The anterior and medial thalamic nuclei and the human limbic system: tracing the structural connectivity using diffusion-weighted imaging

The limbic system is a phylogenetically old, behaviorally defined system that serves as a center for emotions. It controls the expression of anger, fear, and joy and also influences sexual behavior, vegetative functions, and memory. The system comprises a collection of tel-, di-, and mesencephalic structures whose components have evolved and increased over time. Previous animal research indicates that the anterior nuclear group of the thalamus (ANT), as well as the habenula (Hb) and the adjacent mediodorsal nucleus (MD) each play a vital role in the limbic circuitry. Accordingly, diffusion imaging data of 730 subjects obtained from the Human Connectome Project and the masks of six nuclei (anterodorsal, anteromedial, anteroventral, lateral dorsal, Hb, and MD) served as seed regions for a direct probabilistic tracking to the rest of the brain using diffusion-weighted imaging. The results revealed that the ANT nuclei are part of the limbic and the memory system as they mainly connect via the mammillary tract, mammillary body, anterior commissure, fornix, and retrosplenial cortices to the hippocampus, amygdala, medio-temporal, orbito-frontal and occipital cortices. Furthermore, the ANT nuclei showed connections to the mesencephalon and brainstem to varying extents, a pattern rarely described in experimental findings. The habenula—usually defined as part of the epithalamus—was closely connected to the tectum opticum and seems to serve as a neuroanatomical hub between the visual and the limbic system, brainstem, and cerebellum. Finally, in contrast to experimental findings with tracer studies, directly determined connections of MD were mainly confined to the brainstem, while indirect MD fibers form a broad pathway connecting the hippocampus and medio-temporal areas with the mediofrontal cortex.

. List of 29 thalamic nuclei according to [32] ordered in four nuclei group with their full name, abbreviations, and size in voxel.

Material and methods
Data. Data  The DWI data were obtained from the 3 T Human Connectome Project. The data were selected from the HCP-900 sample, and only those volunteers were chosen, who went through the full MRI acquisition pipeline of 2 structural, 4 resting state, 7 tasks and 1 DWI session (s. https ://proto cols.human conne ctome .org/HCP/3T/ imagi ng-proto cols.html). This selection resulted in 730 volunteers, a total of 329 male and 401 female subjects in the age range of 22-37 (693 right-handed and 37 left-handed subjects). Thalamus mask definition. As masks, we used the digital version of "Stereotactic Atlas of the Human Thalamus and Basal Ganglia" 32 , which contains a set of 29 thalamic nuclei assigned to four major groups (anterior, medial, posterior and lateral) (s. Table 1). The digital model of the 3-D anatomy of the thalamus was transformed into a thalamus connectivity-based probability atlas space 21,33 . The transformation has been described in detail in previous studies 13,34 . nuclei native space transformation. The 12 parameter affine transformation 35,36 was computed for each volunteer's non-diffusion image and the MNI-spaced standard brain. The resulting transformation matrix was applied to the left and right anterior thalamic nuclei to transform them into the native diffusion space. The nuclei transformation allowed further diffusion calculations into subject native space while maintaining high data quality and reducing registration interpolation errors 37 . Preprocessing and diffusion fit. The obtained HCP diffusion data were already reconstructed using a SENSE1 algorithm 38 . The DWI data was preprocessed within the HCP pipeline, including distortion correction [39][40][41] and motion correction. The color-coded FA maps were computed for each subject using FDT DT-fit tools and then visually inspected for data quality.

