Topographic organisation of the claustrum–amygdala–prefrontal circuitry in the common marmoset (Callithrix jacchus)

Abstract

Recent rodent studies suggest that the claustrum complex, an evolutionarily conserved structure with widespread cortical connectivity, plays a role in modulation of anxiety-like behaviour via projections to the basolateral amygdala. However, this circuitry remains poorly defined in primates. Here, we investigated structural connectivity between the claustrum complex, amygdala, and prefrontal cortex in the adult common marmoset (Callithrix jacchus) using diffusion-weighted tractography and neuroanatomical tracing. Tracer injections were performed under anaesthesia via stereotaxic surgery. One marmoset received a biotinylated dextran amine injection into the basolateral amygdala, while four others received fluorescent retrograde tracers targeting the frontopolar cortex, orbitofrontal cortex, medial prefrontal cortex, and somatosensory cortex. Brains were processed for histology and tracer visualization. Diffusion weighted imaging and MRI tractography was performed on publicly available data from 24 marmosets from the Marmoset Brain Mapping Project (MBMv4; Tian et al. 2022; www.marmosetbrainmapping.org). The dorsal endopiriform nucleus was the region of the claustrum complex with the highest structural connectivity with both the amygdala and prefrontal cortex, showing particularly strong connectivity with the lateral amygdala and posterior orbitofrontal cortex, and more moderate connectivity with the medial prefrontal cortex. Our findings demonstrate a distinct claustro-amygdalo-prefrontal subcircuit in the marmoset, providing structural foundation for future studies examining the functional relevance of this circuitry in the primate brain.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00429-025-03026-z.

Introduction

The claustrum is a densely interconnected subcortical brain structure with reciprocal connections with nearly all areas of the cerebral cortex (Atlan et al. 2018; Mathur 2014; Narikiyo et al. 2020; Tanné-Gariépy et al. 2002; Torgerson et al. 2015; Zingg et al. 2018). Despite its extensive connectivity, the claustrum remains a relatively understudied region, and its precise functional roles remain poorly understood. In particular, the structural organisation of claustral connectivity with the amygdala, a key anatomically adjacent limbic structure that has extensive prefrontal cortex (PFC) connectivity, has not been systemically examined in primate models.

Rodent studies have investigated the PFC-claustrum–amygdala circuitry, but results are inconsistent and vary across experimental approaches. In rodents, the claustrum complex is defined dorsal-ventrally as comprising the claustrum and the endopiriform nucleus (Smith et al. 2019). However, there is no consensus on how to delineate subdivisions along the anterior-posterior axis which makes comparison between studies difficult (Grimstvedt et al. 2023). For example, the anterior claustrum has been reported to act as a cardiovascular-responsive region in rats, where stimulation reduces arterial blood pressure via medial prefrontal cortex (mPFC) connections (Hatam et al. 2013). Other studies suggest that the anterior claustrum inputs from the basolateral amygdala (BLC) allow bidirectional modulation of anxiety-like responses in acute social stress (Niu et al. 2022; Tanuma et al. 2022). In contrast, lesion studies indicate that anterior claustrum or posterior claustrum damage do not alter anxiety-like behaviours in mice, but the anterior claustrum is involved in fear memory processes such as attenuating retrieval and facilitating extinction (Gu et al. 2024). Moreover, a recent study by Zhao et al. (2024) suggests that local inhibition of the “median” portion of the claustrum reduced anxiety-like behaviours in mice, and the anterior claustrum is involved in reward behaviours. Adding further complexity, molecular studies in mice have identified a role for the transcription factor Tfap2d in the PFC-claustrum–amygdala circuitry; Tfap2d is found to be conserved cross-species including humans and is responsible for producing SOX4 and SOX11 (Kaur et al. 2025). The loss of SOX4/SOX11 was shown to reduce claustrum, piriform cortex, and BLC size. However, the loss of Tfap2d alone disrupts BLC-PFC connectivity and increases anxiety-like behaviours of mice but does not affect claustrum size (Kaur et al. 2025). Taken together, although these findings are divergent and the precise anatomical boundaries remain debated, they suggest that the rodent claustrum exhibits functional specialisation and topographical organisation along the anterior–posterior axis with discrete regional connectivity to the claustrum.

To date, most claustrum research has been undertaken in rodents, and the extent to which these findings apply to humans remains unclear (Striedter 2022). To bridge this evolutionary gap, there is a need to study this structural network in species more phylogenetically, neurobiologically, and behaviourally close to humans particularly as studies have highlighted primate-specific evolution of cortical neurons and claustral gene expression cell-type (Lei et al. 2025; Pham et al. 2019; Watakabe 2017). The marmoset (Callithrix jacchus) has garnered attention as a non-human primate model for neuroscience due to its high fecundity, easier husbandry, and smaller size than Old-World primates while conserving many structural and functional networks (Kishi et al. 2014; Okano et al. 2012). Marmosets exhibit complex social behaviours and emotional repertoires more akin to humans and, unlike macaques, have a small (~ 300–500 g), lissencephalic cortex that greatly facilitates whole-brain imaging and tract tracing (Kishi et al. 2014; Mitchell and Leopold 2015).

In primates, the claustrum complex comprises the insular claustrum and dorsal endopiriform as a largely contiguous structure in the dorsal-ventral axis (Smith et al. 2019). Most primate studies (including human and macaque) refer to only a broad anterior and posterior subdivision (Baizer et al. 2014; Maioli et al. 1983; Pearson et al. 1982; Tanné-Gariépy et al. 2002). At the transcriptomic level, Lei et al. (2025) showed that the macaque claustrum is composed of multiple glutamatergic neuron subtypes arranged into four broad spatial zones that exhibit preferential connectivity with different cortical and subcortical regions. The dorsal-anterior claustral zone in macaques demonstrates preferential connections with PFC and association cortices, whereas the ventral-anterior zone is primarily connected to the anterior cingulate, entorhinal cortex, and amygdala. The dorsal-posterior zone is connected to the primary somatosensory cortex, primary motor, and sensorimotor areas, whereas the ventral-posterior zone is strongly connected with parietal and visual areas (Lei et al. 2025). However, the dorsal-ventral claustral subdivision connectivity profiles have not been described in detail for the marmoset.

