Signal Processing in the Vagus Nerve: Hypotheses Based on New Genetic and Anatomical Evidence

Abstract

Each organism must regulate its internal state in a metabolically efficient way as it interacts in space and time with an ever-changing and only partly predictable world. Success in this endeavor is largely determined by the ongoing communication between brain and body, and the vagus nerve is a crucial structure in that dialogue. In this review, we introduce the novel hypothesis that the afferent vagus nerve is engaged in signal processing rather than just signal relay. New genetic and structural evidence of vagal afferent fiber anatomy motivates two hypotheses: (1) that sensory signals informing on the physiological state of the body compute both spatial and temporal viscerosensory features as they ascend the vagus nerve, following patterns found in other sensory architectures, such as the visual and olfactory systems; and (2) that ascending and descending signals modulate one another, calling into question the strict segregation of sensory and motor signals, respectively. Finally, we discuss several implications of our two hypotheses for understanding the role of viscerosensory signal processing in predictive energy regulation (i.e., allostasis) as well as the role of metabolic signals in memory and in disorders of prediction (e.g., mood disorders).

The vagus nerve is cranial nerve X (Figure 1A). It consists of both ascending afferent sensory signals from the viscera that inform the brain on the sensory state of the body’s internal milieu (Jänig, 2006; Wehrwein et al., 2016) and descending efferent parasympathetic motor signals from the brain to control the viscera and other tissues in the body (e.g., Hammer et al., 2018; Jänig, 2006; Loewy & Spyer, 1990).¹ There is a resurgence of interest in understanding vagus nerve anatomy, in part due to the increased popularity of vagus nerve stimulation as a trans-diagnostic intervention for various mental and physical disorders (e.g., Ben-Menachem, 2002; Nemeroff et al., 2006; Shen et al., 2011). The anatomy of the vagus nerve has also been debated in the context of the polyvagal theory (e.g., Porges, 1995, 2021), given that the anatomical description and evolutionary origins of the vagus nerve that underpin the polyvagal theory has been called into question (Berntson et al., 2007; Doody et al., 2023; Grossman et al., 2023; Grossman & Taylor, 2007; Neuhuber & Berthoud, 2022; Taylor et al., 2022). In this paper, we review new evidence about the molecular and anatomical architecture of vagal fibers and then outline two novel hypotheses about possible computational capacities suggested by that architecture. Our review of the evidence is not proof that these hypotheses are correct, of course, but rather describes how existing empirical data constrains what is functionally possible within the vagus and attests to these hypotheses as worthy of investigation.

Figure 1.

Panel A shows a schematic representation of the gross anatomical distribution of the afferent vagus nerve. The enlargement just above the nodose ganglion represents the jugular ganglion (through which somatic and visceral afferents to the ear, pharynx and larynx pass; our focus here is on the neurons passing through the nodose ganglion). The nodose and jugular ganglia, as well as all cervical and thoracic branches shown are bilateral, with one side omitted here for clarity. Panel B shows a more detailed schematic view of the interface between the brainstem nuclei, vagus nerve, and target organs. The four brainstem nuclei have been spatially oriented to depict their proximity to one another, and their rough ordering relative to the midline of the brainstem; however, this is a simplified view and a precise rendering of the relative lengths of each nucleus and their proximity to other spinal cord structures is not depicted. Created with Biorender.com.

Crucial to our approach is the hypothesis that a central nervous system builds a predictive internal model of its own body in an ever-changing and only partly predictable world, which reduces uncertainty and minimizes metabolic cost while guiding the organism’s behavior (e.g., Sterling & Laughlin, 2015). A predictive strategy allows an animal to reduce uncertainty by generalizing from similar past experiences to prepare responses and anticipate the sensory consequences of those responses, with incoming sensory signals serving to confirm or correct the predictions (for review and additional references, see Hutchinson & Barrett, 2019). This strategy is more energy-efficient and less error-prone than stimulus-response strategies (Sterling & Laughlin, 2015). Energy efficiency is a major consideration both for immediate health but also for longer-term survival and reproductive fitness (Jékely et al., 2015; Niven & Laughlin, 2008; Sterling & Laughlin, 2015). Elsewhere, we have hypothesized that energy-efficient, predictive control of the body forms the anatomical and functional basis for the emergence and functioning of a mind (e.g., Atzil et al., 2018; Atzil & Gendron, 2017; Barrett, 2017; Hutchinson & Barrett, 2019; Katsumi et al., 2022, 2023). From this control perspective, a basic function of a brain is not to “perceive” the world per se, but to build and run an internal model of the body in that world that anticipates and efficiently adjusts for sensory inputs that are relevant to the physiological conditions of the body.² Here, we consider the role of the vagus nerve in this regulatory effort.

