High-resolution 7T fMRI reveals the visual zone of the human claustrum

The claustrum is a thin bilateral subcortical structure that is situated between the putamen and the insula. It has reciprocal connections with most cortical and subcortical brain areas, which suggest its important role in higher-level processing (Edelstein & Denaro, 2004; Jackson et al., 2020; J. B. Smith et al., 2020; Torgerson et al., 2015; Torgerson & Van Horn, 2014). The claustrum has also been shown to have a high density of serotonin 2a receptors (5-HT_2A), the primary targets of psychedelic drugs (McKenna & Saavedra, 1987; Nichols, 2016). Pathology of the claustrum has been linked to several neuropsychiatric disorders such as schizophrenia (Cascella et al., 2011), autism (Wegiel et al., 2014), and attention-deficit hyperactivity disorder (ADHD) (Dickstein et al., 2006). Despite these unique and intriguing properties and a clear clinical relevance, the function of the claustrum remains a rarely investigated topic.

Several hypotheses have been proposed about the claustrum’s role in cognition and brain functioning. Although the original suggestions of it being the site of consciousness (Crick & Koch, 2005) did not find much empirical support (Bickel & Parvizi, 2019; Chau et al., 2015; Duffau et al., 2007), recent research in rodents clearly points to the claustrum’s role in attention and cognitive control (Goll et al., 2015; White et al., 2018). For example, the claustrum was shown to inhibit activity to irrelevant sensory stimuli, making the animal more resilient to distraction (Atlan et al., 2018). Additionally, connections from the anterior cingulate cortex (ACC) to the claustrum were found to be critical when mice are engaged in a cognitively challenging task (White et al., 2018). Furthermore, the claustrum is thought to be involved in coordinating slow-wave cortical activity (Narikiyo et al., 2020), and the claustrum’s homolog in reptiles has recently been shown to play a role in generating slow-wave sleep (Norimoto et al., 2020). In sum, the range of possible claustrum functions is quite diverse, end the evidence supporting them relatively scattered. Furthermore, due to the differences in overall brain anatomy and cognitive abilities across the animal kingdom (Baizer et al., 2014; Croxson et al., 2018), it remains unclear to which extent animal findings are generalizable to humans. Human studies on the function of the claustrum are critical for understanding its contribution to higher cognitive functions.

A well-documented aspect of the claustrum’s functional organization is the existence of sensory zones (Olson & Graybiel, 1980; Remedios et al., 2010; Torgerson & Van Horn, 2014) with corresponding topographically organized claustrocortical connections (Fernández-Miranda et al., 2008; Pathak & Fernandez-Miranda, 2014a; Smythies et al., 2012). Specifically, auditory, visual, and somatosensory zones have been described in the claustrum, with neurons preferring sensory input of the corresponding modality. In the macaque, distinct auditory and visual zones have been described, with the auditory zone located more dorsally and visual zone more ventrally (Gattass et al., 2014; Remedios et al., 2010). The claustrum’s thin shape and its location deep within the brain are challenging for conventional neuroimaging in human participants and until now the existence of the sensory zones of the human claustrum has not been tested.

In this proof-of-concept study, we utilized ultra-high field 7T fMRI to determine whether it is possible to elicit measurable visual and auditory responses in the human claustrum. We presented naturalistic visual, auditory, and audiovisual stimuli to participants in a 7T fMRI experiment. We followed a similar design to Remedios et al. (2010), who also presented naturalistic videos and found modality-specific zones in the claustrum; we therefore expected to find these modality-specific zones within the human claustrum, in which the visual zone would be located more ventrally and the auditory zone more dorsally.

The study’s design and methods were preregistered prior to conducting the experiment on AsPredicted.org (https://aspredicted.org/hj4sr.pdf). Code and data are available on the Open Science Framework (https://osf.io/7ebm2/).

Sixteen healthy participants were recruited to take part in this study. The exact number of participants was determined during preregistration and was informed by previous investigations utilizing 7 Tesla fMRI that focused on subcortical structures and visual perception with a sample size of 6–8 participants (Denison et al., 2014; Poltoratski et al., 2019). One participant had to be excluded due to poor data quality in the functional scans. Therefore, we had a total sample size of 15 (mean age = 24.60 years, SD = 3.33 years, 11 females and 4 males, 14 right-handed). Participants had normal or corrected-to-normal visual acuity, had no history of neurological impairments, and were not taking any medication at the time of participation. All participants gave written informed consent prior to participation. The study was approved by the ethics committee of the Medical University of Vienna and was conducted in accordance with the Declaration of Helsinki. Participants received monetary reimbursement for their participation.

