Human thalamocortical connections and their involvement in language systems.

Human thalamocortical connections and their involvement in language systems Doctoral Thesis by: Mengxing Liu Supervised by: Dr. Pedro M. Paz-Alonso Dr. Garikoitz Lerma-Usabiaga 2022 (cc)2022 MENGXING LIU (cc by-nc-sa 4.0) Mengxing Liu All rights reserved. BCBL Basque Center on Cognition, Brain and Language Paseo Mikeletegi, 69, Donostia-San Sebastian, Spain 1 This work was supported by grants from the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie (grant agreement No. 713673), and from “la Caixa” Foundation (No. 11660016). 2 Human thalamocortical connections and their involvement in language systems Doctoral Thesis by: Mengxing Liu Supervised by: Dr. Pedro M. Paz-Alonso Dr. Garikoitz Lerma-Usabiaga 2022 3 Acknowledgement Thanks to Kepa, for recruiting me at first, and continuously teaching and training me. In the past five years, you keep replying that it is your job every time I say thank you to you. But I know you have done way more than your job asks you to do. If one day I managed to become a qualified researcher and start mentoring students, I hope I could be a mentor like you. Thanks to Gari, for hosting me in a foreign country, and for all the training from you. I learned a lot from you. Thanks to la Caixa Foundation, for the generous fellowship in exchange for only mentioning their names in my publications. Thanks to the admin and lab groups in the BCBL, for all the help they have provided to make my life and research here easier. 4 Content Acknowledgement 4 Content 5 List of abbreviations 6 1 Resumen en castellano 9 2 Abstract 14 3 Background and motivation 18 4 Thalamus 22 4.1 General introduction of the structure 22 4.2 Thalamic atlases and nuclear divisions 23 4.3 Thalamic nuclei segmentation methods 26 4.4 First-order relay thalamic nuclei 29 4.4.1 Lateral geniculate nucleus 30 4.4.2 Medial geniculate nucleus 31 4.4.3 Ventral lateral nucleus 33 4.5 Higher-order relay thalamic nuclei 35 4.5.1 Anterior nuclear complex 35 4.5.2 Mediodorsal nucleus 37 4.5.3 Pulvinar 38 5 Neurobiology of language 41 5.1 History 41 5.2 Modern neuroanatomical models 45 5.2.1 Modality-general models 45 5.2.1.1 Price’s model 45 5.2.1.2 MUC model 47 5.2.1.3 Lau’s model 48 5.2.2 Reading models 49 5.2.3 Speech comprehension models 50 5.2.4 Production models 52 6 Study 1: Structural connection of first-order thalamic nuclei 55 6.1 Methods 56 6.1.1 Subjects 56 6.1.2 Data acquisition 57 6.1.3 Tractography pipeline 58 6.1.3.1 ROI definition 58 6.1.3.2 DWI data preprocessing 60 6.1.3.3 Tract identification and tractometry 60 5 6.1.4 Reproducibility measurement 62 6.2 Results 64 6.2.1 Computational reproducibility 66 6.2.2 Test-retest reproducibility 68 6.3 Discussion 69 7 Study 2: Structural connectivity of higher-order thalamic nuclei: Anterior thalamic complex and mediodorsal nucleus 73 7.1 Methods 75 7.1.1 Subjects and data acquisition 75 7.1.2 Tractography pipeline 75 7.1.2.1 ROI definition 76 7.1.2.2 DWI data preprocessing 77 7.1.2.3 Tract identification and tractometry 77 7.1.3 Reproducibility measurement 78 7.1.4 Post-hoc analysis 78 7.2 Results 79 7.2.1 Computational reproducibility 84 7.2.2 Test-retest reproducibility 85 7.2.3 Post-hoc analysis results 86 7.3 Discussion 87 8 Study 3: task-based fMRI study of thalamic involvement in human language systems 93 8.1 Methods 93 8.1.1 Participants 93 8.1.2 Materials and Experimental Procedure 94 8.1.3 MRI data acquisition 96 8.1.4 MRI data analysis 96 8.2 Results 99 8.2.1 Whole-brain contrasts 99 8.2.2 ROI results 100 8.2.3 Functional connectivity results 102 8.2.4 Structural connectivity 105 8.3 Discussion 105 9 General Discussion 111 10 Bibliography 116 6 List of abbreviations A1 primary auditory cortex ACC anterior cingulate cortex AD axial diffusivity AN anterior nuclear complex ANOVA analysis of variance AR acoustic radiation BCBL Basque Center on Cognition, Brain and Language BOLD blood-oxygen-level-dependent CSD constrained spherical deconvolution dlPFC dorsolateral prefrontal cortex DT dentatothalamic tract DTI diffusion tensor images/imaging DWI diffusion-weighted images/imaging fMRI functional magnetic resonance imaging FODs fiber orientation distributions FoV field of view FWE family wise error FWHM full-width half-maximum GLM general linear model HCP human connectome project HRF hemodynamic response function IFG inferior frontal gyrus LGN lateral geniculate nucleus M1 primary motor cortex MD mediodorsal nucleus MGN medial geniculate nucleus mPFC medial prefrontal cortex MR motor radiation MRI magnetic resonance imaging MUC Memory-Unification-Control OR optic radiation PFC prefrontal cortex pSTG posterior superior temporal gyrus ROI region-of-interest 7 RTP2 reproducible-tract-profiles SMA supplementary motor area SMG supramarginal gyrus T1 T1-weighted structural image TE time-to-echo TR time-to-repetition V1 primary visual cortex V2 secondary visual cortex VLa anterior ventral lateral nucleus VLN ventral lateral nucleus VLp posterior ventral lateral nucleus 8 white-matter tracts and tractometry. We then used computation and test-retest methods to check whether our protocol could reliably reconstruct these tracts of interest and their profiles. Our results demonstrated that the protocol had nearly perfect computational reproducibility and good-to-excellent test-retest reproducibility. This new protocol may be of interest for both neuroimaging and clinical research, and it has been made publicly available to the scientific community. In contrast to the first-order relay nuclei, the “second-order” or “higher-order” relay nuclei of the thalamus are reciprocally interconnected with the cerebral cortex via cortico-thalamo-cortical pathways. For example, the mediodorsal thalamus is interconnected with practically the entire prefrontal cortex (PFC). In addition, it also receives afferents from the anterior temporal lobe and amygdala. The anterior nuclei have interconnections with the cingulate cortex and the retrosplenial cortex. Due to the complex structural connections, there are little tractography studies being able to reconstruct those higher-order thalamic white-matter tracts. In the second study, we focused on the white-matter tracts of the anterior and mediodorsal nuclei of the thalamus and developed a reconstruction protocol to obtain the tracts originating from or terminating at these nuclei. We also tested the reliability of the reconstruction protocol on a relatively large dataset. This protocol has proved to be able to reconstruct those white-matter tracts reliably, with only tracts with specific imaging or anatomical characteristics showing relatively lower reproducibility. A post-hoc analysis was conducted to explore the association of specific imaging and anatomical characteristics with reproducibility. The results revealed a strong negative correlation between both diffusion imaging data noise and tract length with reproducibility, and a positive correlation between streamline count and reproducibility. This protocol is publicly available for both research and clinical use. The reproducibility results also opened new avenues for future studies; for example, for systematically examining the possible factors that could have a stronger impact on tract reproducibility. The first-order thalamic nuclei are traditionally believed to relay sensorimotor information from the periphery and cerebellum to cortical regions. Recent work has shown that the engagement of the first-order thalamic nuclei can be modulated by the cortical regions (Andolina et al., 2007; Cudeiro & Sillito, 2006; von Kriegstein et al., 2008). Neurobiology of language research has been 15 focused on the cortical regions (aside from a few exceptions) and has ignored the contributions of subcortical structures such as the first-order thalamic nuclei. In the third study, we investigated the involvement of the first-order thalamic nuclei in language processes in a task-based functional MRI experiment. In this experiment, the participants performed language tasks that rely on different sensorimotor systems: reading (visual pathway), speech comprehension (auditory pathway) and speech production (motor pathway). These three linguistic tasks rely on different sensorimotor systems recruiting first-order thalamic nuclei during perceptual and motor information processing. In addition, we also included three non-linguistic tasks that were parallel to the linguist tasks: seeing scrambled images (visual pathway); listening to noise audios (auditory pathway) and producing unintellectual sounds (motor pathway). We found modality-specific engagement of the first-order thalamic nuclei in both linguistic and non-linguistic tasks. More importantly, the results revealed a modulation in the engagement of both the visual and auditory thalamic nuclei as a function of the linguistic versus non-linguistic nature of the stimuli. For example, the left lateral geniculate nucleus (LGN) showed stronger activation for reading real words than for seeing scrambled images. This modulation was not observed in the right thalamus. We also found strong functional coupling between the first-order thalamic nuclei and their primary cortical regions for tasks associated with their corresponding modalities. These results suggest a segregation in the implication of different human thalamic sensorimotor nuclei in the main human language systems, and that these nuclei exhibit a functional preference for linguistic versus non-linguistic stimuli. This work raised the possibility that the first-order thalamic nuclei react adaptively to the high level cognitive information of the sensory input. So far very little is known about the role of subcortical structures in high-level cognitive functions. The current study began to extend the traditional view of human language mechanisms from cerebral cortex to a broader scope and started to acknowledge the possible contributions of the thalamus in language processing. Altogether, the current thesis investigated the structural connections between the thalamus and the cerebral cortex, and proposed two reconstruction protocols that are available to the scientific community to obtain reliable first-order and higher-order thalamic white-matter tracts. We also 16 examined some possible factors that are linked with tractography reconstruction reproducibility and provided evidence that some tracts can be less reproducible when having specific imaging or anatomical characteristics. Finally, we showed the involvement of the first-order thalamic nuclei in language processing in a task-based functional MRI experiment, and argued the necessity of considering the role of subcortical structures, like the thalamus, when investigating the neural basis of human language function. 17 3 Background and motivation Current views of cortical function and experimental approaches to understanding the mechanisms of high-level cognitive functions, such as human language, are heavily dominated by what can be described as a corticocentric view. In this view, the cerebral cortex is believed to have the most important role in high-level cognition and behavior, whereas the subcortical structures are seen to have a subservient role, or no role in these functions (Parvizi, 2009). This notion is prevalent enough to raise a concern. In a survey in 2008 about corticocentric trends in the neuroscience field, it was found that 72% of studies in current neuroscience research in which subcortical structures could have been the focus of the study, ended up being ignored. Also, 50% of studies in which subcortical findings were reported chose not to discuss the findings about subcortical structures (Parvizi, 2009). Lacking knowledge about the nature of the connectivity between cortical and subcortical structures might be the main cause of the corticocentric view. As an example, the thalamus has long been referred to as a passive relay, a necessary link in the flow of information from the periphery to the cerebral cortex. The main responsibility of the thalamus is to transmit perceptual information (visual, auditory, somatosensory) and information in other forms (motor instructions from the cerebellum). But this only accounts for a small part of the thalamus (Sherman, 2007). The largest nucleus of the thalamus, the pulvinar, has extensive connections with the visual cortex and also with the extrastriate cortex. It was proved to receive afferent information from the layers V and VI of these visual areas topographically and project back to the superficial layers of these areas (Guillery & Sherman, 2002). These cortico-thalamo-cortical pathways are not only used to relay peripheral information to the cerebral cortex, but rather play a key component in cortico-cortical communications. The pulvinar is not the only structure in the thalamus that receives afferent information from the cerebral cortex. The LGN of the thalamus is the relay of visual information from the retina to the visual cortex, but also receives modulatory axons from layer VI of the primary visual cortex (Sherman & Guillery, 2009). The exact function of the modulatory axons from layer VI of the primary visual cortex to the LGN is still not clear, but it is a useful example of why the thalamus should not be treated as a mere relay. 18 The interconnections between the thalamus and the cerebral cortex are key to understanding thalamic function. An early method to obtain thalamocortical connections was to study the retrograde degeneration of thalamic cells when cortical structures were damaged by local lesions (for a brief introduction, see Sherman & Guillery, 2009). For example, if a retrograde degeneration of the LGN of the thalamus was observed along with a limited lesion in the visual cortex, earlier investigators could speculate that this structure is interconnected with the visual cortex. This method has now been superseded by more modern methods that rely on axonally transported tracers. Restricted cortical injections of transported tracers in specific layers could show which thalamic cells send axons to a given area (e.g., Parent et al., 1999). Nevertheless, it has limited application on human beings as it is an invasive method. Recent advances in non-invasive structural imaging have opened new approaches for investigating in vivo white matter structures in human beings. Among them, DWI allows for indirect estimation of the axon group orientations by measuring the motion of water protons (Bammer, 2003; Mukherjee et al., 2008). This procedure of reconstructing white matter tracts from DWI data is conventionally called tractography. Tractography has proven successful in quantitatively measuring the structural connectivity between different brain structures (Aydogan et al., 2018). It has been used to investigate thalamocortical white matter fibers (Johansen-Berg et al., 2005; Klein et al., 2010; Traynor et al., 2010). More details in this regard will be reviewed in chapter 4. In the first two studies of this thesis, we will use tractography to reconstruct specific thalamocortical white-matter tracts, and propose precise and reproducible protocols that can be used by the scientific community for future studies concerning thalamocortical pathways. The first study will focus on the first-order thalamic tracts that serve to relay peripheral and cerebellar input to the cerebral cortex. The second study will focus on the high-order thalamic tracts that connect the anterior and mediodorsal thalamus with cortical structures. In the two protocols, we will adopt a probabilistic atlas to segment the human thalamus to obtain precise and reliable individual thalamic nuclei, which allows tracking white-matter fibers from DWI data in a more precise and accurate fashion. The two proposed protocols will be validated on a large dataset to prove different reproducibility aspects. In the end, I will wrap the two 19 protocols into containers that are easy to use and guarantee reproducibility for other researchers who are interested in thalamic structure connectivities. In the third study, I will examine the involvement of the thalamus in some main human language systems. As will be introduced in chapter 5, the regionalization of language functions in the human brain has started since the work of Broca and Wernicke in the 19th century. After then, many influential neurobiological models were proposed to address the underpinnings of human language functions, either in a general approach or focused on specific language systems (such as reading models, speech comprehension and production models). Most of these models have focused on cortical structures across the cerebral cortex, while ignoring the contributions of the thalamus in human language function. Given the extensive connections between the thalamus and the cerebral cortex, it is worth investigating the involvement of the thalamus in human language systems. The three main human language systems, reading, speech comprehension and speech production, rely on three sensorimotor information processes: visual, auditory and motor information, respectively. The thalamus is the critical hub to relay this sensorimotor information from the periphery and cerebellum to primary sensorimotor cortices. More specifically, the three first-order thalamic nuclei are responsible for this labor: LGN relays visual information, medial geniculate nucleus (MGN) and ventral lateral nucleus (VLN) relay auditory and motor information respectively. These three nuclei also receive feedback axons from the primary sensorimotor cortices, which provide the neuroanatomical basis for them playing roles beyond being mere relays. Thus, in Study 3, I will examine the involvement of three first-order thalamic nuclei during specific language tasks using task-based functional MRI. In the following chapters, I will first review the structure and function of the thalamus, the white-matter fiber connections of the thalamus with cortical and subcortical structures in chapter 4. Then, in chapter 5, I will provide an overview of the history of the neurobiology of language and some influential neurobiological models of language. Following that, I will present the empirical part of the present doctoral dissertation, which contains three studies. Study 1 (chapter 6) and Study 2 (chapter 7) investigate first-order and higher-order thalamic tracts with the aim of developing reproducible protocols to reconstruct them with DWI data. Study 3 (chapter 8) will be focused on the 20 thalamic involvement in different human language systems, including reading, speech comprehension and production. Finally, chapter 9 will provide a general discussion of the overall work conducted in the present doctoral dissertation. 