Topographic gradients define the projection patterns of the claustrum core and shell in mice

Abstract The claustrum is densely connected to the cortex and participates in brain functions such as attention and sleep. Although some studies have reported the widely divergent organization of claustrum projections, others describe parallel claustrocortical connections to different cortical regions. Therefore, the details underlying how claustrum neurons broadcast information to cortical networks remain incompletely understood. Using multicolor retrograde tracing we determined the density, topography, and co‐projection pattern of 14 claustrocortical pathways, in mice. We spatially registered these pathways to a common coordinate space and found that the claustrocortical system is topographically organized as a series of overlapping spatial modules, continuously distributed across the dorsoventral claustrum axis. The claustrum core projects predominantly to frontal‐midline cortical regions, whereas the dorsal and ventral shell project to the cortical motor system and temporal lobe, respectively. Anatomically connected cortical regions receive common input from a subset of claustrum neurons shared by neighboring modules, whereas spatially separated regions of cortex are innervated by different claustrum modules. Therefore, each output module exhibits a unique position within the claustrum and overlaps substantially with other modules projecting to functionally related cortical regions. Claustrum inhibitory cells containing parvalbumin, somatostatin, and neuropeptide Y also show unique topographical distributions, suggesting different output modules are controlled by distinct inhibitory circuit motifs. The topographic organization of excitatory and inhibitory cell types may enable parallel claustrum outputs to independently coordinate distinct cortical networks.

Here, we systematically measured the density, spatial organization, and collateralization of claustrocortical projections to different cortical regions. In doing so, we found a continuum of overlapping claustrocortical modules organized primarily along the dorsoventral axis.
This topographical organization ensures that spatially distant and weakly connected cortical regions receive inputs from independent claustrum populations, while neighboring and connected cortical regions receive common claustrum inputs. Coupled with this output topography, we found interneurons containing somatostatin and neuropeptide-Y were spatially organized and exhibit a particularly dense labeling in the claustrum relative to surrounding cortical regions. Knowledge of these anatomical motifs will guide future experiments aimed at determining if distinct claustrum populations have unique roles in cognition.

| Tracer injection
Mice were administered carprofen via ad-libitum water 24 hr prior to surgery, and for 72 hr after surgery to achieve a dose of 5 mg/kg. For surgery, mice were initially anesthetized using 4% isoflurane and maintained at 1.0-2.5%. Mice were secured in a stereotaxic frame, with body temperature maintained through an electric heating pad set at 37 C. Local anesthetic (bupivacaine) was applied locally under the scalp, and an incision along midline was made to access bregma and all injection sites. The skin was moved back from the intended injection sites using sterile swabs and kept moist during surgery with sterile 0.9% saline. The skull was leveled between bregma and lambda. Craniotomies were marked and manually drilled using a 400 μm dental drill bit according to stereotaxic coordinates (Table 1), and the dorsoventral measurements made from brain surface. The left hemisphere was used for all injections unless otherwise stated. Pulled pipettes (10-20 μm in diameter) were back filled with mineral oil and loaded with tracers. All injections were made using pressure injection. The glass pipette was lowered into the injection site at 1 mm per minute, and 150-200 nl of each tracer was injected at 50-100 nl/min. The pipette was allowed to rest for 10 min after injecting before removal. Fast blue (Bentivoglio, Kuypers, Catsman-Berrevoets, Loewe, & Dann, 1980;Kuypers, Bentivoglio, Catsman-Berrevoets, & Bharos, 1980) (Polysciences, Pennsylvania) was prepared by dissolving 1 mg of powder in 30 μl 1X phosphate buffered saline (PBS) and 1.5 μl of dimethyl sulfoxide. The solution was warmed and agitated to fully dissolve and was stored at 4 C in 3 μl aliquots. Cholera Toxin subunit-B (Luppi, Aston-Jones, Akaoka, Chouvet, & Jouvet, 1995)  were obtained from Addgene, and aliquoted (3 μl) and stored at −80 C. Prior to surgery an aliquot was thawed on ice. The skin was sutured after completing all injections and sealed with vetbond (3 M).
Mice were returned to fresh cages upon regaining consciousness.

| Perfusion and tissue sectioning
Mice were deeply anesthetized and transcardially perfused 2-3 weeks after injections with ice cold PBS, followed by 4% paraformaldehyde (PFA) in PBS. Brains were extracted and postfixed in 4% PFA for 24-48 hr and stored in PBS at 4 C until sectioning. Brains were mounted in 2% agarose and sectioned at 50 μm using a vibratome (Leica VT1000s, Germany). Coronal sections were used for all brains. The entire brain was sectioned, and every second slice mounted on glass slides and sealed with coverslips using Prolong Gold (ThermoFisher). Slides were kept at 4 C until imaging.