Subjects viewed visual, auditory, and audiovisual naturalistic stimuli with an overlaid central fixation point. The center fixation point changed color every 500 ms between red and nine different shades of green in a pseudorandom order in which we removed consecutive duplicate colors. The participant’s task was to fixate on the central fixation point and to respond with a button press when the fixation point changed to the color red. The center fixation task ensured that participants were paying attention to the stimuli throughout the experiment and fixated their gaze at the center of the screen. Stimuli were presented using PsychoPy v2021.2.3 (Peirce et al., 2019) software on a MacBook pro 13” running macOS 12 Monterey. Visual stimulation was projected on an MRI-compatible rear projection screen using an XGA VPL F X 40 projector (Sony Group Corporation, Minato, Tokyo, Japan). The display was situated inside the scanner bore, which participants viewed through a mirror attached to the head coil at a 45° angle. The display-to-mirror distance was about 148 cm. Video stimuli were displayed at the full size of the MRI-compatible display (46.5 × 37 cm) at a 16:9 aspect ratio and subtended 17.5° × 14° visual angle with a fixation point situated in the center of the screen measuring 4 mm in diameter with a visual angle of 0.15°. Auditory stimulation was delivered using passive noise-cancellation S15 MR compatible in-ear earphones (Sensimetrics Corporation, Woburn, MA, USA). Prior to the beginning of the first functional run, a soundcheck was carried out to ensure that a sufficient level of loudness was achieved despite the ongoing scanner noise during image acquisition. This required fMRI dummy scanning while participants listened to some audio and provided feedback to increase or decrease the volume accordingly.

Stimuli consisted of naturalistic video scenes with a superimposed central fixation point. Naturalistic stimuli have been shown to elicit visual and auditory responses in the primate claustrum (Remedios et al., 2010). The videos were selected from the website Pexels (https://www.pexels.com/), a digital media sharing website. All stimuli materials used in this experiment were published to Pexels under a free-to-use and free-to-modify license (CC0, creative commons zero license). We used 48 video stimuli of different species of animals or natural scenery, and we ensured that the videos contained motion such that they would not be perceived as still images (for an example of the stimuli, see https://osf.io/7ebm2/, “Example Stimuli”). The original video clips were modified using FFmpeg version 2022-07-18-git-cb22d5ea3c-full_build-www.gyan.dev (FFmpeg, 2022) as follows. Each of the 48 selected videos were first cut down to a length of 15 seconds and then modified twice to create a visual only and an auditory only version. For both the audiovisual and auditory only conditions, the audio was normalized, such that the maximum sound volume for each clip did not exceed a max value of 0 dB. These modifications yielded a total of 144 unique 15-second clips each belonging to one of the three stimulus types, corresponding to the three experimental conditions: audiovisual (AV), visual only (V), and auditory only (A).

To ensure consistency of our findings within individuals, each subject took part in two scanning sessions taking place on 2 separate days, with a mean time between session 1 and session 2 of 10.67 days (SD = 6.39 days). Each session contained 6 runs, and in each run 24 trials with stimulation were presented to participants. Each trial belonged to one of the three experimental conditions, with the order of conditions pseudorandomized and counterbalanced using a first-order counterbalanced condition sequence, to minimize trial history effects (Brooks, 2012). On every 7th trial, a baseline (no stimulus) was presented, resulting in a total of 28 trials per run (see Fig. 1a, for an example of the presentation order). Each trial (including baseline) lasted 15 seconds. Each functional run lasted 420 seconds and participants took part in six functional runs per session for a total on-task scan time of 42 minutes. Two of the pilot sessions included in the analysis only involved a single session with 4 runs, and 1 pilot participant carried out 3 sessions resulting in 11 runs. The pilot sessions were not included in the session-to-session consistency analysis.

