeated odor exposures.

(A) The difference in the percent fMRI signal change (ΔS/S₀) between the 1st and 3rd odor exposures for each layer in time. Time courses in each layer were lowpass filtered before subtraction. One outlier was removed from analysis (see arrow in C, Nona), while one 2HA trial did not have a 3rd odor exposure (n = 18 for 2HA and Nona; n = 19 for AA and Lim). The time span when the fMRI signal difference was greater than 80% of its peak (between horizontal dashed lines) was averaged and used in the subsequent ROI (B) and distribution (C) analyses. (B) ROI analysis for the 1st (black line) and 3rd (red line) odor exposures (black horizontal axis on left); and their difference (blue line, blue horizontal axis on right). Significant differences between the 1st and 3rd exposures for each layer were marked by black asterisks. Significant differences between the bulb layers due to repeated odor exposures were marked by blue asterisks (Dunnett’s test with GL as the control layer). (C) Distribution of the fMRI signal changes to the 1st and 3rd odor exposures in GL only. The equality line (yellow line, slope = 1) represents a data distribution without adaptation, while points below the equality line indicate signal attenuation with repeated odor exposure. All graphs are mean ± SEM. * p < 0.05, ** p < 0.01, *** p < 0.001.

Mean fMRI signal changes were calculated for each voxel during the intervals of maximum adaptation (Fig. 6A, dotted lines) and the laminar ROI analysis was performed separately for the 1st (black lines) and 3rd (red) odor exposures (Fig. 6B). A two-way repeated measures ANOVA with Geisser-Greenhouse correction (ε reported where applicable) was used to examine whether the fMRI signal changed between the 1st and 3rd odor exposures for the six bulb layers. For the main effect of odor exposures, significant adaptation was observed for all the odors: AA (F(1, 18) = 11.01, p = 0.004), 2HA (F(1, 17) = 23.87, p < 0.001), Nona (F(1, 17) = 9.525, p = 0.007), Lim (F(1, 18) = 20.75, p < 0.001). We used post hoc Sidak’s multiple comparison tests to examine which layers adapted and observed significant changes in all layers for 2HA and LIM (Fig. 6B, black *’s), all layers except ONL for AA, and in GL, EPL, and core for Nona. We then examined the interaction between the odor exposures and bulb layers to determine whether the adaptation affected the bulb layers differently. This effect was significant for Lim (ε = 0.334, F(1.668, 30.02) = 9.948, p < 0.001), but not for AA (ε = 0.266, F(1.328, 23.91) = 3.200, p = 0.076), 2HA (ε = 0.495, F(2.476, 42.10) = 1.293, p = 0.288), or Nona (ε = 0.349, F(1.746, 29.68) = 3.122, p = 0.065). For Lim, we used a Dunnett’s multiple comparison test and determined that the effects of adaptation were significantly different between GL (control layer) and MCL, GCL, and core (Fig. 6B, blue *’s). To better visualize layer-dependent adaptation, the fMRI signal differences between the 3rd and 1st odor exposures were plotted (Fig. 6B, blue lines). Here, the absolute differences in the fMRI responses to subsequent odor exposures had a similar trend whereby superficial layers (e.g., GL) had larger differences than deeper layers (e.g., GCL). However, evoked responses in GL were larger than in GCL (Fig. 6B, black and red lines). To examine relative laminar differences, the GL-to-GCL ratios for the 1st and 3rd odor exposures (Suppl. Fig. 7) were compared and the differences determined (i.e., 3rd – 1st). AA had a mean ratio difference of 0.355 ± 0.113 and was the only odor where the 3rd exposure ratio was significantly larger than the 1st (p = 0.006, paired t-test, n = 19). Although the GL-to-GCL ratios were not significantly different for the other three odors, the mean ratio differences trended positive (i.e., 3rd ratio > 1st ratio) for 2HA (0.204 ± 0.193, p = 0.306, n = 18) and Lim (0.209 ± 0.149, p = 0.179, n = 19); but not for Nona (0.045 ± 0.078, p = 0.595, n = 17, 1 outlier removed). Therefore, adaptation may result in slightly less feedback to the bulb relative to its input, although more systematic testing is necessary.

Next, we examined the fMRI responses to repeated odor exposures for individual fMRI sessions (Fig. 6C). Only fMRI responses in GL were plotted since it is the input layer, had stronger evoked responses relative to deeper layers, and absolute adaptation effects trended greater here. Equality lines (yellow, slope = 1) were plotted as references, with points below the line indicating adaptation. The percentage of points below the equality line were 73.68% for AA, 83.33% for 2HA, 88.89% for Nona, and 94.74% for Lim. Interestingly, more profound response attenuations were observed with the 3rd odor exposure when the preceding responses to the 1st exposure were higher for all four odors.