21 4 Thalamus 4.1 General introduction of the structure The thalamus refers to a small gray matter structure on each side of the midline, lying on the dorsal diencephalon (Figure 4.1A). Its medial surface makes up the lateral wall of the third ventricle. This oval structure measures about 3 cm in length and makes up 80% of the diencephalon in humans. Although the thalamus mainly consists of gray matter, a Y-shaped white matter structure named internal medullary lamina travels through it from posterior to anterior. It branches at the anterior section, separating the gray matter mass into three parts: the anterior, the medial and the lateral (Figure 4.1B). The customary understanding of thalamus function believes that it serves as a relay from periphery to cortex. Virtually all information reaching the cortex (with the exception of the olfactory information) must pass through the thalamus (Sherman, 2005). Nowadays, the understanding of the thalamus and its functional role has developed and recent findings over the last 15-20 years have shown that the relay function is not the only role played by the thalamus in brain functioning. The thalamus continuously plays a critical role in further cortical information processing, by acting as a hub that transfers information between cortical structures via multiple cortico-thalamo-cortical white-matter routes. The thalamus is not a homogeneous structure; it has a very complex internal organization. It consists of several structurally and functionally distinct cell groups, or nuclei, which are usually named according to their topographic location. 22 Figure 4.1.A) Midsagittal view of the thalamus from a human, a monkey, a cat, and a rat to show the position and relative size of the thalamus (diagonal stripes). Figure adapted from Sherman & Guillery 2006. B) The internal medullary lamina separates the thalamus into three parts: anterior, medial and lateral. Figure adapted from Wikipedia (https://en.wikipedia.org/wiki/Thalamus). 4.2 Thalamic atlases and nuclear divisions The complexity of the internal organization of the thalamus has made it difficult to reach a consistent and derivable scheme about how to categorize and label its nuclei. Figure 4.2 shows some different parcellation schemes in the history of thalamus research. The differences on how to parcellate the main divisions of the thalamus are obvious, regardless of different naming of individual structures. These parcellation differences are mainly due to the existence of great individual variations of thalamic topography, age/disease-related changes, and also differences of perspective among researchers (Mai & Majtanik, 2019). Among these three factors, the largest variation in parcellation schemes might come from the differences in perspective among researchers. Figure 4.3 shows how different specialists delineate and name the thalamic nuclei on the exact same thalamus section. Although it is clear that the thalamic nuclei should be distinguished by anatomically significant features and named intelligibly, appreciable differences with respect to segmentation and terms being used for the thalamic nuclei can be noticed among those maps. Researchers with different training backgrounds, trained in groups with a given history or tradition in the way of segment and classify thalamic nuclei, experience with animal or human brain, and perspective towards thalamic function could possibly lead to diverging parcellation schemes. 23 Figure 4.2. The profile of 12 coronal sections through the thalamus of different brains cut at the level of the posterior commissure. Figure adapted from Mai & Majtanik 2019. Sometimes also the same researcher or research group has adopted different parcellation and terminology throughout their academic life. For instance, in Mai & Forutan’s chapter Thalamus from the book “The Human Nervous System” (Mai & Forutan, 2012), they classify the nuclei into six groups: superior region, medial region, lateral region, intralaminar formation, periventricular formation and posterior formation. In total, they reported 28 anatomically and functionally distinct structures. While in 2019 (Mai & Majtanik, 2019), after quantitatively comparing several atlases, Mai and Majtanik described 9 thalamic nuclear groups that have different structure than it in Mai & Forutan (2012). For example, in Mai & Forutan (2012), the anterior nuclei (anteroventral nucleus, anteromedial nucleus, anterodorsal nucleus, dorsal superficial nucleus) are defined as the superior group, while the same nuclei are grouped into the anterodorsal group in Mai & Majtanik (2019). 24 Figure 4.6. The OR reconstructed in A) Benjamin et al. (2007); B) Sherbondy et al. (2008). 4.4.2 Medial geniculate nucleus The MGN, also known as medial geniculate body, is an oval mass located medially and posteriorly on the ventral lateral surface of the thalamus (Figure 4.1B). It is the synaptic station for acoustic information flowing from the inferior colliculus to the auditory cortex (Winer et al., 2005). Although the MGN can be further subdivided into three nuclei, due to the relatively small volume of human MGN and typically limited spatial resolution of MRI images, the current work will not differentiate the MGN subregions regarding the structure and function, but will take the MGN as a whole. The acoustic radiation (AR) fibers constitute the major input of the MGN to the ipsilateral primary auditory cortex (A1, BA 41, and partly 42 or Heschl’s gyrus) which lies in the temporal operculum. The AR maintains a topographical representation on the cortex, similar to the somatosensory and motor projections, but also presents a tonotopic organization (see Cherches, 2016). Functionally, the AR is basically the main sensory pathway relaying acoustic information from the MGN to the A1 (Berman et al., 2013), and being implicated in multiple high-level behaviors such as speech processing (Ojemann, 1991). A recent study showed that the task-dependent modulation of the 31 left MGN was increased when processing speech with noise background in contrast to the clear speech processing (Mihai et al., 2021). There is an hypothesis that some phonological deficits are caused by abnormal AR or MGN structure. For example, histological alterations were found in MGN after postmortem examination of readers with dyslexia (Galaburda et al., 1994). Additionally, one functional MRI study found that the MGN activation pattern is different when the task required attending to phonemes compared with other speech features in readers with dyslexia, and the activation level is correlated with the scores of the dyslexia diagnosis (Diaz et al., 2012). Diffusion evidence also showed that the connectivity strength of AR is reduced in readers with dyslexia compared to normal readers (Tschentscher et al., 2019). There are several studies trying to track the course of the AR based on in vivo diffusion MRI data (Behrens et al., 2007; Berman et al., 2013; Javad et al., 2014; Maffei et al., 2018, 2019; Profant et al., 2014). The AR, as a non-dominant fiber group, crosses with other fibers, which results in multiple-orientation signals in voxels at the crossing section. Thus, the AR is more difficult to identify with a single direction per voxel tractography model. Behrens et al. (2007) defined the MGN as a cuboid medial to the LGN and started tracking from there to the A1 using a probabilistic algorithm. In Behrens et al’s (2007) study they failed to reconstruct the AR with single fiber tractography, but succeeded when they used multi-fiber tractography. In a more recent study, Maffei et al. (2019) explored how the diffusion MRI acquisition parameters and tracking parameters can affect the reconstruction of AR. They found that higher b-values (≥5,000 s/mm2) and more gradient directions (≥128) increase the accuracy of the reconstruction for both probabilistic and deterministic tracking algorithms, but with low b-values (≤3000 s/mm2) only the probabilistic algorithm can successfully reconstruct the AR (Figure 4.7). 32 Figure 4.7. The effect of DWI acquisition parameters and tracking algorithms on the AR reconstruction. These images from one representative subject showed variant reconstruction of the AR when applying both probabilistic and deterministic algorithms to DWI data with different diffusion b-value shells. The white arrows indicate the false positives of the reconstruction. Figure adapted from Maffei et al. (2019). 4.4.3 Ventral lateral nucleus TheVLN, lying in the ventral lateral part of the thalamus, is an integrative hub for motor control. The major inputs to VLN are from deep nuclei of the cerebellum, pallidum and the substantia nigra. In turn, it projects to the motor areas of the cerebral cortex. On top of that it also receives feedback information from the motor areas. Topographically the VLN can be divided into two subnuclei: the anterior VLN (VLa) and the posterior VLN (VLp), two anatomically, histochemically and functionally different subregions. The involvement of the VLN in motor functions has been demonstrated using simple motor-related tasks, such as finger tapping (Lutz et al., 2000; Mallol et al., 2007). Evidence also found that activation of the VLN in a more complex motor task decreased after intensive learning (Lehéricy et al., 2005). Success of speech production needs motor execution, thus 33 requires the involvement of the VLN. Tourville & Guenther (2011) included VLN as one important node in their speech production neuroanatomy model. Patients with VLN lesions have been found to exhibit difficulties in naming objects and in short-term verbal memory (see review Petrovici, 1980). The main fiber group connecting VLp with cerebellum is known as the dentatothalamic tract (DT). The DT originates from the dentate nucleus in the cerebellum, projects via the superior cerebellar peduncle (i.e., brachium conjunctivum) to terminate in the contralateral VLp after decussating to the contralateral red nucleus (Coenen et al., 2014; Kwon et al., 2011). The DT is the main cerebellar efferent tract and it is mainly involved in movement control and multiple motor behaviors such as speech production (Ojemann, 1975). Lesions to the DT can produce abnormal movement, including ataxia, tremor, and dystonia (Kwon et al., 2011). Some effective surgical interventions for patients with essential tremor often target the VLp and adjacent white-matter tracts (Dallapiazza et al., 2019). The VLN projects to the primary motor cortex (M1) via the motor radiation (MR, Ilinsky & Kultas-Ilinsky, 2002). The VLN also projects to the preSMA (BA 6), which is an area of the motor cortex that traditionally has been assigned as responsible for updating motor plans and learning new motor sequences (Hamani et al., 2006). Functionally, this VLN-M1 primary pathway is thought to be implicated in transmitting cerebellar inputs to the M1. The fact that the DT travels through deep and small nuclei makes it difficult to be identified with DWI techniques. There are only a handful of studies that successfully reconstructed the DT from DWI data in healthy populations (Figure 4.8A; Kwon et al., 2011; Meola et al., 2016) or patients (Figure 4.8B; Coenen et al., 2011, 2014; Nowacki et al., 2019). In Nowacki et al’s (2019) study, they tested four different tracking protocols to identify the DT, which led to divergent results. All the four procedures were based on deterministic algorithms, but no probabilistic algorithms were tested. Some studies also reconstruct the DT and MR as one single tract (Q. Ji et al., 2019; Vo et al., 2015). Hyam et al. (2012) investigated the white-matter fiber bundles between VLN and motor cortex and successfully reconstructed the MR. But this study only used part of the VLN (referred as ventralis intermedius and ventralis oralis in their study) as seed when applying the tractography. 34 Figure 4.8. DT reconstructed in A) Kwon et al., 2011 and B) Nowacki et al., 2019 (four colors indicate the four different methodologies of tracking the DT). 4.5 Higher-order relay thalamic nuclei In contrast to the first-order relay nuclei, which receive afferent inputs from peripheral sensory centers (retina, auditory and somatosensory relays, cerebellum, etc.) and relay information to the cerebral cortex; the higher-order relay nuclei receive few or no afferents from periphery, but instead receive their afferents from the cerebral cortex. The major thalamocortical fibers from the first-order relay nuclei project to the primary cortical areas, such as visual cortex, auditory cortex, somatosensory and motor cortex, whereas the thalamocortical fibers from the higher order nuclei send information to areas that are involved in more complex functions and are traditionally called association cortical areas (Guillery, 1995). Unlike a clear classification of first-order nuclei, the higher-order nuclei are far from being fully defined. Although large inconsistencies remain, the AN, MD and pulvinar are the ones consistently categorized as higher-order relay nuclei. 4.5.1 Anterior nuclear complex The AN is located in the rostral and dorsal part of the thalamus and separated from other dorsal thalamic nuclei by the Y-shaped internal medullary lamina. The AN typically consists of three nuclei: anterior medial, anterior dorsal and anterior ventral nuclei. In the probabilistic atlas proposed by Iglesias et al. (2018), the thalamic nuclei AV actually represents the whole AN. In the 35 current thesis we will use this thalamic nuclei but with the name of AN to describe the whole anterior nuclear complex. The AN is a critical node in the Papez circuit, which begins in the hippocampus and continues through fornix, reaching to the mammillary body (Parmeggiani et al., 1971). The mammillary body connects to AN with the mammillothalamic tract. The AN in turn projects through the cingulum bundle to cingulate cortex, retrosplenial cortex, which connects to hippocampus, thus completing the circuit (Shah et al., 2012). More specifically, the AN receives afferent inputs from the mammillary body and relays it to anterior and posterior cingulate, and retrosplenial cortex. The AN has been found involved in memory-related processes. For example, Aggleton & Brown (1999) proposed that the AN contribute to the information retrieval from memory during item absence through the connection with the hippocampus, and that the direct hippocampal-AN connection and the indirect hippocampal-mamillo-AN connection are involved in allocentric spatial memory. A fMRI study also presented a significant involvement of AN in recall tasks but not in encoding tasks (Pergola et al., 2013). There are some studies that use diffusion data to reconstruct the Papez circuit, which includes the white-matter fibers connection of the AN with other structures involved in this circuit (Concha et al., 2005; Granziera et al., 2011; Wei et al., 2017). However, those studies fail to distinguish the white-matter bundles connecting the AN with other structures separately. For example, the cingulum bundle consists of the white-matter fibers of AN to anterior cingulate, posterior cingulate, and retrosplenial cortex, while in the above-mentioned studies, the cingulum bundle is taken as a whole (Figure 4.9). 36 Figure 4.9. The cingulum bundle reconstructed from diffusion data in A) Wei et al., 2017; B) Concha et al., 2005. 4.5.2 Mediodorsal nucleus The MD is an ovoid structure extending from the level of the intrathalamic adhesion until the level of the habenular commissure. It is relatively easy to distinguish from other structures on most cross-sectional levels since its medial surface borders the third ventricle and the rest issurrounded by the internal medullary lamina with its embedded nuclei. The MD is typically divided into two cytoarchitectonically distinct parts: the medial one-third to one-half is magnocellular in Nissl preparations and the lateral half to two-thirds, which has smaller cells and present parvocellular properties (Jones, 1985). In the present doctoral dissertation, we treat the MD as whole due to the MRI spatial resolution. The MD is robustly connected with the frontal lobe, more specifically the PFC. Indeed the connection of MD and PFC has been playing an important role in the research history of PFC function: the classic definition of PFC was the cortical projection zone of the MD (Mai & Forutan, 2012). The MD not only projects to PFC but also receives afferents from other structures, such as amygdala (Aggleton & Mishkin, 1984), and the temporal pole (Gower, 1989). The MD has been proposed to be involved in higher cognitive functions via its extensive connections with frontal lobe and subcortical structures (Mitchell, 2015; Ouhaz et al., 2018). For example, fMRI research with humans has revealed that the MD-PFC network is activated during successful encoding and retrieval (Pergola et al., 2013). Also, evidence from lesion studies indicates that the damage of MD could 37 disrupt executive functions, attention control, prospective memory, arousal, motivation, language functions (see reviews, Mitchell, 2015; Pergola et al., 2018). A number of studies have used diffusion MRI to reconstruct the projections of MD (Figure 4.10; Eckert et al., 2012; Giraldo-Chica et al., 2018; Jakab et al., 2012; Klein et al., 2010; Lambert et al., 2017; Le Reste et al., 2016; Li et al., 2022). Jakab et al. (2012) adopted a probabilistic algorithm to track the white-matter fibers from the lateral and medial MD to the rest of the brain. The probabilistic tractography showed that the lateral MD is the source of fibers projecting to the superior and middle frontal gyri, while the fibers originated from medial MD mainly terminate in the frontal orbital cortex and various temporal loci. Figure 4.10. The tractography of MD with the rest of the brain from Jakab et al. (2012). Warm color voxels indicate the trajectory of white-matter fibers originated from lateral MD and cold color voxels for the medial MD. Figure adapted from Jakab et al. (2012). 