The first section of this paper briefly reviews the anatomical organization of the vagus nerve and the main brainstem nuclei to which it projects. These details ground discussions in the subsequent sections. The second section reviews two signal processing motifs in the nervous system, one mapping the spatiotemporal correlations between incoming sensory signals (i.e., coding by temporal coincidence over spatially restricted inputs) and the other mapping only temporal correlations (i.e., coding by temporal coincidence over a wide range of spatially distributed inputs; Figure 2). This overview sets the stage for our hypothesis, discussed in the next section, that the sensory vagus nerve may do more than simply relay signals between the body and the brain — its anatomical features may allow for signal processing. We examine new genetic and structural evidence of vagal anatomy as a starting point from which we consider how the vagus may compute both spatial and temporal features of the sensory signals informing on the physiological state of the body, and how the architecture of the brainstem nuclei also may support these computational processing motifs. We also review anatomical evidence of the organization of vagal fibers within the nerve trunk and discuss the possibility that ascending and descending signals may modulate one another, calling into question the strict segregation of sensory and motor signals, respectively. The final section considers the broader implications of the two signal processing hypotheses that we introduced for understanding the role of viscerosensory signal processing in predictive energy regulation (i.e., allostasis) as well as the role of metabolic signals in memory and in disorders of prediction.

Figure 2. A schematic of spatial and temporal mapping in sensory systems.

This figure represents spatial and temporal coding themes across a variety of sensory systems and is thus highly abstracted away from anatomical details in service of a more general depiction. Panel A shows a schematic depiction of nearest-neighbor array-to-array spatial mapping exemplified by the visual system. Neurons in the input field (e.g., retinal ganglion cells) connect to neurons in the target field (e.g., lateral geniculate nucleus neurons) in a manner that preserves the spatial ordering (and spatial correlations) in signals from the input field. Each color represents a different spatial location in the visual field. Signals arising from neurons that are closer to one another in the input field are more highly correlated with one another than are signals arising from neurons that are spatially more distant. Panel B shows a schematic depiction of combinatorial temporal mapping exemplified by the olfactory system. Neurons expressing the same receptor type gene in the input field (e.g., olfactant receptors in the olfactory epithelium) converge onto the target field (e.g., glomeruli in the olfactory bulb) in a manner that preserves the signal co-occurrences in time. A given color represents an olfactory receptor neuron that expresses one specific odorant receptor gene, which converges on the same glomeruli. Local activity patterns for neurons that are co-activated by a common odorant in the input field are highly correlated in time, while activity patterns for other odorants are less correlated. Created with Biorender.com.

A Brief Overview of the Vagus and Its Brainstem Targets

The left and right nodose ganglia sit at the base of the head, while the peripheral processes for these ganglionic neurons are embedded in the visceral organs and other peripheral tissues and serve as the main afferent fibers of the vagus. The nodose ganglia are pseudo-unipolar, meaning that these peripheral processes are functionally ‘dendritic’ but are conventionally considered axons, bringing viscerosensory signals from the body to the nodose ganglion cell bodies (Câmara & Griessenauer, 2015; see Figure 1B). From there, the axons continue and synapse in the nucleus tractus solitarii (NTS; Figure 1B) and in the adjacent area postrema (AP) in the medullary brainstem (see Leslie & Gwyn, 1984 for review).³ In referring to afferent signals from the body before they reach the brain, we use “viscerosensory”; once these signals have reached the brain, we use the term “interoceptive”.