MRI data were acquired using an ultra-high field 7 Tesla Siemens MAGNETOM scanner (Siemens Healthineers, Erlangen, Germany) using a 32-channel head coil (Nova Medical, Wilmington, MA, USA). Blood oxygen level dependent (BOLD) contrast was obtained by using the gradient-recalled echo-planar imaging (GE-EPI) sequence. We acquired 37 sagittal slices at 1.34 mm × 1.34 mm resolution (slice thickness = 0.8 mm; TR = 2000 ms; TE = 23 ms; FA = 62°, GRAPPA acceleration factor = 2). The parameters deviated slightly in one of the sessions of one pilot participant (TR = 2500 ms, 47 sagittal slices). This resulted in a partial brain coverage of either the left or right claustrum, depending on the participant (see Supplementary Fig. 1 for the acquired fields of view of each subject). The EPI slice orientation and anisotropic voxel size were chosen to minimize blurring and maximize resolution along the left-right dimension, where the claustrum is thinnest (Zaretskaya & Polimeni, 2016). The right claustrum was scanned in six of the participants, and the left claustrum was scanned in nine of the participants (including three pilot participants). Each run of the functional scan took 420 seconds to complete, yielding 220 volumes per run. In addition to the main experimental runs, two additional volumes with identical parameters and opposite phase-encoding directions were acquired for subsequent susceptibility distortion correction.

Anatomical images were acquired using a T1-weighted MP2RAGE sequence (Marques et al., 2010) with 0.75 mm isotropic voxel size (matrix size: 320 × 300 slices: TR = 4300 ms; TE = 2.27 ms; FA = 4°, TI1 = 1000 ms, TI2 = 3200 ms, GRAPPA R = 2). The total time taken to complete the MP2RAGE scan was 530 seconds. Anatomical scans were obtained during both the first and second scanning sessions except for pilots with one session and for the pilot with three sessions (we used two of the anatomical scans for the latter).

The preprocessing of the anatomical scans consisted of the following steps.

Each of the two anatomical scans were first processed with the “presurfer” tool (https://github.com/srikash/presurfer) to remove extracerebral noise from the MP2RAGE image by utilizing a bias-corrected second inversion of the MP2RAGE acquisition. After this, the two T1-weighted scans were co-registered using the robust registration method (Reuter et al., 2010), which is part of the FreeSurfer package, and averaged to produce one final structural image per subject. This image was passed to the CAT 12.8.1 toolbox (Dahnke & Gaser, 2017) in SPM12 (7771, 13 Jan., 2020) (http://www.fil.ion.ucl.ac.uk/spm/, 2011) (Penny et al., 2011) running on MATLAB R2019b (MathWorks, 2019). This was done to create a high-quality brain mask by concatenating the white matter (WM) and grey matter (GM) segmentations. Each subject’s structural image was then used to perform cortical surface reconstruction using FreeSurfer’s recon-all stream (Dale et al., 1999; Greve & Fischl, 2009) at native resolution (Zaretskaya et al., 2018), substituting FreeSurfer’s auto-generated brain mask with the CAT12-derived brain mask.

Preprocessing of functional data involved motion correction, distortion correction, co-registration with the structural image, resampling to the anatomical image, and normalization to MNI space. The software used for each of the functional preprocessing steps is listed in Table 1 below.

First, the images were corrected for subject motion using the AFNI 3dvolreg algorithm that is implemented as part of FreeSurfer’s functional analysis stream (FSFAST). After this, the functional images were corrected for susceptibility distortions due to magnetic field inhomogeneities using FSL’s “topup” and “applytopup” (Andersson et al., 2003). Distortion-corrected functional images were co-registered to the averaged anatomical scan using boundary-based registration with 6 degrees-of-freedom (Greve & Fischl, 2009). We then resampled the functional data to the resolution of the structural scan (isotropic voxel size 0.75 mm) using FreeSurfer’s “mri_vol2vol” command. This was done to bring the data from the two sessions into the same space in order to conduct the joint first-level analysis of both sessions. Normalization of the functional data to MNI space was performed by applying the nonlinear transformation derived from the CAT12 toolbox during anatomical processing. MNI-space data were additionally smoothed by convolving each volume with a 3D Gaussian kernel with a full width at half maximum (FWHM) of 1.5 mm using FreeSurfer’s “mri_fwhm” command. Smoothed data were used for the group voxel-wise GLM analysis. Unsmoothed data were used for individual subject analysis (reporting individual peak activity of each subject and session-to-session consistency of voxel selectivity).