4. Discussion

We have demonstrated that high-resolution CBVw fMRI of the olfactory bulb in awake mice 1) is reliable despite rodent motion comparable to previous studies, 2) produces odor-specific activation maps of four odorants similar to existing histological techniques, 3) detects differences in layer-dependent neuronal activity, and 4) distinguishes changes in regional laminar processing due to repeated odor exposure. Thus, the presented work indicates that high-resolution fMRI can be used to longitudinally study laminar activity in the olfactory bulb of awake mice.

4.1. Awake high-resolution fMRI

Awake rodent fMRI has emerged as a reliable alternative to using anesthesia in studying noninvasive brain activity (Chen et al., 2020; Cover et al., 2021; Ma et al., 2022). Specifically, acclimation of rodents to head fixation in a loud fMRI environment and censoring high motion data greatly reduces motion related artifacts. However, it was unknown whether awake imaging at laminar resolutions is practical since it is more sensitive to motion artifacts. While our custom-made cradle and acclimation protocol decreased motion during scans, it is not entirely absent. However, it is unlikely that residual motion impacted the statistical analysis and data interpretation. For instance, motion is classically characterized by a ringing activation along the edges of the brain at high contrast boundaries (Krings et al., 2001), which was not observed in this study. To further mitigate motion artifacts in this study, we added an inter-volume delay so that all fMRI slices were acquired in the shortest time possible (i.e., within 194 ms of the 1000 ms TR) instead of acquiring them evenly throughout the TR, as is typically done. However, it still needs to be determined whether these motion reducing benefits outweigh potential signal loss between slices. Our findings also support that motion predominately occurred in the dorsoventral plane (Chen et al., 2020). This may be due to the design of the horizontal headplate and head-fixation apparatus, which allowed for greater bending in this plane. On average, events with large motion accounted for 5% of any scan, which is far below the 20% threshold we established for the study. In fact, no datasets were omitted from this study due to excessive data censoring. Furthermore, events with large motion appeared sporadically except for at the beginning of the scan when the scanner noise initiated (~1 TR; 2 s) and immediately following odor onset (~3 TRs; 6 s). Motion following odor onset preferentially censored the initial hemodynamic response and weighed the GLM analysis toward the later plateau period of the signal. However, this bias was already present despite increased motion censoring because of the long odor exposure time (64 s). In addition, the number of observations (n) was adjusted for each time point in the SEM calculation to account for motion censoring (Fig. 5B), but noticeable differences in error across time was not observed. Higher censoring during these periods can be offset by including more dummy scans and ensuring the stimulus presentation is sufficiently long to avoid gross time point censoring during stimulation, respectively. It should be noted that odor was not presented during the benchtop acclimation sessions, which may have further reduced motion during odor presentation. Also, we adapted a more conservative acclimation procedure that was shown to reduce stress more adequately in animals (Harris et al., 2015) by using a long benchtop acclimation protocol (4 weeks) plus two 2-h acclimation sessions in a real fMRI environment. Since acclimation reduced motion when other physiological metrics of stress, like corticosterone levels and respiration rate, also returned to baseline levels (Desai et al., 2011; Dinh et al., 2021; Gutierrez-Barragan et al., 2022) Han et al., 2019; Tsurugizawa et al., 2020; Zeng et al., 2022), we used body and head motion as our main metrics of mouse stress during the fMRI scan. Our acclimation procedures likely produced similar reductions in stress since our motion parameters (13.5-μm median framewise displacement) were comparable to values in these other studies (4 – 32 μm) after acclimation. Lastly, the acoustic noise of the GE-EPI sequence has been shown to impact cortical brain activity in resting-state and cognitive-task experiments in humans, though its impact on animal models or the olfactory bulb is unknown (Moelker and Pattynama, 2003; Pellegrino et al., 2022; Tomasi et al., 2005). While earplugs were not used in this study, their use could decrease acoustic-related confounds (Chen et al., 2020).

4.2. Odor-evoked maps

The first aim of the study was to examine whether high-resolution CBVw fMRI could distinguish detailed odor-specific activation patterns in the awake state. This is particularly important to resolve submillimeter-scale active glomeruli because the increased susceptibility of fMRI to motion may cause image blurring within slices, fishtailing between slices, or high rates of volume rejections that reduce the effective fMRI resolution and experimental sensitivity. We, therefore, chose the olfactory bulb because it is well established that different odors produce spatially distinct activation patterns.

We first examined the top 1%, 2.5%, 5%, and 10% of activated voxels and determined that 2HA and Nona had the largest share of uniquely activated voxels, with little overlap between 2HA and any other odor at any threshold (Fig. 3). Comparatively, Lim and AA had the most overlap but became more disparate at more exclusive thresholds. We then separately calculated odor-evoked fMRI flat maps of GL, similar to 2DG maps, and determined the CoM for the medial and lateral loci, respectively (Fig. 4). The CoM in the left and right bulbs were consistent within the same odor (Suppl. Fig. 4b), while the CoM between odors were discrete. Like the top threshold analysis, the CoM of 2HA and Nona loci were the most separable, while those of AA and Lim were more adjacent. Together, we conclude that odor-specific CBVw fMRI activation patterns can be reliably measured in the awake state at high resolutions.