Beyond receiving ascending neural signals directly from the nodose, the area postrema functions like a “circumventricular organ,” by virtue of being highly vascularized by fenestrated⁴ capillaries. As a result, pores between the vasculature and AP proper allow relatively free passage of large macromolecules (e.g., proteins and peptide hormones) into the brain, molecules that would otherwise remain segregated from the brain via the blood brain barrier. In this way, the AP acts as a unique chemoreceptive liaison between the peripheral bloodstream and the brain (Cottrell & Ferguson, 2004; Duvernoy & Risold, 2007; Price et al., 2008). Afferent axons from the nodose ganglia also project directly to the dorsal motor nucleus of the vagus (DMV), which lies adjacent to the NTS dorsally, and was originally assumed to be exclusively a vagal motor nucleus (see Figure 1B; Champagnat et al., 1986; Kalia & Mesulam, 1980; Kalia & Sullivan, 1982; Neuhuber & Sandoz, 1986).⁵ Approximately 80% of vagal fibers ascend to carry viscerosensory signals to the brainstem (Hammer et al., 2018; Jänig, 2006; Loewy & Spyer, 1990; Wehrwein et al., 2016).⁶

The rest of the vagal fibers descend from the DMV and the nucleus ambiguus (nAmb) in the ventrolateral medulla, which are the primary origins of efferent parasympathetic motor control of organs in the periphery (below the neck). These efferent parasympathetic fibers bypass the nodose ganglia on their way to their visceral targets.^7,8 The majority (i.e., 80%) of the neurons in the DMV give rise to vagal efferents to the gastrointestinal tract and abdominal viscera, including the pancreas (e.g., Mussa & Verberne, 2013; Norgren & Smith, 1988), with fewer innervating the heart and lungs (e.g., Cheng & Powley, 2000). The nAmb is divided into a dorsal and ventrolateral portion, with the dorsal division giving rise to fibers innervating the striated muscle of the esophagus, the pharynx, and the larynx (e.g., Bieger & Hopkins, 1987; Hsieh et al., 1998; McLean & Hopkins, 1985b), and the ventrolateral division giving rise to the majority of vagal efferents targeting thoracic organs such as the heart and lungs (Bieger & Hopkins, 1987; Hopkins & Armour, 1998; Kitamura et al., 1993).

The NTS, AP, DMV, and nAmb are themselves densely interconnected. Fibers from the NTS modulate neurons in both the DMV (via DMV dendrites that extend into the NTS; e.g., Davis et al., 2004; McLean & Hopkins, 1985a; Shapiro & Miselis, 1985a) and the nAMB (via NTS projections to the rostral nAmb; Saper & Stornetta, 2015; Stuesse & Fish, 1984). Likewise, there are extensive reciprocal connections between the NTS and adjacent AP neurons (Leslie & Gwyn, 1984; Shapiro & Miselis, 1985b). While there are not significant direct projections between the AP and DMV, it has been suggested that the AP may influence the DMV indirectly, via AP dendrites that synapse in the NTS, which overlap with the dendrites of the DMV synapsing onto the same area in the NTS (e.g., Miselis & Shapiro, 1983; Shapiro & Miselis, 1985a, 1985b). The NTS also sends ascending signals to other brainstem nuclei such as the locus coeruleus (bed nucleus for norepinephrine), the dorsal raphe nucleus (serotonin) and ventral tegmental area (dopamine), as well as targets in the midbrain and forebrain including hypothalamic and amygdalar subnuclei, portions of the basal ganglia (e.g., nucleus accumbens) and the hippocampus via the septal nuclei (Rinaman, 2003, 2010; Rinaman & Schwartz, 2004; Suarez et al., 2018). The DMV receives projections from many of these same NTS targets in the midbrain and forebrain (Loewy & Spyer, 1990; ter Horst et al., 1984). The AP also projects to the parabrachial nucleus in the midbrain and to other nuclei of the medulla (Cunningham et al., 1994; Leslie & Gwyn, 1984; Shapiro & Miselis, 1985b). Taken together, these highly interconnected brainstem nuclei, their vagal inputs, and their shared projections to other brainstem nuclei as well as to midbrain and forebrain targets comprise a crucial infrastructure that is well suited to supporting efficient sensing in the service of metabolic control.