MR data specification.
Multi-shell reconstruction. The data were reconstructed using a sun grid engine to drive multiple CPUs and a GeForce GTX TITAN (Cuda 7.5 and compute capability 3.5) GPU. To enable bedpost processing on the referred GPU 42 , we used a custom-compiled version of the diffusion reconstruction tool, bedpostx_gpu 43 .
In both the CPU and GPU version, similar parameter settings were deployed to run the whole-brain multishell reconstruction. In the multi-shell model 44  connectivity distribution. Probabilistic tractography was performed using FSL-probtrackx 42 . The probability algorithm can build up a histogram of the posterior distribution on the basis of streamline location or connectivity distribution 42 . The entire analysis was performed using a CPU version of probtrackx on the sungrid engine and Nvidia Titan GPU using the GPU version of the code, i.e., probtrackx2_gpu 43 . The probtrackx parameters included the curvature threshold 80° (0.2), sample number 5000, step length 0.5, and a maximum number of steps of 2000. Each seed parcel's tractogram was confined to the ipsilateral hemisphere. In the direct diffusion tractography, all streamlines passing through other thalamic nuclei were excluded to depict only directly routed connections to the ipsilateral cortex. The resulting tractograms were each normalized by dividing them by the way total and multiplying them by 100.
Group fixed effect analysis. The B0 volume of each subject was registered to MNI 1 mm brain space. The resulting transformation matrix was then applied to the tractograms to align them in the MNI space. Group fixed effects analysis was performed across all subjects for each parcel. Line artifacts generated by the seed restriction mask were removed by manual masking. Finally, the fixed effect maps were thresholded at a level of 0.25 and visualized in FSLeyes. The rendered visualization of the group connectivity distribution was achieved using Freeview (with minimal threshold 0.1) 45 .
Anatomical Atlas label assignments. The threshold group fixed effect maps (≥ 0.25) were investigated for their specific cortical, sub-cortical and cerebellar projected assignments. Labeling (≥ 0.2 atlas thresholds) was performed using the Harvard-Oxford Cortical and Subcortical Structural Atlas 46-49 as well as the Jülich Histological Atlas [50][51][52] . The assignments were computed separately for left and right group maps.

Results
Anterodorsal nucleus AD. The anterodorsal nucleus AD is the smallest nucleus of the ANT (21 voxels).
It lies most medially and adjacent to the AV but is separated from AV, the stria medullaris thalami, and the ventricular surface of myelinated fibers and a glial lamella. Our results show that the major AD pathways reveal dominant projections to the left hemisphere. An overview of our results is given in axial, coronal, sagittal and 3D rendered views in Fig. 3 and described here in detail: The AD tracts connect: In the right hemisphere most tracts are much less pronounced, but in general take the same route anteriorly along the mammillary tract to the MB and via the anterior commissure to the hippocampus, where they are then more confined to medial aspects of the inferior temporal gyrus and project along the hippocampal gyrus without reaching its head and the amygdala. The posterior route along the body of the fornix and retrosplenial cortex into the lingual and occipital cortex is quite similar to the left hemisphere. However, on the right, the MTT tract is only recognizable at a lower threshold, and it remains unclear whether it also terminates in the VTG.
Anteromedial nucleus AM. The anteromedial nucleus AM is more ventrally and medially located than AD and appears as a ventromedial extension of AV with its lower part extending towards the third ventricle (s. Fig. 1). AM is also small in size (33 voxels). It is separated from AV by a thin fiber lamina and shows a somewhat higher density of myelinated fibers and cells. AM tracts cover similar diencephalic structures and areas as the AD. However, the tracts are much more extended in the frontal and occipital cortex and are also more prominent in the brainstem and the cerebellum. Nevertheless, again a left-sided dominance is maintained (s. Fig. 4). The AM tracts connect: 1. Anteriorly they run much like the AD, but more prominently along the FX and MT to the MB, again including the hypothalamus (HT), all septal nuclei, and also the nucleus accumbens (NA). They then connect: (a) bilaterally via the anterior commissure to the hippocampus and via the ansa peduncularis (AP) to the entire amygdala (AG), which is also right dominant connected via the stria terminalis (ST), (b) posteriorly running via the retrosplenial cortex (RSC) and inferior longitudinal fascicle (ILF) back medio-frontally and then via the parahippocampal gyrus to the hippocampus, and (c) from the RSC to the visual cortices. 2. In the frontal direction, the tracts propagate from the MB and the septum bilaterally via the medial forebrain bundle (MFB) and the anterior thalamic peduncle (ATP), and from the temporal lobe via the uncinate fasciculus (UF) to the orbito-frontal cortices (OFC) until their frontal poles. For the most part they cover the orbito-frontal and ventromedial cortices (Brodmann area 10, 11 and 25) with a preponderance to the left. The MFB contains fibers running in both directions from the olfactory apparatus and the HT via the DLF to the brainstem nuclei, interchanging fibers with many nuclei along its way 53  Anteroventral nucleus (AV). Despite its name, the anteroventral nucleus AV is more dorsally located than the AD and AM. AV is the largest ANT nucleus (94 voxels), and its cells are medium-sized, pale, and moderately densely packed. Anteriorly, it usually bulges into the lateral ventricle. Posteriorly, it is pushed deeper into the thalamus by the lateral dorsal nucleus LD. Compared to AM, the AV tracts show a similar but even a more extended pattern, again with a left-sided dominance (s. Fig. 5).
The major AV tracts connect: 1. Anteriorly like the AM along the FX and MT to the MB including the septal nuclei, the nucleus accumbens (NA), the preoptic area and the hypothalamus running via the anterior commissure to the hippocampus, including the whole amygdala (AG) and connecting more dorsally along the entire superior and medial temporal lobe via the fimbria hippocampi and inferior longitudinal fascicle. Habenula (Hb). In mammals, the Hb anatomically splits into two main subregions: the medial and lateral habenula, display distinct individual anatomical connectivity, see 55 . However, due to its small size, MRI usually cannot differentiate between the medial and lateral habenula 56 . Similarly, our Hb template was small (only 26 voxels) and contained both subdivisions. The significant tracts as visualized in our study are displayed in Fig. 7.
The major Hb tracts connect:

