A taxonomy of the brain’s white matter: twenty-one major tracts for the 21st century

Abstract

The functional and computational properties of brain areas are determined, in large part, by their connectivity profiles. Advances in neuroimaging and network neuroscience allow us to characterize the human brain noninvasively, but a comprehensive understanding of the human brain demands an account of the anatomy of brain connections. Long-range anatomical connections are instantiated by white matter, which itself is organized into tracts. These tracts are often disrupted by central nervous system disorders, and they can be targeted by neuromodulatory interventions, such as deep brain stimulation. Here, we characterized the connections, morphology, traversal, and functions of the major white matter tracts in the brain. There are major discrepancies across different accounts of white matter tract anatomy, hindering our attempts to accurately map the connectivity of the human brain. However, we are often able to clarify the source(s) of these discrepancies through careful consideration of both histological tract-tracing and diffusion-weighted tractography studies. In combination, the advantages and disadvantages of each method permit novel insights into brain connectivity. Ultimately, our synthesis provides an essential reference for neuroscientists and clinicians interested in brain connectivity and anatomy, allowing for the study of the association of white matter’s properties with behavior, development, and disorders.

Introduction

The brain acts as an ensemble of distal computing centers located in both cortical and subcortical structures (Bullmore and Sporns 2009). The functions and computational profiles of these brain areas are determined not only by their local properties but, importantly, by their connectivity profiles. Most long-range connectivity is composed of ensembles of axonal projections wrapped in myelin, generally referred to as white matter bundles or tracts. We can now confidently predict that to fully understand the functions of the human brain, neuroscientists will have to develop an account of the connections and tissue properties of these active wires. Indeed, it has been proposed that some brain disorders are best considered as disruptions of connectivity (Geschwind 1965; Catani and Ffytche 2005; Catani and Mesulam 2008a).

The recent expansion of network neuroscience and neuroimaging has brought renewed interest to the study of long-range white-matter bundles and brain connectivity. Yet, despite the theoretical expansion of our understanding of the brain, there is a major roadblock ahead: We must first build an anatomically accurate connectivity map of the human brain. Unfortunately, a complete map of the brain connections is not yet available. Frustratingly, although deep knowledge of major white matter bundles inarguably exists, it is mostly dispersed across sub-fields of neuroscience spanning comparative histological tract-tracing, dissections, medicine, and neuroimaging.

Here, we are particularly interested in how tract-tracing, diffusion-weighted tractography, and dissection results (performed across species) bear on one another to yield novel insights. Ultimately, these insights will allow for the widespread adoption of investigations of white matter as a standard component of neurobiological accounts of behavior, brain disorders, and changes in cognition across the lifespan (Box 1).

Box 1

Implications for Human Health and Interventions: Neuromodulation of White Matter. Targeted changes to axonal bundles have been used to try treating brain disorders, including both neurological diseases and mental illness. Ablation of the white matter of the internal capsule, for example, is known as a capsulotomy and has been used to treat major depressive disorder and obsessive–compulsive disorder (Oliver et al. 2003; Christmas et al. 2011; Hurwitz et al. 2012; Pepper et al. 2019). Deep brain stimulation is a treatment used widely for movement disorders and sparingly for psychiatric disorders (Perlmutter and Mink 2006; Heilbronner et al. 2016). It requires placing an electrode semipermanently in the brain. Although initially conceived of as operating on nearby cell bodies, we now understand that the efficacy of DBS is mostly dependent on stimulating myelinated axons (Ashkan et al. 2017). Advances in electrode technology mean that neurosurgeons can not only target different physical locations but also steer current to selectively stimulate axons of different orientations (Chaturvedi et al. 2012).

Defining White Matter

Nearly 50% of total brain volume is composed of white matter tissue. Notably, the total brain volume occupied by gray and white matter scales universally across the mammalian species and is related to the complexity in

in cortical folding (Azevedo et al. 2009; Herculano-Houzel et al. 2010; Herculano-Houzel 2014).

What Is a White Matter Tract?

The brain’s gray matter can be divided into regions and subregions. These are not simply used as delineations of convenience but are instead taken to reflect meaningful biological units with distinct functional profiles. Can the same be said of white matter?

The taxonomic analogue of a gray matter brain region is a white matter tract or bundle: a collection of myelinated axons that share a meaningful combination of properties that may include connectivity, volume, gross morphology, trajectory, ontogeny, phylogeny, function, and susceptibility to disease or injury processes. The actual implementation of this notion appears to be bifurcated into two implicitly distinct approaches to delineating white matter tracts.

One of the principles that can define white matter tracts is their “volumetric” characteristics, such that a tract is defined as coursing through a specific white matter volume. For example, the internal capsule could be considered the white matter lateral to the caudate and thalamus but medial to the putamen and globus pallidus. Somewhat distinct from this is a second set of principles used to define tracts, namely, their connective characteristics. With this approach, the defining features are the regions that the tract connects to—“origins” or “terminations.” For example, the internal capsule could be considered the majority of the neuronal projections between the cortex and the thalamus and brainstem. Sometimes these two approaches conflict: There are striato-pallidal axons that traverse the volume occupied by the internal capsule, but are these part of the internal capsule itself?

Most definitions comprise some combination of both volumetric and connective features. To foreshadow several discussions that will follow in this review, many of the controversies surrounding the definition of a tract may be the result of unacknowledged tensions between these definitions. The volume-based definitions may stem from the common practice of using volumetric boundaries to delineate the boundaries of gray matter brain regions. Of course, this approach faces challenges in areas with crossing fibers. Connectivity-based approaches to the delineation of white matter structure, though also present in historical dissections and tract-tracer studies, are becoming more prominent with the rise of diffusion tractography.

Finally, there are other terms relevant to white matter organization, many of which can have multiple definitions within the field. For example, a bundle, connection, or tract may refer to a white matter structure or a particular subset of connectivity within a larger structure. For example, the cingulum bundle is a bundle (or tract) unto itself, but, at the same time, the axons connecting the anterior cingulate cortex with the posterior cingulate cortex (which constitutes just a subcomponent of the cingulum bundle) are also a bundle (or tract). Individual axons can also be referred to as fibers; in a tractography setting, connections are expressed as streamlines, but these are not equivalent to single axons.

Methods Used to Study White Matter Bundles

Understanding white matter necessitates bridging the gaps between nonhuman animal model species and humans. As a result, the process of mapping white matter inevitably involves different measurement modalities, and each modality comes with opportunities and challenges (Table 1). In the sections below, we provide an overview of four of the major categories of methods used to investigate white matter and its connectivity, in a rough chronological order.

Table 1.

An evaluation of common methods for assessing white matter anatomy and connectivity

Method	Advantages	Disadvantages
Gross dissection	Overt and compelling morphology, extended historical precedent, face validity	Ex vivo, low spatial resolution (difficulty resolving intermingled bundles, difficulty resolving smaller bundles), labor intensive, not replicable within subject
Tract-tracing	High spatial resolution, high accuracy	Limited by injection site, requires significant data aggregation, invasive, ex vivo, labor intensive, limited to nonhuman animals
Tractography	Whole-brain, in vivo or ex vivo, noninvasive, high-throughput, associability with other imaging data modalities	Low spatial resolution (difficulty resolving intermingled bundles, difficulty resolving smaller bundles), indirect/inferred, lack of methodological consensus
Label-free imaging	Applicable to humans and nonhuman animals, moderate spatial resolution, whole-brain	Ex vivo, unclear whether it can successfully replicate areal connectivity, limited use (expense and expertise)

Gross Dissection

Scientists have been dissecting brain tissue for centuries; however, the Klingler technique for white matter dissection was only developed in the 1930s (Silva and Andrade 2016). This ex vivo technique requires brain fixation, freezing, and thawing. The formation of ice loosens the axons, allowing them to be pulled apart while maintaining their structural integrity. This technique can be applied to both human and nonhuman animal brains. Dissection has been useful for determining the broad shapes of the largest and most prominent white matter bundles. That being said, dissection cannot distinguish directionality of fibers and presents difficulties in determining details of where the fibers originate and terminate (Beevor and Ferrier 1891). Furthermore, dissection is limited to the major structures. Finally, it is likely that the Klingler method is biased toward the dominant bundle within a volume of white matter. This problem is similar to that attributed to deterministic tractography based on the diffusion tensor model method (Pestilli et al. 2014; Takemura, Caiafa, et al. 2016; Caiafa and Pestilli 2017).