| Immunohistochemistry
Mice were perfused and coronal sections obtained as above. Slices were first washed with 1X PBS (3 × 10 min) and then blocked using 2% PBST (3 x 10 min) and then 1X PBS (3 × 10 min), slices were mounted onto slides and cover slipped. Confocal images were obtained on a Leica SP5 or SP8 using a ×10, ×20, or ×25 objectives as described below. injections) were imaged in each brain. Images were taken at 2048 × 2048 pixels, accumulation = 2x, bidirectional x, pinhole set to 1 airy unit, and a z-stack of 4 images over a 12 μm volume were taken.

| Imaging
Each scan was set to image fast blue and AF-647 simultaneously, with EGFP and tdTomato imaged sequentially. Images were loaded into FIJI and converted to maximum intensity z-projection for analysis. We found that retrograde labeling of neurons in the claustrum was spatially sparse enough in the z-imaging plane to enable analysis using the maximum intensity projection (over this small volume), as the manual assessment of co-localization using multiple z-axis imaging planes or the maximum intensity projection yielded the same rate of co-projections on a subset of images analyzed.

| Analysis
Only pathways where the injection site was confirmed to reside in the target region were used for analysis. Before quantification, images were  to obtain a single value reflecting the relative connectivity strength between regions. The table contains source-target connectivity density information for most cortical regions. However, data for our ALM coordinate and different rostrocaudal levels of the RSP were not differentiated in this data. Therefore, for our cortical connectivity analysis, the RSP was considered a single structure, and ALM was not The spatial registration of the CLA PL pathway in this example image, using the CLA RSP pathway as a reference. The magenta polygon outlines the spatial extent of CLA RSP labeling (see Methods and Materials). Neurons inside/outside of the CLA RSP region are classified as core/shell, respectively. For each image, 10% of CLA RSP neurons most distant from the CLA RSP centroid were removed before calculating the claustrum core polygon, in order to reduce the effect of spatial outliers. (f) The average spatial density of CLA RSP neurons (magenta, left), and the density of all 14 claustrocortical pathways studied (cyan, middle). The overlay of the two plots shows regions classified as the dorsal shell, core, and ventral shell (far right). Otsu's method (see Methods) was used to calculate the boundaries of the core and shell for these density plots [Color figure can be viewed at wileyonlinelibrary.com] included. Consequently, 19 pairs of cortical regions were compared, rather than the original 27.

| Statistics
The mean and SD (across mice or slices) are shown in all figures, unless otherwise stated. Pairwise t tests or Wilcoxon rank sum tests were used and corrected for multiple comparisons with the Bonferroni correction. p-values of <.05 were deemed statistically significant.