Resting state marmoset functional data have suggested that activity in the claustrum is strongly correlated with the amygdala (Holzscherer et al. 2025), however, the structural basis of the claustrum–amygdala pathway remains unclear. Anatomical work by Pham et al. (2019) subdivided the marmoset claustrum complex into four distinct subdivisions based on differences in myelinated axon orientation and cell density. These divisions include the insular claustrum and the dorsal (DEnD), intermediate (DEnI), and ventral subdivisions (DEnV) of the dorsal endopiriform. In this study, we examined the topography of connections between subdivisions of the claustrum, BLC, and PFC.

Here we applied dual-modality mapping specifically to the PFC-claustrum–amygdala network in the marmoset brain. Retrograde anatomical tracers are considered the gold-standard for identifying monosynaptic projections (Majka et al. 2016; Reser et al. 2017; Zeater et al. 2019). However, they present practical limitations when applied to subcortical structures in small primate brains such as the marmoset. The thin, sheetlike morphology of the claustrum complex makes direct targeting via tracer injections difficult, as there is a high propensity for tracer leakage into the surrounding white matter. Targeting of specific amygdala subnuclei through tracer injections also requires a difficult surgical approach, with frequent dispersion of tracer into the cannula track. The amygdala is a multinuclear complex with multiple functionally distinct subdivisions (Yang et al. 2017) located deep in the primate brain, making precise targeting of individual subnuclei difficult. As a result of these morphologic constraints, we principally examined tracer connections originating in targeted areas of the PFC and a single injection into the amygdala. The unavailability of tracer injections into the claustrum was offset by seed-based region of interest (ROI) studies performed using diffusion weighted imaging (DWI) tractography in a large cohort of marmoset brains (n = 24). Tractography allows for non-invasive ROI-based analysis, enabling selective seeding from both cortical and subcortical targets, including areas not accessible to tracers due to surgical constraints (Donahue et al. 2016). However, tractography-based population templates and atlases for the marmoset are relatively underdeveloped compared to those for macaques and humans, with only a few direct comparisons against structural mapping, and most have focused largely on cortico-cortical circuits with limited exploration of subcortical circuits (Hikishima et al. 2011; Tian et al. 2022; Skibbe et al. 2023). In this report, we employed the complementary strengths of anatomical tracers and DWI tractography to map the topographic organisation of the claustrum–amygdala–prefrontal pathway in the common marmoset.

Methods

Subjects

Twenty-nine adult common marmosets (Callithrix jacchus; 11 females; 18 males) were used (Table 1). Five animals were used for tracer and histological studies: four from the Rosa Laboratory at Monash University, and one from Cambridge University. Tractography was performed on data from the Marmoset Brain Mapping Project Version 4 (MBMv4, n = 24; www.marmosetbrainmapping.org). Publicly available data from the Marmoset Brain Architecture Project (www.marmosetbrain.org; (Majka et al. 2016; Majka 2020) were used for analysis of cortico-cortical connectivity and 2-D flat-map display.

Table 1.

Details of marmoset cases

Animal ID	Source	Sex	Age (months)	Weight (g)
CJ102	Rosa Laboratory	F	47	350
CJ181	Rosa Laboratory	M	26	368
CJ148	Rosa Laboratory	F	19	430
CJ178	Rosa Laboratory	F	26	344
F15	Cambridge	F	NA	NA
sub-NIHm14	MBM4	M	62.86	406
sub-NIHm15	MBM4	M	24.36	484
sub-NIHm16	MBM4	F	41.69	533
sub-NIHm17	MBM4	M	73.05	446
sub-NIHm19	MBM4	M	32.71	554
sub-NIHm20	MBM4	M	27.42	473
sub-NIHm21	MBM4	M	65.03	414
sub-NIHm22	MBM4	M	31.82	380
sub-NIHm23	MBM4	F	59.24	543
sub-NIHm24	MBM4	M	28.60	359
sub-NIHm25	MBM4	F	44.88	437
sub-NIHm26	MBM4	F	49.51	345
sub-NIHm27	MBM4	M	37.08	513
sub-NIHm29	MBM4	M	32.09	433
sub-NIHm30	MBM4	F	22.55	463
sub-NIHm31	MBM4	M	28.80	483
sub-NIHm32	MBM4	M	62.99	562
sub-NIHm33	MBM4	M	40.50	461
sub-NIHm34	MBM4	M	107.97	528
sub-NIHm35	MBM4	F	75.78	414
sub-NIHm36	MBM4	M	78.41	270
sub-NIHm37	MBM4	M	30.81	625
sub-NIHm38	MBM4	F	77.16	448
sub-NIHm39	MBM4	M	70.52	472

Neuroanatomical tracers

Five marmosets (4 female, 1 male) received neuroanatomical tracer injections. In case F15 (Cambridge University), biotinylated dextran amine (BDA) was injected into the BLC. The remaining cases received fluorescent retrograde tracers targeting three PFC regions—area 10 (frontopolar cortex; case: CJ178; tracer: cholera toxin subunit b [CTB] combined with Alexa 594), area 11 (orbitofrontal cortex (OFC); case: CJ181; tracer: CTB- Alexa 594), and area 32 V (mPFC; case: CJ148; tracer: diamidino yellow [DY]). An additional injection was placed into the ventral part of area 3b (primary somatosensory cortex; case: CJ102; tracer: DY). Area 3b is used as a negative control as previous studies have shown negligible connections between area 3b and any of the targeted areas of PFC (Majka 2020; Reser et al. 2014).

Tracer injections were performed via stereotaxic surgery following laboratory protocols detailed previously (Reser et al., 2017). Briefly, marmosets were weighed, then premedicated with intramuscular atropine (0.2 mg/kg) and diazepam (2 mg/kg), followed 30 min later by alfaxalone (10 mg/kg, i.m.) for anaesthesia. Dexamethasone (0.3 mg/kg, i.m.) and amoxicillin (50 mg/kg, i.m.) were administered before placing the animal in a stereotaxic frame. Body temperature, heart rate, and PO₂ were continuously monitored, and supplemental anaesthesia was given as required to maintain areflexia. A craniotomy was performed over motor areas using stereotaxic coordinates (Paxinos et al., 2012). Marmosets were euthanised with alfaxalone (10 mg/kg IV) followed by sodium pentobarbitone (100 mg/kg IV) 10–15 days after tracer injection. Cardiac perfusion was performed using heparinised saline followed by 4% paraformaldehyde. The brains were extracted, cryoprotected in a sucrose gradient (10%, 20%, 30%), and sectioned coronally at 40 μm using a cryostat. DWI tractography was performed in three brains (CJ208, CJ196, CJ197) using the imaging protocol described below prior to sectioning for histological analysis. The anatomical scans were directly compared to the individual histological sections and reviewed by experts in marmoset brain anatomy (MGPR and DHR) to arrive at consensus regarding anatomical landmarks and morphological features used for parcellation of nuclear boundaries (claustrum and amygdala) and cortical areas (PFC). The individual animal tractography and high-spatial resolution imaging data are included as supplemental data to this report.