Histological Tract-Tracing

Neuroanatomists of the 20th century devoted considerable effort to developing methods for precisely determining region-to-region connectivity in the brain (Nauta and Gygax 1951, 1954; Fink and Heimer 1967; Heimer 1970; Wouterlood and Groenewegen 1991; Nauta 1993). Modern-day histological tract-tracing techniques rely on dyes (or, more recently, viruses (Cushnie et al. 2020; Lanciego and Wouterlood 2020)) that are taken up by cells and/or terminating axons and actively transported in a retrograde (from the terminal field to the cell body) or anterograde (from the cell body to terminal fields) manner. This reliance on active transport means that most tract-tracing involves a delay (days to weeks) during which the tracer remains in the live brain. Due to this combination of intracranial surgery and timed sacrifice, tract-tracing is impossible in the human brain.

Uptake of tract-tracers within the white matter is unreliable; therefore, it is generally not possible to inject any part of a bundle to understand what connections it carries. Our understanding of the fine details of white matter organization is therefore due in large part to the combination of hundreds of tract-tracing cases from injections into specific gray matter regions, across many laboratories. A downside to this approach is that it is decidedly not “whole-brain.” In addition, common laboratory animal models, such as of mice and rats, tell us very little about human white matter organization (Coizet et al. 2017). This leaves us with costly and limited nonhuman primate experiments (Heilbronner and Chafee 2019). Another major limitation of tract-tracing is that negative results (absent connections) are difficult to interpret without many supporting studies. That is, a tract-tracing study may fail to find a connection between two specific regions, but it may be that another part or layer of the region does connect. Only with overwhelming evidence from many cases across laboratories are we confident in a true lack of connection.

Diffusion-Weighted MRI

With tract-tracing limited to nonhuman animals and gross dissection lacking detail and resolution, there has long been a need for a noninvasive method of determining white matter organization in humans. Neuroimaging, and specifically diffusion-weighted magnetic resonance imaging (dMRI), has presented itself as a viable means of meeting this need. dMRI is sensitive to water molecule displacement (diffusion) within the fluids in the brain tissue. The displacement of water molecules is constrained by the microscopic tissue structure of the brain, and in particular the myelin-wrapped, long-range axonal projections (Basser and Pierpaoli 1996; Le Bihan and Iima 2015). Moreover, this imaging modality is sensitive to a combination of white matter tissue features, including axonal configuration (e.g., axonal direction and crossing) as well as the packing density of the said axons and the properties of the myelin tissue itself (Basser and Pierpaoli 1996; Assaf and Pasternak 2008; Tournier et al. 2008).

Multiple computational models have been developed to represent the process of water diffusion within brain tissue (Panagiotaki et al. 2012; Wandell 2016; Jelescu and Budde 2017; Rokem et al. 2017; Shi and Toga 2017). Among these is the diffusion-tensor imaging (DTI) model (Sakuma et al. 1991; Pierpaoli et al. 1996; Basser and Pierpaoli 1998), which represents water diffusion as a “single” Gaussian process in three-dimensional space. Although quite elegant, DTI is one of the simplest models, a property that limits the complexity of bundle configurations that can be modeled by this method. More recent and complex models (Tournier et al. 2007; Aganj et al. 2010) can capture nuanced aspects of diffusion processes, and thereby model more complex white matter configurations.

To reconstruct long-range axonal fibers using dMRI, the discrete, voxel-wise output features from a diffusion model must be interpolated to model the contiguous white matter as streamlines. The systematic production of such elements corresponds to the process of tractography generation which, when performed for the entire volume of the white matter, results in a whole brain tractogram. A variety of methods have been proposed to generate fiber tractography (Mori et al. 1999; Jiang et al. 2006; Descoteaux et al. 2009; Smith et al. 2012; Tournier et al. 2012; Takemura, Caiafa, et al. 2016; Sarwar et al. 2019). Among the many distinguishing characteristics of such models is whether they approach tractography generation in a probabilistic or deterministic fashion. Deterministic tractography follows the parameters from diffusion models fit to the signal of individual voxels in such a way that starting in the same voxel always proceeds along the same path. Probabilistic tractography does not trust the model fits in individual voxels; instead it chooses the path from a distribution of potential paths with each step and therefore can result in different outcomes from within the same voxel. In this way, it captures the inherent uncertainty regarding aggregate local fiber distributions and allows the fiber tracking process, through stochastic variation, to eventually model the range of possible configurations.

dMRI is typically collected at a resolution in the range of 1.5–2 mm³ (van Essen et al. 2012; Sotiropoulos et al. 2013). In contrast, the typical axon diameter is less than 1 μm. Due to this, and because of the number of presumptions made in models of diffusion, tractography is necessarily an indirect and inferential method. In turn, fiber pathway identification can suffer from several issues that undermine the biological plausibility of the aggregate tractogram unless informed by relevant anatomical priors. Different types of priors can potentially improve the accuracy of tractography at different levels. The anatomical information can be very generic, such as tissue types and boundaries (gray/white matter, cerebral spinal fluid, etc.), and serve as a global constraint for tracking algorithms (Smith et al. 2012; Girard et al. 2014). More specific information of fiber morphology, such as sparsity and smoothness (Wassermann et al. 2016; Bullock et al. 2019), can help refine the bundles to be more biologically plausible. These can serve as adjustable parameters in automated modeling and tracking algorithms. Such generic priors will reduce the biases in position, shape, size, and length of the streamline distribution and improve the bundle segmentation, though they cannot fix region-specific problems (such as those caused by crossing fibers). An even more specific and stronger type of prior is the ground-truth fiber trajectory information provided by post-mortem dissection and animal tract-tracing (Silva and Andrade 2016; Heilbronner and Chafee 2019; Schilling et al. 2020). This is the gold standard to eliminate technical mistakes in tractography, although such information is not available in all brain regions and is labor intensive to obtain. Nonetheless, without anatomical priors, tractography may suggest pathways that are nonexistent (“false positives”) or miss pathways that do, in fact, exist (“false negatives”) (Fig. 1). Of particular concern is the appropriate reproduction of unknown connections (Maier-Hein et al. 2017) and the biased density of the coverage and penetration of the cortical mantle by tractography (Thomas et al. 2014; Jbabdi et al. 2015; Reveley et al. 2015; Schilling et al. 2018; Grier et al. 2020). Nevertheless, dMRI-based techniques and tractography are the only methods available to track major fiber bundles in living human brains. Unfortunately, where there is no rich account of a white matter structure’s morphology and anatomy available, the accuracy of the output of tractography is uncertain. It is for this reason that better anatomical models are paramount to understanding brain connectivity.

Figure 1.

Using tractography to establish brain connectivity. (A–C) depictions of (A) dMRI data, modeling of the diffusion processes giving rise to these data (B), and the generation of streamline tractography using these models (C). The generation of streamline tractography models of white matter is an inherently iterative and inferential process. This process, framed as a signal detection problem, can result in the accurate generation of streamlines (green streamline, panel C) as well as the accurate failure to generate streamlines (blue outline, panel C), but also the generation of inaccurate streamlines (yellow streamline, panel C) as well as inaccurate failures to generate streamlines (red outline, panel C). (D) Distinct tractography generation approaches (e.g., models, parameters, thresholds, priors, etc.) result in different levels of specificity and sensitivity for detecting white matter anatomy connectivity. The resultant tractograms (represented by the three bar columns) can be composed of streamlines and tracts with varying levels of validity, and thus reflects a signal detection problem. Assessing these qualities remains a challenge, and there is currently no known optimal tractography method.

One of the primary uses of dMRI in the literature is to understand the relationship between white matter properties and human behavior, development, aging, and disease (Sullivan and Pfefferbaum 2003; Clark et al. 2011; Thomason and Thompson 2011). For example, fractional anisotropy refers to the degree of anisotropic diffusion and is often roughly interpreted as white matter integrity. It is a widely used measurement to assess white matter abnormalities in brain disorders. Here, we are concerned primarily with using dMRI to determine anatomical connectivity and not to assess how these values are different across populations. However, the two goals can be mutually beneficial such that more accurate connectivity maps make specific white matter abnormalities more meaningful and reproducible.

Label-Free Imaging Technologies

Polarization-sensitive optical coherence tomography (PS-OCT) and polarized light imaging (PLI) both take advantage of birefringence (an optical property of materials) in brain tissue to differentiate between axons that are parallel versus perpendicular to the plane of light (Larsen et al. 2007; Axer et al. 2016; Wang, Akkin, et al. 2016; Jones et al. 2020). PS-OCT is undistorted (it is performed on block-face images before sectioning) whereas PLI is performed on distorted histological slices after sectioning. Both are ex vivo techniques that can be performed on unlabeled human tissue. Crucially, this means they do not require the injection of a tract-tracer. Spatial resolution tends to be between tract-tracing and dMRI, in the tens to hundreds of microns range (Axer et al. 2011; Schmitz et al. 2018). Therefore, although too big to visualize single axons, PLI and PS-OCT provide a valuable intermediate step between the spatial resolutions of tract-tracing and dMRI/dissection. They may or may not identify some important details (such as particular termination points or crossing fibers) present in tract-tracing but missed by dMRI.

