Functional imaging studies of the last decade have thoroughly investigated the importance of cortical networks in highly differentiated subfunctions of human behavior. Interaction, co-play and functional alterations of these networks have been related to various higher order cognitive functions, behavior and disease. While a continuing debate on segregated networks generally focuses on cortical regions (Barabasi and Albert, 1999; Hagmann et al., 2008; Bullmore and Sporns, 2009), the interaction of subcortical structures—foremost thalamus and basal ganglia—with these cortical networks, their influence and control has hardly been investigated and therefore remains poorly understood. Such a cortico-subcortical interaction seems, however, likely given anatomical and stimulation studies in animals (for review see Haber, 2003) and post-mortem studies in humans (reviewed by Romanelli et al., 2005) and has been previously suggested by Alexander and colleagues (1986). Moreover, cortical-subcortical interaction have been indicated to be especially important for fear-conditioning and emotional behavior (e.g., LeDoux, 1993; Price and Drevets, 2010). Due to the invasiveness of most functional investigations of such cortico-subcortical networks, studies in humans are so far limited to methods such as functional magnetic resonance imaging (fMRI) or—in very rare cases—deep brain stimulation (DBS). For fMRI, insufficient spatial resolution in most studies limited the interpretation of thalamic activation, while continuous innovation in high resolution fMRI (hr-fMRI) now enables the functional investigation of small, anatomically well-described subcortical structures including the thalamus—also in humans. The thalamus has been described as the “gate to the cortex,” encompassing well-characterized, anatomically and functionally segregated nuclei as described in a massive amount of animal (Alexander et al., 1986; McFarland and Haber, 2002; Zikopoulos and Barbas, 2007; Xiao et al., 2009) and human post-mortem studies (Cullen et al., 2003; Danos et al., 2003; Pakkenberg et al., 2009). From these studies, it is known that thalamic nuclei are embedded in so-called cortico-striatal-thalamo-cortical circuits or “loops” (Alexander et al., 1986), making the thalamus a major relay station for cortical input.

The high specificity and behavioral relevance of these circuits in humans makes most invasive neuroanatomical approaches in the human central nervous system impossible and the study of these circuits thus highly depends on translations between species. With the aid of the new techniques of hr-fMRI, the functional investigation of such cortico-subcortical interactions and related follow-up studies are now possible. This review focuses on the description and discussion of specific thalamic imaging findings in high-resolution fMRI in humans and their embedding in brain network interaction by giving the reader a large overview on specific and unspecific thalamic findings in animals and humans.

Previous descriptions of thalamic activation can be found in a large number of human imaging studies of cognition (Minzenberg et al., 2009; Ide and Li, 2011; Li et al., 2012), emotion (Mouras et al., 2003; Moulier et al., 2006; Kober et al., 2008; Kessler et al., 2011; Kühn and Gallinat, 2011, 2012; Sabatinelli et al., 2011; Grimm et al., 2012), health and disease (Dickstein et al., 2006; Maisog et al., 2008; Ide and Li, 2011; Passamonti et al., 2011).

While cortico-subcortical circuits have been hypothesized to be involved in very diverse functions like movement, attention, cognition and emotion (Alexander et al., 1986), these integrated loops may account for different facets of basic human behavior. Sexual processing comprises a suitable example paradigm to investigate such diversity, given its characteristic multidimensionality. Spanning various functional systems, brain areas, and networks, processing of sexually salient information involves cognitive, motivational, emotional, and autonomic processes, which contribute to the subjective experience of sexual arousal (Redouté et al., 2000).

We will therefore in the next section link findings from animal anatomical and human high-resolution studies to explain functionality and anatomy of thalamic nuclei involved in sexual processing. For (1) the mediodorsal nucleus and (2) the centromedian/parafascicular complex the animal anatomical and functional literature will be described, followed by related functional studies in humans at low field strength and consecutive studies at high-field strength. For anatomical guidance, all described nuclei are visualized in Figure 1.

Thalamic connections have been extensively described in animals and are revised in the previous paragraph sorted by nuclei (Reep and Winans, 1982a,b; Mufson and Mesulam, 1984; Royce and Mourey, 1985; Vogt and Pandya, 1987; Vogt et al., 1987; Cornwall and Phillipson, 1988; Nakano et al., 1990; Berendse and Groenewegen, 1991; Sadikot et al., 1992).