Histological processing and analysis

Tissue sections were prepared in five series: unstained sections for fluorescence microscopy, sections stained for myelin, Nissl, and NeuN, and BDA histochemistry for anterograde transport visualisation. Tracer-labelled cell bodies and axon terminals were mapped using a Zeiss Axioplan 2 epifluorescence microscope and X-Y digitizer system (MD-3, Accustage) in MDPlot (v5.4). Fluorescent-labelled neurons were identified based on nuclear presence and pyramidal morphology. Only ipsilateral tracer distribution was analysed. For BDA tracers in case F15, both retrograde-labelled neurons and anterograde-labelled axonal projections were documented. To segment the claustrum complex, tracer plots were aligned with Nissl-stained sections and classified into the insular claustrum and dorsal endopiriform, which was further subdivided using the myeloarchitectonic and cytoarchitectonic features described by Pham et al. (2019) as outlined below. The overlap of labelled neurons within claustral subdivisions across all cases was assessed to identify areas with monosynaptic neuronal connections.

Anatomical parcellation of the claustrum complex

The claustrum was identified as a slender layer of grey matter embedded between white matter fiber tracts in Nissl, myelin, and NeuN stained specimens. Based on myelination, axon orientation and cell density, the claustrum complex was divided into the insular claustrum (dorsal region) and the dorsal endopiriform (ventral region). The dorsal endopiriform was parcellated into dorsal (DEnD), intermediate (DEnl), and ventral (DEnV) using the criteria of Pham et al. (2019). Dorsally, the ventral region of the insular claustrum can be distinguished as the triangular area of grey matter immediately lateral to the DEnV. For anatomical localisation within the marmoset brain, we utilised stereotaxic coordinates referencing the interaural line. The claustrum complex’s rostral-caudal extent was demarcated from approximately + 13.0 mm to + 6.20 mm anterior-posterior (A-P) interaural distance. To facilitate regional comparisons, the claustrum complex was segmented into anterior (+ 13.0 mm to + 10.73 mm A-P), central (+ 10.73 mm to + 8.47 mm A-P), and posterior (+ 8.47 mm to + 6.20 mm A-P) sectors.

Quantification of tracer labelling

For quantitative analysis of retrograde tracers, labelled neurons were counted in the claustrum complex using Adobe Illustrator. Counts were normalized as a proportion of total claustral neurons per case to account for variability in tracer uptake. Anterograde tracer analysis was qualitative, focusing on axonal terminal field distribution. Axonal fields were distinguished from fibers of passage by orientation and morphology, and the absence of axonal boutons.

Diffusion-Weighted imaging and tractography

Multi-shell diffusion-weighted magnetic resonance imaging (MRI) data for the adult marmosets (n = 24) were obtained as part of the in-vivo dataset found in MBMv4 (www.marmosetbrainmapping.org). The marmosets were scanned in a two-dimensional diffusion-weighted spin-echo echo-planar imaging sequence on a 7T horizontal MRI (Bruker, Billerica, USA) equipped with a 30-mm quadrature coil and a 15 cm customized gradient set capable of 450 mT/m gradient strength. Scanning parameters are described in detail by Tian et al. (2022). A total of 400 DWI was collected with each acquisition having 3 diffusion shells: 8 volumes at b = 0 s/mm², 64 volumes at b = 1000 s/mm², and 128 volumes at b = 2000 s/mm². High-resolution 80 μm T2*-weighted images available from MBM version 2 and the Paxinos Marmoset Atlas (Paxinos et al., 2012) were used to guide parcellation of the ROIs. The claustrum, amygdala subnuclei, OFC (areas 11 and 13), and primary somatosensory cortex (area 3b) were manually parcellated via ITK-SNAP. The mPFC parcellation was guided by the MBM version 3 atlas (www.marmosetbrainmapping.org) and includes area 32, 32 V, 24, and 25 (Liu et al. 2021). The ROI parcellations were registered to diffusion space using Advanced Normalization Tools. Data preprocessing and fiber orientation distributions were computed using multi-shell, multi-tissue constrained spherical deconvolution in MRtrix3 (Tournier et al. 2004).

Additionally, individual subject tractography was performed on three in-house marmosets from the Rosa Laboratory (CJ196, CJ197, CJ208), acquired using different imaging protocols. These datasets (Supplementary Figure) were used to assess intersubject variability and reproducibility prior to the population-level analyses using the MBMv4. Individual structural MRI data was used for alignment of imaging and histological sections with tracer cases and to improve the precision of anatomical parcellations for the grouped MRI data. The animals included one adult male (CJ208; 60 months, 347 g), one young adult male (CJ196; 24 months, 361 g), and one young adult female (CJ197; 24 months, 395 g). Scans were acquired using a 40-mm Bruker volume resonator with the following parameters: multi-shell diffusion with 81 directions, b-value = 5000 s/mm², isotropic resolution of 125 × 125 × 125 μm³, and a total scan time of 18 h. Pre-processing and analysis were conducted identically to the MBMv4 dataset.

For the MBMv4 dataset, ROI-based tractography was performed using streamlines generated between parcellated ROIs defined in the population template space, which were subsequently warped into each individual subject’s diffusion space for tractography analyses. Streamlines were generated using the -seeds_random_per_voxel option set to 10,000 streamlines per voxel. The -seed_unidirectional flag was applied to constrain propagation to a single direction from each seed voxel. One ROI is used as the seed region and the other as the inclusion ROI specified with the -include option, and streamlines were terminated upon exiting the brain mask (-stop). Tracking parameters were: FOD amplitude cutoff = 0.065, maximum curvature angle = 45°, step size = 0.125 mm, maximum length = 25 mm, with the default minimum length of 1.25 mm retained. The cutoff threshold was empirically determined based on pilot tractography runs using whole-brain tractography runs and using SIFT2-weighted tckedit which enabled us to identity a value that optimised the balance between sensitivity and specificity, such that fiber bundles between prefrontal and amygdala regions (considered biologically plausible) were preserved, while spurious prefrontal streamlines to area 3b were negligible or absent.