To determine whether activations we observed in the functional experiment are located within the claustrum, it was necessary to accurately define the left and the right claustrum in each individual. Standard atlases either do not contain a claustrum label (Fischl et al., 2002, 2004; Rolls et al., 2020) or label only the dorsal claustrum part (Ewert et al., 2018). Therefore, we manually labeled the left and right claustrum twice in an ultra-high resolution (0.1 mm isotropic) 7T post-mortem brain dataset, which is available in MNI space (Edlow et al., 2019). The union of the left and right claustrum label was then resampled to the anatomical space for use as a mask for statistical inference in group analysis and as masks for plotting the individual subject data, as described in our previous study (Coates & Zaretskaya, 2024).

To ensure that the claustrum label derived from MNI space did not include any of the neighboring structures in individual subject space (e.g., insula, putamen), we projected the MNI-space label into each subject’s individual space by applying the inverse transformation generated by CAT12. FreeSurfer’s cortical-subcortical segmentation (aseg.mgz) was then used to visually check whether any of the claustrum voxels overlap with structures other than voxels labeled as white matter (because the claustrum label is not available in FreeSurfer 7.1.0). We found that often the putamen and insula labels overlapped with the claustrum label. However, a closer visual inspection of the correspondence between native T1-weighted images and generated labels showed that this was due to poor automatic segmentation of the insula and putamen rather than the inaccuracy of the claustrum label. Poor segmentation of the putamen and insula in FreeSurfer has been shown previously (Perlaki et al., 2017; Yaakub et al., 2020), and is likely to be caused by the partial volume effects between these structures and the claustrum’s gray matter (Tian et al., 2020). As we did not find any overlap between the claustrum label and the gray matter of the putamen or insula in T1-weighted images, claustrum labels were unedited in all subjects.

Individual-level analysis was performed using FreeSurfer’s FSFAST using a standard GLM approach with Visual (V), auditory (A), audiovisual (AV), and baseline conditions as regressors of interest, which were convolved with the canonical hemodynamic response function. In addition, run-specific offsets, scanner drifts (modeled with a quadratic polynomial term), and the first 4 timepoints (to ensure that the scanner reached magnetic equilibrium) were modeled as nuisance regressors. To identify voxels within the claustrum that showed preference for either visual or auditory stimuli, we compared the beta estimates for the corresponding regressors, yielding the following contrasts: V − baseline and A − baseline. To identify voxels responding to both modalities, we also calculated a multisensory contrast. Following Noppeney (2012), we looked for a super-additive effect of the multisensory condition by calculating the contrast (AV + baseline) − (A + V). A GLM fit and contrast calculation was first performed individually for each session. The result of each session was then combined by carrying out a subject-level fixed-effects GLM analysis for each contrast.

For the second-level group analysis, we used the contrast estimates from the individual-level GLMs to perform one-sample t-tests for nonzero effects using the random-effects GLM. Since the visual and auditory conditions are expected to activate a small part of the whole claustrum, we corrected the results for multiple comparisons within the claustrum label using cluster-wise permutations (Greve & Fischl, 2018). We carried out 1000 permutations using a cluster-forming threshold of p < 0.01 and a cluster significance level of p < 0.05 (Hagler et al., 2006) for each contrast. Since the left and the right claustrum data cannot be combined in this approach, group analysis was performed twice, once for subjects with the left claustrum scans (N = 9) and once for subjects with the right claustrum scans (N = 6).

In order to determine whether the visual and auditory responses within the claustrum of each individual appear consistently at the same location across both session 1 and session 2, we performed an analysis in which we used data from either session to define a region of interest (ROI) and data from the contrary session to measure the responses in that ROI. For example, to measure the visual response in session 1 we used session 2 to define the visually responsive voxels (contrast “V − baseline”, p < 0.05 uncorrected) and then extracted the average contrast estimates from these voxels in session 1. This procedure was repeated with session 2 by defining visually responsive voxels using data from session 1. The same analysis was performed for auditory responses. The whole procedure yielded 4 values for each of the 12 subjects: visual activations in session 1 and 2 and auditory activations in session 1 and 2. Additionally, we calculated session-to-session consistency for auditory deactivations in session 1 and 2 for the six subjects with only the right claustrum scanned. Subjects with either their left or right claustrum scanned were pooled together. Session-to-session consistency was assessed by testing the ROI responses against zero using a one-sample t-test.