Mediodorsal nucleus (MD).
The mediodorsal nucleus is a large nucleus which is easily identified by histology 57 and in MRI 58,59 . It is located on the medial wall of the thalamus, abutting the third ventricle and possessing extensive connectivity to the prefrontal cortex, cingulate gyrus, and insula; see 57,60,61 . Based on cell morphology in rodents, MD can be divided into three different parts including medial MD, central MD, and lateral MD, with further subdivisions in primates; see 62 . In our thalamic template, MD-lacking further subdivision-was the largest nucleus (246 voxels). However, in contrast to the reported literature, our direct tracking  www.nature.com/scientificreports/ revealed that all major connections are devoted to the mesencephalon and brainstem (s. Fig. 8). Therefore, we also analyzed all indirect tracts by tracing all MD connections which are routed via other thalamic nuclei (s. Fig. 9).

Connections of the major direct MDd tracts:
1. From the nucleus, the tracts run anteriorly via the MT and the MB to the optic chiasm (OC) and from MB left laterally via the AC to the anterior temporal pole without connections to DG and HP. www.nature.com/scientificreports/ 2. Posteriorly they run via the optic tract (OT) and LGN to the SC and IC of the lamina tecti, most probably including the Hb. They then propagate prominently but predominantly on the left to the dorsal mesencephalon and the brainstem. 3. Medially a subtle tract connects the hypothalamus via the MTT to the VTG, where it then joins the major mesencephalic pathways.