| RESULTS
The claustrum was studied using a series of coronal brain sections from across the rostrocaudal axis (Figure 1a), giving high spatial resolution in the dorsoventral and mediolateral axes (see Materials and   (Dillingham et al., 2019;Druga, Chen, & Bentivoglio, 1993;Mathur, Caprioli, & Deutch, 2009;Wang et al., 2017;White et al., 2017;Zingg et al., 2018). Comparing these two markers in dorsoventral, mediolateral, and rostrocaudal axes, showed a highly correlated spatial overlap, indicating that both methods identify a common region of the claustrum (Figure 1a-c). Therefore, we chose to use the For each brain, one tracer was injected into the RSP at an intermediate location along the rostrocaudal axis (−1.5 mm from bregma), and all other tracers deposited into anatomically distinct areas of the cortex (Table 1) Table 1). Averaging across all brains we found retrograde labeling from these cortical injection sites showed a considerable spatial spread, beyond the border defined by CLA RSP and PV labeling (Figure 1f, middle). Thus, we adopted the term "core" and "shell" to provide coarse-grained classification of the spatial location of retrogradely labeled neurons (Figure 1f, right) in accordance with the core-shell nomenclature used previously (Atlan et al., 2017;Real, Dávila, & Guirado, 2006).  Table 2). However, co-projecting neurons were common among specific pathways including claustrocortical outputs to ALM/MOp, MOs/RSP, RSP/PL, pSUB/RSP, and pSUB/ENTm, whereas low co-projection rates were found in experiments labeling inputs to sensory cortex (SSbfd, AUDp, and VISp) (Figure 8a, c, e, g).
The upper limit on the detectability of co-projection between pairs of tracers was found to be~50-60% (Figure 9a-f), suggesting that rates of 10-20% indicate a high rate of co-projection given these methods.
As suggested by the data in Figure 8, the co-projection rate depended on the topography of individual claustrocortical modules ( Figure 10a). The co-projection rate was positively correlated with the spatial overlap between claustrum modules ( Figure 10b) and negatively correlated with the distance between downstream cortical targets ( Figure 10c). Therefore, spatially separated cortical regions, particularly in the rostrocaudal axis, receive input from largely interneurons was greater than PV (43.8 ± 8.23 cells/mm 2 ) (Figure 13f).
There was a 25% overlap of NPY and SST neurons, particularly in the claustrum shell, but only a 1.6% overlap between PV and NPY ( Figure 13g). We measured the ratio between different interneuron subtypes in the claustrum and in neighboring brain regions. The SST/PV and NPY/PV ratio was particularly high in the claustrum relative to other brain regions (Figure 13h) suggesting an inhibitory neuron signature that aligns more closely to that found in association cortex (Kim et al., 2017). This differential pattern of neuropil and cell body labeling through the claustrum suggests different claustrum output modules are differentially controlled by PV, SST, and NPY mediated inhibition (Figure 14).  The correlation between the spatial overlap of claustrocortical modules, and the percentage of co-projecting neurons for each pair of claustrocortical pathways. The spatial overlap between claustrum modules was calculated by dividing the area jointly occupied by both pathways by the sum total of both individual pathways. (c) The correlation between co-projection rate and the distance between cortical injection sites. (d) The correlation between co-projection rate and the average (bidirectional) connectivity between each pair of cortical regions. Cortical connectivity was estimated using the data from the Allen brain institute (Oh et al., 2014).  (Kitanishi & Matsuo, 2017;Macchi et al., 1983;Minciacchi et al., 1985;Sadowski et al., 1997;Smith & Alloway, 2014), while in other reports, very little topographical organization was identified (Sloniewski, Usunoff, & Pilgrim, 1986;White et al., 2017). Likewise, claustrum neurons have been reported to co-project to multiple cortical regions (Smith et al., 2012;Wang et al., 2019;Zingg et al., 2018), whereas in other instances little to no co-projections between different claustrocortical pathways were identified (Sloniewski et al., 1986;White et al., 2017).
The discrepancy between studies can be accounted for by several factors. First, claustrocortical mapping has usually focused on a small number of projections in each experiment. Therefore, differences in topography and co-projection rate would depend on the choice of cortical injection target. For example, injections in multiple areas of midline cortex would lead to a high rate of co-projecting neurons and a lack of topographical difference between pathways, whereas injections into temporal lobe and frontal motor areas would lead to low co-projection rate and major topographical differences in claustrum labeling. Our approach involved the study of multiple cortical injection sites and registering the data to a common pathway. This approach has not been used previously, but we find it is essential for accurate registration across experiments where small differences in the location of claustrocortical projection modules arise. Another issue giving rise to discrepancies between studies is the species-specific organization of claustrocortical projections. The original work on claustrocortical connections was performed in cat and primate which show clear topographical zones that project to specific areas of visual, auditory, and somatosensory cortex (LeVay & Sherk, 1981;Olson & Graybiel, 1980;Pearson et al., 1982;Remedios et al., 2010;Witter et al., 1988). However, in rodents, the majority of claustrum neurons project to association cortex, rather than primary sensory cortex (White et al., 2017;White & Mathur, 2018;Zingg et al., 2018). As mice may now provide an essential model system to study the function of the claustrum, the data we present here will enable neural activity of specific claustrocortical pathways to be manipulated or measured while taking into consideration the crosstalk with other projection streams. However, there are species differences in the anatomical organization of the claustrum (Edelstein & Denaro, 2004;Orman et al., 2017;Pham et al., 2019;Smith et al., 2019;Witter et al., 1988) and therefore the results in mice may not generalize to other species.
A recent study showed that the intrinsic electrical properties of PV, SST, and VIP interneurons in the claustrum were distinct from each other (Graf et al., 2020), similar to cortex. However, to the best of our knowledge, no study has tested or compared the connectivity of SST, NPY, or VIP cells with different claustrocortical connections. Future studies will be critical to test the hypothesis that different output streams are controlled by different inhibitory circuit motifs.
In conclusion, claustrocortical connections are comprised of several overlapping spatial modules arranged in a dorsoventral continuum, topographically aligned with separate cortical networks.
Claustrum neurons innervate many functionally related and anatomically connected cortical regions, but claustrum modules projecting to weakly connected and spatially diffuse cortical regions are non-overlapping. This organizational framework may enable distinct behaviors and brain states to be supported by independent claustrum circuits.

CONFLICT OF INTEREST
The authors declare that they have no conflicts of interest.
F I G U R E 1 4 Summary of the topographic mapping between claustrum and cortex. Claustrocortical projections are mainly organized across the dorsoventral axis which maps onto the rostrocaudal axis of the cortex. Interneuron subtypes are differentially localized to the core and shell of the claustrum. The lines linking claustrum and cortex highlight some of the most numerous and divergent claustrocortical pathways identified in this study [Color figure can be viewed at wileyonlinelibrary.com]

AUTHOR CONTRIBUTIONS
Brian A. Marriott designed the project, collected the data, analyzed the data, and wrote the manuscript. Alison D. Do and Ryan Zahacy collected data, performed immunohistochemistry, and edited the manuscript. Jesse Jackson designed the project, analyzed the data, wrote the paper, and supervised the project.

PEER REVIEW
The peer review history for this article is available at https://publons. com/publon/10.1002/cne.25043.