Since we did not find auditory activations within the claustrum, we wanted to ensure that the auditory stimuli used in the experiment were sufficient to activate the auditory cortex. To achieve this, we defined voxels corresponding to the transverse temporal gyrus (Heschl’s gyrus) in every participant from the automatic segmentation generated by the FreeSurfer recon-all stream, for example, aparc+aseg.nii.gz (Dale et al., 1999; Fischl et al., 2002) using the “mri_extractlabel” command. Using the ROI, we extracted average beta estimates for each condition and compared the auditory condition with baseline.

To ensure that the participants maintained stable gaze fixation throughout the experiment, we examined behavioral responses of each participant. To do this, we examined a time period between 0 and 2000 ms after the onset of each red color to determine whether a response was given. A response made in this time window was classified as a “hit.” We removed accidental duplicate responses and kept only the first response participants made. We then calculated the percentage of hits $(numberofcorrectresponsesnumberoftotalredfixations×100)$ and mean reaction time for each participant. For two participants we encountered a technical issue with the button response box that failed to record responses during two runs of the first session. Therefore, for these two participants behavioral analysis was performed with the remining four unaffected runs only. Moreover, analysis of behavioral responses for the first pilot subject was not possible due to a technical issue with fixation color changes not being logged by the script. Since we did record button presses from this subject and the number of recorded presses was in the range of the remaining participants, it is unlikely that the participant did not perform the task properly. We thus report behavioral performance for 14 out of 15 subjects.

Our aim in the current study was to determine whether high-resolution 7T fMRI would allow us to identify visual and auditory sensory zones of the human claustrum, and whether these or any other regions of the claustrum show signs of audio-visual integration. To test this, we presented participants with naturalistic video clips containing visual only, auditory only, or audiovisual information (see Fig. 1a for an example), while measuring fMRI activity in either the left or the right claustrum. To unambiguously assign functional activations to the claustrum, we manually labeled the left and right claustrum in an ultrahigh-resolution post-mortem MRI image that was mapped to MNI space (see Fig. 1b).

We performed a group analysis across all voxels within the claustrum for the left and right claustrum separately, comparing responses of each unisensory condition with baseline. Our analysis revealed 2 significant clusters of voxels that responded stronger to visual stimulation compared to baseline in the left claustrum (most significant voxel in cluster 1; z-statistic = 3.3, p < 0.001, d = 1.7, MNI x= -32.2, y= -3.75, z= -12, cluster size = 132.9 mm³, most significant voxel, cluster 2; z-statistic = 3.2, p < 0.01, d = 1.6, MNI x= -37.5, y= -1.5, z= -18.8, cluster size = 21.9 mm³). We also found 1 significant cluster in the right claustrum that responded stronger to visual stimulation compared to baseline (most significant voxel in cluster; z-statistic = 3.9, p < 0.001, d = 4.4, MNI x= 32.2, y= -5.25, z= -11.2, cluster size = 52.3 mm³) (Fig. 1c). To demonstrate that the group results originate from the claustrum and do not result from the averaging of neighboring structures such as the putamen and insula, we provide unmasked statistical maps of individual subjects overlaid on their individual anatomical images in Supplementary Figure 2.

We then ensured that significant visual activity observed in the claustrum at the group level is present in each individual subject at a similar spatial location in unsmoothed data. To do this, we analyzed session 1 and session 2 of each subject together and then extracted MNI coordinates of the most significant voxel for the visual versus baseline comparison. Peak coordinates of each subject as well as the mean and SD are shown in Table 2. Supplementary Figure 3 shows the number of subjects with significant visual claustrum activity for every voxel.

We found no voxel clusters responsive to the auditory condition after the multiple comparison correction in either hemisphere. Interestingly, we observed a significant cluster of voxels that showed a suppression of activity relative to baseline during the auditory stimulation, but only in the right hemisphere (most significant voxel in cluster: z-statistic = -3.4, p < 0.001, d = 3.1, MNI x = 34.50, y = 0.75, z = -5.25, cluster size = 55.7 mm³, Fig. 2a). There were no voxels suppressed by the auditory stimulation in the left hemisphere even with a more liberal cluster-forming threshold of p < 0.05. The auditory deactivations were not overlapping with visual activations (Fig. 2b).