Interpreting Convergent and Discordant Evidence

Mapping anatomical connectivity of the human brain remains an urgent challenge, and no single method can achieve this goal. Integrating (nontractographic) anatomical knowledge from human and monkey studies into tractography offers the best path forward to accurately map human brain connectivity.

Unsurprisingly, there are major challenges with identifying the potential source of discrepancies across accounts of tracts as attributable to species versus methodological differences. This is largely because the most common method used in nonhuman primates is tract-tracing whereas noninvasive tractography is commonly used in humans. When tract-tracing in monkeys versus tractography in humans identify apparent discrepancies, should this be assumed to reflect a species difference or a methodological error? One crucial step is to examine tractography results in monkeys (Table 2). In some cases, a method-based, within-species discrepancy may arise. For example, the tractography in monkeys is found to be consistent with the results seen in humans for a given tract, but nonetheless those same results are found to be inconsistent with tract-tracing findings. In such a case, under the assumption of homology between species, we can infer that the mismatch between human tractography and monkey tract-tracing is due to a methodological error in tractography and not a species difference. In contrast, if tractography in monkeys matches tract-tracing in monkeys (but not tractography in humans), there may be a true species difference at work leading to the novel human tractography result. A similar logic holds when comparing tract-tracing and dissection.

Table 2.

Interpreting multimodal, multispecies, discordant versus concordant evidence

Monkey tract-tracing	Monkey diffusion tractography	Human diffusion tractography	Appropriate inference
√	√	√	True connection in monkeys and humans
		X	Potential species difference
	X	√	Unclear (not observed)
		X	False negative in diffusion tractography
X	√	√	False positive in diffusion tractography
		X	Unclear (not observed)
	X	√	Potential species difference
		X	True lack of connection in monkeys and humans

√ = tract found; X = tract not found

White Matter Taxonomy of the Brain

For each white matter structure in the brain, we will review what is known about its location, fiber orientation(s), specific connections, reconstruction using dMRI, and functional associations. We will also touch upon abnormalities in brain disorders and any neuromodulatory treatments targeted at the bundle. An overview can be found in Supplementary Table 1.

Here, rather than provide a systematic review or formal meta-analysis, we have focused on assessing white matter structures and papers for which cross-modal and/or cross-species (human—nonhuman primate) evidence exists and is enriching to the discussion at hand. In doing so, we have focused particularly on areas in which there are problems to be solved. There are also additional, important white matter tracts not covered here. For those, we refer the reader to outside work on a handful of additional tracts: acoustic radiations (Maffei et al. 2018); cerebellar peduncles (van Baarsen et al. 2016); stria terminalis (Kwon et al. 2011; Mori et al. 2017); frontal aslant tract (Pascual-Diaz et al. 2020; La Corte et al. 2021); and basal ganglia pathways (Johnson et al. 2021).

Figure 2.

The corpus callosum, anterior commissure, and internal capsule. (A) Sagittal view of a tractography model of the corpus callosum (orange), along with a (B) corresponding midline voxel mask indicating a volumetric criterion for it. Even for the well-understood corpus callosum, the terminology may be used differently across investigations. “Corpus callosum” may refer to only the volumes at the midline of the brain (B) or might extend to include the entirety of the axons traversing that region (A). (C) A sagittal view of a tractography model of the anterior commissure. (D) A sagittal view of the internal capsule. This depiction excludes some known fibers from the ventral prefrontal cortex. The inset shows segmentation of the anterior limb of the internal capsule on the basis of constituent prefrontal cortical fibers, from Safadi et al. (2018). Tractography models derived from TractSeg (Wasserthal et al. 2018) for this and all figures unless otherwise indicated.

Corpus Callosum

Perhaps the best understood white matter structure, the corpus callosum (Fig. 2) connects the left and right cerebral cortices (although not all the cerebral cortex: see the description under Anterior Commissure). Its anterior end is the genu, the posterior end is the splenium, and its central portion is the body.

Figure 3.

A tractography model of the superior longitudinal fasciculus (SLF). (A) Depicts a sagittal view of the three subcomponents of the SLF (I, pink; II, green; III, blue). (B) Depicts anteroposterior positions from which slices in (C) are drawn. (C) Depicts coronal views of the cores of the subcomponents, using the same color convention.

The corpus callosum has a topographic organization (Sunderland 1940; Pandya et al. 1971; de Lacoste et al. 1985). A dMRI study in which analyses were matched for rhesus macaques and humans found strikingly similar organization across the two species (Hofer et al. 2008). Crossing fibers utilizing the corpus callosum have several targets. First, they target homotopical cortex (the mirror image location) very strongly. Second, they target heterotopical cortex (other cortical locations on the contralateral side), generally with the same pattern (region and strength) as seen in the ipsilateral cortical connections. Finally, some fibers do travel through the corpus callosum to reach contralateral subcortical targets, including the striatum and claustrum, though likely not the thalamus (Locke et al. 1964; Jones and Powell 1969; Cavada and Goldman-Rakic 1991).

The broad pattern of anterior–posterior topography through the corpus callosum can be straightforwardly reconstructed in human dMRI. However, early, tractography-based examples of the corpus callosum (Catani et al. 2002; Abe et al. 2004; Huang et al. 2005; Hofer and Frahm 2006; Catani and de Schotten 2008) were often not whole brain, were generated using deterministic diffusion-weighted tractography, and/or were focused on an extended U-shaped morphology. Therefore, resultant tractography models of the corpus callosum were biased toward homotopic connections. A more recent and comprehensive dMRI investigation of connection motifs of the corpus callosum explored more complex pathways of the corpus callosum (De Benedictis et al. 2016).

Agenesis of the corpus callosum, a condition in which the corpus callosum never develops, and split-brain patients, in whom the corpus callosum has been severed, provide opportunities to study the functions and plasticity of this bundle. Both sets of subjects demonstrate “disconnection syndrome,” which involves a lack of interhemispheric transfer of sensory information and difficulties with the performance of bilaterally coordinated motor tasks (Zaidel and Sperry 1977). Extensive work performed by Sperry and Gazzaniga demonstrated the cognitive consequences of callosotomy (Baynes et al. 1998; Gazzaniga 2005).

Finally, due to the strong topography present in the corpus callosum, the location of a callosal abnormality can and has been used to infer which parts, broadly speaking, of the cortex are involved in a particular brain condition. For example, obsessive–compulsive disorder is associated with reduced fractional anisotropy in a restricted portion of the anterior body of the corpus callosum (Nakamae et al. 2011), an area that carries crossing fibers from the dorsolateral prefrontal and mid-cingulate cortices. In contrast, schizophrenia is associated with reduced fractional anisotropy in the genu of the corpus callosum (Foong et al. 2000). These results are suggestive of distinct circuits in these two pathologies (although it is not clear how central the corpus callosum itself is to their respective etiologies).

Anterior Commissure

The anterior commissure (Fig. 2) is the second largest cerebral commissure. Its decussation is a prominent landmark in the brain and crosses through the striatum and internal capsule. Moving posteriorly, it is situated lateral and ventral to the striatum and medial to the claustrum.

The anterior commissure is responsible for carrying crossing fibers from the basal surface of the cerebral cortex, from the zone extending from the temporal pole and posterior orbitofrontal cortex to the occipitotemporal boundary (Jouandet and Gazzaniga 1979; Barbas and Pandya 1984). Its dorsal boundary for fibers is the inferior insula. Although most areas that send crossing fibers through the anterior commissure do so exclusively, the posterior basal temporal lobe does appear to send crossing fibers through both the splenium of the corpus callosum and through the anterior commissure (Zeki 1973). Unlike the corpus callosum, the anterior commissure appears to have only a weak topography (Schmahmann and Pandya 2006).

Finally, the anterior commissure can be reconstructed in human dMRI using regions of interest (ROIs) around its lateral branches, excluding the most lateral areas to avoid contamination with the uncinate fasciculus, inferior frontooccipital fasciculus, and external capsule (Catani and de Schotten, 2008). Alternative reconstruction methods using the third ventricle as a landmark have also been presented (Wang et al. 2008). A common feature of these approaches is that the reconstruction methods used targeted seeding-based approaches as opposed to whole brain tractography. This trend may be explained by the likely difficulty of reconstructing a very small tract (the anterior commissure is only ~ 1% the size of the corpus callosum). As such, whole brain tractography approaches to segmentation do not typically include this tract.

Internal Capsule

The internal capsule (Fig. 2) is positioned between the caudate and thalamus medially and the putamen and globus pallidus laterally. It can be divided into the anterior and posterior limbs. The dividing line is the genu, which, in a transverse brain slice, is shaped like a “V.”