In addition, the neural-specificity of the fMRI odor maps can be further verified by comparing them to previous 2DG flat maps, which also reflect activity in the awake state. Specifically, both AA and Lim produced bilateral fMRI activation in the dorsolateral regions of anterior bulb and the ventromedial regions of posterior bulb, consistent with 2DG (Johnson et al., 1998, 2002, 2009) (Figs. 3 and 4). Interestingly, the CoM of fMRI loci were slightly more ventral for Lim relative to AA (Fig. 4C), which may be supported by a rat 2DG study (Johnson et al., 2002). The AA activation pattern in awake mice was also consistent with BOLD and CBVw fMRI in anesthetized rodents (Poplawsky et al., 2015; Sanganahalli et al., 2009; Xu et al., 2003). Nona activated the ventrolateral regions of anterior bulb and ventromedial regions of posterior bulb for both CBVw fMRI and 2DG (Johnson et al., 2009). For 2HA, the fMRI activation was highly localized to the dorsolateral regions of anterior bulb and dorsomedial regions of posterior bulb, similar to 2DG activation of acetophenone (2HA maps not found) in rats (Johnson et al., 2005). 2HA is also a potent activator of M72 glomeruli, which are located near both fMRI activation loci (Smear et al., 2013). It is noted that odor-specific activation patterns are species- and concentration-dependent, while 2DG and fMRI measure metabolic and hemodynamic changes, respectively. Therefore, the general consistency of the 2DG and CBVw fMRI activation maps supports that neural-specific fMRI can be performed at high resolutions in the awake mouse.

4.3. Layer specificity of the awake fMRI responses

The second aim of the study was to examine whether layer-specific activity can be measured with CBVw fMRI in awake mice. The olfactory bulb is an advantageous model to examine this since its layers are anatomically identifiable, can be independently activated, and have fewer reciprocal connections between layers compared to the neocortex. As such, we previously showed that CBVw fMRI responses were well localized to the individual layers of targeted synaptic activity. Notably, we measured peak fMRI and neural responses in the superficial GL for odor stimulation, in the middle EPL for electrical stimulation of the lateral olfactory tract, and in the deep GCL layer with electrical stimulation of the anterior commissure in anesthetized rats (Poplawsky et al., 2015). We further found that the CBVw fMRI responses to stimulation of the lateral olfactory tract spread only 50 – 100 μm beyond one side of the evoked EPL (Poplawsky et al., 2019b); suggesting that responses in GL and GCL are separable in the current study at our current in-plane resolution (100 × 100 μm²). This is of particular interest since responses in bulbar input and feedback, processes segregated to GL and GCL, respectively, may change with repeated odor stimulation (i.e., adaptation). It is also noted that odor-evoked layer-dependent changes in awake mice were greatest in superficial ONL and GL that decreased with laminar depth, similar to anesthetized rats (Poplawsky et al., 2015).

4.4. Changes in laminar responses due to olfactory adaptation

Repeated exposure to the same odor causes attenuation of neural and fMRI responses in the olfactory system (i.e., odor adaptation or habituation) (Chaudhury et al., 2010; McNamara et al., 2008). In particular, fMRI responses in the olfactory bulb and higher olfactory regions, like the piriform cortex and the anterior olfactory nucleus, both attenuated with repeated odor stimulation in anesthetized rats; although less adaptation was observed in the bulb (Zhao et al., 2016, 2017). We had similar findings in the awake mouse bulb where modest, but significant, adaptation was observed for all four odors. However, it is possible that decreased MION concentrations between the 1st and 3rd odor exposures contributed to some of the fMRI signal attenuation. MION washout is evident in the significant 5.866 ± 1.852% increase (one-sample t-test, p < 0.002, all runs pooled, n = 74) in the whole-bulb tSNR between the 1st (8.723 ± 0.350) and 3rd (9.08 ± 0.351) odor exposures. With the effects of MION washout considered, several features of the fMRI signal attenuation suggest neural-specific adaptation: 1) The temporal shape of the fMRI response differences between the 3rd and 1st odor exposures were unique for each odor (Fig. 6A), which included different peak and duration times that often diminished before odor offset. The peak times of the signal differences (Fig. 6A, 14–32 s after odor onset) were also slower than the odor-evoked responses (Fig. 5B, 6–8 s after odor onset). This suggests that early CBVw responses, which are less neural specific (Jin and Kim, 2008), for the 3rd and 1st exposures had similar response magnitudes that were canceled by subtraction despite MION washout. Initial neural adaptation, therefore, may also be masked in the CBVw responses. 2) The laminar response differences (Fig. 6B) were larger for superficial layers (e.g., GL) compared to deeper layers (e.g., GCL), but only reached significance for Lim odor; indicating that bulb input may be attenuated slightly more than bulb feedback. However, odor-evoked responses are generally ~2-times larger in GL compared GCL (Fig. 5A), which may bias the layer-dependent differences caused by repeated odor exposure. To remove this laminar bias and any possible confounds associated with changing MION concentrations, we calculated the GL-to-GCL ratio for individual odor exposures (Suppl. Fig. 7). This ratio trended larger for the third exposure compared to the first, but only reached significance for AA, similar to an anesthetize rat study (Zhao et al., 2016). This indicates that relative decreases are greater in GCL compared to GL; and that feedback from higher order olfactory regions is diminished relatively more than bulb input. 3) Attenuation in GL seemed greater during the third odor exposure when the activation to the first exposure was larger (Fig. 6C), suggesting greater adaptation effects with increased initial responses (Lecoq et al., 2009). Nonetheless, a more systematic study is needed to specifically test olfactory habituation on the different bulb layers and piriform cortex.