4.5.3 Pulvinar The pulvinar lies in the posterior part of the thalamus and it is the largest thalamic nuclear complex, reaching about 30% of its volume in humans (Mai & Forutan, 2012). The pulvinar is typically subdivided into four subnuclei based on neuroanatomical properties (Jones, 1985): anterior, inferior, lateral and medial. This pulvinar subdivision was adopted by the probabilistic atlas used in the current work (Iglesias et al., 2018). Aside from its size, the pulvinar is widely connected with all 38 the cerebral lobes, which makes it one of the most complex higher-order thalamic structures (Mai & Forutan, 2012). Specific functions associated with each cortico-pulvino-cortical pathway have been explored in neuroimaging research, but still they are far from being clear (Fiebelkorn & Kastner, 2019). Research have linked the human pulvinar to multiple cognitive functions, such as emotion recognition (Ward et al., 2007), visual attention (Fischer & Whitney, 2012), visual motion (Villeneuve et al., 2005), and fear recognition (McFadyen et al., 2019). Also, there is discussion about the notion that the pulvinar behaves as a ‘connectional hub’ integrating convergent information and then transmitting processed signals to cortical and subcortical structures (Bridge et al., 2016). Yet, little is known regarding whether the pulvinar relay function is to help communication between cortical areas without changing the information being relayed, or the pulvinar also manipulates and processes that information (Sherman & Guillery, 2009). To date, there are few tractography studies investigating the connections of human pulvinar, mainly due to its complex white-matter connectivity with extensive cerebral cortex. Leh et al. (2008) used diffusion tensor imaging tractography and reconstructed pulvinar tracts (Figure 4.11A). The reconstructed tracts were found to project to the frontal eye fields, prefrontal areas, visual cortex, and parietal association areas. Another tractography study focused on the optic nerve that connects the optic chiasm with pulvinar (Maleki et al., 2012), in which they used the optic chiasm and pulvinar as inclusion ROIs and LGN and primary visual cortex (V1) as exclusion masks (Figure 4.11B). The investigation of the structural connectivity of the pulvinar and its functional role in language-related processes are beyond the scope of the present doctoral work, due in part to its complexity and the limited time. However, this is one of the first research lines we are planning to pursue after the doctoral dissertation. 39 Figure 4.11.A) Reconstructed tracts that connect the pulvinar with cortical and subcortical structures in Leh et al., (2008). B) Optic nerve from optic chiasm to the pulvinar reconstructed in Maleki et al. (2012). 40 Figure 5.4. Neurological and cognitive model of language proposed in Price 2000. Figure adapted from Price 2000. 5.2.1.2 MUC model Both the Classic model and Price’s model constrain their framework at word-level language processing and preclude any language components beyond word processing. Hagoort (2005, 2013) proposed a Memory-Unification-Control (MUC) model that takes into account what goes on beyond production and comprehension of single words. The MUC model divides language processing into three components: Memory, Unification, and Control (Figure 5.5A). The Memory component represents the linguistic information that gets encoded and consolidated memory during language acquisition, such as phonology and phoneme knowledge, semantic memory and syntactic properties of words. This component projects to the left temporal cortex according to the MUC model. The Unification in this context is the operation of unifying lexical information into overall representations that span multi-word utterances, which is critical in higher level language processing. The Unification takes place in the left inferior frontal cortex, with a spatial gradient (Figure 5.5B). Depending on the type of information, semantic information is unified in pars orbitalis; syntactic unification recruits pars triangularis; phonology unification involves the pars opercularis. The Control component refers to attentional control and action planning during a conversational setting, which, for example, allows a bilingualism to switch to the correct language during conversation. The model suggests that the Control component involves the anterior cingulate cortex (ACC) and the dorsolateral prefrontal cortex (dlPFC). The MUC model is a substantial augmentation of the Classic model with three major additions: first, the connectivity of the critical language regions is more expanded and not restricted to the arcuate fasciculus, which is proposed by the Classic model. Second, the components in that model are not separated in terms of production and comprehension as in the Classic model, but instead are divided into memory, unification and control. Third, the network proposed is more extended than it in the Classic model, which was mainly based on evidence from single word processing. Although the network is not exclusively for language processing, it is necessary to be recruited for the sake of successful language processing. 47 Figure 5.5.A) The MUC model proposed by Hagoort (2005, 2013). Memory (yellow) in the left temporal cortex, Unification (blue) in left IFG, and Control (pink) in the dlPFC. The ACC (part of the Control component) is not shown. Figure adapted from Hagoort (2013). B) The unification gradient in the left inferior frontal cortex. 5.2.1.3 Lau’s model Similar to the MUC model, Lau and colleagues proposed a semantic processing in context or sentence processing model that divides semantic processing into four components: lexical storage and access, lexical retrieval, lexical selection and combinatorial semantics (Figure 5.6). Both the semantic information storage and combinatorial semantics components are shared between the MUC model and Lau’s model, while the neuroanatomical labor divisions are slightly different. In Lau’s model, lexical information is stored in posterior MTG. The anterior temporal cortex and angular gyrus are involved in combining incoming lexical information with existing semantic and syntactic representation. In the MUC model the semantic memory storage was predicted to recruit the whole temporal cortex, and the combinatorial semantic operations were predicted to take place in IFG. In contrast, the IFG is involved in different functions in Lau’s model: the anterior IFG (aIFG) is recruited in lexical representation retrieval and control, and the posterior IFG (pIFG) mediates lexical selection from multiple active candidates. 48 Figure 5.6.A) Schematic model for semantic processing of words in context proposed by Lau et al. 2008.B) The corresponding functional neuroanatomic framework. Figure adapted from Lau et al. 2008. 5.2.2 Reading models A reading dual pathway model was proposed by Pugh and colleagues (Pugh et al., 2000, 2001) based mainly on neuroimaging evidence of single word reading from reading-imparied and normal populations (Figure 5.7). In this model, the reading networks comprise anterior and posterior circuits. The anterior circuits are located in the IFG and are associated with phonological recoding during reading. The posterior reading circuits include both dorsal (angular gyrus, SMG and pSTG) and ventral (occipito-temporal junction) components. The dorsal circuits are critical in mapping the visual input of printed words to phonology and the ventral circuits are involved in visual orthographic information processing. 49 Figure 5.7. Three critical components in the dual pathway model proposed by Pugh and colleagues. Figure adapted from Pugh et al. (2001). 5.2.3 Speech comprehension models There are two popular neuroanatomical models explaining speech comprehension. Both of them could be simply referred to as dual stream models, proposed by Hickok and Poeppel (2000, 2007), and by Friederici (2002, 2011, 2012). 50 Figure 5.8. The dual stream model proposed by Hickok and Poeppel. Figure adapted from Hickok & Poeppel 2007. According to the Hickok and Poeppel dual stream model, the speech comprehension starts from the bilateral auditory cortex, which involves spectrotemporal analysis (Figure 5.8). The phonological processing involves the bilateral mid-post STS. Afterwards, two streams emerge and carry the phonological information to the frontal lobe through different pathways. The dorsal pathway goes through the Sylvian fissure at the parieto-temporal boundary and reaches the IFG and premotor cortex. This pathway mainly maps phonological information onto articulatory representations in the frontal cortex, which is critical for speech development and production. The ventral pathway goes through the posterior MTG and ITS after leaving the mid-post STS, and reaches the anterior MTG and ITS. This pathway is responsible for accessing semantic representation from the phonological information (posterior MTG and ITS) and combinatorial semantics (anterior MTG and ITS). In general the dorsal pathway is left hemisphere-dominant while the ventral pathway is bilateral. Friderici’s dual stream model shares neuroanatomical regions with Hickok and Poeppel’s model (Figure 5.9). This model is derived mainly from speech sentence processing and proposes that both the dorsal pathway and the ventral pathway serve more functions than the ones originally proposed by Hickok and Poeppel. For example, the dorsal pathway not only subserves mapping from auditory to motor, but can also be involved in syntactic processing in sentence comprehension. Taking 51 into account white-matter fibers involved in the speech comprehension network, two dorsal pathways are proposed in this model: one connecting the pSTG/STS to the premotor cortex (auditory-motor mapping), the other pathway reaching to the posterior IFG (BA44) from pSTG/STS via the arcuate fasciculus (syntactic processing). Similarly, the ventral stream can be also subdivided into two pathways: one pathway connects the temporal cortex with BA45 and BA47, which is supporting auditory-semantic mapping; the other pathway connects the anterior STG to frontal operculum and seen as supporting the combinations of adjacent elements in a sequence, which is important in sentence comprehension. Figure 5.9. The dual stream model proposed by Friderici. Figure adapted from Friderici 2012. 5.2.4 Production models Indefrey and Levelt proposed a word production model based on the time course of critical component processes and the involvement of brain regions in word production (Indefrey, 2011; Indefrey & Levelt, 2004). The word production networks consist of the left posterior IFG, the left precentral gyrus, the supplementary motor area (SMA), the left mid and posterior parts of the STG and MTG (Figure 5.10). The thalamus and cerebellum are also included in this model, being involved in phonetic encoding and articulation processes. 52 Figure 5.10. Neuroanatomical model of word production proposed by Indefrey and Levelt. The numbers within regions indicate median peak activation time estimates in milliseconds. Figure adapted from Indefrey (2011). Tourville and Guenther developed the DIVA model of speech production (Figure 5.11; Tourville & Guenther, 2011). This model is built on a computational model under the same name, and unified neuroanatomical evidence that involves speech acquisition and production. This model comprises multiple components that contribute the successes of speech production and assign them to corresponding brain regions (see anatomical labels in Figure 5.11). In the DIVA model the thalamus also plays a critical role with its projections to the motor cortex serving as gates on the outflow of motor commands, and being involved in representing the feedforward motor programs. 53 Figure 5.11. The DIVA model of speech acquisition and production. In the box there are the cognitive components of the model and the corresponding anatomical regions. Figure adapted from Tourville & Guenther (2011). In sum, since Broca and Wernicke’s pioneering work linking human language functions and specific brain regions, the knowledge about the neurobiological underpinning of language has been tremendously expanded. Many neurobiological models have been proposed since then to account for the representations of language functions in different brain regions. The influential models reviewed above made connections between critical components of language processing and brain structures. However, most of these models are focused on cortical structures, neglecting to some extent the role of thalamus in language (only the DIVA model on speech production included the thalamus in the speech articulation phase) despite its widespread connections with the cerebral cortex. One of the main goals of the present doctoral work is to investigate the role that thalamus is playing in some of the main language systems. 54 6 Study 1: Structural connection of first-order thalamic nuclei The present study was aimed at developing and testing a reproducible protocol for obtaining four first-order relay thalamic input and output white-matter tracts. The novelty of this protocol capitalizes on 4 aspects: (1) it is focused on well-known white-matter tracts constituted by myelinated axons that originate and/or target the first-order relay nuclei of the thalamus, testing them within the same study, and using similar methods and reconstruction procedures across them; 2) different from most previous studies, here we specifically investigated in a large dataset the reliability of the protocol in terms of both computational and test-retest reproducibility; 3) the present protocol uses state-of-the-art MRI protocols (multiband, multi-shell) and tractography methods with the aim of developing an advanced protocol that can be applied to current ongoing studies and future research; and, 4) the protocol is designed to be reproducible, easy to use and automatized, which unfortunately has not been the norm in the past. Also, it builds on previous well-validated tools including the first probabilistic atlas of the thalamus based on combining high-resolution ex vivo MRI and histology (Iglesias et al., 2018) and the reproducible-tract-profiles (RTP2) containerized tool which is based on state-of-the-art techniques implemented on top of Vistasoft's code, which have been tested and used in many publications over the last 15 years (https://www.github.com/vistalab/vistasoft; Lerma-Usabiaga et al., 2022). The ultimate goal of this work was to provide a reliable protocol for obtaining and estimating first-order relay thalamic pathways for basic research and clinical studies. 55 To this end, we first defined multiple parameters to optimally reconstruct the above-mentioned four thalamic pathways in left and right hemispheres. Second, we tested the computational and test-retest reproducibility of our protocol by examining a range of white-matter proxies related to the microstructural and macrostructural properties of these tracts. To examine the reliability of the protocol we obtained tracts from DWI of 113 normal adults. The protocol consisted of three components: Defining the regions-of-interest (ROI); preprocessing DWI data; modeling white-matter tracts and tractometry. Reproducibility was tested using two approaches: 1) Computational reproducibility, tested by identifying each tract using the same parameters 10 independent times for all 113 subjects, and 2) Test-retest reproducibility, tested by re-scanning a subset of 24 participants using the same MRI protocol twice within an average interval of 15 days. Our hypothesis was that we would obtain a high degree of reproducibility for the microstructural and macrostructural properties of these tracts. However, we expected some variability in specific tracts, such as the DT which crosses hemispheres and it is relatively long, and hypothesized that this variability would be higher for test-retest than for computational reproducibility. 6.1 Methods 6.1.1 Subjects A total of 113 healthy volunteers (mean age = 24.5 years, SD = 4.33 years; 65 females) participated in the study. Twenty-four of the volunteers (mean age = 24.7 years, SD = 4.06 years; 13 females) returned for a second session in which they were scanned using exactly the same MRI protocol (mean interval = 15 days, SD = 21.82 days, range: 7-104 56 To evaluate these two types of reproducibility at the microstructural scale, we performed pairwise correlations on tract profiles for all possible pairs from the 10 repeated computations to measure computational reproducibility, and across test and retest to measure test-retest reproducibility. For simplicity, we only show the correlations for FA values. At the macrostructural level, we quantitatively analyzed tract volume overlap, streamline density and distance to check the reproducibility of tract shapes. These analyses included: (1) Dice similarity index to check for volume-based overlap of all tract pairs; (2) density correlation for the voxel-level streamline density of all tract pairs; and, (3) bundle adjacency, the average distance between streamlines from two tracts. The measurements used to examine computational reproducibility and rest-retest reproducibility were computed using the package scilpy (see details in Schilling et al. 2021 and https://github.com/scilus/scilpy). These analyses were conducted across all possible pairs of computational reproducibility and rest-retest reproducibility, as well as for each tract. 