Since our main analysis did not reveal any significant activity in response to auditory stimulation that survived multiple comparisons correction, we ensured that our auditory stimuli were efficient in evoking auditory activity by analyzing responses of the auditory cortex. We extracted the mean beta estimates for each condition (visual, auditory, audiovisual, and baseline) from the transverse temporal gyrus (Heschl’s gyrus), which corresponds to the primary auditory cortex. As expected, the primary auditory cortex exhibited no significant activation in response to visual stimuli (V-baseline, M = -0.84, SE = 0.25; t(14) = 0.24, p = 0.81, d = 0.02), but a significant activation in response to auditory stimuli (A-baseline, M = 2.13, SE = 0.62; t(14) = 5.4, p < 0.001, d = 1.92), to audiovisual stimuli (AV-baseline, M = 2.17, SE = 0.6; t(14) = 5.50, p < 0.001, d = 1.98) and when comparing auditory and visual stimuli (V-A, t(14) = 5.23, p < 0.001). These results are summarized in Figure 2c. It is, therefore, unlikely that the absence of auditory activity within the claustrum is related to the inefficient auditory stimulation in our experiment.

We also looked at multisensory responses within the claustrum at the group level. We wanted to determine if there are claustrum regions beyond the unisensory zones that show multisensory responses. Although we found a cluster that showed an activity pattern consistent with the superadditive response (AV + baseline) – (A + V), a closer examination of the cluster location and the beta estimates for each condition revealed that first, the cluster largely overlaps with the location of auditory deactivations and second, the multisensory contrast effects are driven by suppression of auditory responses below the baseline (Supplementary Fig. 4). The latter is inconsistent with a superadditive multisensory effect, which requires unisensory responses to be higher than the baseline (Beauchamp, 2005; Noppeney, 2012). We, therefore, did not observe any signs of multisensory integration within the claustrum.

Our main analysis thus reveals consistent bilateral visual responses at a specific location within the human claustrum. To check whether the location of visual activity observed at the group level was consistent within individual subjects across different scanning days, we performed a session-to-session consistency analysis of visual activity, additionally including auditory effects for completeness. We used one session to define a group of visual/auditory-selective voxels and measured the responses of these same voxels in the other session, statistically comparing the responses with zero.

We found that visual responses were consistent from session-to-session, as indicated by above-zero contrast estimates for both sessions (Fig. 3a). A one-sample t-test comparing the effects in each session against 0 revealed a significant effect for session 1 (t(11) = 4.56, p < 0.001, d = 0.71) and a significant effect for session 2 (t(11) = 4.86, p < 0.001, d = 0.59). In contrast, there was no significant auditory activity, neither for session 1 (M = 0.21, SE = 0.12; t(11) = 1.69, p = 0.12, d = 0.22) nor for session 2 (M = 0.17, SE = 0.16; t(11) = 1.06, p = 0.31, d = 0.16) which is consistent with the absence of auditory activations within the left and right claustrum at the group level (Fig. 3b). For the auditory deactivations, we looked at the sessions-to-session consistency only within the right hemisphere that showed significant deactivation effects at the group level (Fig. 3c). However, we could not confirm the session-to-session consistency of these effects, as there was no significant difference from 0, neither for session 1 (M = -0.22, SE = 0.12; t(5) = -1.80, p = 0.13, d = 0.27), nor for session 2 (M = -0.24, SE = 0.14; t(5) = -1.78, p = 0.13, d = 0.29).

To ensure that participants were keeping their gaze on the fixation point throughout the experiment, we calculated percentage of hits and reaction times. As expected, we found a high average percentage of hits (M = 86.95%, SD = 17.93%, IQR = 79.96%— 97%) across all participants and reaction times were less than 1 second (M = 0.67 s, SD = 0.15 s, IQR = 0.65 s— 0.70 s).

In this study, we aimed to investigate the visual, auditory, and audiovisual sensory responses within the human claustrum using ultra-high resolution 7T fMRI and naturalistic video clips. We found visual responses within the claustrum that were consistent across sessions and appeared at a similar location across participants in both hemispheres. We did not find significant auditory activity and no response pattern consistent with audiovisual integration. These results provide the first insight into the modality-specific sensory responses within the human claustrum that have otherwise been demonstrated only in animal models.