A parcellation of the human thalamus based on in vivo imaging has developed from a rough division based on anatomical and functional connectivity (Johansen-Berg et al., 2005; Fair et al., 2010; Zhang et al., 2010) over smaller more accurate parcellations (Klein et al., 2010; Kim et al., 2012) to parcellations on single nuclei level (Eckert et al., 2011). These nuclear parcellations might have the closest link to the observations in animals.

The following paragraph will give an overview of the development of anatomical thalamic parcellation in human imaging studies over the last years.

As one of the first studies, Johansen-Berg and colleagues divided the thalamus based on fiber connections with large predefined cortical areas like the prefrontal cortex, temporal- and occipital cortex and three specific gyri like the somatosensory-, premotor- and motor cortex (Johansen-Berg et al., 2005).

A similar approach was used by Zhang and colleagues, when dividing the thalamus by its functional connectivity in resting state fMRI with the same cortical lobes and areas and comparing it to fiber counts assed by diffusion tensor imaging (DTI) in the same subjects and a human histological talairach atlas (Zhang et al., 2010). Both studies by Johansen-Berg and Zhang used very large cortical areas, therefore receiving a very rough thalamic parcellation. However, these studies added to the field in confirming anatomical knowledge from animal and human post-mortem studies by using non-invasive brain imaging in humans in vivo by two commonly used brain imaging techniques—namely DTI and resting state fMRI.

In line with these large-scale parcellations, Fair and colleagues specified changes of thalamic connectivity with these six cortical divisions over development and livespan (Fair et al., 2010).

Based on resting state fMRI data, Kim and colleagues parcellated the thalamus and basal ganglia in various functional subdivisions (Kim et al., 2012). Therefore, they used an independent component analysis, which groups all subregions in the brain according to their similarity of behavior during resting state. Despite using a low resolution, this study could find a functional thalamic parcellation, similarly diverse as described in human post-mortem atlases.

All three studies revealed preferential connectivity of the mediodorsal thalamic portion with the prefrontal cortex. This relationship was specified by Klein and colleagues, who divided the mediodorsal thalamus according to its preferential connectivity with subdivisions of the prefrontal cortex in monkeys and humans (Klein et al., 2010). They confirmed a specific diversity within the mediodorsal thalamic nucleus, which was previously reported in animal and post-mortem studies (Haber and McFarland, 2001; Haber and Calzavara, 2009).

Based on the findings by Metzger and colleagues, the anatomical underpinnings of the differential functional connectivity of MD and CM/PF under task circumstances were investigated by DTI (Eckert et al., 2011). Looking at preferential anatomical connectivity between MD, CM/PF and cortical as well as subcortical targets, Eckert and colleagues described a relative preference of fiber tract distributions for both nuclei in humans. While MD was preferentially connected with cortical structures (with the exception of the caudate nucleus), CM/PF showed preferential connectivity with subcortical structures, with the exception of the anterior insula (Figure 2).

Based on these preferential connectivities, this study could indicate the structural basis of putative networks, with MD connecting to prefrontal cortical areas and CM/PF interconnecting with subcortical areas. Therefore the authors suggested MD and CM/PF as two important neighboring hubs of structurally and functionally segregated networks in humans. Their result supported the segregation of thalamic nuclei into functional units of distinct cortico-subcortical systems. In this view, the involvement of the cingulate cortex and the anterior insula in distinct attentional processes like attentional set and switching of attention (Yamasaki et al., 2002; Menon and Uddin, 2010), supports the idea of cortico-striatal-thalamo-cortical loops involved in segregated but integrated circuits of information processing.

The described developments in the field of human thalamic functional connectivity in vivo, from unspecific findings in activation studies to more and more detailed parcellations, form the bases for ultra high resolution studies, which describe specific thalamic activation and therefore function, together with specific connectivity of certain nuclei and their integration into cortico-striatal-thalamo-cortical loops. The following paragraph gives an overview of findings in the animal literature on thalamocortical loops, and related human studies in vivo at low and high field strength.

The concept of segregated but integrated circuits, forming functional subsystems or cortico-striatal-thalamo-cortical “loops,” was suggested by Alexander and others, mainly based on animal studies (1986).