63 Figure 6.1. The reproducibility measurement scheme. A) Computational reproducibility (reproducibility across computations); test-retest reproducibility (reproducibility across test and retest sessions). B) The Dice overlap of MR reconstructed at the first second computations, from subject S038. C) Correlation of the FA profile of MR from subject S038’s test and retest sessions. 6.2 Results In the present study, we obtained and measured fibers bundles connecting three first-order sensory (LGN, MGN) and motor (VLN) thalamic nuclei with their main corresponding cortical target areas. In addition, we reconstructed the subcortical input pathway to VLp from the dentate nucleus of the cerebellum. These four tracts were identified as homologous tract pairs in the left and the right hemisphere. Figures 6.2 and 6.3 show these tracts in a representative subject. To examine the reproducibility of our protocol, we followed a double analytical approach testing: (1) computational reproducibility by repeating the computation on the same diffusion data 10 times and quantifying changes from computation to computation for the same tract; and, (2) test-retest reproducibility, by obtaining DWI data 64 from the same subjects and using the same MRI protocols across two different sessions to quantify test-retest changes in the same tracts. Figure 6.2. The OR (A) and AR (B) reconstructed in a representative subject. A1 and B1 show the 3D representations of the OR and AR in yellow. A2 and B2 show the positions from which the slices in A3 and B3 are respectively drawn. A3 and B3 depict axial and coronal views of the core subcomponents of OR and AR, respectively. Green color indicates the cortical ROIs V1/V2 and A1. 65 Figure 6.3. The MR and DT reconstructed in a representative subject. A shows the 3D representations of the MR and DT. B shows the positions from which the slices in C and D are drawn. C and D depict coronal views of the core subcomponents of MR and DT. Green color indicates the cortical ROI M1. Yellow streamlines are MR and blue streamlines represent the DT. D shows the axial view of the core subcomponents of the DT. 6.2.1 Computational reproducibility For the four pairs of white-matter fibers with the established protocol, the repeated computations on same diffusion data resulted in mostly identical tract profiles and high agreement of streamlines. The mean correlations of FA profile were above 0.99 for all the tracts examined (see Table 6.2; Figure 6.4B). At individual level, for all the possible pairs of computation, most correlation coefficients were higher than 0.97 for each fiber, except for left AR and right DT, which have a long tail towards 0.82. Bearing this in mind, with 10 repeated 66 computations, there will be at least 9 relatively low coefficients if only one computation resulted in a different tract than all the others. Agreement indices also showed that the identified white-matter fibers have consistent shapes and density across repeated computations (Table 6.2). Figure 6.4C shows agreement indices for each individual pair of computations. These three agreement indices revealed the same pattern as the one observed in the correlation coefficient of FA profile, with more variability in left AR. And the same pattern was also found for the right homologous AR. It is noteworthy that some of this variability in agreement indices derives from the same single subject (e.g., outlying clusters for bundle adjacency and Dice coefficient in right AR, and for Dice coefficient and density correlation in left OR). Table 6.2. Reproducibility indices and their standard deviations (in parentheses) for all measures and fiber bundles. computational test-retest FA profile correlation bundle adjacency dice index density correlation FA profile correlation bundle adjacency dice index density correlation L OR 0.9996(0.0016) 0.09(0.01) 0.92(0.01) 0.998(0.001) 0.9956(0) 0.11(0.01) 0.90(0.01) 0.990(0.005) R OR 0.9996(0.0005) 0.09(0.01) 0.92(0.01) 0.998(0.001) 0.9944(0) 0.11(0.01) 0.90(0.01) 0.989(0.005) L AR 0.9976(0.0074) 0.18(0.09) 0.85(0.05) 0.990(0.008) 0.9265(0.12) 0.39(0.27) 0.76(0.09) 0.907(0.082) R AR 0.9989(0.0013) 0.11(0.09) 0.90(0.04) 0.997(0.002) 0.9473(0.07) 0.34(0.41) 0.80(0.12) 0.940(0.047) L MR 0.9985(0.0012) 0.09(0.01) 0.91(0.01) 0.993(0.001) 0.9787(0.02) 0.13(0.05) 0.89(0.01) 0.971(0.010) R MR 0.9982(0.0017) 0.09(0.01) 0.91(0.01) 0.993(0.001) 0.9713(0.03) 0.12(0.03) 0.89(0.01) 0.971(0.014) L DT 0.9988(0.0038) 0.07(0.03) 0.93(0.02) 0.994(0.004) 0.9638(0.03) 0.21(0.11) 0.82(0.08) 0.853(0.097) R DT 0.9986(0.0032) 0.06(0.01) 0.94(0.01) 0.994(0.004) 0.9458(0.06) 0.22(0.14) 0.82(0.09) 0.782(0.182) 67 Figure 6.4. Evaluation of computational reproducibility for the OR, AR, MR and DT. A) Examples of group average FA profiles for left MR from the first (gray continuous line) and second (green dashed line) computations. The light green shaded area indicates the standard deviation. B) Strip plots showing the distribution of correlation coefficients between all possible pairs computed for each tract and each subject (lighter color columns represent the left hemisphere, darker color columns represent the right hemisphere). Each dot represents the correlation coefficient for a specific computation pair for one participant. C) Agreement indices distribution: bundle adjacency (top), Dice coefficient (middle), and density correlation (below) for all possible computation pairs for each tract and each subject (light color columns represent the left hemisphere, darker color columns represent the right hemisphere). 6.2.2 Test-retest reproducibility To examine test-retest reproducibility, 24 participants came back for a retest session where we used exactly the same MRI protocol. The mean of FA profile correlations was above 0.9 across the ten tracts of interest, although as expected the values were numerically lower than those observed in the computational reproducibility analyses. As in the computational reproducibility analysis, the left AR also showed higher variability in the test-retest reproducibility analysis, with lower values within the mean correlation coefficients (0.93, Table 6.2 and Figure 6.5A & 6.5B). Nevertheless, it is important to highlight that all of these values reflect a high degree of reproducibility. 68 Test-retest reproducibility was also confirmed by the agreement indices. The group averages for bundle adjacency were all under 0.4 for the ten tracts, indicating that the streamlines identified in the test were very close to the streamlines identified in the retest (see Table 6.2 and Figure 6.5C). High reproducibility was also reflected by the Dice index and streamline density correlation. Among the ten tracts, the AR and DT tended to show more variability bilaterally. Figure 6.5. Evaluation of test-retest reproducibility for the OR, AR, MR and DT. A) Examples of group average FA profiles for left MR from test (gray continuous line) and retest (green dashed line). The light green shaded area indicates the standard deviation. B) Strip plots showing the distribution of the correlation coefficients between test and retest for each tract and each subject (lighter color columns represent the left hemisphere, darker color columns represent the right hemisphere). C) Agreement indices distribution: bundle adjacency (top), Dice coefficient (middle), and density correlation (below) for test and retest for each tract and each subject (lighter color columns represent the left hemisphere, darker color columns represent the right hemisphere). 6.3 Discussion We present a reproducible protocol for tractography reconstruction of first-order human thalamocortical tracts, which play a critical role in sensory and motor information relay between the thalamus and cortex. We tested the reproducibility of our protocol for 69 obtaining these tracts of interest by examining their microstructural tractometric properties and volume-based macrostructural similarity across repeated computations and test-retest sessions. Results showed nearly perfect computational reproducibility across ten repetitions and high-to-excellent test-retest reproducibility. In terms of computational reproducibility, it is worth highlighting that across ten separate and independent computations using the same raw data and protocol, reproducibility was nearly perfect; for example, providing an average FA profile correlation of 0.99 (individual-subject values ranging from 0.82 to 0.99) and an average Dice similarity value of 0.91 (individual Dice index values ranged from 0.66 to 0.97) across all the bundles examined. We expected computational reproducibility would be high, but it was important to demonstrate that the protocol is reliable and appropriate for obtaining the tractography measures of interest. Concerns about this type of reproducibility, which we refer to as computational reproducibility, have grown in recent years (e.g., Theaud et al., 2020). One goal of our protocol is to offer neuroscientists and medical practitioners efficient and reproducible processing guidelines to reconstruct first-order thalamocortical tracts. Hence, we packed our solution in software containers that can be run using Singularity or Docker technologies. In both cases, exactly the same set of algorithms, associated libraries and operating systems can be run in a computationally reproducible manner. This allows researchers to run previously as well as recently acquired data using exactly the same software and configuration options many times. Test-retest reliability, which ensures stability across time, is one of the most widely used measures for a protocol or tool. Since test-retest reproducibility entails inputting different data, we expected to obtain lower numerical tractometric and volume-based similarity values than those observed in our computational reproducibility calculations. At the microstructural level, our data showed an average 0.97 (individual-subject values ranged 70 from 0.50 to 0.99) test-retest reproducibility for FA profile correlations after an average of 2 weeks. These correlation coefficient values for tract profiles are high and consistent with the values obtained in a previous study aimed at validating test-retest reliability in classic fiber bundles (Kruper et al., 2021). At the macrostructural level, the Dice index is commonly used to assess the overlap of bundles reconstructed at two time points (Besseling et al., 2012; Boukadi et al., 2019; Cousineau et al., 2017). In the present study, Dice index values averaged 0.85 (individual-subject values ranged from 0.39 to 0.94) for all white-matter bundles examined. Based on the Dice index values reported in previous studies focused on white-matter bundles (e.g., minimum group-level Dice index of 0.70 in Besseling et al. 2012 and Cousineau et al. 2017; 0.71 in Boukadi et al. 2019), the range of the Dice index reported here (i.e., 0.76-0.90 across tracts at group level) indicates that all the white-matter bundles we investigated had high test-retest reliability. It is important to understand the nature of the variability observed in these two types of reproducibility measurements. Our results showed that computational reproducibility is nearly perfect, but there is some variance across repeating computations. The main source of this variance is random seed generation. Basically, there are steps in our reconstruction pipeline that involve non-deterministic processes, which will be different across computations. A fully reproducible pipeline can be achieved by fixing the random seed initialization, but we decided not to proceed in this way. We used a probabilistic algorithm to generate the streamlines, whose advantage has been discussed by many researchers (see for instance (Bonilha et al., 2015; Grisot et al., 2021; Khalsa et al., 2014). Indeed, fixing random seed initialization would work against the probabilistic nature of the algorithm and the philosophy of probabilistic tractography. In addition, fixing the random seed is not compatible with using multi-threaded steps that allow for faster computation. Multi-threading 71 introduces randomness in terms of the order of step execution, which will generally affect the reproducibility of results. It is relevant to note that previous studies that focused on probabilistic tractography have not examined computational reproducibility. Together with random seed generation, there could possibly be other factors contributing to some extent to the computational reproducibility variance. Instead of assuming no or slight changes among computations, researchers should be aware that computational variance exists depending on the parameters chosen to conduct the tractography. It is important that future research further investigates, in a systematic manner, factors that might be associated with computational variance in probabilistic tractography beyond random seed generation. The test-retest reproducibility reported here showed some variability especially for specific bundles. We observed overall more variability in test-retest reproducibility for the AR and DT tracts. This can be in part explained by specific characteristics of these tracts, for instance: i) the small seed used for reconstruction (i.e., MGN/A1, dentate nucleus); ii) long streamlines (i.e., DT); and iii) anatomical complexity (i.e., DT). Previous studies have related lower reproducibility to smaller seed size (Bonilha et al., 2015; Buchanan et al., 2014; Zhang et al., 2019) and longer streamlines (Bonilha et al., 2015; Mori & Van Zijl, 2002; Tsai, 2018). Fiber tracking from small seeds may be influenced by systematic errors and noise, leading to spurious findings. Moreover, smaller seeds are likely to generate less fibers and therefore decrease the likelihood of successful tracking. Similarly, streamlines with longer paths lead to more interruptions in fiber tracking, which can also lead to larger variation. Also, the anatomically complex DT connects small and deep nuclei, making it more vulnerable to partial volume effects between different tissues (Mori & Van Zijl, 2002). Small ROIs, long fibers, and anatomical complexity affect computational reproducibility in the same manner as to the test-retest reproducibility. Nevertheless, our protocol was able to reconstruct the four pairs of tracts in all 113 subjects. This is an important achievement considering the 72 For each tract and each subject, we extracted the DWI noise by first creating a mask of the tract and corresponding seed and target; then we applied this mask to the noise image generated in the DWI preprocessing step to extract the average noise within this mask. The streamline length was calculated by averaging the length of each streamline in one specific tract. Both the streamline length and amount were calculated for each tract and each subject. The tracts used in this post-hoc analysis were generated by the first computation. Tract-wise correlation analysis was conducted between the three possible factors and the four reproducibility indices for computational and test-retest reproducibility separately. In order to do the correlation analyses, three reproducibility indices, FA profile correlation, density correlation and Dice index were transformed to a normal distribution. Fisher-Z transform was adopted for the two correlation indices, to transform the Pearson correlation coefficient to a Z score. The Dice index has a restricted range of [0,1] and is often close to the value of 1. A logit transform was applied to the Dice index, where logit(Dice) = ln(Dice/(1-Dice)). This monotone transformation maps the Dice range of [0,1] to [-∞, +∞], and logit(0.5) = 0. This distribution is close to a normal distribution for a large sample size (Zou et al., 2004). For each tract, the group level average of the possible factors and reproducibility indices (after transformation if applied) were calculated across subjects. Then, paired-sample correlation analyses were conducted for each pair of factor and reproducibility index values at tract level. 