Discussion
Although the mammillothalamic tract was already described in the eighteenth century by Vicq d' Azyr 63 , the thalamus was subsequently long neglected and regarded as simply being a part of the limbic system until Rose and Woosley 64 correlated the "limbic cortex" with specific thalamic nuclei, based on experimental findings in rabbits. These results were later confirmed in numerous studies in various species 65 . Our study continues in this vein, aiming to determine how and to what extent selected human thalamic nuclei are connected in vivo to brain midline structures, which have been assigned to the limbic system 66 since Broca (1878) 67 . An overview of the different connectivity patterns of the ANT nuclei is given in Fig. 10 and for all nuclei in Fig. 11. Moreover, Table 2 summarizes the major target structures for each nucleus according to the Juelich histology atlas, which is available electronically.
ANT. An evaluation of the connections reveals that all ANT nuclei are indeed associated with the limbic system, as they mainly include the hippocampal-diencephalic and parahippocampal-retrosplenial network dedicated to memory and spatial orientation (s. Fig. 10). In addition, AM, AV, and LD encircle the temporoamygdala-orbitofrontal network involved in the integration of visceral sensation and emotion with semantic memory and behavior 82 . More specifically, all ANT nuclei encircle the MB, HT, and AC on their route to the hippocampus, but differ in their extent to the pre-and subcommissural septal areas as well as in their projections to the temporal pole and the adjacent amygdala. The AD in particular is mostly confined to the left anterior temporal lobe, and its connection via the MTT and VTG to the MLF and the brainstem is subtle. AM and AV both display broad connections to the dorsal brainstem but differ with respect to their connections with the cerebellum (only AM) and with the cingulum (only in right AV and LD); only AV includes the perigenual anterior cingulate cortex-a part of the default mode network (DMN). LD tracts-similar to AM-involve the right cingulum but offer-similar to AD-only a subtle connection via the right MTT to the brainstem. LD, however, differs from the other ANT nuclei in that it has a projection to the left parietal cortex, an area commonly not assigned to the limbic system. However, van Groen and Wyss 68 reported after using anterograde tracer that in rats the lateroventral parts of LD project to the parietal cortex and that LD is attributed to spatial learning and memory 69 . This finding may correspond to the multimodal nature of the parietal cortex 70 .
Hb. In contrast to the ANT, before MacLean 71 the habenula was rarely recognized as part of the limbic system. Until today, the Hb is anatomically classified under the heading of the epithalamus [15][16][17] . The Hb is a highly conserved nucleus across vertebrates and has often been overlooked by neuroscientists. Its function was initially thought to be related to the regulation of the nearby pineal gland. Anatomically the Hb splits into two subregions in mammals: the medial and lateral habenula, which display distinct gene expression profiles and anatomical connectivity and hence are thought to subserve different functions. The medial habenula primarily receives input from the medial and lateral septal nuclei. Its output is almost entirely confined to the interpeduncular nucleus of the midbrain. On the other hand, the lateral habenula connects various structures, including the septum, hypothalamus, basal forebrain, globus pallidus, and prefrontal cortex, with the dopaminergic, serotonergic and noradrenergic systems 72 . Clinically Hb is of relevance to psychiatric disorders as a number of studies have associated it with dysregulated reward circuitry function, mood disorders, schizophrenia, and substance use disorder 9 . We found that the Hb mainly connects via the stria medullaris and fornix with the hypothalamus and orbitofrontal cortices and posteriorly with the tectum and visual cortices. Caudally it is in contact with all major mesencephalic and brainstem nuclei and connects left dominantly with the cerebellum. Overall, these connections are consistent with experimental findings, in which most afferents to the habenular nuclei arrive via the stria medullaris. Afferents arising predominantly in limbic brain regions are directly or indirectly innervated by www.nature.com/scientificreports/ ANT versus Hb and MD. In contrast to the ANT nuclei, the Hb tracts are sparse with respect to medio-temporal and occipital connections, but their brainstem and cerebellar connections are roughly as prominent as AM tracts (s. Fig. 11). The direct MD tracts are confined to the dorsal mesencephalon and spinal cord and lack significant cortical connections, in contrast to all other nuclei. However, allowing the MD tracts to reach the cortex via other nuclei reveals extensive bilateral connections to the medio-temporal lobe and the medio-frontal cortices (s. Fig. 9).