These cortico-striatal-thalamo-cortical loops connect functionally segregated, higher order cortical regions with the basal ganglia and form functional entities for motor and sensory processes as well as for higher cognitive functions like cognitive and emotional processes. This concept has been further put forward by research in animals, healthy humans, and diseased subjects (Price, 1999; Haber and Calzavara, 2009).

In humans such a distinction of segregated networks has been described on cortical level between two major networks, namely the default mode- and the attention network, comprising several subnetworks (Buckner et al., 2008; Biswal et al., 2010).

The attention network—or task-positive network—is always active, when a cognitive or attentionally demanding task is performed. In contrary, the default-mode network—or task-negative network is related to self-referential, interoceptive processing and is more active when no specific task is required and has been associated with increased mind-wandering. Both networks are anticorrelated, meaning high activity in one network is related to low activity in the other and vice versa.

Characteristic hubs of the task-positive network lie in the inferior parietal sulcus, superior parietal lobule and frontal eye-field for top-down control, in the temporo-parietal junction and ventral frontal cortex for stimulus-driven attention and in the anterior insula and the dorsal anterior cingulate cortex for continuous attention and salience processing (Corbetta and Shulman, 2002; Seeley et al., 2007; Dosenbach et al., 2008), while the task-negative default mode network encompasses parts of the medial prefrontal cortex (BA10), the posterior cingulate cortex extending toward the precuneus as well as temporo-parietal and lateral temporal regions.

While in humans such networks have been extensively described on cortical level, not so much is known about subcortical hubs orchestrating these cortical networks.

As there is striking evidence from a large number of animal studies that anatomical thalamic parcellation into thalamic nuclei also mirrors functional diversity (Groenewegen and Berendse, 1994; Haber, 2003), such functional integration of thalamic nuclei into cortico-striatal-thalamo-cortical loops has been proposed.

The study by Klein and colleagues focused on connectivity between anatomically predescribed subparts of the mediodorsal thalamic nucleus and subregions of the prefrontal cortex using diffusion tensor tractography (Klein et al., 2010). Using DTI in macaque and humans, they could confirm a topographical division within the MD thalamus for preferential prefrontal connectivity previously described by tracer studies in monkeys (Ray and Price, 1993). This study, however, focused on anatomical connectivity patterns. To assess functional network segregation on thalamic and basal ganglia level, more functional studies, involving both default mode- and attention network, looking at predefined thalamic nuclei would be essential. Therefore, the study by Metzger and colleagues focused on functional connectivity during task performance of specific thalamic nuclei like MD, CM/PF, VA and LD/PT, addressing the question of functional network affiliation of these nuclei (Metzger et al., 2010). They found a coactivation of CM/PF with the task-positive salience network, while MD coactivated with more rostral parts of the anterior cingulate cortex, namely the pregenual anterior cingulate cortex (pgACC), which was previously characterized to show default-mode like behavior in functional studies (Northoff et al., 2007; Grimm et al., 2009; Walter et al., 2009; Biswal et al., 2010). Moreover, and in line with the described findings by Fair and colleagues (2010), the VA showed a coactivation with the premotor cortex, also supported by anatomical VA connectivity patterns in animals (Katayama et al., 1986; Fang et al., 2006). This finding of cortico-striatal-thalamo-cortical integration was further substantiated by the anatomical findings of preferential connectivity of MD and CM/PF with the predefined network hubs like dACC, pgACC and insula (Eckert et al., 2011).

Functional connectivity, substantiated by anatomical connectivity of these cortical network hubs together with distinct thalamic nuclei and basal ganglia components thus support the distinction of affective and cognitive cortical subdivisions based on segregated cortico-striatal-thalamo-cortical loops (Devinsky et al., 1995; Bush et al., 2000).

While these studies form a first step toward the discovery of cortico-striatal-thalamo-cortical integration in humans, further evidence by functional and connectivity studies is needed to substantiate these first results.