7.2 Results In the present study, we reconstructed 42 pairs of fiber bundles connecting two higher thalamic nuclei with their main corresponding cortical areas (3 AN-related and 39 MD-related fiber bundles) from DWI data. To examine the reproducibility of our protocol, we followed a double analytical approach testing: (1) computational reproducibility by repeating the computation on the same diffusion data 10 times and quantifying changes from computation to computation for the same 79 tract; and, (2) test-retest reproducibility, by obtaining DWI data from the same subjects and using the same MRI protocols across two different sessions to quantify test-retest changes in the same tracts. The results are presented with the tracts of interest divided into 5 groups: (1) AN related tracts (Figure 7.1); (2) tracts of MD with dorsolateral prefrontal regions (dlPFC, Figure 7.2); (3) tracts of MD with medial prefrontal regions (mPFC, Figure 7.3); (4) tracts of MD with orbital and frontal polar regions (Figure 7.4); (5) tracts of MD with inferior gyrus regions (IFG, Figure 7.5). Figure 7.1. The computational and test-retest reproducibility of tracts connecting AN with cingulate and retrosplenial cortex. A) The lateral (A1) and medial (A2) views of the reconstructed tracts in a representative subject. Color scheme is the same as used in B and C. B) Computational reproducibility: B1) strip plots showing the distribution of correlation coefficients between all possible pairs computed for each tract and each subject (lighter color columns represent the left hemisphere; darker color columns represent the right hemisphere). Each dot represents the correlation coefficient for a specific computation pair for one participant; Box-and-whisker plots are overlaid on top to show the quartiles of the distribution. B2) Agreement indices distribution: bundle adjacency (top), Dice coefficient (middle), and density correlation (below) for all possible computation pairs for each tract and each subject (lighter color columns represent the left hemisphere, darker color columns represent the right hemisphere). C) Test-retest reproducibility with the same layout as panel B. The y axes of both panel B and C show the targets of those reconstructed tracts. 80 Figure 7.2. The computational and test-retest reproducibility of tracts connecting MD with dlPFC subregions. Details about each panel can be found in Figure 7.1 legend. 81 Figure 7.3. The computational and test-retest reproducibility of tracts connecting MD with mPFC subregions. Details about each panel can be found in Figure 7.1 legend. 82 Figure 7.4. The computational and test-retest reproducibility of tracts connecting MD with subregions of orbital and polar frontal cortex. Details about each panel can be found in Figure 7.1 legend. 83 Figure 7.5. The computational and test-retest reproducibility of tracts connecting MD with IFG subregions. Details about each panel can be found in Figure 7.1 legend. 7.2.1 Computational reproducibility For the 42 pairs of white-matter fibers with the established protocol, the repeated computations on the same diffusion data resulted in mostly identical tract profiles and high agreement of streamlines. The group mean correlations of FA profiles were above 0.97 for all the tracts examined. At the macrostructural level, the reproducibility is measured by calculating the adjacency, the streamline density correlation, and the Dice index of two repeated tracts. The results showed that the tracts have high reproducibility at macrostructural level in all three indices. The adjacency, which reflects the average distance between the two 84 tracts, ranges from 0.05 to 0.20 across all tracts. In all the tracts of interest, similar to it in FA profile correlation, the tracts connecting the right AN with the posterior cingulate and retrosplenial cortex have the highest adjacency, which are both 0.20. Similar high reproducibility is found in results of streamline density correlation and Dice index. The streamline density correlation ranges from 0.94 to 0.99. The Dice index showed that all tracts had high overlap between repeating computations, ranging from 0.84 to 0.95. Among all the tracts, the tracts connecting the right AN with the posterior cingulate and retrosplenial cortex showed good reproducibility but relatively low Dice index of 0.84 and 0.85 respectively. 7.2.2 Test-retest reproducibility Twenty four participants were scanned for a second time with the exact same MRI protocol as the first scanning session. In this 24 participant test-retest subset, we measured how the protocol could reconstruct the tracts of interest reproducibly across test and retest. The same indices reflecting reproducibility at the microstructural and macrostructural levels were calculated. In general, the tracts of interest showed good test-retest reproducibility regarding the mean of FA profile correlations, although as expected the values were numerically lower than the ones observed in the computational reproducibility analyses. There are tracts showing relatively high variability, mostly the tracts connecting MD with orbital frontal cortex, such as the bilateral MD-Area25, with the group mean of 0.64 (left, the lowest correlation among all the tracts of interest) and 0.71 (right). Nevertheless, it is important to highlight that all the tracts have a high degree of reproducibility with a correlation average above 0.80, with only a few exceptions. At the macrostructural level, the agreement indices showed similar patterns of reproducibility across tracts of interest. The group average of bundle adjacency, were all under 1 for all the tracts except the left MD-Area 25, which has an adjacency of 1.2, 85 indicating the streamlines identified in the test have very close distance to the streamlines identified in the retest. High reproducibility is also reflected by the Dice index, and streamline density correlation. Most tracts have the test-retest group average Dice index and density correlation above 0.7. Similar to the microstructural level of reproducibility, the tracts connecting MD with the orbital frontal cortex showed a relatively lower Dice index and density correlation. 7.2.3 Post-hoc analysis results Post-hoc analysis was conducted separately for computational and test-retest reproducibility, to investigate the associations between the reproducibility and three factors of interest (diffusion data noise, tract streamline length and quantity). The results related to computational reproducibility showed that the diffusion noise has little correlation with the reproducibility indices except the FA profile correlation (Figure 7.6A), which has a moderate negative correlation (-0.38, higher noise in diffusion data with lower FA profile correlation). The streamline length and count both show high correlation with all the four reproducibility indices, only with the opposite direction. The length is negatively correlated with computational reproducibility: the longer the streamlines, the less reproducible across computations. Streamline count has the opposite pattern: the more streamlines one tract is composed of, the higher the reproducibility across computations. The correlation results for the test-retest reproducibility have similar direction as in the computational reproducibility (Figure 7.6B). The noise has a negative correlation with test-retest reproducibility, ranging from -0.69 to -0.48. Opposite to the strong correlation between the tract streamline length and the computational reproducibility, the length has a null correlation with test-retest reproducibility. Regarding the correlation of the tract 86 streamline count, it has a similar strong correlation with test-retest reproducibility as it does with computational reproducibility. Figure 7.6. DWI noise, streamline length and streamline count were found to be correlated with reproducibility across tracts. A) The correlation between the three factors with four computational reproducibility indices. B) The correlation between the three factors with four test-retest reproducibility indices. *p< .05, **p< .01, ***p< .001. The symbols of the correlation coefficients with bundle adjacency were reversed to simplify the illustration, as the value of bundle adjacency indicates the variance between tracts instead of similarity. 7.3 Discussion The current study proposes a reproducible protocol to reconstruct the tractography of the AN and MD from the human thalamus. These nuclei play important roles in various cognitive processes, in particular during rapid integration of new learning, working memory, decision making and beyond. The reproducibility of the protocol for reliably obtaining these tracts of interest was measured by examining their microstructural tractometric properties and volume-based macrostructual similarity across repeating computations and test-retest sessions. Results revealed nearly perfect computational reproducibility across ten repetitions and high-to-excellent test-retest reproducibility of the AN and MD related thalamocortical tracts. The structural connectivity profile of the higher-order thalamic nuclei, AN and MD, has been examined in numerous experiments on non-human primates (Bay & Çavdar, 2013; 87 Groenewegen, 1988; Lozsádi, 1995; Ray & Price, 1992; Vann et al., 2007) and postmortem studies on humans (Blennow et al., 2000; Cullen et al., 2003). However, the in vivo examinations of the white-matter connection of those nuclei with subcortical and cortical regions in humans are rare. Due to the lack of a well defined and validated thalamic segmentation, most tractography studies on humans used the whole thalamus (Fan et al., 2014; O’muircheartaigh et al., 2015; Pelzer et al., 2017) or segmentation in a common space (Klein et al., 2010) to estimate the structural connections of the thalamus. Unlike the first-order thalamic nuclei having relatively straightforward structural pathways with the sensorimotor cortical regions, the AN and especially MD, have more extensive and complex efferent and afferent pathways with cortical structures (Aggleton & Mishkin, 1984; Gower, 1989; Mai & Forutan, 2012; Shah et al., 2012). The current study adopted the first probabilistic atlas combining ex vivo MRI and histological data to define the thalamic seeds, which is implemented in Freesurfer (Iglesias et al., 2018). It provides a precise, reliable and automatic thalamic ROI definition, which is critical for the very basis of a reproducible and reliable protocol of thalamic tractography. Moreover, the major thalamic tracts of AN and MD were reconstructed from state-of-art DWI sequences according to the physiological descriptions from literature and guidance from an expert anatomist. The reproducibility of the reconstruction procedures is demonstrated in the current study. The reconstruction procedures are implemented in a containerized pipeline that allows other researchers to reconstruct the same tracts with their data. In the next section I will discuss the reproducibility of the protocol from two different approaches. The computational reproducibility measures how reliable the protocol is to reconstruct the tracts of interest across multiple computations with the same data and same computation parameters. As expected, across ten separate and independent computations using the same raw data and protocol, the reproducibility is nearly perfect. Most tracts have high 88 presented with an instruction asking them to produce noun words with their eyes closed. Participants were instructed to produce one word during each silence of the sparse-sampling fMRI protocol. Similarly, in the non-linguistic motor condition participants were asked to produce 12 unintelligible sounds during each silence break while having their eyes closed. In every activation block during which the participants had their eyes closed (i.e., auditory linguistic, auditory non-linguistic, motor linguistic and motor non-linguistic conditions), there was a bell sound playing through the scanner sound system to signal participants the end of the activation block and that they have to open their eyes again. A total of 168 Basque noun words (e.g. poltsa,bag in English) were selected for the visual and auditory tasks. Half of the words were presented visually in the reading task, and the remaining half were presented auditorily in speech comprehension tasks, in which the audio was recorded by a female Basque native speaker. Scrambled words in the visual non-linguistic task were designed by creating 10 × 10 pixel tiles and mixing them randomly, using the same word images from the reading task. The noise audios in the auditory non-linguistic task were created by randomly shifting the auditory signal in time domain, using the same speech words used in the speech comprehension task. The stimuli in visual and auditory modalities were counterbalanced across participants such that the visual word images and corresponding perceptual stimuli used in half of the participants will be presented in auditory, and vice versa for auditory modality stimuli. In the linguistic motor or speech word production task, participants were instructed to produce object names existing in their familiar environment, such as a table or a keyboard in an office. In the non-linguistic motor, participants have to produce unintelligible sounds that the experimenter showed to the participant before undergoing MRI scanning. These unintelligible sounds (e.g., palatal click sound) had no semantic meaning. Before participants underwent MRI scanning, they practiced a behavioral version with the six main conditions of the fMRI experiment to familiarize them with the procedure and the instructions presented during the tasks. 95 8.1.3 MRI data acquisition Whole-brain MRI data acquisition was conducted on a 3-T Siemens PRISMA Fit whole-body MRI scanner (Siemens Medical Solutions) using a 64-channel whole-head coil. Functional images were acquired with a sparse-sampling paradigm (effective repetition time (TR) = 2.9 s, real TR = 1.7 s) in a single gradient-echo echo-planar multiband pulse sequence with the following acquisition parameters: TE =35 ms; MB acceleration factor = 5; 65 axial slices with a 2.4 mm3voxel resolution; no inter-slice gap; flip angle = 56º; FoV = 210 mm; 1169 volumes in 7 runs. High-resolution MPRAGE T1-weighted structural images were also collected for each participant with the following parameters: TR = 2530 ms; TE = 2.36 ms; flip angle = 7°; FoV = 256 mm; voxel resolution = 1 mm3; 176 slices. In total 100 diffusion weighted images were acquired with the anterior to posterior phase-encoding direction and 50 isotropically distributed diffusion-encoding gradient directions. The 100 diffusion weighted images included 50 images with b-values of 1000 s/mm2and 50 images with b-values of 2000 s/mm2. Twelve images with no diffusion weighted (b-values of 0 s/mm2) were obtained for motion correction and geometrical distortion correction, which comprised five images with the same phase-encoding direction as the DWI images and seven images with the reversed phase-encoding direction (posterior to anterior). Both DWIs and b0 images shared the following parameters: TR = 3600 ms, TE = 73 ms, FA = 78°, voxel size = 2 mm isotropic, 72 slices with no gap and a multiband acceleration factor of 3. 8.1.4 MRI data analysis For structural image analysis, the T1w images were processed using RTP-anatROIs, which involves processing the subjects’ anatomical T1w image with recon-all from freesurfer (http://surfer.nmr.mgh.harvard.edu) and extract ROIs. First, Freesurfer was used to perform cortical/subcortical segmentation and parcellation. Next, the thalamic nuclei were obtained by running the thalamic segmentation module implemented in Freesurfer on a probabilistic atlas built based on histological and high-resolution ex vivo MRI data (Iglesias et al., 2018). For this study, we only considered first-order relay nuclei as ROIs: the LGN, MGN and VLN. For functional and structural 96 connectivity analyses, we also extracted the cortical regions V1/V2, A1 and M1. To parcellate the visual cortex we ran the Neuropythy (Benson & Winawer, 2018; https://github.com/noahbenson/neuropythy) tool on the Freesurfer results. A combination of the resulting V1 and V2 ROIs was used for our visual cortex ROI. A1 and M1 were converted from the human connectome project (HCP) atlas (Glasser et al., 2016). To convert them to individual subject space, we performed a non-linear registration of a 1 mm3MNI template using Advanced Normalization Tools (ANTs, http://stnava.github.io/ANTs/). For functional images preprocessing, we used SPM12 (Wellcome Center for Human Imaging, London) preprocessing routines and analysis methods. Images were corrected for differences in slice acquisition timing across every functional scan and then realigned for motion correction. Afterwards, each subject’s functional volumes were smoothed using a 2 mm full-width half-maximum (FWHM) Gaussian kernel. Motion parameters were extracted from the realignment step to inform a volume repair procedure (ArtRepair; Stanford Psychiatric Neuroimaging Laboratory) that identified bad volumes on the basis of scan-to-scan movement (>0.5 mm) and signal fluctuations in global intensity (>1.3%) and corrected bad volumes via interpolation from the nearest non-repaired scans. Five participants with more than 15% to-be-corrected outlier functional volumes were excluded. After volume repair, functional volumes were separately coregistered in two different ways: 1) to MNI space in order to conduct whole-brain contrasts in normalized MNI space at the group level; 2) to high-resolution anatomical T1 images and resliced from the original 2.4 mm3functional voxel dimensions to 1 mm3voxels in anatomical T1 space for ROI analysis and functional connectivity. Finally, time series were temporally filtered to eliminate contamination from slow frequency drift (high-pass filter: 128s). Statistical analyses were performed on each subject data from both the MNI space and individual space using the general linear model (GLM). A series of impulses convolved with a canonical hemodynamic response function (HRF) were used to model the fMRI time series data. The six main experimental conditions in our design (i.e., 2 Task X 3 Modaly) were modeled as epochs from the onset of the first trial within each block until the last trial within the block, resulting in 20.4s 97 (12 x 1.7) periods. These functions were used as covariates in the GLM. The motion parameters for translation (i.e., x, y, z) and rotation (i.e., yaw, pitch, roll) were used as covariates of non-interest in the GLM. SPM12 FAST was used for temporal autocorrelation modeling in this GLM due to its optimal performance in terms of removing residual autocorrelated noise in first-level analyses (Olszowy et al., 2019). The least-squares parameter estimates of the height of the best-fitting canonical HRF for each condition were used in pairwise