Methodological limitations.
This study had to deal with significant methodological problems, so the results are subject to major limitations. www.nature.com/scientificreports/ First, the atlas of thalamic nuclei which we used is based on six series of maps derived from stacks of histologically processed brain sections by combining three different series of the right and left hemisphere to construct a unique three-dimensional surface rendered model of 29 major thalamic nuclei 32 . Therefore, the anatomical templates for each nucleus apply to both hemispheres, but cannot be seen as a representative sample for a larger population since they do not take the normal structural variation and hemispheric differentiation into account [87][88][89] . However, with respect to the connections to the frontal and temporal lobe, our tracking results are consistent with other human probabilistic tractography approaches 6,90 .
The human brain possesses a considerable variable organization within both hemispheres. Such as, the occipital cortex is more extended in the left hemisphere compared to the right and reversely the frontal cortex 91 . Such a slight asymmetric organization also appears for the thalamus 34,92 . For example, the left thalamus is more extended in the posterior direction in contrast to the right thalamus. To reflect this slight variability we now added bilateral images of the six examined nuclei for 2 subjects as an additional image in the supplement (S1) and the bilateral tacking results as additional images in the supplement (S2-S7). In a visual comparison, the maps matched closely with the group fixed effect maps.
Secondly, although DWI tractography is an important tool for determining structural pathways of the whole brain in vivo, uncertainty exists about the evaluation of spatial accuracy and anatomical assignment due to the inter-subject variability 93,94 . Seed-based probabilistic tractography can yield a mix of multi-brain area projections, in which a pathway connects from one node to the next and so on. Thus, the path can follow the probabilistic maxima defined by trajectories to the multiple brain areas. Intuitively, it is possible to estimate the direct path by adding the anatomically constrained node information into the seed-based tractography. However, we still lack a valid and detailed map of subcortical and brainstem structures and pathways, which are based on larger samples. Nevertheless, we hope that the obtained results can serve as roadmaps for a more detailed connectivity profile of the limbic system and the involved thalamic nuclei.
Thirdly, the anatomical assignment, especially of subcortical structures is limited as significant differences exist in the nomenclature and concepts for naming subcortical tracts and nuclei due to historical and experimental reasons 95,96 . Nonetheless, the described whole brain ANT connections extend existing experimental and anatomical findings on the limbic system, in particular concerning the brainstem. It seems that the anterior thalamus serves as a gating and control unit that transmits elementary functions associated with vision, hearing, motor control, sleep and wake cycles, alertness, and vegetative function from various brainstem centers to the diencephalon as well as to temporal and frontal cortices 11,12,97-100 . interpretation. The interpretation of our results is imperfect in two ways. First, we had to link the connection between two different neuroscientific items: the thalamus-a circumscribed anatomical structure-and the limbic system-for the most part a behaviorally defined system whose components have evolved and increased over time 4,82 and which is not universally accepted as a separate entity in the neurosciences [101][102][103] . Secondly, we had to compare our macroscopic whole brain results from a large sample of subjects with particular microscopic findings mainly determined in animals. The classical limbic circuit of Papez defines a loop from the hippocampal formation (dentate gyrus and subiculum) and parahippocampal gyrus via the post-commissural fornix (FX) to the mamillary body (MB) and then projects (a) via the mammillothalamic tract (MT) to the ANT 63 , (b) from ANT via the subcallosal and precallosal portions of the septum to septal and preoptic areas and via the cingulate bundle (CB) to the cingulate gyrus (CG), and (c) from the CG via the fimbria hippocampi and tractus perforans back to the hippocampus 5,12,104-106 . MacLean 107,108 has widened this concept for a unitary model of the limbic system by incorporating both the Papez circuit and Yakovlev's view 109 of an amygdala-orbitofrontal network to develop a concept of the `visceral brain` considering that stimulation of the cingulate cortices can evoke autonomic changes that are linked to emotion (s. Fig. 2). As the ANT nuclei are considered as part of the limbic thalamus and a central component of the circuit of Papez 10 with extensive direct and indirect hippocampal-anterior thalamic connections [11][12][13] , we were stimulated to analyze their connection profiles in a whole-brain approach. The inclusion of Hb and MD nuclei in our study is based on the fact that the Hb, due to its unique position, serves as a crossroad between the forebrain and midbrain regions [110][111][112] and acts as a critical neuroanatomical hub that connects and regulates motivated behavior, affective states, cognition, and social behavior. Similarly, the MD serves as a primary cortical relay for the limbic system in offering major connections to the prefrontal cortex 113-115 . conclusion This study presents the first approach in humans to examine and verify the structural connectivity between diverse components of the limbic system and selected thalamic nuclei in a whole-brain approach. Despite methodological discrepancies between diffusion-guided fiber tracking and experimental connectivity studies using ante-and retrograde tracers in animals, we were able to confirm that the ANT, Hb, and MD nuclei connect-to different extents-to major limbic components and the mesencephalon and brainstem. While ANT nuclei broadly connect the hypothalamus, septal and prefrontal areas via fornix and cingulum with the retrosplenial area and the hippocampus, the Hb links the hypothalamus and orbitofrontal cortices via the stria medullaris and fornix to the tectum and visual cortices. Finally, only indirect MD tracts (i.e., with connectivity via other thalamic nuclei) show extensive bilateral connections to the medio-frontal cortex. The tracts of the six nuclei examined will be made available at https ://githu b.com/vinkr ishna /Limbi c_Thala mus.

Data use of the Human Connectome Project
The study was performed in agreement with the WU-Minn HCP Consortium Open Access Data Use Terms of the Human connectome project. The study used datasets from the Human connectome project (HCP). We obtained