The internal capsule is a major ascending and descending fiber structure, carrying fibers between the cortex and the thalamus, brainstem, spinal cord, and subthalamic nucleus. This means that the internal capsule includes subcomponents or the entirety of many bundles and connections with their own terminology: thalamic radiations, pyramidal tracts, and the corticospinal and corticobulbar tracts (the pyramidal tracts). Finally, although many brainstem–cortical connections traverse the internal capsule, others do not, such as those that use the medial forebrain bundle.

Like the corpus callosum, the internal capsule is topographic. For the anterior limb of the internal capsule, the cortical fibers originate/terminate in the prefrontal cortex; for the posterior limb, the cortical fibers originate/terminate outside the prefrontal cortex. The posterior limb contains the majority of the pyramidal tracts—the axons connecting cortical motor neurons with the brainstem and spinal cord (Beevor and Horsley 1890; Ross 1980). Importantly, according to our knowledge, fibers in the internal capsule are associated with the ipsilateral cortex only (see Aggleton et al. 1986; Stanton et al. 1988; Smith et al. 1990; Morecraft et al. 2002). Although fibers do project between the cortex and the contralateral thalamus and brainstem, they do so inside white matter bundles contained within these structures rather than through the contralateral internal capsule. Capsule fibers also travel outside of the main outline of the bundle, in small fascicles embedded within the central striatum, globus pallidus, anterior commissure, and basal forebrain (Lehman et al. 2011). Whether these axons should rightfully be considered part of the internal capsule per se harkens back to volumetric versus connectionist approaches to tract delineation as discussed earlier.

A series of cross-species histological tract-tracing and dMRI studies have demonstrated the relatively strict topography of the anterior limb of the internal capsule (Lehman et al. 2011; Jbabdi et al. 2013; Safadi et al. 2018). For the most part, dMRI in monkeys and humans is able to replicate this topography. One manner in which tractography struggles is in capturing ventromedial prefrontal cortical fibers that use small fascicles embedded in larger structures mentioned above (Lehman et al. 2011). In addition to the fine topography of the anterior limb of the internal capsule, tractography can accurately map the topography and connections of the posterior limb. Its ability to do so has been greatly improved over time due to the advent of probabilistic tractography methods (Parker et al. 2002; Behrens et al. 2007; Sherbondy et al. 2008; Descoteaux  et al. 2009; Tournier et al. 2012) and high-angular resolution data (Frank 2002). Whereas early mapping of the corticospinal tract using deterministic tracing based on tensor models failed to represent the full extent of the tract (reconstructing only the vertical fibers: Basser et al. 2000; Lazar et al. 2003; Yeatman et al. 2012), more recent data and tractography methods allow a more complete mapping of the corticospinal tract that represent fibers spanning the full extent of the motor cortex (Behrens et al. 2007; Smith et al. 2015; Takemura, Caiafa, et al. 2016; Aydogan and Shi 2021).

The anterior limb of the internal capsule has been used as a target of neuromodulatory therapies (capsulotomy and deep brain stimulation) for obsessive–compulsive disorder, major depressive disorder, and Tourette syndrome (Flaherty et al. 2005; Greenberg et al. 2010; Hurwitz et al. 2012). Due to the highly topographic nature of the internal capsule, stimulation or lesion therapy will affect a very different set of fibers depending on where treatment is applied. Ventral anterior limb stimulation will affect the ascending and descending fibers of the orbitofrontal cortex and ventromedial prefrontal cortex; central anterior limb stimulation will affect fibers of the orbitofrontal and anterior cingulate cortex; and dorsal anterior limb stimulation will affect fibers of the dorsal prefrontal cortex (Widge et al. 2016; Safadi et al. 2018; Baldermann et al. 2019, 2021; Haber et al. 2020).

In addition, like the corpus callosum, the topography of the internal capsule may help us to deduce which specific sets of fibers are abnormal in disorders. For example, the aforementioned cross species tract-tracing and tractography studies were used to pinpoint that it is the ventrolateral prefrontal cortex projections to and from the thalamus and brainstem that are abnormal in patients with bipolar disorder based on a very regionally specific fractional anisotropy reduction (Safadi et al. 2018).

Superior Longitudinal Fasciculus

The superior longitudinal fasciculus (SLF, Fig. 3) runs anterio-posteriorly through the white matter of the dorsal parietal and frontal lobes. It has multiple separable components, although how many is a matter of dispute. The terms SLF IV and V are sometimes used to refer to what we (and others) call the arcuate fasciculus and posterior arcuate fasciculus, respectively (Wu, Sun, Wang, Wang, Wang 2016; Mandonnet et al. 2018; Panesar and Fernandez-Miranda 2019). Easily identified across species and techniques are SLF II and III. Tract-tracing in monkeys shows that SLF II contains fibers from the posterior inferior parietal lobule (in humans, the angular gyrus), intraparietal sulcus, and middle portion of the lateral prefrontal cortex, whereas SLF III (lateral and ventral to SLF II) contains fibers from the anterior inferior parietal lobule (in humans, the supramarginal gyrus), parietal operculum, and ventrolateral prefrontal cortex (Schmahmann and Pandya 2006).

Figure 4.

A tractography model of the middle longitudinal fasciculus (MdLF) and arcuate fasciculus (arc). (A) Depicts a sagittal view of the MdLF (pink) and arc (gold). (B) Depicts anteroposterior positions from which slices in (C) are drawn. (C) Depicts coronal views of the cores of the subcomponents, using the same color convention.

More controversial of these is SLF I. Schmahmann and Pandya (2006) claim that in monkeys it is the longest, most dorsal, and most medial component of the SLF, sitting above the cingulum (also see Petrides and Pandya 1984). According to this account, SLF I contains fibers from the precuneus, posterior superior parietal lobule, supplementary motor area, and dorsomedial frontal cortex. However, recent human tractography and dissection studies have failed to find a long anterio-posterior bundle in this dorsomedial cortex (Wang, Pathak, et al. 2016; Mandonnet et al. 2019). Others show that the so-called “SLF I” is continuous with and identical to (connectivity wise) the cingulum bundle, and thus should not be considered its own tract. This seems to broadly agree with the monkey tract-tracing data, which show continuity of connectivity between the cingulum bundle and “SLF I” (see commentary in Panesar and Fernandez-Miranda 2019). Nevertheless, the SLF I could be considered separable from the cingulum bundle volumetrically, and the debate is not fully resolved.

In humans, the SLF complex is associated primarily with language but also with musical ability (Oechslin et al. 2009), tool use (Hecht et al. 2015), and working memory (Karlsgodt et al. 2008; Vestergaard et al. 2011; Rizio and Diaz 2016). The SLF may act, via the supramarginal gyrus, as a relay between the frontal and temporal language regions (Catani et al. 2005; Catani and Mesulam 2008b). Finally, patients with schizophrenia have abnormalities in the SLF (Karlsgodt et al. 2008; Davenport et al. 2010), which is consistent with the broad fronto-parietal abnormalities observed in this disorder (Chang et al. 2014; Sheffield et al. 2016).

Middle Longitudinal Fasciculus

The middle longitudinal fasciculus (MdLF, Fig. 4) connects the parietal and temporal lobes traversing the white matter medial to the lateral fissure. The details of its connections, however, have historically been the subject of debate. Early tractography reports of the MdLF alternatively depicted it as primarily extending to either the superior parietal lobule (Oishi et al. 2008; Maldonado et al. 2013; Wang et al. 2013; Jang et al. 2015; Wu, Sun, Wang, Wang, Wang 2016) or inferior parietal lobule (Makris et al. 2009; de Champfleur et al. 2013; Martino, da Silva-Freitas, et al. 2013) (but not both). Subsequent studies synthesized these reports and presented findings indicating the presence of terminations in both parietal lobules (Makris et al. 2013; Kamali, Flanders, et al. 2014; Kamali, Sair, et al. 2014; Bajada et al. 2015, 2017; Bullock et al. 2019).

Figure 5.

A tractography model of the posterior arcuate fasciculus (pArc), vertical occipital fasciculus (VOF), and temporo-parietal connections to the superior parietal lobule (TP-SPL). (A) Depicts a sagittal view of the pARC (red), VOF (light pink) and TP-SPL (dark pink). (B) Depicts dorsoventral positions from which slices in (C) are drawn. (C) Depicts horizontal views of the cores of the subcomponents, using the same color convention. Images derived from WMA_segmentation available on brainlife.io.