contrasts. At the group level, whole-brain contrasts between conditions were computed in MNI space by performing paired t-tests on these images from different conditions, treating participants as a random effect. Three whole-brain contrasts were conducted: Visual tasks - (Auditory + Motor tasks), Auditory tasks - (Visual + Motor tasks) and Motor tasks - (Visual + Auditory tasks). These three contrasts were selected to examine the modality-specific regions at the whole-brain level. Our standard statistical threshold for whole-brain maps was a Family Wise Error (FWE) set to p< .05 at the voxel level. Individual ROI analysis was performed on GLM results from the individual space with the MARSBAR toolbox for use with SPM12. Given that this study focused on the involvement of first-order thalamic nuclei in language processing, the LGN, MGN and VLN were selected from both hemispheres as ROIs. Then, in line with our main experimental design, we extracted parameter estimates (i.e., scaled % signal change values) for each single region and subject individually and used them as dependent variables in 2 (Task: linguistic and non-linguistic) by 3 (Modality: visual, auditory and motor) repeated measures ANOVAs. As we were interested in whether there is a dissociation of thalamic involvement between linguistic and non-linguistic tasks, we tested the left and right thalamic ROIs separately and compared their involvement between linguistic and non-linguistic tasks in their corresponding modality based on our hypotheses. Functional connectivity was examined between each first-order thalamic nuclei of interest (LGN, MGN and VLN) and the site of cortical termination of the corresponding sensorimotor pathways (V1/V2, A1 and M1). The functional connectivity analyses were conducted using the beta-series correlation method (Rissman et al., 2004), implemented in SPM12 with custom Matlab scripts. The canonical HRF in SPM was fit to each trial from each experimental condition and the 98 resulting parameter estimates (i.e., beta values) were sorted according to the study conditions to produce a condition-specific beta series for each voxel. Pairwise functional connectivity analysis between thalamic nuclei and cortical regions were conducted at the individual-subject level for each task in the corresponding sensorimotor modality (LGN-V1V2 in visual, MGN-A1 in auditory and VLN-M1 in motor). The beta-series correlation values (r values) were transformed to Fisher’s z values by applying an arc hyperbolic tangent transform (Fisher, 1921) at the subject level for each pair of ROIs and each experimental condition. Since the correlation coefficient is inherently restricted to range from −1 to +1, this transformation ensured the null hypothesis sampling distribution approached that of the normal distribution. To assess the significance of the correlation findings at the group level, the z-transformed correlation of the individual subjects were compared against zero at group level. Structural connectivity analysis was conducted by using the tractography protocol proposed in Study 1 to reconstruct the first-order relay thalamic tracts on the DWI data collected in this study. Three pairs of first order thalamic tracts were reconstructed: bilateral OR, AR and MR. The FA profile of each tract from each subject was obtained using RTP-pipeline. More details can be found in chapter 6, section 6.1.4. The mean of the FA profile was calculated as the index by averaging the 100 FA values in the FA profile for each tract and each subject. For each first-order thalamocortical pathway of interest, to examine the associations between the structural connectivity and functional connectivity, correlation analyses were conducted between the FA values from structural connectivity and z values from functional connectivity of the corresponding modality task. 8.2 Results 8.2.1 Whole-brain contrasts To identify brain regions associated with processing specific modalities across all participants and linguistic-non linguistic tasks, we computed three whole-brain contrasts: visual > (auditory + motor); auditory > (visual + motor) and motor > (visual + auditory). These contrasts revealed the involvement of the first-order thalamic relay nuclei and the primary corresponding sensorimotor cortical regions in modality specific tasks. The visual > (auditory + motor) contrast showed increased 99 activation in bilateral LGN and primary and secondary visual cortex in visual tasks compared to tasks in other two modalities (Figure 8.1A). Similarly, the auditory > (visual + motor) contrast revealed the auditory thalamic nuclei bilateral MGN and A1 (Figure 8.1B). The motor > (visual + auditory) contrast showed higher involvement of the motor thalamic nuclei (bilateral VLN) and M1. Vermis III in the cerebellum also showed specific involvement in the motor contrast (Figure 8.1C). Using a mask of the entire thalamus with these same contrasts revealed the specificity of each of them showing functional activation of the expected thalamic nuclei: LGN for visual > (auditory + motor) contrast; MGN for auditory > (visual + motor) contrast; and, VLN extended to mediodorsal nucleus for the motor > (visual + auditory) contrast (Figure 8.1D) Figure 8.1.A) Brain sections showing activations for the visual > auditory + motor whole-brain contrast across all subjects. B) Brain sections showing activations for the auditory > visual + motor whole-brain contrast across all subjects. C) Brain sections showing activations for the motor > visual + auditory whole-brain contrast across all subjects. D) Brain sections showing the same three contrasts using a mask of the entire thalamus (green = visual > auditory + motor, red = auditory > visual + motor, blue = motor > visual + auditory). All brain sections presented here are in MNI space. The statistical threshold was p < 0.05 FWEcorrected at the voxel level. 8.2.2 ROI results ROI analyses were conducted to characterize the activation profile of the three thalamic ROIs (LGN, MGN and VLN) bilaterally for the main experimental tasks in three perceptual modalities. We 100 extracted fMRI parameter estimates from these six ROIs and conducted hypothesis-driven analyses based on planned comparisons between conditions. LGN. Results from bilateral LGN are presented in Figure 8.2A. A repeated measures analysis of variance (ANOVA) was conducted for the LGN parameters estimates from both hemisphere separately showed a main effect of Modality, for left LGN [ F(2,228) = 26.39; p< 0.001, ηp 2= 0.19] and right LGN [F(2,228) = 23.30; p< 0.001, ηp 2= 0.17], with stronger activation for the visual tasks relative to the auditory and motor tasks. Planned comparisons revealed that LGN showed higher activation in both linguistic and non-linguistic tasks in visual modality compared to their counterparts in auditory and motor modalities, with the exception of the right LGN activation to non-linguistic tasks having no difference between visual and motor modalities. More importantly, since we are interested in possibly different involvement of the first-order thalamic ROIs between linguistic and non-linguistic tasks in the corresponding modality, simple-effects comparisons between linguistic and non-linguistic tasks in visual modality were conducted for both left and right LGN. The results revealed that the left LGN showed higher involvement for visual linguistic task compared to visual non-linguistic task (p< 0.05), while this difference was not observed for right LGN (p= 0.17). MGN. The repeated measures ANOVA conducted for bilateral MGN showed the main effect of Modality, for left MGN, [F(2,228) = 3.77; p< 0.05, ηp 2= 0.03]; for right MGN, [F(2,228) = 5.27; p< 0.01, ηp 2= 0.04], revealing stronger activation for auditory tasks relative to the motor tasks (Figure 8.2B). Simple effects comparisons revealed that left MGN showed higher activation for linguistic tasks in the auditory modality compared to both visual and motor modalities. Similar patterns were also found in right MGN, along with higher activation in non-linguistic tasks in the auditory modality than in the visual modality. Finally, the left MGN showed stronger engagement in the auditory modality for the linguistic task than for the non-linguistic task (p< 0.01), and this effect was not observed for the right MGN (p= 0.79). VLN. As expected, same ANOVA conducted for bilateral VLN showed the main effect of Modality for left VLN, [F(2,228) = 48.8; p< 0.001, ηp 2= 0.30] and right VLN, [F(2,228) = 49.3; p< 101 0.001, ηp 2= 0.30], with stronger activation for motor tasks relative to the visual and auditory tasks (Figure 8.2C). Simple-effects comparisons conducted separately for linguistic and non-linguistic revealed higher activation in both left and right VLN in the motor modality compared to both visual and auditory modalities. Different from the results of left and right LGN and MGN, comparison in the VLN between activation of linguistic and non-linguistic tasks in motor modality did not reveal statistically significant differences (ps > 0.12). Figure 8.2. ROI analyses for three first-order relay thalamic nuclei: A) bilateral LGN; B) bilateral MGN and C) bilateral VLN. Bar graphs show averaged parameter estimates (% signal change) of each thalamic nuclei for linguistic and non-linguistic tasks in the three modalities: visual, auditory and motor. *p< 0.05, **p< 0.01. 8.2.3 Functional connectivity results Pairwise functional connectivity analyses of three first-order relay thalamic pathways for the linguistic and non-linguistic tasks of the corresponding modalities were examined: LGN-V1/V2 visual pathway for linguistic and non-linguistic tasks in the visual modality; MGN-A1 auditory pathway for 102 linguistic and non-linguistic tasks in the auditory modality; and, VLN-M1 motor pathway for linguistic and non-linguistic tasks in the motor modality. To examine the significance of the correlation findings at the group level, the z-transformed correlation values of the individual subjects were compared against zero. The results revealed that predicted functional connections between regions based on known neuroanatomy were statistically significant for both linguistic (Figure 8.3A) and non-linguistic tasks (Figure 8.3B). For the visual pathway, functional coupling of LGN with V1/V2 for linguistic tasks (z=0.56, p< 0.001 for left and z=0.58, p< 0.001 for right) and non-linguistic task (z=0.52, p< 0.001 for left; z=0.51, p< 0.001 for right). Similarly significant functional coactivation was found between MGN and A1 for auditory linguistic and non-linguistic tasks, as well as between the VLN and M1 for motor linguistic and non-linguistic tasks. Simple-effect analyses were conducted to examine if there was any statistically significant difference between the functional connectivity in linguistic and non-linguistic tasks for each of these first-order thalamic pathways, with none of these comparisons revealing statistically significant differences (ps ≥ 0.54). 103 Figure 8.3. Functional connectivity analyses of three first-order relay thalamic pathways in A) linguistic and B) non-linguistic tasks for each of the corresponding modalities. Lines with an arrow represent the connection of the first-order thalamic relay nuclei with the cortical primary sensorimotor regions. Thalamic nuclei, cortical regions and the lines are colored as a function of sensorimotor modality: green indicates visual modality, red indicates auditory modality and blue indicates motor modality. The values along the arrow lines are the Fisher-transformed z values. All the reported z values are significant against zero with p< 0.001 after Bonferroni FWE correction for multiple comparisons. 104 9 General Discussion The current dissertation is focused on the human thalamus structure and function. In the first two empirical studies, I examined the white-matter fiber connections between the thalamus and cortical structures and subcortical structures, more specifically the first-order thalamic tracts of LGN, MGN and VLN in Study 1, and the higher-order thalamic tracts of AN and MD in Study 2. The protocols of the reconstruction of those tracts from DWI data have proved to have high reproducibility, and are made public for the scientific community to reproducibly reconstruct the same tracts in their own data. Moreover, as pioneering work, in study 3 we examined the function of human thalamus in language functions, more specifically, the involvement of the first-order thalamic nuclei in some of the main human language systems. We showed that the LGN and MGN activation is modulated as a function of the stimuli type. These results suggest that the thalamus is involved in human language processing and underscore the need for more systematic studies of the involvement of the thalamus in language processing. In Study 1 of this doctoral dissertation, we focused on the first-order thalamic white-matter tracts, namely, the OR, AR, MR and DT. These thalamic tracts of interest are critical structures in relaying information from periphery or cerebellum to cortical structures. They are contrasted to higher-order thalamic white-matter tracts, such as those reported in Study 2, which are believed to serve as links in cortico-thalamo-cortical pathways that continue the information flow between cortical structures (Ramcharan et al., 2005). The tractography technology allows to visualize white matter fibers in vivo and offers the opportunity of extracting microstructural information to perform quantitative analyses. On the other hand, it is also well known that probabilistic tractography can come with false positive or negative results due to data noise, partial volume effects, and complex anatomical properties such as crossing fibers (Pierpaoli et al., 2001; Wiegell et al., 2000). In fact, the reproducibility of tractography has been under discussion for a long time, but there are no general answers to how to improve reproducibility of tractography, as it depends on the MRI sequences, tracking parameters and the specific tracts under investigation. Previous studies have used in vivo 111 tractography on DWI data to reconstruct these first-order thalamic tracts, but only a few of them have reported reproducibility measures about the reconstruction of the specific white-matter tracts. Different from previous work, this study tested the first-order thalamic tracts in the same study using similar methods and reconstruction procedures across all them. Each tract has gone through multiple parameter iterations to obtain the most optimal trajectory that aligns with the extant neuroanatomical knowledge. More importantly, we specifically investigated in a large dataset the reliability of the reconstruction protocol in terms of both computational and test-retest reproducibility. The reproducibility was measured at both microstructural and macrostructural levels for each tract. Results from Study 1 revealed that the proposed protocol could reliably reconstruct these first-order thalamic tracts, and obtain reproducible microstructural and macrostructural measurements that can reflect the characteristics of these white-matter bundles. This protocol has been implemented in a docker container and made publicly available, so it can be easily used by other researchers in the community. This unique tool allows research to test hypotheses as to whether any of these specific first-order thalamic tracts are related with cognitive functions or diseases of interest, with the accuracy and reproducibility of the tract reconstruction guaranteed. The Study 2 extends the rationale of Study 1 to higher-order thalamic white-matter bundles. Conventionally, the thalamic relays can be classified into firstand second (or higher)-order. First-order thalamic tracts connect these first-order thalamic relays with their corresponding sensorimotor cortical areas. In contrast, higher-order thalamic tracts relay information between cortical areas, such as the white-matter tracts connecting the MD and PFC, representing a cortico-thalamo-cortical circuit (Mitchell, 2015; Sherman, 2017). The higher-order thalamic tracts often have more structural and functional complexity and are far from being fully understood in humans. For example, the MD has extensive connections with practically the whole PFC, which is involved in numerous cognitive functions, such as working memory, attention, and decision making (Clark et al., 2010). Each subregion of the PFC has typically specific roles in cognition. These PFC subregions receive afferent fibers from the MD in the thalamus via the anterior thalamic radiation. In Study 2, the afferent fiber tracts from AN and MD were reconstructed from DWI data and the 112 corresponding computational and test-retest reproducibility was examined. Our protocol included 42 pairs of left and right hemispheric tracts, with 3 tracts being examined for the AN and 39 tracts being investigated for the MD). The reproducibility results showed that in general these tracts can be reliably reconstructed from DWI data with the proposed protocol. There were specific tracts showing relatively lower reproducibility, for example the AN related tracts and tracts connecting MD with orbital frontal cortex. To explore the associations between the reproducibility and the possible influential factors on reproducibility, a post-hoc analysis was conducted. The results unveiled three factors strongly associated with reproducibility of specific tracts. For example, noise in the diffusion data has strong negative correlation with the test-retest reproducibility. Also, the computational reproducibility is linked with the streamline length and count. If one tract has longer or less amount of streamlines, it might lead to relatively lower computational reproducibility. To the best of our knowledge, no study has investigated so far the