Signal Processing Motifs in the Nervous System

The vagus nerve projects to the central nervous system (CNS), which is a sophisticated signal processor that continually receives ensembles of signals from the sensory surfaces of the body (in humans, some examples are the retina in each eye, the cochlea in each ear, pressure receptors in the skin, the taste buds, receptors in the olfactory epithelium of the nose, glucose sensors in the skeletal muscles, baroreceptors in the large arteries, and nutrient sensors in the gastrointestinal tract and other viscera). The signals from these surfaces continually stream to the brain to inform on the changing sensory state of the body. Neuron structure – such as neuron size, axonal diameter, degree of myelination, etc. – as well as the architecture of neuronal connectivity and the supporting glial cells afford signal transformations that enable an organism to survive, thrive, and reproduce in a highly variable and continually changing environment. A variety of signal processing computations exist in the CNS, including signal integration, modulating signal strength or duration, adding or reducing stochastic noise, transducing signals into a common format, and so on (e.g., Traynelis & Jaramillo, 1998; Wolff, 2016).

One particularly important type of computation for our purposes is signal compression, which involves removing redundancies for signals that are correlated in time and/or space to increase signal quality while decreasing processing cost. Signal compression effectively creates efficient summaries of the original signals. These summaries are sometimes referred to as “abstractions” of the original signals, not because they remove the sensory details entirely, but because they summarize those details in a more efficient format. The CNS has been viewed as a system for removing redundancies among incoming sensory signals at least since the time of Claude Shannon’s information theory (Attneave, 1954; Barlow, 1961; Shannon & Weaver, 1964/1949). Its architecture affords compression along multiple gradients, across which signal ensembles of higher dimensionality (e.g., for many small, specific sensory details) are increasingly summarized into ever more efficiently packaged ensembles of lower dimensionality. Compression gradients have been identified within the cerebral cortex (e.g., Bethlehem et al., 2020; Bastos et al., 2020; Chanes & Barrett, 2016; Finlay & Uchiyama, 2015; Katsumi et al., 2022, 2023; Margulies et al., 2016; Paquola et al., 2019), the hippocampus (Kharabian Masouleh et al., 2020; Przeździk et al., 2019; Vos de Wael et al., 2018), and the cerebellum (Dong et al., 2020; Guell et al., 2018; see Katsumi et al., 2023 for discussion). There is also some evidence that somatosensory signals are compressed along a gradient as they arise in the dorsal column of the spinal cord (e.g., Cappe et al., 2009; Castejon et al., 2021; Saadon-Grosman et al., 2020), and signal compression may occur in the peripheral nervous system as well (Copelli et al., 2005).

There is spatial and temporal structure in the sensory signals that are transduced by a body’s sensory surfaces (i.e., the “where” and “when” of signal co-occurrences). Distinctive architectural arrangements in the CNS give rise to signal processing, including signal compression, within spatial and temporal domains (see Figure 2). To illustrate, we use the processing of light signals in the vertebrate visual system as an example of spatial mapping (coding the temporal co-occurrence of signals in a manner that preserves the spatial organization of receptors), and chemosensory signals in the olfactory system as an example of temporal mapping (coding the temporal co-occurrence of signals across receptors that are widely distributed). To be clear, our goal is not to provide comprehensive descriptions of the visual and olfactory systems here, but to focus on those details that are most relevant to our discussion of possible signal processing motifs in the vagus nerve. Nor we are claiming a strict double-dissociation; the world is filled with signals that vary in a spatiotemporal fashion and spatially coded signals can be transformed into temporal codes and vice versa (Buzsáki & Tingley, 2018). The visual system certainly processes the temporal relations between signals (e.g., Rucci et al., 2018; Price & Gavornik, 2022) and olfactory cues are used in spatial navigation (e.g., Jacobs et al., 2015; Reddy et al., 2022). Nonetheless, the architectural distinctions between these two different signal processing motifs may be a useful lens for considering the sorts of signal processing that might be possible in the afferent vagus nerve, given its anatomy.