The concept of functional integration of segregated thalamo-cortical loops—and therefore thalamic nuclei—could for a long time not be substantiated by human imaging findings. Therefore, clinical concepts of thalamo-cortical integration were often based on animal work, post-mortem or human lesion studies, but lacked the integration of human functional (imaging) findings. As one example, the controversial findings of stronger or weaker functional connectivity between different parts of the cingulate cortex and “the thalamus” as a whole—as described in various patient populations (Anand et al., 2005; Greicius et al., 2007; Minzenberg et al., 2009; Walter et al., 2009)—could not be traced back to distinct thalamic nuclei. Supported by animal studies, some thalamic structures are, however, targeted by deep brain stimulation in various diseases like depression, obsessive compulsive disorder (see Lujan et al., 2008 for an extensive review) and Parkinson's disease (Fasano et al., 2012).

While circumscribed anatomical localization of thalamic structures still remains a challenge in MRI-guided neurosurgery, thalamic nuclei represent anatomically well-defined structures with small inter-subject variability—as compared to that found to a much larger extent in cortical brain regions. As one example, Sartorius and colleagues recently showed an antidepressant effect of electrode placement near the lateral habenular complex for deep brain stimulation purposes (Sartorius and Henn, 2007; Sartorius and Meyer-Lindenberg, 2009; Sartorius et al., 2010).

The lateral habenular complex neighbors the thalamic structures described above, and deep brain stimulation is not restricted to one point in the brain, but rather acts in a radius within a sphere of brain tissue. Therefore, the functional diversity and connecting fibers of neighboring thalamic nuclei in humans requires a larger knowledge.

As for MD and CM such functional diversity has been introduced as described in the previous paragraphs, this sheds a different light on the explanation of effectivity of thalamic targets in deep brain stimulation. While CM is a thalamic region targeted to treat epilepsy with the help of deep brain stimulation (Cukiert et al., 2009), its stimulation has also been described as an effective treatment option against Parkinson's disease (Stefani et al., 2009) and a promising target in Tourette's syndrome, movement disorders (Krauss et al., 2002; Ackermans et al., 2011) and pain syndromes (Weigel and Krauss, 2004).

As thalamus and the basal ganglia form the major targets of neurostimulation in humans—the investigation of their integration into cortico-striato-thalamo-cortical loops seems crucial not only for the optimization of current treatment practice, but also for the development of new treatment options. With the help of high-resolution imaging at ultra-high fields, such crucial knowledge can now be gained (Duchin et al., 2012) and provide new insights in human brain function during health and disease.

Based on the finding by Eckert and colleagues, Osoba and colleagues looked at differences in MD and CM fiber tracts between patients with major depression and healthy controls (Osoba et al., 2013). As stated above, the MD was more closely linked to emotional perception, while attentional processing was more associated with CM thalamus (Metzger et al., 2010). As both emotion and attention processes have been found abnormal in MDD patients, the comparison of fiber tracts of MD and CM with their putative preferential targets seemed quite intuitive, especially as altered function and histology of the MD has been previously shown in MDD (Drevets et al., 1992; Steiner et al., 2008). Indeed, the distinction of structural disconnections of CM and MD turned out as a meaningfull level of observation. Osoba and colleagues reported significant differences for fractional anisotropy (FA) in an MDD group compared to healthy controls, however, only in tracts of interest that were defined by the previously reported preferential networks of MD and CM (Eckert et al., 2011). Reductions in FA were thus restricted to those tracts originating from CM or MD, which connected them with cortical and subcortical target structures that would also preferentially connect to these nuclei in healthy controls.

At sufficiently high resolution, consistency between human and animal anatomy and function increases. Therefore, and for the fact that high resolution imaging is reliable and feasible, it should be aimed to more and more implement high resolution imaging in human basic and clinical imaging—especially as the anatomical basis for human stimulation studies, is built on models on nuclei level. Current work at spatial resolutions of 2 mm³ voxel size or lower provided support for a meaningful level of observation that is not accessible at lower resolutions. This applied for both basic physiological studies in healthy populations as well as for patient populations. The functional distinctions, revealed by such experiments further converged with structural findings, which, also at lower fields are able to investigate thalamic connectivity at the level of anatomically separable nuclei. Especially when higher resolution is needed for subcortical structures, group studies with an acceptable between subject alignment seem possible. When investigating cortico-striato-thalamo-cortical circuits, future research should thus incorporate the notion that such spatial resolution, so far sufficient to discern cortical subregions of larger extent, need to be adapted for the smallest functional units included in the circuit. This should enable and encourage more investigations that focus on related thalamo-cortical networks in humans and should help to disambiguate findings arising from studies with insufficient resolution for subcortical network components.