computational reproducibility of tractography methods, as researchers assume that the same data and same methods mostly lead to the same results. These insights are valuable as it should promote the systematic investigation on computational reproducibility of tratography. In the third and final empirical study, we investigated the involvement of the three first-order thalamic nuclei (LGN, MGN and VLN) in three main human language systems: reading, speech comprehension and production. The results revealed stronger engagement of the LGN, MGN and VLN for both linguistic and non-linguistic tasks in their corresponding modalities. More importantly, we found stronger activation for linguistic versus non-linguistic stimuli in reading and speech comprehension for LGN and MGN, respectively. For example, the LGN showed higher activation for reading words than for seeing scrambled pixels that are perceptually equal to visual words. Very few studies have investigated the subcortical contributions on high-level cognitive functions such as language (Parvizi, 2009), with most of the studies to date examining the neurobiology of language being focused on the cortical areas (Friederici, 2002; Hickok & Poeppel, 2007; Price, 2000). To the best of our knowledge, this study is the first to investigate the three modalities in relation to the involvement of first-order thalamic relays in their respective human language systems, and proved the 113 associations between linguistic and non-linguistic tasks in the LGN and MGN. Our findings showed that the responses in LGN and MGN during the processing of visual and auditory stimuli are task dependent, which implies the existence of a feedback mechanism that supports the recognition of visual and auditory information at the level of sensory thalamic nuclei. Given that this association has been found in the left thalamic nuclei but not on the right, we postulate that this feedback mechanism can be, very possibly, language-specific. The studies in the current dissertation have some limitations. The first two studies have tested the reproducibility of the proposed protocol using a state-of-art DWI sequence, which includes, for example, multiband and multi-shell techniques. Although we cannot guarantee that the exact same computational and test-retest reproducibility will be obtained with different DWI acquisition protocols, we do not expect that the reproducibility profiles described here for these thalamocortical projections may change dramatically when using other DWI protocols widely used in the past, such as monoband or single-shell data. Despite the fact that in this study we decided to go with state-of-the-art DWI sequences, the current protocols can be easily adapted to different sequences. Furthermore, the reproducibility of the proposed protocol was measured on DWI acquired from a healthy population. It would be also relevant in the future to examine the reproducibility to reconstruct the thalamic tracts from DWI data in clinical populations. In Study 2, the reproducibility has been linked to some factors of the DWI data or the neuroanatomical characteristics of specific tracts. Three factors were under investigation in this work, and in future studies other possible factors that could be associated with reproducibility, such as pathological features or infant populations, should be examined. In the task-based fMRI Study 3, the experimental design examined the activation profile of the first-order thalamic nuclei for linguistic and for non-linguistic tasks attending to three modalities (visual, auditory, motor). This study paves the road for follow-up analyses and experiments. First, an open question from this study is the task-dependency on first-order thalamic nuclei in their language-specific modality. We hypothesized that this is very likely language-specific as we observed differences as a function of the linguistic nature of the stimuli in LGN and MGN only in the left hemisphere, which is the dominant hemisphere for language function. We plan to further explore these 114 differences and try to answer this question. Second, the role that task modulation on first-order thalamic nuclei plays in high-level cognitive functions, such as language, is still not clear. Future studies including more systematic manipulations might shed light on the mechanisms underlying task modulation on first-order thalamic nuclei. In sum, the current dissertation successfully reconstructed first-order and higher-order thalamic white-matter tracts from DWI data, and has proved high reproducibility of the reconstruction protocol. This protocol could benefit the tractography community to better understand the structural connectivity of the thalamus with cortical and subcortical structures and facilitate the research on thalamocortical pathways in humans. We also found evidence for differences in the processing of linguistic and nonlinguistic stimuli in first-order thalamic nuclei through a task-based fMRI study. These results suggest that the first-order thalamic nuclei play roles in human language that are beyond relaying sensory information from periphery to cerebral cortex. These findings are important to push forward our understanding on the role of subcortical structures, such as the thalamus, in human language functions, and to urge a revisitation of existing language models taking the thalamus into consideration. 115 10 Bibliography Abivardi, A., & Bach, D. R. (2017). Deconstructing white matter connectivity of human amygdala nuclei with thalamus and cortex subdivisions in vivo. Human Brain Mapping,38(8), 3927–3940. https://doi.org/10.1002/hbm.23639 Aggleton, J. P., & Brown, M. W. (1999). Episodic memory, amnesia, and the hippocampal–anterior thalamic axis. Behavioral and Brain Sciences,22(3), 425–444. https://doi.org/10.1017/S0140525X99002034 Aggleton, J. P., & Mishkin, M. (1984). Projections of the amygdala to the thalamus in the cynomolgus monkey. Journal of Comparative Neurology,222(1), 56–68. Alvarez, I., Schwarzkopf, D. S., & Clark, C. A. (2015). Extrastriate projections in human optic radiation revealed by fMRI-informed tractography. Brain Structure and Function, 220(5), 2519–2532. https://doi.org/10.1007/s00429-014-0799-4 Anastasiades, P. G., Collins, D. P., & Carter, A. G. (2021). Mediodorsal and Ventromedial Thalamus Engage Distinct L1 Circuits in the Prefrontal Cortex. Neuron,109(2), 314-330.e4. https://doi.org/10.1016/j.neuron.2020.10.031 Andolina, I. M., Jones, H. E., Wang, W., & Sillito, A. M. (2007). Corticothalamic feedback enhances stimulus response precision in the visual system. Proceedings of the National Academy of Sciences of the United States of America,104(5), 1685–1690. https://doi.org/10.1073/pnas.0609318104 Arrigo, A., Calamuneri, A., Mormina, E., Gaeta, M., Quartarone, A., Marino, S., Anastasi, G. P., & Aragona, P. (2016). New insights in the optic radiations connectivity in the human brain. Investigative Ophthalmology and Visual Science,57(1), 1–5. https://doi.org/10.1167/iovs.15-18082 Aydogan, D. B., Jacobs, R., Dulawa, S., Thompson, S. L., Francois, M. C., Toga, A. W., Dong, H., Knowles, J. A., & Shi, Y. (2018). When tractography meets tracer injections: A systematic study of trends and variation sources of diffusion-based connectivity. Brain Structure and Function,223(6), 2841–2858. 116 https://doi.org/10.1007/s00429-018-1663-8 Bammer, R. (2003). Basic principles of diffusion-weighted imaging. European Journal of Radiology,45(3), 169–184. https://doi.org/10.1016/S0720-048X(02)00303-0 Bassi, L., Ricci, D., Volzone, A., Allsop, J. M., Srinivasan, L., Pai, A., Ribes, C., Ramenghi, L. A., Mercuri, E., Mosca, F., & others. (2008). Probabilistic diffusion tractography of the optic radiations and visual function in preterm infants at term equivalent age. Brain, 131(2), 573–582. Bay, H. H., & Çavdar, S. (2013). Regional connections of the mediodorsal thalamic nucleus in the rat. Journal of Integrative Neuroscience,12(02), 201–219. https://doi.org/10.1142/S021963521350012X Behrens, T. E. J., Berg, H. J., Jbabdi, S., Rushworth, M. F. S., & Woolrich, M. W. (2007). Probabilistic diffusion tractography with multiple fibre orientations: What can we gain? NeuroImage,34(1), 144–155. https://doi.org/10.1016/j.neuroimage.2006.09.018 Behrens, T. E. J., Johansen-Berg, H., Woolrich, M. W., Smith, S. M., Wheeler-Kingshott, C. A. M., Boulby, P. A., Barker, G. J., Sillery, E. L., Sheehan, K., Ciccarelli, O., Thompson, A. J., Brady, J. M., & Matthews, P. M. (2003). Non-invasive mapping of connections between human thalamus and cortex using diffusion imaging. Nature Neuroscience,6(7), 750–757. https://doi.org/10.1038/nn1075 Benjamin, C. F. A., Singh, J. M., Prabhu, S. P., & Warfield, S. K. (2014). Optimization of tractography of the optic radiations. Human Brain Mapping,35(2), 683–697. https://doi.org/10.1002/hbm.22204 Benson, N. C., & Winawer, J. (2018). Bayesian analysis of retinotopic maps. ELife,7, 1–29. https://doi.org/10.7554/eLife.40224 Berker, E. A., Berker, A. H., & Smith, A. (1986). Translation of Broca’s 1865 Report: Localization of Speech in the Third Left Frontal Convolution. Archives of Neurology, 43(10), 1065–1072. https://doi.org/10.1001/archneur.1986.00520100069017 Berman, J. I., Lanza, M. R., Blaskey, L., Edgar, J. C., & Roberts, T. P. L. (2013). High angular resolution diffusion imaging probabilistic tractography of the auditory radiation. 117 American Journal of Neuroradiology,34(8), 1573–1578. https://doi.org/10.3174/ajnr.A3471 Besseling, R. M. H., Jansen, J. F. A., Overvliet, G. M., Vaessen, M. J., Braakman, H. M. H., Hofman, P. A. M., Aldenkamp, A. P., & Backes, W. H. (2012). Tract specific reproducibility of tractography based morphology and diffusion metrics. PLoS ONE, 7(4). https://doi.org/10.1371/journal.pone.0034125 Binder, J. R. (2015). The Wernicke area: Modern evidence and a reinterpretation. Neurology, 85(24), 2170–2175. https://doi.org/10.1212/WNL.0000000000002219 Binder, J. R., Desai, R. H., Graves, W. W., & Conant, L. L. (2009). Where is the semantic system? A critical review and meta-analysis of 120 functional neuroimaging studies. Cereb Cortex,19(12), 2767–2796. https://doi.org/10.1093/cercor/bhp055 Binder, J. R., Rao, S. M., Hammeke, T. A., Frost, J. A., Bandettini, P. A., Jesmanowicz, A., & Hyde, J. S. (1995). Lateralized Human Brain Language Systems Demonstrated by Task Subtraction Functional Magnetic Resonance Imaging. Archives of Neurology, 52(6), 593–601. https://doi.org/10.1001/archneur.1995.00540300067015 Blennow, K., Bogdanovic, N., Heilig, M., Grenfeldt, B., Karlsson, I., & Davidsson, P. (2000). Reduction of the synaptic protein rab3a in the thalamus and connecting brain regions in post-mortem schizophrenic brains. Journal of Neural Transmission,107(8), 1085–1097. Bonilha, L., Gleichgerrcht, E., Fridriksson, J., Breedlove, J. L., Rorden, C., Nesland, T., Paulus, W., Helms, G., & Focke, N. K. (2015). Reproducibility of the structural brain connectome derived from diffusion tensor imaging. PLoS ONE,10(9), 1–17. https://doi.org/10.1371/journal.pone.0135247 Boukadi, M., Marcotte, K., Bedetti, C., Houde, J. C., Desautels, A., Deslauriers-Gauthier, S., Chapleau, M., Boré, A., Descoteaux, M., & Brambati, S. M. (2019). Test-retest reliability of diffusion measures extracted along white matter language fiber bundles using Hardi-based tractography. Frontiers in Neuroscience,13(JAN). https://doi.org/10.3389/fnins.2018.01055 118 Bridge, H., Leopold, D. A., & Bourne, J. A. (2016). Adaptive Pulvinar Circuitry Supports Visual Cognition. Trends in Cognitive Sciences,20(2), 146–157. https://doi.org/10.1016/j.tics.2015.10.003 Brugge, J. F. (2013). Anatomy and physiology of auditory pathways and cortex. Handbook of Clinical Neurophysiology,10, 25–59. Buchanan, C. R., Pernet, C. R., Gorgolewski, K. J., Storkey, A. J., & Bastin, M. E. (2014). Test-retest reliability of structural brain networks from diffusion MRI. NeuroImage,86, 231–243. https://doi.org/10.1016/j.neuroimage.2013.09.054 Büchel, C., Turner, R., & Friston, K. (1997). Lateral geniculate activations can be detected using intersubject averaging and fMRI. Magnetic Resonance in Medicine,38(5), 691–694. Bullier, J., & Kennedy, H. (1983). Projection of the lateral geniculate nucleus onto cortical area V2 in the macaque monkey. Experimental Brain Research,53(1), 168–172. Chen, W., Kato, T., Zhu, X.-H., Ogawa, S., Tank, D. W., & Ugurbil, K. (1998). Human primary visual cortex and lateral geniculate nucleus activation during visual imagery. Neuroreport,9(16), 3669–3674. Chen, W., Kato, T., Zhu, X.-H., Strupp, J., Ogawa, S., & Uǧurbil, K. (1998). Mapping of lateral geniculate nucleus activation during visual stimulation in human brain using fMRI. Magnetic Resonance in Medicine,39(1), 89–96. Cherches, I. M. (2016). Clinical neuroanatomy. In Neurology Secrets: Sixth Edition. https://doi.org/10.1016/B978-0-323-35948-1.00002-4 Clark, D. L., Boutros, N. N., & Mendez, M. F. (2010). The brain and behavior: An introduction to behavioral neuroanatomy. Cambridge university press. Coenen, V. A., Allert, N., & Mädler, B. (2011). A role of diffusion tensor imaging fiber tracking in deep brain stimulation surgery: DBS of the dentato-rubro-thalamic tract (drt) for the treatment of therapy-refractory tremor. Acta Neurochirurgica,153(8), 1579–1585. https://doi.org/10.1007/s00701-011-1036-z Coenen, V. A., Allert, N., Paus, S., Kronenbu¨rger, M., Urbach, H., Ma¨dler, B., 119 Kronenbürger, M., Urbach, H., & Mädler, B. (2014). Modulation of the Cerebello-Thalamo-Cortical network in thalamic deep brain stimulation for tremor: A diffusion tensor imaging study. Neurosurgery,75(6), 657–669. https://doi.org/10.1227/NEU.0000000000000540 Concha, L., Gross, D. W., & Beaulieu, C. (2005). Diffusion tensor tractography of the limbic system. American Journal of Neuroradiology,26(9), 2267–2274. Cousineau, M., Jodoin, P. M., Morency, F. C., Rozanski, V., Grand’Maison, M., Bedell, B. J., & Descoteaux, M. (2017). A test-retest study on Parkinson’s PPMI dataset yields statistically significant white matter fascicles. NeuroImage: Clinical,16(March), 222–233. https://doi.org/10.1016/j.nicl.2017.07.020 Cudeiro, J., & Sillito, A. M. (2006). Looking back: Corticothalamic feedback and early visual processing. Trends in Neurosciences,29(6), 298–306. https://doi.org/10.1016/j.tins.2006.05.002 Cullen, T. J., Walker, M. A., Parkinson, N., Craven, R., Crow, T. J., Esiri, M. M., & Harrison, P. J. (2003). A postmortem study of the mediodorsal nucleus of the thalamus in schizophrenia. Schizophrenia Research,60(2–3), 157–166. Dallapiazza, R. F., Lee, D. J., De Vloo, P., Fomenko, A., Hamani, C., Hodaie, M., Kalia, S. K., Fasano, A., & Lozano, A. M. (2019). Outcomes from stereotactic surgery for essential tremor. Journal of Neurology, Neurosurgery and Psychiatry,90(4), 474–482. https://doi.org/10.1136/jnnp-2018-318240 D’Arceuil, H., & de Crespigny, A. (2007). The effects of brain tissue decomposition on diffusion tensor imaging and tractography. NeuroImage,36(1), 64–68. https://doi.org/10.1016/j.neuroimage.2007.02.039 De Benedictis, A., Duffau, H., Paradiso, B., Grandi, E., Balbi, S., Granieri, E., Colarusso, E., Chioffi, F., Marras, C. E., & Sarubbo, S. (2014). Anatomo-functional study of the temporo-parieto-occipital region: Dissection, tractographic and brain mapping evidence from a neurosurgical perspective. Journal of Anatomy,225(2), 132–151. https://doi.org/10.1111/joa.12204 120 structural and functional connectivity of the posterior cingulate cortex: Comparison between deterministic and probabilistic tractography for the investigation of structure-function relationships. NeuroImage,102(P1), 118–127. https://doi.org/10.1016/j.neuroimage.2013.12.022 Klein, J. C., Rushworth, M. F. S., Behrens, T. E. J., Mackay, C. E., de Crespigny, A. J., D’Arceuil, H., & Johansen-Berg, H. (2010). Topography of connections between human prefrontal cortex and mediodorsal thalamus studied with diffusion tractography. NeuroImage,51(2), 555–564. https://doi.org/10.1016/j.neuroimage.2010.02.062 Krauth, A., Blanc, R., Poveda, A., Jeanmonod, D., Morel, A., & Székely, G. (2010). A mean three-dimensional atlas of the human thalamus: Generation from multiple histological data. NeuroImage,49(3), 2053–2062. https://doi.org/10.1016/j.neuroimage.2009.10.042 Kruper, J., Yeatman, J. D., Richie-Halford, A., Bloom, D., Grotheer, M., Caffarra, S., Kiar, G., Karipidis, I. I., Roy, E., Chandio, B. Q., Garyfallidis, E., & Rokem, A. (2021). Evaluating the Reliability of Human Brain White Matter Tractometry. Aperture Neuro, 1(1). https://doi.org/10.52294/e6198273-b8e3-4b63-babb-6e6b0da10669 Kwon, H. G., Hong, J. H., Hong, C. P., Lee, D. H., Ahn, S. H., & Jang, S. H. (2011). Dentatorubrothalamic tract in human brain: Diffusion tensor tractography study. Neuroradiology,53(10), 787–791. https://doi.org/10.1007/s00234-011-0878-7 Lambert, C., Simon, H., Colman, J., & Barrick, T. R. (2017). Defining thalamic nuclei and topographic connectivity gradients in vivo. NeuroImage,158(September 2016), 466–479. https://doi.org/10.1016/j.neuroimage.2016.08.028 Lau, E. F., Phillips, C., & Poeppel, D. (2008). A cortical network for semantics: (De)constructing the N400. Nature Reviews Neuroscience,9(12), 920–933. https://doi.org/Doi 10.1038/Nrn2532 Le Reste, P. J., Haegelen, C., Gibaud, B., Moreau, T., & Morandi, X. (2016). Connections of the dorsolateral prefrontal cortex with the thalamus: A probabilistic tractography 127 study. Surgical and Radiologic Anatomy,38(6), 705–710. https://doi.org/10.1007/s00276-015-1603-8 Lebel, C., Caverhill-Godkewitsch, S., & Beaulieu, C. (2010). Age-related regional variations of the corpus callosum identified by diffusion tensor tractography. NeuroImage,52(1), 20–31. https://doi.org/10.1016/j.neuroimage.2010.03.072 Leh, S. E., Chakravarty, M. M., & Ptito, A. (2008). The connectivity