4.5. Awake vs. anesthetized olfactory bulb fMRI

Although the current study did not examine the influences of anesthesia on odor-evoked responses, our general observations in awake (current study) and anesthetized (our prior studies) rodents were consistent with studies that directly compared these states. However, the following comparisons of our studies are limited by species (awake mice vs. anesthetized rats) and MION concentration (25 vs. 15 mg/kg, respectively) differences; and that odor-evoked fMRI responses greatly depend on the choice of anesthetic and dose (Zhao et al., 2020 and odor concentration (Xu et al., 2000). First, we calculated a ~6 s faster CBVw peak time in our awake mice (Fig. 5B, AA in GL) compared to our α-chloralose anesthetized rats (Fig. 8C, AA odor in GL, CBVw; (Poplawsky et al., 2015)), which is similar to a 2 – 3 s faster BOLD peak time to visual stimulation in awake mice compared to ketamine/xylazine anesthetized mice (Dinh et al., 2021). This is also consistent with the CBVw responses of all four odors peaking 12–16 s sooner than a hemodynamic response function determined in anesthetized rats (Silva et al., 2007) (Fig. 5B insets, red vs. black lines). Next, the mean CBVw fMRI response magnitude in GL was ~2.6-times larger in awake mice (Fig. 5B, AA) compared to anesthetized rats (Fig. 8C, AA odor in GL, CBVw; (Poplawsky et al., 2015)); although this enhancement is likely exaggerated by the larger MION dose in awake mice. This is similar, however, to a >3-times larger BOLD response in awake vs. isoflurane anesthetized rats and mice (Desai et al., 2011; Liang et al., 2015), but not in ketamine/xylazine anesthetized mice that had similar BOLD response magnitudes as when awake (Dinh et al., 2021). Finally, more brain regions are activated in the awake state, especially in subcortical and downstream regions (Desai et al., 2011; Dinh et al., 2021; Ferenczi et al., 2016 as cortico-cortical and cortical feedback connections are more susceptible to anesthesia (Keller et al., 2020). In our studies, CBVw fMRI responses in GCL were nearly absent in anesthetized rats compared to awake mice. Specifically, the GL-to-GCL ratio was calculated to normalize the bulb input responses to its feedback responses for each fMRI session. This mean ratio was ~2.7-times smaller in our awake mice (Fig. 5A) compared to our α-chloralose anesthetized rats (Fig. 4C, AA odor, CBVw; (Poplawsky et al., 2019a)), indicating relatively stronger GCL responses in awake mice. Since GCL receives feedback from olfactory cortices and CBVw fMRI responses occur at active synaptic sites (Poplawsky et al., 2015; 2021), the lack of responses here likely indicates interference of anesthesia on downstream processes and its confounding effect on normal laminar processing.

A limitation of the current study is that isoflurane (1.5%, <15 min duration) anesthesia was used during MION contrast agent injection. Although mice recovered for 30 – 45 min before fMRI acquisition, isoflurane was shown to alter resting-state fMRI and neurophysiological dynamics ~3 h to one month after administration (Magnuson et al., 2014; Stenroos et al., 2021). While this likely did not significantly impact our results (see previous paragraph), layer-specific fMRI sequences with endogenous contrast, such as vascular-space-occupancy (VASO) (Huber et al., 2018; 2021) and spin-echo fMRI (Kenkel et al., 2016), should be explored to avoid confounds associated with residual anesthesia and tail-vein injections. However, these techniques have reduced sensitivity compared to exogenous MION contrast (Kim et al., 2013).