The superior versus inferior parietal lobule debate bears on questions about direct anatomical connectivity and species differences. Tract-tracing studies do not show strong anatomical projections between the superior temporal cortex and the superior parietal lobule in monkeys; in contrast, the connectivity between the superior temporal cortex and the inferior parietal lobule is strong (Seltzer and Pandya 1978, 1984, 1991, 1994; Petrides and Pandya 1984; Padberg et al. 2019). This raises the possibility that the often-observed connectivity between the superior temporal cortex and the superior parietal lobule in human tractography and dissection may be a false positive or a species difference. However, matched cross-species tractography and dissection studies would be needed to confirm or deny such an assertion. There has been one such study of the MdLF (Roumazeilles et al. 2020). However, these authors do not note a superior parietal lobule connection in humans, limiting how much we can say about potential species differences. To our knowledge, there has not been a Klingler dissection study in monkeys equivalent to Kalyvas et al. (2020), which found evidence of a superior parietal lobule component, that could be compared with the monkey tract-tracing data. Due to the reported issues with dissections, given that tract-tracing studies have failed to find a pathway between the superior parietal lobule and the superior temporal gyrus, if a novel monkey MdLF dissection study were to reveal one, it would be taken as a false positive. However, if a monkey MdLF dissection study showed a tract stopping at the inferior parietal lobule, there may be a uniquely human component to the MdLF.

The middle longitudinal fasciculus may, along with the arcuate fasciculus, connect the major language areas of Wernicke’s area and the angular gyrus (Makris et al. 2009). Indeed, this bundle does seem to show a left lateralization (de Champfleur et al. 2013) as would be expected for a language tract. However, electrical stimulation of the MdLF did not appear to interfere with language (Hamer et al. 2011). Other possible functions include attentional processing (Makris et al. 2009), auditory perception, and/or auditory–visual integration (Kalyvas et al. 2020).

Arcuate Fasciculus

The arcuate (“arched”) fasciculus (Fig. 4), alternatively referred to as SLF IV and direct SLF (Catani et al. 2005; Bernal and Altman 2010; De Benedictis and Duffau 2011; Mandonnet et al. 2018; Panesar et al. 2019), is a long, anterio-posteriorly directed bundle. It runs laterally in the temporal, parietal, and frontal lobes. It carries fibers between the temporal cortex (as opposed to the parietal cortex, like the superior longitudinal fasciculus) and some parts of the frontal lobe.

Matched tractography in macaques and humans shows a species difference in the specific connections constituting the arcuate fasciculus, with more extensive temporal (middle and inferior temporal gyri) and frontal (ventrolateral prefrontal cortex) connectivity in the human than the macaque (Rilling et al. 2008). These extensive human tractography connections have been replicated multiple times (Babo-Rebelo et al. 2021; Schilling et al. 2021). Furthermore, the monkey tractography seems to closely match the tract-tracing results. Thus, it seems quite likely that the human arcuate fasciculus does indeed carry connections that do not exist in the macaque.

The more extensive connectivity in the human (vs macaque) arcuate fasciculus is also more pronounced in the left hemisphere (Rilling et al. 2008). Due to the combination of species and laterality differences and the proximity of the bundle to traditional language areas in humans, the arcuate fasciculus is thought to be essential for language function. Indeed, the arcuate fasciculus has been associated with conduction aphasias (Geschwind 1965; Anderson et al. 1999; Catani et al. 2005; Catani and Mesulam 2008b; Bernal and Ardila 2009; Dick and Tremblay 2012), a stereotyped form of aphasia with etiology associated with tissue damage in the perisylvian area of the brain. Lesions of the arcuate fasciculus following stroke result in language deficits (Marchina et al. 2011). Fractional anisotropy values and number of streamlines in the arcuate fasciculus predict the degree of aphasia in stroke patients (Hosomi et al. 2009).

Posterior Arcuate Fasciculus

Although its connectivity is often characterized as being part of structures like the middle longitudinal fasciculus (Frey et al. 2008), arcuate fasciculus, and inferior longitudinal fasciculus (Schmahmann and Pandya 2006; Frey et al. 2008), there is evidence of a separate, vertically oriented bundle, termed the posterior arcuate fasciculus (pArc, Fig. 5).This bundle connects the inferior parietal lobule with the middle and inferior temporal lobes (Catani et al. 2005; Catani and Mesulam 2008b; Kamali, Flanders, et al. 2014; Kamali, Sair, et al.  2014; Weiner et al. 2017). It has also been called SLF-V (Koutsarnakis et al. 2015; Wu, Sun, Wang, Wang, Wang 2016), temporoparietal connection with the inferior parietal lobule (Makris et al. 2005; Kamali, Flanders, et al. 2014; Kamali, Sair, et al. 2014), vertical arcuate fasciculus (Makris et al. 2005; Panesar and Fernandez-Miranda 2019), vertical SLF (Martino, da Silva-Freitas, et al. 2013; Martino and De Lucas 2014), perisylvian SLF (Catani et al. 2005; Martino, Hamer, et al. 2013), posterior SLF (Martino, Hamer, et al. 2013), indirect arcuate fasciculus (Turken and Dronkers 2011), temporo-parietal aslant tract (Panesar et al. 2019), parietotemporal long association fibers (Oishi et al. 2008), and anterior vertical occipital fasciculus (Choi et al. 2020). It has received reduced attention in nonhuman animal work. The pArc’s dorsal terminations are in the inferior parietal lobule and superior to the posterior portion of the lateral fissure. The pArc runs roughly parallel, although anterior, to the vertical occipital fasciculus (Weiner et al. 2017).

Figure 6.

A tractography model of the inferior longitudinal fasciculus (ILF) and inferior fronto-occipital fasciculus (IFOF). (A) Depicts a sagittal view of the ILF (lime green) and putative IFOF (light purple). (B) Depicts anteroposterior coordinates from which slices in (C) are drawn. (C) Depicts coronal views of the cores of the subcomponents, using the same color convention.

Vertical Occipital Fasciculus

The vertical occipital fasciculus (VOF, Fig. 5) (Yeatman et al. 2014; Takemura, Rokem, et al. 2016; Takemura et al. 2017), identifiable in both human and macaque dMRI, is a dorsal-ventrally running tract lateral to the calcarine sulcus and the optic radiation. It connects dorsal and ventral regions of the occipital lobe (Takemura, Rokem, et al. 2016; Takemura et al. 2017). The fact that the VOF connects primarily cortical structures within the occipital lobe distinguishes it from several other bundles discussed here, which typically connect cortical structures spanning relatively distal portions of the brain. The VOF has had a circuitous history and has been associated with several other terms including fasciculus occipitalis verticalis, stratum profundum convexitatis (Yeatman et al. 2014). It was first characterized around the turn of the 20th century, fell out of discussion, and was recently recharacterized (Yeatman et al. 2014; Takemura, Rokem, et al. 2016). Efforts have also been made to differentiate the VOF from the pArc (Weiner et al. 2017). Interestingly, this issue is predicated upon the assumption that the VOF extends outside the occipital lobe, potentially contrary to its nomenclature. A series of other white matter bundles connecting the parietal cortex and ventral cortex have also been clarified (Kamali, Flanders, et al. 2014; Kamali, Sair, et al. 2014; Bullock et al. 2019; Panesar et al. 2019), for example, posterior arcuate and temporo-parietal connection to the superior lobule. This clarification delineates these structures from the VOF suggesting that the VOF is indeed constrained to within the occipital cortex. More broadly, although connections between the dorsal and ventral early visual cortex have been reported in major tract-tracing work (Felleman and van Essen 1991; Ungerleider et al. 2008), there has been a relative dearth of investigation of this bundle as its own entity.

The VOF may help to connect the early dorsal visual stream, which is associated with the representation of space, and the ventral visual stream, which is responsible for object identification (Mishkin and Ungerleider 1982; Goodale and Milner 1992).

Temporo-Parietal Connections to the Superior Parietal Lobule

The temporo-parietal connection to the superior parietal lobule (TP-SPL), also referred to as the temporo-parietal connection (Wu, Sun, Wang, Wang, Wang 2016), SLF TP (SPL) (Kamali, Flanders, et al. 2014; Kamali, Sair, et al. 2014) or inferior parietal sulcus–frontal–fusiform gyrus connector (IPS-FG) (Jitsuishi and Yamaguchi 2020), is a vertically oriented tract connecting the superior parietal lobule to the fusiform gyrus and posterior inferior temporal gyrus. The TP-SPL’s superior and inferior terminations are generally located medially to the pArc, whereas its trunk exhibits extensive volumetric overlap with that tract. Its relative positioning with the pArc mirrors (at least in the dorsal white matter) the relationship between the angular gyrus and putative superior parietal lobule components of the MdLF. In the ventral white matter, however, the pArc and TP-SPL are both found to have terminations posterior to the MdLF subcomponents, thus distinguishing the two pairings. In this way, the structure of the pArc, TP-SPL and MdLF constitutes an interesting motif for brain connectivity between the posterior dorsal and ventral cortices that has been postulated to play a role in the development of the dorsal and ventral pathways of visual information processing (Choi et al. 2020; Vinci-Booher et al. 2021).