of the human pulvinar: A diffusion tensor imaging tractography study. International Journal of Biomedical Imaging,2008(1). https://doi.org/10.1155/2008/789539 Lehéricy, S., Benali, H., Van De Moortele, P. F., Pélégrini-Issac, M., Waechter, T., Ugurbil, K., & Doyon, J. (2005). Distinct basal ganglia territories are engaged in early and advanced motor sequence learning. Proceedings of the National Academy of Sciences of the United States of America,102(35), 12566–12571. https://doi.org/10.1073/pnas.0502762102 Lerma-Usabiaga, G., Liu, M., Paz-Alonso, P. M., & Wandell, B. A. (2022). Reproducible Tract Profiles (RTP2): From diffusion MRI acquisition to clinical practice and research. BioRxiv. https://doi.org/10.1101/2022.03.17.484761 Lerma-Usabiaga, G., Mukherjee, P., Perry, M. L., & Wandell, B. A. (2020). Data-science ready, multisite, human diffusion MRI white-matter-tract statistics. Scientific Data, 7(1), 1–9. https://doi.org/10.1038/s41597-020-00760-3 Lerma-Usabiaga, G., Mukherjee, P., Ren, Z., Perry, M. L., & Wandell, B. A. (2019). Replication and generalization in applied neuroimaging. NeuroImage,202(December 2018), 116048. https://doi.org/10.1016/j.neuroimage.2019.116048 Li, K., Fan, L., Cui, Y., Wei, X., He, Y., Yang, J., Lu, Y., Li, W., Shi, W., Cao, L., Cheng, L., Li, A., You, B., & Jiang, T. (2022). The human mediodorsal thalamus: Organization, connectivity, and function. NeuroImage,249, 118876. https://doi.org/10.1016/j.neuroimage.2022.118876 Lichtheim, L. (1885). On aphasia. Brain,7, 433–484. Llano, D. A. (2013). Functional imaging of the thalamus in language. Brain and Language, 128 126(1), 62–72. https://doi.org/10.1016/j.bandl.2012.06.004 Lozsádi, D. A. (1995). Organization of connections between the thalamic reticular and the anterior thalamic nuclei in the rat. Journal of Comparative Neurology,358(2), 233–246. Lutz, K., Specht, K., Shah, N. J., & JaÈncke, L. (2000). Tapping movements according to regular and irregular visual timing signals investigated with fMRI. Neuroreport,11(6), 1301–1306. Maciewicz, R. J. (1975). Thalamic afferents to areas 17, 18 and 19 of cat cortex traced with horseradish peroxidase. Brain Research,84(2), 308–312. https://doi.org/10.1016/0006-8993(75)90985-3 Maffei, C., Jovicich, J., De Benedictis, A., Corsini, F., Barbareschi, M., Chioffi, F., & Sarubbo, S. (2018). Topography of the human acoustic radiation as revealed by ex vivo fibers micro-dissection and in vivo diffusion-based tractography. Brain Structure and Function,223(1), 449–459. https://doi.org/10.1007/s00429-017-1471-6 Maffei, C., Sarubbo, S., & Jovicich, J. (2019). Diffusion-based tractography atlas of the human acoustic radiation. Scientific Reports,9(1), 1–13. https://doi.org/10.1038/s41598-019-40666-8 Mai, J. K., & Forutan, F. (2012). Thalamus. In The Human Nervous System. https://doi.org/10.1016/B978-0-12-374236-0.10019-7 Mai, J. K., & Majtanik, M. (2019). Toward a common terminology for the Thalamus. Frontiers in Neuroanatomy,12(January), 1–23. https://doi.org/10.3389/fnana.2018.00114 Maleki, N., Becerra, L., Upadhyay, J., Burstein, R., & Borsook, D. (2012). Direct optic nerve pulvinar connections defined by diffusion MR tractography in humans: Implications for photophobia. Human Brain Mapping,33(1), 75–88. https://doi.org/10.1002/hbm.21194 Mallol, R., Barrós-Loscertales, A., López, M., Belloch, V., Parcet, M. A., & Ávila, C. (2007). Compensatory cortical mechanisms in Parkinson’s disease evidenced with fMRI during the performance of pre-learned sequential movements. Brain Research, 129 1147(1), 265–271. https://doi.org/10.1016/j.brainres.2007.02.046 Mang, S. C., Busza, A., Reiterer, S., Grodd, W., & Klose, A. U. (2012). Thalamus segmentation based on the local diffusion direction: A group study. Magnetic Resonance in Medicine,67(1), 118–126. https://doi.org/10.1002/mrm.22996 Manger, P. R., & Rosa, M. G. P. (2005). Visual thalamocortical projections in the flying fox: Parallel pathways to striate and extrastriate areas. Neuroscience,130(2), 497–511. https://doi.org/10.1016/j.neuroscience.2004.09.047 McFadyen, J., Mattingley, J. B., & Garrido, M. I. (2019). An afferent white matter pathway from the pulvinar to the amygdala facilitates fear recognition. ELife,8, 1–51. https://doi.org/10.7554/eLife.40766 Meola, A., Comert, A., Yeh, F. C., Sivakanthan, S., & Fernandez-Miranda, J. C. (2016). The nondecussating pathway of the dentatorubrothalamic tract in humans: Human connectome-based tractographic study and microdissection validation. Journal of Neurosurgery,124(5), 1406–1412. https://doi.org/10.3171/2015.4.JNS142741 Mihai, P. G., Tschentscher, N., & von Kriegstein, K. (2021). Modulation of the Primary Auditory Thalamus When Recognizing Speech with Background Noise. The Journal of Neuroscience,41(33), 7136–7147. https://doi.org/10.1523/jneurosci.2902-20.2021 Mitchell, A. S. (2015). The mediodorsal thalamus as a higher order thalamic relay nucleus important for learning and decision-making. Neuroscience and Biobehavioral Reviews,54, 76–88. https://doi.org/10.1016/j.neubiorev.2015.03.001 Mori, S., & Van Zijl, P. C. M. (2002). Fiber tracking: Principles and strategies—A technical review. NMR in Biomedicine,15(7–8), 468–480. https://doi.org/10.1002/nbm.781 Mukherjee, P., Berman, J. I., Chung, S. W., Hess, C. P., & Henry, R. G. (2008). Diffusion tensor MR imaging and fiber tractography: Theoretic underpinnings. American Journal of Neuroradiology,29(4), 632–641. Müller-Axt, C., Anwander, A., & von Kriegstein, K. (2017). Altered Structural Connectivity of the Left Visual Thalamus in Developmental Dyslexia. Current Biology,27(23), 3692-3698.e4. https://doi.org/10.1016/j.cub.2017.10.034 130 Nowacki, A., Schlaier, J., Debove, I., & Pollo, C. (2019). Validation of diffusion tensor imaging tractography to visualize the dentatorubrothalamic tract for surgical planning. Journal of Neurosurgery,130(1), 99–108. https://doi.org/10.3171/2017.9.JNS171321 O’Connor, D. H., Fukui, M. M., Pinsk, M. A., & Kastner, S. (2002). Attention modulates responses in the human lateral geniculate nucleus. Nature Neuroscience,5(11), 1203–1209. https://doi.org/10.1038/nn957 Ojemann, G. A. (1975). Language and the thalamus: Object naming and recall during and after thalamic stimulation. Brain and Language,2(C), 101–120. https://doi.org/10.1016/S0093-934X(75)80057-5 Ojemann, G. A. (1991). Cortical Organization of Language and Verbal Memory Based on Intraoperative Investigations.August, 193–230. https://doi.org/10.1007/978-3-642-75964-2_4 Olszowy, W., Aston, J., Rua, C., & Williams, G. B. (2019). Accurate autocorrelation modeling substantially improves fMRI reliability. Nature Communications,10(1), 1220. https://doi.org/10.1038/s41467-019-09230-w O’muircheartaigh, J., Keller, S. S., Barker, G. J., & Richardson, M. P. (2015). White matter connectivity of the thalamus delineates the functional architecture of competing thalamocortical systems. Cerebral Cortex,25(11), 4477–4489. https://doi.org/10.1093/cercor/bhv063 Ouhaz, Z., Fleming, H., & Mitchell, A. S. (2018). Cognitive functions and neurodevelopmental disorders involving the prefrontal cortex and mediodorsal thalamus. Frontiers in Neuroscience,12(FEB), 1–18. https://doi.org/10.3389/fnins.2018.00033 Parent, M., Lévesque, M., & Parent, A. (1999). The pallidofugal projection system in primates: Evidence for neurons branching ipsilaterally and contralaterally to the thalamus and brainstem. Journal of Chemical Neuroanatomy,16(3), 153–165. Parmeggiani, P. L., Azzaroni, A., & Lenzi, P. (1971). On the functional significance of the circuit of Papez. Brain Research,30(2), 357–374. 131 https://doi.org/10.1016/0006-8993(71)90086-2 Párraga, R. G., Ribas, G. C., Welling, L. C., Alves, R. V., & De Oliveira, E. (2012). Microsurgical anatomy of the optic radiation and related fibers in 3-dimensional images. Neurosurgery,71(SUPPL.1), 160–172. https://doi.org/10.1227/NEU.0b013e3182556fde Parvizi, J. (2009). Corticocentric myopia: Old bias in new cognitive sciences. Trends in Cognitive Sciences,13(8), 354–359. https://doi.org/10.1016/j.tics.2009.04.008 Pelzer, E. A., Melzer, C., Timmermann, L., von Cramon, D. Y., & Tittgemeyer, M. (2017). Basal ganglia and cerebellar interconnectivity within the human thalamus. Brain Structure and Function,222(1), 381–392. https://doi.org/10.1007/s00429-016-1223-z Pergola, G., Danet, L., Pitel, A. L., Carlesimo, G. A., Segobin, S., Pariente, J., Suchan, B., Mitchell, A. S., & Barbeau, E. J. (2018). The Regulatory Role of the Human Mediodorsal Thalamus. Trends in Cognitive Sciences,22(11), 1011–1025. https://doi.org/10.1016/j.tics.2018.08.006 Pergola, G., Ranft, A., Mathias, K., & Suchan, B. (2013). The role of the thalamic nuclei in recognition memory accompanied by recall during encoding and retrieval: An fMRI study. NeuroImage,74, 195–208. https://doi.org/10.1016/j.neuroimage.2013.02.017 Petrovici, J.-N. (1980). Speech disturbances following stereotaxic surgery in ventrolateral thalamus. Neurosurgical Review,3(3), 189–195. Pierpaoli, C., Barnett, A., Pajevic, S., Chen, R., Penix, L., Virta, A., & Basser, P. (2001). Water Diffusion Changes in Wallerian Degeneration and Their Dependence on White Matter Architecture. NeuroImage,13(6), 1174–1185. https://doi.org/10.1006/nimg.2001.0765 Price, C. J. (2000). The anatomy of language: Contributions from functional neuroimaging. Journal of Anatomy,197(3), 335–359. https://doi.org/10.1017/S0021878299006901 Prochnow, D., Kossack, H., Brunheim, S., Müller, K., Wittsack, H.-J., Markowitsch, H.-J., & Seitz, R. J. (2013). Processing of subliminal facial expressions of emotion: A behavioral and fMRI study. Social Neuroscience,8(5), 448–461. 132 Profant, O., Škoch, A., Balogová, Z., Tintěra, J., Hlinka, J., & Syka, J. (2014). Diffusion tensor imaging and MR morphometry of the central auditory pathway and auditory cortex in aging. Neuroscience,260, 87–97. https://doi.org/10.1016/j.neuroscience.2013.12.010 Pugh, K. R., Mencl, W. E., Jenner, A. R., Katz, L., Frost, S. J., Lee, J. R., Shaywitz, S. E., & Shaywitz, B. A. (2000). Functional neuroimaging studies of reading and reading disability (developmental dyslexia). Mental Retardation and Developmental Disabilities Research Reviews,6(3), 207–213. https://doi.org/10.1002/1098-2779(2000)6:3<207::AID-MRDD8>3.0.CO;2-P Pugh, K. R., Mencl, W. E., Jenner, A. R., Katz, L., Frost, S. J., Lee, J. R., Shaywitz, S. E., & Shaywitz, B. A. (2001). Neurobiological studies of reading and reading disability. Journal of Communication Disorders,34(6), 479–492. https://doi.org/10.1016/S0021-9924(01)00060-0 Ramcharan, E. J., Gnadt, J. W., & Sherman, S. M. (2005). Higher-order thalamic relays burst more than first-order relays. Proceedings of the National Academy of Sciences of the United States of America,102(34), 12236–12241. https://doi.org/10.1073/pnas.0502843102 Ray, J. P., & Price, J. L. (1992). The organization of the thalamocortical connections of the mediodorsal thalamic nucleus in the rat, related to the ventral forebrain—Prefrontal cortex topography. Journal of Comparative Neurology,323(2), 167–197. Rissman, J., Gazzaley, A., & D’Esposito, M. (2004). Measuring functional connectivity during distinct stages of a cognitive task. NeuroImage,23(2), 752–763. https://doi.org/10.1016/j.neuroimage.2004.06.035 Rokem, A., Takemura, H., Bock, A. S., Scherf, K. S., Behrmann, M., Wandell, B. A., Fine, I., Bridge, H., & Pestilli, F. (2017). The visual white matter: The application of diffusion MRI and fiber tractography to vision science. Journal of Vision,17(2), 4. https://doi.org/10.1167/17.2.4 Saalmann, Y. B., & Kastner, S. (2011). Cognitive and Perceptual Functions of the Visual 133 Thalamus. Neuron,71(2), 209–223. https://doi.org/10.1016/j.neuron.2011.06.027 Schilling, K. G., Rheault, F., Petit, L., Hansen, C. B., Nath, V., Yeh, F. C., Girard, G., Barakovic, M., Rafael-Patino, J., Yu, T., Fischi-Gomez, E., Pizzolato, M., Ocampo-Pineda, M., Schiavi, S., Canales-Rodríguez, E. J., Daducci, A., Granziera, C., Innocenti, G., Thiran, J. P., … Descoteaux, M. (2021). Tractography dissection variability: What happens when 42 groups dissect 14 white matter bundles on the same dataset? NeuroImage,243(November 2020), 118502. https://doi.org/10.1016/j.neuroimage.2021.118502 Shah, A., Jhawar, S. S., & Goel, A. (2012). Analysis of the anatomy of the Papez circuit and adjoining limbic system by fiber dissection techniques. Journal of Clinical Neuroscience,19(2), 289–298. https://doi.org/10.1016/j.jocn.2011.04.039 Sherbondy, A. J., Dougherty, R. F., Napel, S., & Wandell, B. A. (2008). Identifying the human optic radiation using diffusion imaging and fiber tractography. Journal of Vision,8(10), 1–11. https://doi.org/10.1167/8.10.12 Sherman, S. M. (2005). Thalamic relays and cortical functioning. Progress in Brain Research,149, 107–126. https://doi.org/10.1016/S0079-6123(05)49009-3 Sherman, S. M. (2007). The thalamus is more than just a relay. Current Opinion in Neurobiology,17(4), 417–422. https://doi.org/10.1016/j.conb.2007.07.003 Sherman, S. M. (2012). Thalamocortical interactions. Current Opinion in Neurobiology, 22(4), 575–579. https://doi.org/10.1016/j.conb.2012.03.005 Sherman, S. M. (2016). Thalamus plays a central role in ongoing cortical functioning. Nature Neuroscience,19(4), 533–541. https://doi.org/10.1038/nn.4269 Sherman, S. M. (2017). Functioning of circuits connecting thalamus and cortex. Comprehensive Physiology,7(2), 713–739. https://doi.org/10.1002/cphy.c160032 Sherman, S. M., & Guillery, R. W. (2009). Exploring the Thalamus and Its Role in Cortical Function. In Tohoku Mathematical Journal (Vol. 35, Issue 1). The MIT Press. https://doi.org/10.7551/mitpress/2940.001.0001 Small, S. L., & Hickok, G. (2016). Chapter 1—The Neurobiology of Language. In G. Hickok & 134 S. L. Small (Eds.), Neurobiology of Language (pp. 3–9). Academic Press. https://doi.org/10.1016/B978-0-12-407794-2.00001-8 Smith, S. M., Jenkinson, M., Woolrich, M. W., Beckmann, C. F., Behrens, T. E. J., Johansen-Berg, H., Bannister, P. R., De Luca, M., Drobnjak, I., Flitney, D. E., Niazy, R. K., Saunders, J., Vickers, J., Zhang, Y., De Stefano, N., Brady, J. M., & Matthews, P. M. (2004). Advances in functional and structural MR image analysis and implementation as FSL. NeuroImage,23(SUPPL. 1), 208–219. https://doi.org/10.1016/j.neuroimage.2004.07.051 Steiner, L., Federspiel, A., Slavova, N., Wiest, R., Grunt, S., Steinlin, M., & Everts, R. (2020). Functional topography of the thalamo-cortical system during development and its relation to cognition. NeuroImage,223(September), 117361. https://doi.org/10.1016/j.neuroimage.2020.117361 Takagi, T., Nakamura, M., Yamada, M., Hikishima, K., Momoshima, S., Fujiyoshi, K., Shibata, S., Okano, H. J., Toyama, Y., & Okano, H. (2009). Visualization of peripheral nerve degeneration and regeneration: Monitoring with diffusion tensor tractography. NeuroImage,44(3), 884–892. https://doi.org/10.1016/j.neuroimage.2008.09.022 Theaud, G., Houde, J. C., Boré, A., Rheault, F., Morency, F., & Descoteaux, M. (2020). TractoFlow: A robust, efficient and reproducible diffusion MRI pipeline leveraging Nextflow & Singularity. NeuroImage,218(September 2019). https://doi.org/10.1016/j.neuroimage.2020.116889 Theisen, F., Leda, R., Pozorski, V., Oh, J. M., Adluru, N., Wong, R., Okonkwo, O., Dean, D. C., Bendlin, B. B., Johnson, S. C., Alexander, A. L., & Gallagher, C. L. (2017). Evaluation of striatonigral connectivity using probabilistic tractography in Parkinson’s disease. NeuroImage: Clinical,16(July), 557–563. https://doi.org/10.1016/j.nicl.2017.09.009 Tournier, J. D., Calamante, F., Connelly, A., & others. (2010). Improved probabilistic streamlines tractography by 2nd order integration over fibre orientation distributions. Proceedings of the International Society for Magnetic Resonance in Medicine,1670. 135 Tournier, J. D., Smith, R., Raffelt, D., Tabbara, R., Dhollander, T., Pietsch, M., Christiaens, D., Jeurissen, B., Yeh, C. H., & Connelly, A. (2019). MRtrix3: A fast, flexible and open software framework for medical image processing and visualisation. NeuroImage, 202(August), 116137. https://doi.org/10.1016/j.neuroimage.2019.116137 Tourville, J. A., & Guenther, F. H. (2011). The DIVA model: A neural theory of speech acquisition and production. Language and Cognitive Processes,26(7), 952–981. https://doi.org/10.1080/01690960903498424 Traynor, C., Heckemann, R. A., Hammers, A., O’Muircheartaigh, J., Crum, W. R., Barker, G. J., & Richardson, M. P. (2010). Reproducibility of thalamic segmentation based on probabilistic tractography. NeuroImage,52(1), 69–85. https://doi.org/10.1016/j.neuroimage.2010.04.024 Tremblay, P., & Dick, A. S. (2016). Broca and Wernicke are dead, or moving past the classic model of language neurobiology. Brain and Language,162, 60–71. https://doi.org/10.1016/j.bandl.2016.08.004 Tsai, S. Y. (2018). Reproducibility of structural brain connectivity and network metrics using probabilistic diffusion tractography. Scientific Reports,8(1), 1–12. https://doi.org/10.1038/s41598-018-29943-0 Tschentscher, N., Ruisinger, A., Blank, H., Díaz, B., & von Kriegstein, K. (2019). Reduced structural connectivity between left auditory thalamus and the motion-sensitive planum temporale in developmental dyslexia. Journal of Neuroscience,39(9), 1720–1732. https://doi.org/10.1523/JNEUROSCI.1435-18.2018 Vann, S. D., Saunders, R. C., & Aggleton, J. P. (2007). Distinct, parallel pathways link the medial mammillary bodies to the anterior thalamus in macaque monkeys. European Journal of Neuroscience,26(6), 1575–1586. Veraart, J., Novikov, D. S., Christiaens, D., Ades-aron, B., Sijbers, J., & Fieremans, E. (2016). Denoising of diffusion MRI using random matrix theory. NeuroImage,142, 394–406. https://doi.org/10.1016/j.neuroimage.2016.08.016 Villeneuve, M. Y., Kupers, R., Gjedde, A., Ptito, M., & Casanova, C. (2005). Pattern-motion 136