Nearest-Neighbor Spatial Mapping

Light signals from neighboring areas in the visual field activate receptors in neighboring areas of the retina, creating a retinal map of visual space. Corresponding retinotopic maps exist in the superior colliculus in the midbrain, in the lateral geniculate, pulvinar, and the lateral posterior nuclei of the thalamus, and in the early portions of visual cortex, including primary visual cortex (e.g., Chklovskii & Koulakov, 2004; Luo & Flanagan, 2007). The positioning of neuronal receptive fields in each of these regions mirrors the spatial array in the retina (Kandel et al., 2014). This is called “nearest-neighbor, array-to-array” mapping, whereby the spatial relationships in the signal array at their origin (the retina) correspond to the array at their termination (see Figure 2A; for discussion, see Finlay & Uchiyama, 2015; MacIver & Finlay, 2022). The nearest-neighbor mapping for the spatial relationships among signals does not reference the spatial relations in the external world in any absolute, body-independent manner. Rather, the spatial mapping is relative to the sensory surface of the body (in this case, the retina), which necessarily depends on the body’s position in the external world (e.g., when you turn your head from left to right, or move your eyes from left to right, your retinas change position in relation to the world, and so your visual field necessarily changes). It is this spatial array that is recapitulated in multiple parts of the brain. Nearest-neighbor mapping of the visual field occurs not because the brain is tracking the outside world, but because it is mapping what is happening on the retina. Other classic exteroceptive sensory systems also use nearest-neighbor mapping: in the somatosensory system, nearest-neighbor mapping is relative to positions on the skin, in the joints and muscles, etc. (e.g., Sanchez-Panchuelo et al., 2010; Servos et al., 1998), and in audition, nearest-neighbor mapping is relative to the tonotopic map of the cochlea (Allen et al., 2022; Mesik et al., 2022; Saenz & Langers, 2014; Thomas et al., 2015; von Békésy & Peake, 1990).

A complete discussion of the details establishing nearest-neighbor, array-to-array mapping that preserves the spatial topography of the retina across levels of the neuraxis is beyond the scope of this paper, but some details provide a useful starting point for formulating hypotheses about possible signal processing capabilities within the vagus by virtue of its molecular and anatomical architecture (details summarized from Reese, 1987, 1993, 1996, 2011; Reese et al., 1991; Reese & Cowey, 1988). Retinal ganglion cells of all types and sizes with different neural features and response profiles are intermingled across the retina. As axons leave the retina within the optic nerve (cranial nerve II), they are bundled together into heterogeneous collections of fibers called fascicles, which are separated by connective tissue. Each fascicle contains axons of all types from a radial sector of the retina. As a result, the optic nerve has a mixed distribution of large and small diameter axons intermingled at every location along its extent. These axons are organized in a crude retinotopic manner.

Once retinal axons cross the optic chiasm, they are no longer bundled into fascicles and are referred to as the optic tract. Post-chiasm, axons become reordered according to their diameter, with finer axons positioned deeper within the tract and thicker axons positioned superficially. Axon diameter (and therefore axon position in the optic tract) is organized by the embryonic birthday of the retinal ganglion cell from which the axon derives. This organization is called chronotopy (Reese, 1993). This reorganization of axons by diameter crudely approximates the fact that axons of small diameter terminate in the deep layers of the LGN whereas axons of larger diameter terminate in the superficial layers.¹⁰ In addition, the axons also undergo a retinotopic reorganization across the medio-lateral axis of the optic tract (corresponding to the dorso-ventral axis of the retina). Importantly, both reorganizations are insufficient to produce the retinotopic maps found elsewhere in the brain, including where retinal axons first synapse in target nuclei further up the neuraxis.