Inferior Longitudinal Fasciculus

The definition of the inferior longitudinal fasciculus (ILF, Fig. 6) has been the subject of contradiction. Historically one view has been that the ILF is not a long association bundle but instead consists of a series of short U-fibers that interconnect nearby regions (Tusa and Ungerleider 1985; Bajada et al. 2015) and thus cannot be properly considered a bundle at all. An alternative view was that the bundle consists of fibers descending to subcortical regions (Redlich 1905). We are now certain that the ILF is a long anterior–posterior association (cortical–cortical) bundle with embedded short fibers (Dejerine and Dejerine-Klumpke 1895). However, which particular cortical connections define the ILF, as well as its precise location, remain in dispute.

Figure 7.

Cingulum bundle, uncinate fasciculus, external capsule, Muratoff’s bundle, and extreme capsule. (A) Depicts a sagittal view of the cingulum bundle (orange). (B) a segmentation of the cingulum bundle in the macaque according to constituent fibers. Many fibers use all four subcomponents of the cingulum bundle; others (in boxes) are only present in a subset. Reproduced from (Heilbronner and Haber 2014) (C) depicts a sagittal view of the uncinate fasciculus (cyan). (D) External capsule, Muratoff’s bundle, and extreme capsule. Volumetric locations shown on high-resolution brain from (Edlow et al. 2019).

In their comprehensive tract-tracing study in monkeys, (Schmahmann and Pandya (2006) did see many examples of long corticocortical axons in the ILF, in addition to U-fibers, but no subcortically projecting fibers. However, their definition of both the placement (volume) of the ILF and its connectivity are somewhat unique. All authors agree that the ILF carries temporo-occipital fibers. More controversial (Mandonnet et al. 2018) is whether parietal fibers belong to this bundle. Tract-tracing makes clear that there are connections between the parietal cortex and inferior temporal cortex, and that some of those axons occupy the same volume as canonical ILF temporo-occipital fibers. Schmahmann and Pandya (2006) include these fibers, as well as a vertical branch that impinges upon the white matter of the parietal lobe, in their definition of the ILF (see also Seltzer and Pandya 1984). Some others also do show more dorsal cortical connections at the posterior end, including to the angular gyrus (Rushworth et al. 2006; Frey et al. 2008; Seghier 2013). In contrast, others argue that these more vertically oriented tracts belong to the MdLF and pArc (Davis 1921; Catani et al. 2002, 2003; Catani and de Schotten 2008; Turken and Dronkers 2011). On such an account, the ILF would be in the ventral temporal lobe and would exclusively contain anterio-posteriorly oriented temporo-occipital fibers. This is a clear case of tension between a volume versus connection approach to tract definition. Finally, the ILF’s temporo-occipital fibers likely connect the entire occipital lobe (Martino, da Silva-Freitas, et al. 2013; Wang et al. 2013; De Benedictis et al. 2014; Kamali, Flanders, et al. 2014) rather than just the ventral (Catani et al. 2002, 2003; Catani and de Schotten 2008; Turken and Dronkers 2011; de Champfleur et al. 2013).

Functions of the ILF map to known functions of the ventral visual stream (Herbet et al. 2018). For example, lesions or degeneration of the ILF lead to deficits in object recognition (Benson et al. 1974; Meichtry et al. 2018), alexia (Gaillard et al. 2006; Epelbaum et al. 2008), and visual memory (Shinoura et al. 2010). The ILF is likely also involved in face recognition, which depends upon intact connections between the occipital face area and the fusiform face area (Herbet et al. 2018).

Inferior Fronto-Occipital Fasciculus

The inferior fronto-occipital fasciculus (IFOF, Fig. 6) has had a contentious history in the literature, particularly as it relates to cross species considerations. Putatively, it is located medial to the insula and ventral to the extreme capsule and runs from the occipital lobe to the frontal lobe. It is thought to directly connect ventral visual cortices with the frontal lobe.

Based on their tract-tracing data, Schmahmann and Pandya (2006, 2007a) assert that the IFOF does not exist, at least in monkeys. They observed no direct connections between the ventral visual cortices and the basal frontal cortex. Therefore, they believe the IFOF to be a false positive when identified in tractography, a nonexistent tract. Although they themselves did not explicitly consider the existence of the IFOF in humans, their statements were cited in subsequent literature as evidence of a potential species difference: Perhaps the IFOF exists in humans but not in monkeys (Catani 2007; Sarubbo et al. 2013; Forkel et al. 2014).

When tractography and dissection techniques have been more recently applied to monkey data, a tract that clearly matches the human IFOF emerges (Mars et al. 2016; Feng et al. 2017; Decramer et al. 2018; Sarubbo et al. 2019; Bryant et al. 2020), arguing against a species difference. This raises the distinct possibility that the IFOF is a false positive. Perhaps tractography and dissection studies are essentially fusing multiple distinct bundles (potentially the uncinate fasciculus and inferior longitudinal fasciculus) creating a polysynaptic pathway between the frontal and occipital lobes (Sarubbo et al. 2019). We would not consider a series of polysynaptic connections to constitute a discrete bundle.

However, evolving endpoint definitions complicate this view. Early tractographic depictions of the IFOF, primarily derived from deterministic tractography, depict a tract with posterior endpoints in the ventral occipital and frontal lobes (Catani and de Schotten 2008; de Champfleur et al. 2013). More recent studies reflect increasingly permissive definitions of ventral occipitofrontal connectivity, resulting in broader connectivity profiles between these regions as compared to earlier accounts (Panesar et al. 2017; Conner et al. 2018; Sarubbo et al. 2019). This broadening is relevant because direct anatomical connections (as shown with monkey tract-tracing) are known to exist between lateral frontal cortex (such as the frontal eye fields) and occipital areas (such as V2 and V4) (Gerbella et al. 2010; Markov et al. 2014). Other bundles terminating in the frontal eye fields or ventral visual areas do not provide an obvious route for a direct connection between the two. Resolving these issues will likely require meticulous tractography, tract-tracing, and dissection studies across species.

Cingulum Bundle

The cingulum bundle (Fig. 7) is another of the brain’s major anterior–posterior bundles and is intimately tied to limbic circuitry (Yakovlev and Locke 1961; Mufson and Pandya 1984; Bubb et al. 2018). Heilbronner and Haber (2014) have used monkey tract-tracing data to define the cingulum bundle as including not only its prominent dorsal component but also a subgenual component curving around the genu of the corpus callosum and continuing posteriorly, tucked up against the gray matter of the subgenual cingulate cortex, as well as a temporal component that curves around the splenium and extends anteriorly into the most medial white matter of the medial temporal lobe (Seltzer and Pandya 1984). According to our work, the volume associated with the cingulum bundle contains three different types of connections. First, fibers from the adjacent regions of cortex that travel short distances (or not at all) anterio-posteriorly within the cingulum bundle must nevertheless cross through it in order to reach their targets. From a connectivity approach, we would not consider these to be properly part of the cingulum bundle, but they cannot be ignored because of their shared volumes. Second, the bundle contains anterio-posteriorly directed cingulate fibers: those emanating from the cingulate, those traveling to the cingulate, and both. Third, there are some fibers traveling long distances within the cingulum bundle with no relationship to the cingulate cortex itself. These connections allowed us to segment the cingulum bundle into four separable components in monkeys: subgenual, anterior dorsal, posterior dorsal, and temporal.

Figure 8.

Fornix, medial forebrain bundle (MFB), ventral amygdalofugal pathway (VAF), and uncinate fasciculus (UF). Volumetric locations shown on high-resolution brain from (Edlow et al. 2019). (A) Sagittal view and (B) coronal view.

Tractographic reconstructions of the cingulum bundle can also capture its full length in humans (Jones et al. 2013). It is also possible to segment the cingulum bundle in humans using dMRI tractography and/or gross dissection according to its constituent connections, as in monkeys. Usually, these correspond roughly to the monkey subdivisions shown above, but with additional components identified because of the whole-brain nature of the investigation (Jones et al. 2013; Wu, Sun, Wang, Wang, Ou 2016; Sweet, Beylergil, et al. 2020). In fact, in a direct human dMRI tractography-based reconstruction of the monkey tract-tracing experiment described above, constituent fibers using each portion of the bundle could be replicated with 72% accuracy (Sweet, Beylergil, et al. 2020). Nevertheless, the cingulum bundle poses significant challenges for tractography. Tract-tracing suggests that fibers appear to enter and exit the bundle at different points anterio-posteriorly, but then there is only a weak topography mediolaterally or dorsoventrally as they travel within the bundle. Thus, precise endpoints may be difficult to track as any given voxel contains a heterogeneous composition of connectivity profiles (Donahue et al. 2016; Grier et al. 2020).