Our findings of visually evoked activity within the claustrum are in line with what is known about claustrum physiology from animal models. For example, similar to Remedios et al. (2010) that found visually responsive neurons at the more ventral claustrum locations, we also observed visual activity within a ventral claustrum site. Notably, a more ventral location of visual activity is also expected based on the topography of claustrocortical connections known from human tractography studies (Fernández-Miranda et al., 2008; Pathak & Fernandez-Miranda, 2014b). While that study used natural movie clips, earlier studies using more controlled visual stimulation could also narrow down the feature-specific nature of the cells within the visual region of the cat claustrum (Sherk & LeVay, 1981). Neurons in the visual zone of the cat claustrum showed preference for elongated moving bar stimuli as opposed to stationary stimuli and were particularly selective for the orientation of the bar stimuli. These animal model studies suggest that neurons in the visual claustrum zone may show a preference for specific features of the visual stimuli. We intentionally used natural stimuli that contain a wide range of visual features. It, therefore, remains unknown if particular features of the stimuli used may have evoked a stronger response in the visual zone of the claustrum compared to other features. Future research that uses controlled manipulation of individual stimulus features (color, contrast, form, and motion) will have to determine what stimuli features the visual zone is more selective to in humans.

In contrast to our expectations based on primate literature (Remedios et al., 2014), we did not find any auditory activation within the claustrum in response to auditory stimulation. There may be different reasons as to why auditory responses could not be found. Firstly, it is likely that the auditory zone is located in a thinner part of the claustrum compared to the visual zone. The auditory zone identified, described in primates, was located more dorsally compared to the visual region (Remedios et al., 2010). Naturally, the claustrum’s structure becomes much thinner in the dorsal areas (Kapakin, 2011). The auditory zone is thus expected to be more susceptible to partial volume effects and to produce a weaker functional signal, which we may have been unable to detect in the current experiment even using high spatial resolution at 7 Tesla. Future investigations that aim to differentiate between the auditory and the visual zone activity may require strategies that further increase the resolution of functional images (Goense et al., 2016; Pohmann et al., 2016; Uğurbil, 2021; Viessmann & Polimeni, 2021).

Another potential reason for the lack of auditory response is the noise generated by imaging gradients, which could have attenuated the auditory activity. Although measures were taken to ensure that the auditory volume was loud enough for participants to hear the sounds despite the scanner noise, the overall level of gradient noise may have led to a saturation of auditory activity, preventing us from detecting more subtle differences between the auditory stimulation and its absence. In an additional control analysis, we confirmed that our auditory stimuli evoked activity in the auditory cortex. However, the corresponding effects in the claustrum could have been smaller and thus harder to detect. Future studies aiming at measuring reliable auditory activity within the human claustrum could therefore take advantage of dedicated quiet EPI acquisition techniques that are tailored for fMRI studies of auditory processing (De Martino et al., 2015; Peelle et al., 2010).

Finally, it is possible that the temporal pattern of auditory responses in the claustrum neurons is different from that of the auditory cortex. A single-cell physiology study investigating claustrum responses to natural vocalizations observed that claustrum responses were highest when the vocalizations occurred immediately following silence, which may point to the claustrum’s role in change detection, an aspect closely related to attention (Remedios et al., 2014). Accordingly, the claustrum may only show a response in the auditory modality when there is a salient change from one stimulus to the next. In our experiment, the auditory stimuli, which consisted of natural sounds (e.g., waterfalls, cat vocalizations, cowbells), may have not been sufficiently salient and behaviorally relevant to induce a response within the auditory claustrum. Future studies that focus on auditory processing within the claustrum could test the role of saliency in evoking stronger auditory responses using, for example an auditory oddball paradigm.

Surprisingly, we found a deactivation in response to auditory stimuli in the right claustrum. This effect was observed in the right hemisphere only and was less consistent between sessions compared to visual activations. It should, therefore, be interpreted with caution and followed up in future studies. At this point, we can only speculate about the potential functional significance of these deactivations. One possibility is that deactivation in response to auditory stimuli mirrors the well-known multisensory effects in the primary sensory cortices. It has been repeatedly shown that the presentation of stimuli in one modality leads to a deactivation in primary sensory cortex of the other modality (Driver & Noesselt, 2008; Gau et al., 2020). In this scenario, we would expect visual activations and auditory deactivations to coincide spatially, which is not the case (Fig. 2b). In addition, multisensory effects in the sensory cortices alone do not explain why deactivation only occurred in the right claustrum but not in the left one.