For each pair of regions, tractography was performed separately in both directions: once with ROI_i as the seed and ROI_j as the inclusion mask, and once with ROI_j as the seed and ROI_i as the inclusion mask. Each .tck file therefore represented a unidirectional tractography run. As diffusion tractography is inherently non-directional, examining both directions was essential to fully capture the distribution of streamline terminations. ROI tractography was conducted per hemisphere and did not include self-connections.

To generate track density images, streamlines were converted to voxel-wise maps using the tckmap function in MRtrix3, with the -ends_only flag engaged. The endpoints of a streamline refer to both its seed region and termination point, so there are 2 endpoints for each streamline connection (one in ROIi and one in ROIj). To account for variability in brain size and morphology, diffusion images were spatially normalised using a population-averaged white matter fiber orientation distribution template. Using tract density images, we examined the streamline endpoints within specified ROI pairs. Independent coronal sections of the claustrum complex (27 sections from rostral to caudal) were created as individual ROI masks. For each section, a tract density image (TDI) map was constructed to visualise streamline endpoints between the claustrum to one other ROI region. The TDI endpoint density maps were manually inspected to ensure streamline endpoints were confined within the defined claustrum boundary mask.

After thresholding, the total number of streamlines within the claustrum mask was calculated for each ROI pair–yielding a final streamline count. The proportion of spurious streamlines (pre-thresholded streamline count minus corrected streamline count/ pre-thresholded streamline count) for each ROI pair was then applied as a correction factor in subsequent calculations. The estimated proportions of spurious streamlines for the ROI pairs are as follows: 0.009 for claustrum-BLC, 0.05 for claustrum-PFC, 0.008 for claustrum-area 3b. Previous studies have threshold at 0.01 to reduce false negatives in macaques and human brains (Warrington et al. 2020). The number of endpoints in the claustrum complex at each coronal section was calculated allowing us to assess the proportion of claustral connections at different rostral-caudal distances.

There is no current consensus on the best method to quantify white matter connectivity via probabilistic tractography. Streamline count (SC) is highly dependent on parameters set such as seed volume, tract selection, number of diffusion directions, image resolution, length of anatomical tract, and the microstructural measurement of interests (Reid et al. 2020). Probabilistic tractography can yield varying results with each execution, resulting in microstructural measurements or anatomical boundaries that lack consistency. To mitigate this issue, connectivity weights were normalised using fractional scaling (FS) of streamline counts across all ROI pairs (Donahue et al. 2016; Gu et al. 2021).

The resulting streamline count is then scaled using the following calculation to give an adjusted streamline count (). The seed ROI region is denoted i and j is the target ROI.

)

ε_ij = estimated spurious proportion.

SC_ij = streamline count from ROI i to ROI j.

The fractional scaled streamline count (FS) for any ROI pair is calculated by the ratio of the adjusted streamline count of i-j divided by the sum of the total streamline count of all connections where i is the seed region and the total streamline count of all connections that terminates at j, excluding self-connections.

As tractography does not provide a reliable measure of directionality, the i-j and j-i fractional scaled value needs to be averaged for symmetrisation as the measured connectivity from region i to j often differs from that of j to i. Averaging these values provides a more consistent and undirected representation.

A symmetrical connectivity matrix is computed from the normalised fractional scaled value of the connection weights.

Fixel-based density and cross-section (FDC) values were computed separately for the left and right hemispheres. This was achieved by extracting FDC measurements using standard MRtrix3 fixel-based analysis pipeline for each ROI pair for both hemispheres for all marmoset brains. The FDC values were averaged across the marmoset brains to generate a group-level mean for each ROI-pair, separately for the left and right hemispheres.

Results

Cortical–claustrum connectivity

The anatomical boundaries of the claustrum complex in the marmoset brain were delineated using histology to define subdivisions and to assess for topographical organisation in tracer studies (Fig. 1). Tracer injections across PFC subdivisions revealed differential connectivity patterns with the claustrum complex. The proportions described refer to the distribution of total labelled cells within the claustrum complex for each injection. The CTB injection in area 10 (case CJ178) exhibited sparse labelling (32 cells), primarily in the insular claustrum (75.0%) and secondarily in DEnD (25.0%), with no labelling in other dorsal endopiriform subdivisions (Fig. 1C). According to the approximate inter-aural A-P level most of the labelling from area 10 was in the anterior sector of the claustrum, followed by the central sector. Due to sparse labelling of cell bodies in the claustrum complex, the frontopolar cortex was not generated as an ROI for tractography.

Fig. 1.

Anatomical organisation and tracer distribution within the claustrum complex of the common marmoset. A Coronal sections from Case CJ102 (+ 11.80 mm interaural anterior–posterior [A–P] distance) stained for Nissl and myelin, showing internal subdivisions of the claustrum complex. The dorsal endopiriform nucleus (dorsal, intermediate, and ventral subdivisions) is outlined in yellow. The insular claustrum is outlined in blue. An adjacent NeuN-stained section from Case F15 at the same A–P level is also shown. Scale = 1 mm. B Lateral view of the marmoset brain illustrating the location of four coronal sections spanning from + 13.00 mm to + 7.00 mm A–P. Representative sections from Case F15 show the rostral-to-caudal extent of the left claustrum complex. The insular claustrum (blue); the dorsal endopiriform nucleus (yellow). At + 9.50 mm and + 9.00 mm, the ventral portion of the insular claustrum (IC) appears as a blue triangular region lateral to the dorsal endopiriform. Subnuclei of the basolateral amygdala complex is labelled (LA; lateral amygdala, BL; basolateral amygdala, and BM; basomedial amygdala).C Stacked bar chart showing the relative proportion of labelled cells across the four subdivisions of the claustrum complex: insular claustrum (blue), dorsal endopiriform nucleus dorsal subdivision (DEnD; gold), intermediate subdivision (DEnI; red), and ventral subdivision (DEnV; purple). Bottom right: bar graph showing total retrogradely labelled cell counts across all cases in the insular claustrum (blue) and dorsal endopiriform nucleus (yellow)

Area 11 (CJ181) exhibited the dense CTB labelling (612 cells) within the claustrum complex, predominantly in DEnD (53.0%), with additional distribution in the insular claustrum (15.8%), DEnV (11.8%), and DEnI (5.3%). These connections were densest in the anterior claustrum region. Tractography analysis of the claustrum–OFC connectivity (OFC ROI: areas 11 and 13) predominantly arises from the anterior to central portions of the claustrum complex along the interaural anterior–posterior (A–P) axis. These projections were localised primarily to the insular claustrum and DEnD along the dorsal–ventral axis (Fig. 2A), consistent with our retrograde tracer injections into area 11.