Two components of the cingulum bundle are targets for neuromodulatory treatments. The anterior dorsal component is a target for lesions (cingulotomies) of treatment-resistant patients with major depressive disorder, obsessive–compulsive disorder, and chronic pain. More anteriorly placed lesions are associated with better response to cingulotomy in major depressive disorder and obsessive–compulsive disorder (Steele et al. 2008; Yang et al. 2014). Based on the monkey tract-tracing data, this location would involve more amygdala and lateral orbitofrontal cortex fibers than its more posterior neighbors. Volumetric changes post cingulotomy as well as tractographic analyses also suggest that fibers connecting the anterior cingulate with the striatum and thalamus may also be critical to this treatment’s efficacy, emphasizing the importance of the fibers crossing over the cingulum bundle described above (Rauch et al. 2000; Schoene-Bake et al. 2010). The subgenual cingulum bundle is the target of deep brain stimulation for major depressive disorder (Mayberg et al. 2005). Although originally conceived of for correcting a hyperactive subgenual cingulate cortex in patients (Mayberg et al. 1999), it is now clear that this target involves electrical stimulation of the white matter of the subgenual cingulum bundle, along with fibers from the uncinate fasciculus and corpus callosum (Riva-Posse et al. 2014). One intriguing possibility is that a cingulum bundle target situated between the traditional cingulotomy and subgenual DBS locations may be able to optimally capture both dorsally and ventrally directed fibers (Sweet, Thyagaraj, et al. 2020).

Uncinate Fasciculus

The uncinate fasciculus (Fig. 7) connects the prefrontal cortex with the anterior temporal lobe (Ebeling and v. Cramon 1992). It is noted for its hook- or arclike shape. At its anterior and superior end, the uncinate fasciculus is situated dorsal to the gray matter of the orbitofrontal cortex. Progressing posteriorly and inferiorly, it moves laterally around the striatum, situated underneath the external and extreme capsules, before curving ventrally into the temporal lobe. It is well known for bidirectionally connecting the prefrontal cortex with the amygdala and anterior temporal cortex. In addition, the portion of the uncinate fasciculus sitting dorsally to the orbitofrontal cortex carries fibers traveling within the ventral prefrontal cortex (Lehman et al. 2011). These patterns of connectivity have been captured using tract-tracing and tractography in monkeys, as well as dissections and tractography in humans (Folloni et al. 2019).

One point of contention is whether the uncinate fasciculus extends posteriorly enough to connect the prefrontal cortex with the hippocampal formation (Pribram et al. 1950) or whether such fibers are carried by the nearby inferior longitudinal fasciculus. In addition, some papers refer to the connection as being primarily one between the temporal lobe and the orbitofrontal cortex rather than the prefrontal cortex more broadly (de Schotten et al. 2012). Other accounts have divided the hooklike superstructure of the uncinate into a layered organized series of subcomponents with distinct topographical connectivity profiles (Hau et al. 2017).

Von Der Heide et al. (2013) suggest that the function of the uncinate fasciculus is to allow mnemonic information (from the anterior temporal lobe) to influence valence for decision-making (from the orbitofrontal cortex) and vice versa. Injury to the uncinate fasciculus can cause isolated retrograde amnesia (Levine et al. 1998). dMRI has also demonstrated reduced fractional anisotropy in the uncinate fasciculus in Alzheimer’s disease patients, a change that may be related to memory deficits (Yasmin et al. 2008).

Cortico-Striatal Connections: Muratoff’s Bundle and the External Capsule

Nearly the entire cerebral cortex projects to the striatum, the entryway to the basal ganglia. Cortico-striatal bundles use many routes to reach their targets, but the majority at some point use the external capsule and/or Muratoff’s bundle (Fig. 7). The external capsule is located just lateral to the caudate nucleus. Curving around the lateral edge of the putamen, the external capsule is separated from its more lateral partner, the extreme capsule, by the claustrum (Berke 1960; Petrides and Pandya 2006).

Muratoff’s bundle is situated dorsal to the caudate nucleus, curving around its upper edge. Monkey tract-tracing studies have identified fibers bound for the striatum originating in preoccipital cortices (Yeterian and Pandya 2010), anterior and posterior cingulate cortices (Heilbronner and Haber 2014), dorsal prefrontal cortex, and other association and limbic areas (Schmahmann and Pandya 2006). Although both bundles carry fibers from the cortex to the striatum, fibers passing through the external capsule more commonly terminate in the putamen, whereas fibers passing through Muratoff’s bundle more commonly terminate in the caudate nucleus. However, this rule is not strictly observed, and axons do transfer between these two bundles as well. Extensive literature on the reconstruction of Muratoff’s bundle using dMRI is lacking other than differentiating it from the putative superior frontal occipital fasciculus (Makris et al. 2007).

Extreme Capsule

The extreme capsule is situated between the insula and claustrum (Fig. 7). Although the claustrum is traditionally used as the boundary between the external and extreme capsules, tract-tracing shows that the boundaries across the three structures are not strict, and both capsules also send fibers streaming across the claustrum (Lehman et al. 2011). The extreme capsule primarily runs anterio-posteriorly. However, the volumes occupied by the extreme capsule also contain many fibers traveling to and from the insula simply because of its position. As with the cingulum bundle, from a connectionist perspective, we would not consider these to be part of the extreme capsule. Instead, monkey tract-tracing demonstrates that the extreme capsule connects the frontal cortex with the superior temporal gyrus and sulcus (Schmahmann and Pandya 2007b). It targets relatively more dorsal aspects of the frontal cortex and posterior aspects of the temporal cortex than the uncinate fasciculus does.

Tractography investigations of the extreme capsule in humans as well as monkeys have replicated the connections found in monkey tract-tracing (Frey et al. 2008; Mars et al. 2016) but also suggested that the bundle reaches back to the visual cortex (Mars et al. 2016). Importantly, the relationship between the extreme capsule and the inferior fronto-occipital fasciculus (IFOF) is not clear (Bajada et al. 2015) not least because some investigators use the term “extreme capsule fiber complex,” which may be broader than just the extreme capsule. One parsimonious explanation of the various discrepancies is that the IFOF and extreme capsule are separable bundles (IFOF and its “neck” is positioned ventrally to the extreme capsule), with the IFOF reaching back to the occipital cortex, and the extreme capsule terminating in the superior temporal cortex. However, given the controversies surrounding IFOF, further study on the differentiation between these two bundles is necessary (Makris and Pandya 2009).

Superior Fronto-Occipital Fasciculus

Schmahmann and Pandya (2006) described a bundle in monkeys, which they refer to as the fronto-occipital fasciculus (which here we will refer to as the superior fronto-occipital fasciculus, SFOF, to distinguish it from IFOF), as adjacent to the corpus callosum, SLF II, Muratoff’s bundle, and cingulum bundle. It carries fibers between the parietal and dorsal frontal lobes. Older dissection studies on human brains were mixed in their descriptions of this bundle (Forel 1881; Dejerine and Dejerine-Klumpke 1895; Schröder 1901). Importantly, modern human investigations, using both tractography and Klingler dissections (Türe et al. 1997; Forkel et al. 2014; Meola et al. 2015; Liu et al. 2020), have failed to replicate this structure. Instead, a structure consistent with descriptions of the SFOF can be found in individuals with agenesis of the corpus callosum, but this observation may not apply to white matter architecture more broadly (Schmahmann and Pandya 2006, 2007a; Forkel et al. 2014). Furthermore, even Schmahmann and Pandya (2006) do not identify occipital connections of SFOF; instead, the connectivity of their bundle appears quite similar to that of SLF II. Even in cases where a structure resembling accounts of SFOF can be reconstructed in humans via tratographic approaches (Makris et al. 2007; Uddin et al. 2010; Liu et al. 2020), the posterior connectivity observed is with the parietal lobe (and thus more consistent with accounts of SLF II), and not with the occipital lobe as would presumably be required of a fronto-occipital fasciculus (Mandonnet et al. 2018). One possibility is that SLF II extends more medially in the monkey than in humans, thus generating the observed tract-tracing results. Perhaps there are constituent frontal or parietal connections that distinguish SFOF from SLF II. Regardless, it is unclear whether there is a separate, distinct bundle with its own connectivity patterns in either species that should be referred to as SFOF.