Another potential explanation is that these findings are due to attention-related processing within the claustrum and reflects distractor suppression, similar to what has been described in rodent literature (Atlan et al., 2018; Goll et al., 2015). In our case, the central fixation task was used primarily as a way to ensure participants were fixating their gaze at the center of the screen. However, it is possible that because participants had to pay attention to the task, and hence to the visual modality, auditory stimulation served as a distractor stimulus, leading to suppression of the corresponding auditory representation within the claustrum whenever auditory stimulation occurred. This idea is consistent with fMRI findings in humans, which linked claustrum to within-modal and cross-modal divided attention (Vohn et al., 2007) and task control (Barrett et al., 2020) using conventional fMRI.

Our study did not yield any evidence for multisensory responses within the claustrum. A previous study in primates also did not find any evidence of multisensory responses, at least in the visual and auditory claustrum zones (Remedios et al., 2010). Single-cell recordings as used in that study are a unique possibility to measure activity of individual neurons with an unprecedented spatial and temporal resolution, but at the same time they limit the spatial extent of brain tissue that can be examined within the same individual. Given the topography of the claustrocortical connections, multisensory responses could reside outside of the sensory zones, for example in areas that project to or receive inputs from the classical multisensory cortical regions such as the temporo-parietal junction (TPJ) and the intraparietal sulcus (IPS) (Anderson et al., 2010; Geers & Coello, 2023; Ionta et al., 2011; Regenbogen et al., 2017). High-resolution functional MRI allowed us to measure activity throughout the whole claustrum, yet we did not find any evidence for multisensory effects, even beyond the visual zone. As with the lack of auditory activity, we cannot entirely rule out that some subregion of the claustrum shows multisensory responses, but we were unable to detect them due to the limitations of our method. In future high-resolution studies, a detailed map of the claustrum’s connection topography using other imaging modalities and techniques (resting-state functional MRI, DWI-based tractography) could be established to constrain the search of multisensory responses to a more specific location within the claustrum.

Our study presents significant group-level claustrum activity, which is also evident at the level of individual subjects and is consistent across two sessions for the same subjects. Although this ensures the reliability and generalizability of our results in terms of the presence of visual effects (Editorial, 2020; P. L. Smith & Little, 2018), due to our moderately sized sample we are less confident in our estimates of the underlying effect size (i.e., the magnitude of visual claustrum activity). Therefore, future studies that intend to detect claustrum activity and are interested in estimating its magnitude should consider conducing a careful a priori power analysis assuming medium effect sizes, rather than basing their sample planning on what is reported here (Poldrack et al., 2017).

Here, we present the first proof-of-concept study that has investigated the fine-scale functional response of the human claustrum using fMRI with high spatial resolution at ultra-high magnetic field. We demonstrate that it is possible to detect evoked visual activity within the human claustrum. Although further studies are needed to determine the functional role of auditory deactivations, our current results for the visual modality open the possibility of studying the claustrum’s contribution to visual processing.

In accordance with the declaration of Helsinki, all participants who partook in the study provided written informed consent and the study was approved by the local ethics committee.

Data and available scripts are provided at Code, and data are available on the Open Science Framework (https://osf.io/7ebm2/).

A.C. contributed to conceptualization, data analysis, original draft writing, reviewing and editing, and methodology. D.L. contributed to data collection, review and editing, and methodology. C.W. contributed to review and editing and methodology. A.I. contributed to supervision, review and editing. N.Z. contributed to conceptualization, funding resources, project administration, resources, methodology, supervision, and review and editing.

The authors declare no competing interests.

The authors would like to thank Denis Chaimow for suggestions on MP2RAGE preprocessing, Joana Leitao for discussion of the multisensory integration analysis and Maximilian Gerschütz for help with proofreading. This study was funded by the BioTechMed-Graz Young Researcher Group Grant to N.Z. and was supported by the Austrian Science Fund (FWF PAT 8722623 and FWF P 35583). The authors acknowledge the financial support by the University of Graz.

Supplementary material for this article is available with the online version here: https://doi.org/10.1162/imag_a_00327.