Fig. 2.

Cortical streamline endpoints in the claustrum complex. The distribution of streamline endpoints (including both seed and termination points of the streamlines) are shown for each cortical-claustrum streamline, plotted across the anterior–posterior (AP) axis of the claustrum complex (top row). The regions of interest (ROIs) are shown in 3D population template brain space with orientation indicated (R, right; L, left; S, superior; I, inferior) (middle row). The two columns (bottom) for each cortical-claustral connection include the corresponding tracer injection (left column) and streamline endpoint density maps (right column) showcased in the representative right claustrum complex. The streamline density scale ranges from 5 to 100 endpoints per voxel. Endpoint distributions are localised within the claustrum complex, including the insular claustrum and dorsal endopiriform nucleus [dorsal (DEnD), intermediate (DEnI), and ventral (DEnV) subdivisions]. A Orbitofrontal cortex (OFC). Tracer (case CJ181) = area 11; ROI = OFC (areas 11 and 13). Streamline endpoints are distributed across anterior and mid–AP levels of the claustrum complex, localised within the insular claustrum and dorsal endopiriform subdivisions. B Medial prefrontal cortex (mPFC). Tracer (case CJ148) = area 32 V; ROI = mPFC (areas 24, 25, 32, and 32 V). Streamline endpoints predominantly found in anterior claustrum complex in dorsal endopiriform. C Primary somatosensory cortex (area 3b). Tracer (case CJ102) = area 3b; ROI = primary somatosensory cortex (area 3b). Streamline endpoints are concentrated in insular claustrum throughout the anterior and central regions of the claustrum complex in the anterior-posterior axis

The area 32 V injection revealed 920 DY-labelled cells, which were located primarily in the insular claustrum (49.8%), followed by DEnD (28.1%), DEnI (20.1%), and DEnV (2%). Along the rostral-caudal axis, the connections revealed by this injection were densest in the anterior-central claustrum region is the claustral region in the A-P axis with the densest connections and cell bodies were found in the DEnD predominantly (Fig. 1C). In contrast, our tractography ROI encompassed a broader set of mPFC subdivisions (areas 32, 32 V, 24, and 25) and revealed that the anterior claustrum is the primary region of connections with the mPFC (Fig. 2B). Streamline endpoints concentrated the dorsal endopiriform in rostral sections and extended into the insular claustrum and DEnD within the central claustrum.

Tracer injection in area 32 V and area 11 both showed large numbers of tracer-labelled neurons in the amygdala, whereas area 10 injection revealed minimal amygdala connections. However, these differences could be in part attributable to injection size effectiveness, since the same area 10 injection labelled 4,755 neurons throughout the cortex (Majka et al. 2020; data retrieved from www.marmosetbrain.org), which is significantly less than the injections in area 11 (26,738 labelled cells) and area 32 V (42,187 labelled cells).

Primary somatosensory cortex (Area 3b)—claustrum connectivity

In contrast to PFC injections, the primary somatosensory cortex (area 3b; CJ102) exhibited sparse labelling, confined exclusively to the insular claustrum (total 235 labelled cells; for comparison: labelled cells throughout the cortex = 37,560) in the anterior and central claustrum. No labelled cells were observed in the dorsal endopiriform or the amygdala, supporting the specificity of connectivity observed in PFC cases. Similarly, tractography seeded in area 3b, used as a negative control, were confined to the dorsal insular claustrum (Fig. 2C), in agreement with our findings from the area 3b injection that demstrated a lack of dorsal endopiriform - area 3b connections. While the 3b tracer injection showed connections mainly in the anterior followed by central claustrum, our 3b ROI showed the central claustrum slightly more connected than anterior claustrum, with connections present in the posterior claustrum.on

Amygdala–Claustrum connectivity

In case F15, a biotinylated dextran amine (BDA) injection targeted the amygdaloid complex, resulting in extensive labelling within the BLC and minimal diffusion into the central amygdala. Microscopic examination confirmed that the injection cannula traversed the primary somatosensory cortex (area 3b) and basal ganglia, with some tracer deposition along the track. Retrograde cell-body labelling was observed across all four subdivisions of the claustrum complex, with the highest density of labelled neurons in the DEnD (54 cells, 42.0%), followed by the insular claustrum (42 cells, 32.6%). The remaining labelling was distributed across DEnI and DEnV (25.4%). Along the rostral-caudal axis, a left-skewed distribution was noted, peaking at + 11.30 mm interaural distance, and extending caudally. BLC projections to the dorsal endopiriform were primarily between + 10.50 mm and + 12.00 mm, while insular claustrum connectivity spanned + 13.00 mm to + 8.70 mm, with the highest density rostrally (Fig. 1C). Analysis of labelled axonal terminal fields revealed broad distributions across the claustrum complex. Terminals were concentrated in the rostral DEnD but extended into nearly all subdivisions caudally. However, the widespread and diffuse nature of axonal arborisation made it difficult to quantify local density. Consequently, tracer data provided confirmation of connectivity but limited resolution in terms of density weights of specific claustral and amygdalar subdivisions. Light terminal labelling from tracer case F15 was also present in areas 10 and 32 V, with denser fields in medial area 32. The densest terminals were found in area 11, followed by area 32 V, whereas area 10 exhibited fewer terminals. Area 3b displayed the sparsest axonal terminal labelling among the areas compared (Fig. 3).

Fig. 3.