Optic Radiation

The optic radiation consists of feed-forward axonal projections connecting the lateral geniculate nucleus (LGN) with the visual cortex (Ebeling and Reulen 1988). It is deeply embedded in the corona radiata and sagittal stratum. There is also evidence that fibers to/from structures other than the LGN and V1 (such as the pulvinar, V2, V3) contribute small numbers of axons to this structure (Yoshida and Benevento 1981; Párraga et al. 2012; Alvarez et al. 2015).

A characteristic morphological aspect of OR corresponds to the fibers in the anterior portion that progress with a sharp turn as they move towards the posterior of the brain. Some have used the term Meyer’s loop to define only the more anteriorly arching, morphological feature of OR (Ebeling and Reulen 1988; Sarubbo et al. 2015) whereas others use the same term to refer to the entire length of the axons that traverse the anterior of the thalamus, which includes their posterior expanse as well (Párraga et al. 2012; Pescatori et al. 2017). Another interesting terminological phenomenon has been associated with the posterior/dorsal component of the optic radiation. As noted (Knipe et al. 2020), a number of recent publications including, most notably, the 2019 edition of Gray’s Surgical Anatomy (Brennan et al. 2019) have made reference to this structure as the eponymously named “Baum’s loop.” Though this structure may in fact exhibit a dorsal arc, and thus be sensibly labeled as a “loop,” the actual provenance of this full term is decidedly suspicious. Indeed, the source has been traced (Knipe et al. 2020) to a 2009 edit to the Wikipedia article for “optic radiation” (specific edit here) and, perhaps even more stunningly, may have even traced it to the apparent individual who coined the eponym implicitly referencing themself.

OR has been routinely studied for its involvement in supporting vision in humans, and recent evidence shows major effects to the white matter of the optic radiation due to eye and visual disease (see BOX 2).

Box 2

Implications for Human Health and Interventions: Effect of Eye and Visual Disease on the Brain’s White Matter. Recently, researchers interested in diseases and disorders of the human eye have reported associations between these conditions and changes to the brain white matter tissue. In this way, such disorders uniquely demonstrate how the alteration of external sensory organs (and thus the signals they transduce) can result in downstream alterations, that is a demonstration of plasticity. Indeed, investigations of a diverse array of pathogenic processes involving eye motor coordination (Allen et al. 2015; Ashkan et al. 2017), intraocular pressure (Hanekamp et al. 2021), and the retina itself (Ogawa et al. 2014; Yoshimine et al. 2018) have demonstrated that a reduction in sensory inputs results in alterations of the brain’s white matter tissue. Understanding the extent to which such changes are reversible will be key to identifying viable and effective paths to intervention. For example, is it the case that once the white matter tissue is affected because of sensory disease no rehabilitation will allow effective recovery of function?

Fornix

A complex, C-shaped structure with three major components (crus, body, and columns), the fornix is situated at the brain’s midline and is the major output structure for the hippocampus (Saunders and Aggleton 2007) although it also contains hippocampal afferents (Fig. 8). Fibers leave the hippocampus, course underneath the lateral ventricle, and form the fornix. At the level of the splenium, these fibers form the crus of the fornix; fibers run beneath the corpus callosum and eventually form the body of the fornix; fibers then descend near the anterior commissure to form columns (Poletti and Creswell 1977). The fornix contains both crossing fibers (at the body) and ipsilateral projection fibers. Through the fornix, hippocampal formation fibers reach subcortical structures like the mammillary bodies, anterior thalamic nuclei, and the septum.

The fornix can be reconstructed with tractography using targeted approaches (Rheault et al. 2018; Milton et al. 2020) as opposed to whole brain tractography. Its cross-sectional area in some locations has been estimated to be approximately 2 mm, which is challenging for nontargeted tractography methods, particularly given the curvature of the structure. Furthermore, there is evidence that adjacent cerebrospinal fluid might contaminate the signal, such that suppression improves reliability (Concha et al. 2005), which is an example of a partial volume effect (Alexander et al. 2001). Perhaps for this reason, the fornix has received little attention in the tractography literature. Therefore, it is difficult to assess whether there are potential methodological and/or species differences associated with this bundle.

The fornix is central to the classic Papez circuit (Papez 1937) responsible for emotion, learning, and memory (Delay and Brion 1969). Fornix damage is well known to result in deficits in these processes (see (Thomas et al. 2011)). For example, fornix injury has resulted in anterograde amnesia (Gaffan et al. 1991). Properties of the fornix revealed by diffusion imaging are found to be associated with clinical status (as compared to healthy controls) in a multitude of diseases, including Alzheimer’s Disease, multiple sclerosis, and schizophrenia (Douet and Chang 2014).

Medial Forebrain Bundle

The medial forebrain bundle is a complex network of fibers connecting parts of the brainstem and basal forebrain with the frontal cortex (Garver and Sladek 1976; Felten and Sladek 1983; Oades and Halliday 1987; Moore and Bloom 2003; Parent et al. 2011) (Fig. 8). It is not technically white matter because its axons tend to be unmyelinated or lightly myelinated, but we will review it here because of widespread interest. This bundle’s axons ascend from the brainstem and course through the basal forebrain, ventral to the decussation of the anterior commissure, collecting fibers along the way. Notably, the medial forebrain bundle contains many dopaminergic fibers, although not exclusively.

There is substantial confusion in the dMRI literature regarding the position of the medial forebrain bundle (Haber et al. 2020). This bundle does not ascend into the internal capsule; instead, it remains ventrally positioned until reaching its targets in the frontal lobe. Seeding certain brainstem regions that contribute to the medial forebrain bundle results in internal capsule connectivity; however, this is because of the complexity of the brainstem-cortical connection. For example, dopaminergic fibers do not traverse the internal capsule, but use the medial forebrain bundle to reach their targets.

The medial forebrain bundle is widely known for effective self-stimulation in rodents; this is likely due in part to the stimulation of dopaminergic fibers within the bundle. More recently, it has come to the attention of clinicians interested in DBS. However, because of the previously mentioned issues with tracking brainstem fibers, there is a widespread misunderstanding of the types of connections DBS in different locations can affect. DBS of the internal capsule is not capturing the medial forebrain bundle, and therefore not capturing dopaminergic axons (Coenen et al. 2009; Haber et al. 2020).

Ventral Amygdalofugal Pathway

The ventral amygdalofugal pathway contains fibers connecting the amygdala with the rest of the brain (Nauta 1961; Aggleton et al. 1980; Porrino et al. 1981; Amaral and Price 1984; Mori et al. 2017) (Fig. 8). A careful study comparing anatomical tract-tracing in macaques, diffusion tractography in macaques, and diffusion tractography in humans found broad consistency in the location and connectivity of the ventral amygdalofugal pathway (Folloni et al. 2019). This tract exits the amygdala, ascends dorsally and medially, and travels ventral to the anterior commissure and striatum and medial to the uncinate fasciculus. At its most medial projection, the pathway bifurcates into ascending and descending branches. The ascending branch enters the nucleus accumbens followed by the prefrontal cortex. The descending branch projects to the anterior hypothalamic nuclei (Kamali et al. 2016). At frontal locations, it is interwoven with the uncinate fasciculus.

Conclusion

Despite substantial efforts, it remains inarguable that we do not have a complete map of human brain connections. This stands as a major obstacle for turning our field’s already appreciable collective anatomical knowledge into actionable insights for understanding human behavior, development, and disordered processes. One challenge we will be forced to face as a field is a reconsideration of what we actually know and what is instead a matter of convention. By annealing our diverse methodological backgrounds, we have come to realize that understanding can become siloed, cut off from contributing to the formation of a discipline-wide synthesis. In other words, we strive to understand white matter anatomy, agnostic of approach or method. Unfortunately, current accounts are occasionally divergent or even inconsistent. Moreover, it is not always clear if divergent accounts can accommodate one another. This raises the uncomfortable possibility that some amount of research work, perhaps even nontrivial amounts thereof, may ultimately need to be reconsidered. On an optimistic note, though, this necessary self-reflection may also stand as an opportunity to learn about how anatomical evidence accumulates and evolves.

Funding

National Institutes of Health (R01MH118257 to S.R.H.); National Institute on Drug Abuse (T32DA007234 to P. Mermelstein (UMN) to M.D.G.); National Institute of Mental Health (5T32MH103213-05 to W. Hetrick (IU) to D.N.B.); National Science Foundation ((IIS-1636893), National Science Foundation (BCS-1734853), National Science Foundation (IIS-1912270), National Institutes of Health NIBIB (1-R01-EB029272-01) to F.P.; and a Microsoft Investigator Fellowship a Microsoft Investigator Fellowship to F.P.

Notes

We thank Sylvie Bindas for assistance with reference curation. Competing interests: S.R.H. has received teaching fees from Medtronic, Inc. The remaining authors have no conflicts to disclose.