Histological comparison of axonal fields of Case F15. Low- (0.05 mm) and high-magnification (0.01 mm) coronal sections showing biotinylated dextran amine (BDA)-labelled neurons and terminal axonal fields. Corresponding low-magnification (0.05 mm) Nissl-stained sections are shown for key regions of interest: frontopolar cortex (area 10), orbitofrontal cortex (area 11), medial prefrontal cortex (area 32 V), the dorsal endopiriform nucleus of the claustrum complex, and primary somatosensory cortex (area 3b)

Tractography provided convergent evidence and extended these findings by enabling subdivision-level mapping across the BLC subnuclei (Fig. 4). Whereas tracer injections labelled widespread cell bodies throughout the BLC, tractography could be constrained within ROI-to-ROI tracks, excluding intra-amygdalar connections and isolating claustrum–amygdala interactions directly. ROI analyses showed that the strongest claustrum-BLC connectivity is with the LA and the anterior-central claustrum complex, predominantly within the DEnD and DEnI subregions dorsal-ventrally. BL-claustral connections were relatively weaker, localising to the DEnD of central and posterior sections of the claustrum complex, while BM projections were least dense and were concentrated in the DEnD/DEnI of the central–posterior claustrum.

Fig. 4.

Connectivity between the amygdala and the claustrum complex. A Coronal sections from Case F15 following a biotinylated dextran amine (BDA) injection into the basolateral amygdala complex (BLC), showing retrogradely labelled cell bodies (red) and terminal axon fields (pink) within the claustrum complex. The insular claustrum (IC) is outlined in blue; the dorsal endopiriform nucleus [dorsal (DEnD), intermediate (DEnI), and ventral (DEnV) subdivisions] in yellow. B Streamline endpoint maps derived from the population template of diffusion-weighted tractography (Marmoset Brain Mapping Atlas), showing streamline endpoints from lateral (LA), basolateral (BL), and basomedial (BM) amygdala subnuclei - claustrum streamlines in the claustrum complex. C Left: anatomical parcellation of LA, (BL, and (BM amygdala subnuclei based on T2*-weighted MRI represented in 3D space with orientation (R; right, L; left, S; superior, I; inferior). Right: distribution of BLC–claustrum streamline endpoints along the anterior–posterior (A-P) axis of the claustrum complex

Together, these data show that amygdala–claustrum connectivity is densest in the anterior–central claustrum, particularly within the DEnD, with a consistent hierarchy of input strength (LA > BL > BM).

Prefrontal–Claustrum and Prefrontal–Amygdala connectivity

Connectivity patterns were overall similar across hemispheres, though marginally denser in the left, particularly for PFC-claustrum and claustrum-BLC pathways (Fig. 5A). Claustrum connectivity was strongest with the LA, followed by the BL, and the BM connections. OFC–BLC pathways were weaker than claustrum–BLC connections, consistent with prior diffusion tractography studies underestimating long-range pathways, as these trajectories typically intersect with other tracts running in different directions, complicating their detection and reconstruction (Zalesky and Fornito 2009). FDC is a composite metric that quantifies white matter connectivity by integrating the fiber density and fiber cross-section, which is useful for detecting subtle connectivity differences and assessing the lateralisation of neural circuits (Raffelt et al. 2017). Fiber density measures the intra-axonal volume within each individual fiber population within a voxel, reflecting white matter microstructural integrity. Conversely, fiber cross-section measures changes in the cross-sectional area of a fiber bundle, indicating macrostructural morphology differences. FDC analysis revealed broadly similar connectivity profiles across hemispheres, with slightly higher values in the left hemisphere for several PFC-BLC pathways (e.g., BL–mPFC: left = 0.48, right = 0.40; OFC–BM: left = 0.43, right = 0.36) (Fig. 5B). The strongest FDC values were observed for BL–mPFC, OFC–mPFC, and OFC–BL/LA connections. Claustrum–PFC and claustrum–BLC pathways were present but comparatively weaker, while 3b–claustrum tracts showed high FDC. Overall, these results suggest largely symmetrical connectivity with modest leftward asymmetry in prefrontal–amygdala tracts.

Fig. 5.

A Fractionally scaled streamline counts between left and right hemispheres across regions of interest (ROIs). B Fixel-based density and cross-section (FDC) values for the same ROIs in both hemispheres. Regions of interests (ROI): 3b – primary somatosensory cortex; BM – basomedial amygdala; BL – basolateral amygdala; LA – lateral amygdala; CLA – claustrum complex; mPFC – medial prefrontal cortex; OFC – orbitofrontal cortex

For the mPFC–claustrum connection (Fig. 6A), endpoints were densest in the anterior dorsal half of the claustrum complex, primarily within the insular claustrum, and in the anterior ventral half of the mPFC. Anteriorly, endpoints were concentrated in ventral mPFC (area 32), whereas posteriorly they were localised to posterior 32 V and ventral area 25 (Fig. 6G). The mPFC–claustrum endpoints overlapped with mPFC–BLC connection endpoints mainly in the posterior aspect of the ventral mPFC (area 32 V and 25). Within the claustrum complex, mPFC–claustrum endpoints were distributed mainly across the insular claustrum and DEnD along the A–P axis, but were most prominent anteriorly. For OFC–claustrum connection (Fig. 6C), endpoints extended across the anterior–posterior extent of the OFC, with highest density in posterior OFC. The OFC–BLC projections were similarly concentrated in posterior OFC, where they overlapped with OFC–claustrum endpoints (Fig. 7B). In the claustrum, both OFC–claustrum and OFC–BLC endpoints overlapped predominantly within the insular claustrum and DEnD, spanning anterior and posterior sections (Fig. 7D). By contrast, area 3b–claustrum projections (Fig. 6E) were restricted to the dorsal insular claustrum. Overlap with claustrum–BLC endpoints was limited, confined to the dorsal-most region of the insular claustrum and none in the dorsal endopiriform (Fig. 7E). Within the BLC, claustrum projections terminated most densely in the LA, followed by the BL, which showed widespread coverage across the claustrum complex (Fig. 6F). BM connections were weaker and preferentially localised to the DEnD. Taken together, the cortical (mPFC and OFC) and BLC pathways endpoints overlap most predominantly in the claustrum complex in the DEnD, whereas area 3b-claustrum endpoints remain largely segregated.

Fig. 6.

3D renderings of tract density maps showing streamline endpoints (including both seed and termination points) between selected regions of interest (ROIs). For each connection, tractography was performed in both directions, and the combined endpoints are displayed. ROI orientation is indicated (A, anterior; P, posterior; D, dorsal; V, ventral). Endpoint density is shown on a scale of 1–10 endpoints per voxel for panels (A–F). Panels display the spatial location of endpoints using colour coding, which indicates anatomical position but not density. A Medial prefrontal cortex (mPFC) – claustrum. B mPFC – basolateral amygdala complex (BLC). C Orbitofrontal cortex (OFC) – claustrum. D OFC – BLC. E Primary somatosensory area 3b – claustrum. F Claustrum – BLC, including projections to BLC subnuclei [lateral (LA), basolateral (BL), and basomedial (BM)]

Fig. 7.

Spatial Distribution of Streamline Endpoints displayed in 3D brain space. Each coloured voxel corresponds to at least one streamline endpoint, as the analysis focuses exclusively on endpoint location overlap between ROI pairs. A Overlap (purple) between mPFC–claustrum (blue) and mPFC–BLC (red) on the mPFC; coronal sections illustrate subregions (areas 32, 32 V, 24, and 25). B Overlap (green) between OFC–claustrum (yellow) and OFC–BLC (blue) on the OFC; coronal sections illustrate OFC subregions (areas 11 and 13). C Overlap (purple) between mPFC–claustrum (blue) and claustrum–BLC (red) on the claustrum complex; coronal sections illustrate subdivisions including the insular claustrum and dorsal endopiriform nucleus [dorsal (DEnD), intermediate (DEnI), and ventral (DEnV) parts]. D Overlap (brown) between OFC–claustrum (green) and claustrum–BLC (red) on the claustrum complex; coronal sections illustrate subdivisions including the insular claustrum and dorsal endopiriform nucleus [DEnD, DEnI, DEnV]. E Overlap (pink) between area 3b–claustrum (white) and claustrum–BLC (red) on the claustrum complex; coronal sections illustrate subdivisions including the insular claustrum and dorsal endopiriform nucleus [DEnD, DEnI, DEnV]

Discussion

Our study provides a dual-modality characterisation of the claustrum–amygdala–prefrontal network in the common marmoset, integrating retrograde tracers with population-level diffusion tractography. We demonstrated anatomical evidence that the claustrum connectivity is topographically organised along both the anterior-posterior axes and the dorsal-ventral axes, with the densest PFC and amygdala connections localised to the DEnD of the dorsal endopiriform (dorsal-ventral axis) in the central claustrum (anterior-posterior axis). For the purposes of analysis, we defined rostral–caudal claustral subdivisions as anterior, central, and posterior thirds based on approximate stereotaxic interaural distance, providing a reproducible framework for regional comparison.

The claustrum–amygdala and claustrum–PFC pathways identified in this study formed a network anatomically distinct from insular claustrum–area 3b projection. This aligns with previous marmoset data showing minimal overlap between sensory and prefrontal inputs (Majka 2020; Reser et al. 2014). In macaques, retrograde tracer injections into the anterior and extreme dorsal claustrum predominantly labelled motor and prefrontal neurons with little to no amygdala input, particularly contralaterally (Lei et al. 2025). The homologous region in our marmoset dataset corresponds to the site of area 3b–insular claustrum connectivity. In contrast, the posterior-dorsal and middle-ventral claustrum tracer injections in macaques revealed strong amygdala connectivity, including the contralateral BLC input and bilateral connections within the ventral claustrum (Lei et al. 2025). Taken together, our findings and previous studies support a conserved dorsoventral organisation of the claustrum across species: the dorsal claustrum preferentially connects with sensorimotor areas, whereas the ventral claustrum is linked to limbic–associative regions in macaques (Gamberini et al. 2017, 2021; Lei et al. 2025; Pearson et al. 1982), humans (Fernández-Miranda et al. 2008; Qadir et al. 2018; Torgerson et al. 2015), and rodents (Grimstvedt et al. 2023; Kitanishi and Matsuo 2017).

Our findings also highlight a claustrum–BLC network in which connectivity is strongest with the LA, followed by BL and BM. Across subjects, tractography revealed dense claustrum–LA projections, localised primarily to the anterior–central DEnD and DEnI. In contrast, BL connections were less dense and concentrated within the central–posterior claustrum, while BM projections were weakest overall and preferentially localised to the DEnD/DEnI of central–posterior regions. This hierarchy (LA > BL > BM) aligns with the established organisation of the BLC. The LA functions as the principal input nucleus of the amygdala, receiving convergent afferents from cortical and thalamic sources (LeDoux and Daw 2018). The BL serves as an integrative hub, linking LA inputs to intra-amygdala circuits and cortical association targets (Etkin et al. 2009) The BM is more strongly associated with efferent projections to prefrontal and limbic regions, including the hippocampus and mPFC (Etkin et al. 2009). Within this framework, our data suggest that the claustrum interacts most strongly with the amygdala primary input subnucleus (LA), while also engaging, to a lesser degree, with the integrative (BL) and output-oriented (BM) subnuclei.

In our study, both the DEnD and the LA exhibited their strongest PFC connectivity with the ventral-posterior sectors of the OFC and mPFC, particularly areas 11, 13, and 25. Along the anterior-posterior axis, this connectivity was concentrated primarily in the central region of the claustrum, followed by the anterior sector.

Previous studies in primates have shown that lesions in the posterior OFC (area 13) diminish the microstructural integrity of the uncinate fasciculus, a major tract that bidirectionally links the OFC and the amygdala (Kenwood et al. 2023). Our data also show that the posterior OFC is the main OFC region that has connections to the BLC.

In marmosets, inactivation of either the anterior (area 11) or posterior (area 13) OFC alters behavioural responses to distal threats, distinguishing these regions from adjacent PFC subdivisions such as area 14 which is involved in immediate threat processing (Stawicka et al. 2022). Marmoset imaging studies. Shiba et al. (2021) have implicated an insular–amygdala-PFC pathway, specifically involving areas 11 and 13, in the regulation of negative emotion in response to innate and social threat. Our data extend this framework by identifying claustrum connectivity with areas 11/13 and the BLC, suggesting that the claustrum is part of the same broader primate OFC–amygdala network. Historically, functional imaging studies have largely failed to localise activity to the claustrum (Smith et al. 2017; Smith, Watson, Smith et al. 2019a, b) largely due to signal contamination from adjacent regions, such as the insular cortex, an area also implicated in salience detection with the same shared vasculature (Krimmel et al. 2019). Given their overlapping connectivity profiles, further investigation is warranted to clarify the distinct contributions of these structures.

Moreover, cross-species comparisons should be interpreted cautiously. Although the rodent prelimbic and infralimbic cortices are often mapped onto primate Brodmann areas 32 and 25, respectively (Vogt and Paxinos 2014) functional data do not always support direct homology (Wallis et al. 2017). In our study, the claustrum–amygdala–mPFC circuit was found to predominantly involve the posterior portion of area 25, an mPFC subregion that, when overactivated in marmosets, has been shown to induce a profound negative affective state—marked by heightened responses to both certain and uncertain threats and reduced anticipatory motivation (Alexander et al. 2020). Conversely, the opposite effect on negative emotion is observed following infralimbic cortex inactivation in rodents (Wallis et al. 2017). Additionally, while rodents possess agranular OFC regions that share certain connectivity patterns with the caudal agranular areas of the primate OFC, they lack the granular anterior regions present in primates (Preuss and Wise 2022). Moreover, the BLC in primates and rodents differs notably, with primate BLC neurons showing more differential and segregated projections to specific frontal regions, while rodent BLC neurons more often project to multiple frontal cortical targets (Zeisler et al. 2024).

Historically, the rodent claustrum complex is described as comprising the claustrum and the dorsal endopiriform nucleus, with further dorsal endopiriform subdivisions being a matter of debate (Kowiański et al. 1999; Smith et al. 2019a). Accumulating evidence suggest that the rodent DEnV is a separate structure, and this distinction is supported by a high degree of overlap in neurochemical and genetic markers between the claustrum and dorsal endopiriform, which are distinct from those of the DEnV (Smith et al. 2019). These cross-species structural differences indicate that, although there is partial homology between primate and rodent regions, the extent to which this translates into functional analogy remains uncertain, highlighting the need for more primate research to clarify the organisation and function of the claustrum–amygdala-prefrontal circuitry.

Limitations

Several methodological and interpretational limitations should be considered in evaluating the present findings. First, although the integration of retrograde tract-tracing and diffusion-weighted tractography offers complementary perspectives on structural connectivity, each approach has inherent constraints. Tracer studies are spatially and directionally precise but limited by surgical accessibility of target areas, injection spread, and the number of sites that can be evaluated per subject. In our study, only a single tracer injection was performed per cortical site and for the amygdala. To address this, we supplemented our tracer data with population-level diffusion tractography, which enabled broader, whole-brain analysis across multiple individuals. While tractography lacks cellular specificity and directionality, it complements tracer findings by providing a non-invasive estimate of structural connectivity patterns at scale. However, diffusion tractography is susceptible to false positives, particularly in regions with complex fibre crossings. So, while we used cytoarchitectural criteria and histological verification to delineate subdivisions of the claustrum complex, interindividual variability in anatomical boundaries, particularly at interfaces with adjacent regions such as the insular cortex and striatum, may introduce error in ROI placement and interpretation. Finally, while the current study identifies structural connectivity between the claustrum, amygdala, and PFC in the marmoset, no functional data were collected. The behavioural relevance of these connections remains to be validated through targeted manipulations, inactivation studies, or task-dependent imaging approaches.

Conclusion

This study provides a detailed characterisation of the structural organisation of the claustrum–amygdala–prefrontal network in the common marmoset, revealing a topographically specific pattern of connectivity. Using a combined approach of retrograde tracing and DWI tractography, we demonstrate that the dorsal endopiriform nucleus of the claustrum, particularly the anterior and central region of the anterior-posterior axis, serves as a key zone that shares connections with both the BLC complex and PFC. This connectivity was distinct from that of the primary somatosensory cortex, supporting a structural segregation within the claustrum complex. These findings advance our understanding of claustrum structural connectivity in the marmoset brain and provide a structural foundation for investigating its functional role(s) in the primate brain. Future work combining functional and behavioural approaches will be essential to determine the relevance of this neural circuitry.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

A-P

Anterior–Posterior

BDA

Biotinylated Dextran Amine

BL

Basolateral Nucleus

BLC

Basolateral Amygdala Complex

BM

Basomedial Nucleus

CTB

Cholera Toxin Subunit B

DEnD

Dorsal Endopiriform Nucleus, Dorsal

DEnI

Dorsal Endopiriform Nucleus, Intermediate

DEnV

Dorsal Endopiriform Nucleus, Ventral

DWI

Diffusion-Weighted Imaging

DY

Diamidino Yellow

FDC

Fixel-Based Fibre Density and Cross-section

FS

Fractionally Scaled Streamline Count

LA

Lateral Amygdala

MBMv4

Marmoset Brain Mapping Version 4

mPFC

Medial Prefrontal Cortex

MRI

Magnetic Resonance Imaging

OFC

Orbitofrontal Cortex

PFC

Prefrontal Cortex

ROI

Region of Interest

SC

Streamline Count

TDI

Track Density Imaging

Author contributions

The authors have no relevant financial or non-financial interests to disclose. All authors contributed to the study conception and design. The manuscript was principally written by B.N.T.H. and D.H.R. All co-authors (D.K.W., A.Z., A.C.R., M.G.P.R., and D.H.R.) provided supervision, feedback, and critical revisions. All authors reviewed and approved the final manuscript. The authors would also like to thank Cirong Liu for valuable discussions and support throughout the research.

Funding

This work was supported by the Australian Government Research Training Program (RTP) Scholarship to BNTH. Additional funding was provided by The Australian National Health and Medical Research Council to the following authors: DHR- Project APP1068140; DKW Project APP1174040; MGPR- Investigator Grant APP1194206; and by the Australian Research Council to MGPR- DP210103865, DP210101042.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Conflict of interest

The authors have no relevant financial or non-financial interests to disclose.

Ethics statement

Project number 26071 Monash Animal Research Platform (MARP). All data collected for this study were obtained from animals used from Monash University and Cambridge University-approved projects and publicly available data from publicly available data from the Marmoset Brain Architecture Project (www.marmosetbrain.org), and the National Institutes of Health (USA) and National Natural Science Foundation (China) Marmoset Brain Mapping Project (www.marmosetbrainmapping.org). The materials and procedures used in the present study conformed with the Australian Code of Practice for the Care and Use of Animals for Scientific Purposes, which explicitly encourages the use of post-mortem tissue scavenged from animals used for approved research purposes.