The retinotopic maps in the superior colliculus and thalamic nuclei take shape, at least in part, via a gradient of genetically-expressed guidance molecules, producing what is called gradient-matching. A gradient of gene expression in the retinal ganglion cells matches a corresponding gradient in the neurons of the superior colliculus and in the relevant thalamic nuclei. The matching gradients allow axons that originate in specific genetic classes of retinal ganglion cells to find their appropriate targets in the collicular or thalamic neurons, thereby producing retinotopic maps (Feldheim et al., 1998; Frisén et al., 1998; Triplett et al., 2012). The retinotopic maps in early visual cortex (i.e., in V1, V2, and so on) are established, in large part, by Hebbian learning. Here, signals from collicular and thalamic neurons that monitor neighboring regions of visual space fire in a coordinated fashion, sending coordinated signals to neurons in early visual cortex during development in the weeks and early months after birth. That is, retinotopic maps in early visual cortex emerge from spatially correlated, nearest-neighbor signals. Gradient-matching and Hebbian learning may not be the only processes at play in setting up nearest-neighbor, array-to-array (spatially correlated) maps, but they are important ones.

The processing of visual signals, including the compression of redundant spatial correlations (to increase transmission and computational efficiency), begins in the retina (e.g., the signals from about 120 million rods and 6 million cones in the human retina are transmitted out of the retina by about 1.2 million retinal ganglion cells; Kim et al., 2021). The degree of compression in retinal ganglion signals as they reach the LGN is still a matter of considerable research (e.g., Morgan et al., 2016). Importantly for our purposes, it is well known that a good deal of visual signal compression occurs along the main cytoarchitectural gradient of the cerebral cortex, anchored at one end by neurons of primary visual, auditory, or somatosensory cortex (computing specific sensory details that are relatively closer to the dimensionality of the sensory surfaces). At the other end, this gradient is anchored by neurons of the subgenual and pregenual anterior cingulate cortex (sgACC, pACC), entorhinal cortex (EC), the ventral anterior insula (vAI; which is adjacent to or part of the posterior orbital frontal cortex, pOFC; Öngür & Price, 2000) and other cortical limbic regions that assemble multimodal compressed summaries (for discussion, see Barbas, 2015; Chanes & Barrett, 2016; Katsumi et al., 2023, 2023 and references therein). Not surprisingly, these regions are some of the most well-connected hubs within the cerebral cortex (e.g., van den Heuvel & Sporns, 2013). These cortical regions are also heavily involved in allostasis and are the cortical origins of parasympathetic visceromotor regulation signals (e.g., Beissner et al., 2013; Vogt, 2016; Vogt et al., 2003), a point that we return to later in the paper.

This main gradient of signal compression in the cerebral cortex has been characterized in terms of evolutionary and developmental changes in cortical expansion, as changes in allometric scaling across species, and is manifest in several biological features that afford signal compression, including successive increases in intracortical myelination and in pyramidal neuron size and connectivity, successive decreases in neuronal density, and decreases in laminar development, as well as increases in cerebral metabolism in the upper layers of the cerebral cortex. A quick review of the cortico-cortical connections in this gradient is helpful to understand how signal compression is achieved.

In the visual system, the main gradient begins in V1, with many small, relatively less connected pyramidal neurons located in the upper cortical layers carrying signals for specific retinotopic details of high dimensionality. As signals flow to early visual cortical areas (e.g., Felleman & Van Essen, 1991; Rao & Ballard, 1999) and eventually to the multimodal integration regions of the cortex (e.g., Braga et al., 2013; Sepulcre et al., 2012; Szinte & Knapen, 2020), they converge on successively fewer and fewer upper layer pyramidal neurons with progressively larger cell bodies with increasing connectivity (e.g., Rao & Ballard, 1999). This convergence successively removes redundant correlations. The signal compression is lossy, meaning that redundant information is lost during compression. This ‘many-to-fewer’ cytoarchitectural pattern that achieves efficient signal compression is consistent with anatomical findings from Barbas (2015; Barbas & Rempel-Clower, 1997), Markov (e.g., Markov et al., 2013), and others (e.g., Hilgetag & Goulas, 2020) showing that patterns of cortico-cortical connections predict signal flow within the cerebral cortex. As signals traverse this cortical compression gradient, many signals of high dimensionality are reduced to multimodal summaries of lower dimensionality (described in Barrett, 2017; Chanes & Barrett, 2016; Finlay & Uchiyama, 2015; Katsumi et al., 2022, 2023; for related review, see Bastos et al., 2020). Such compression effectively computes “abstract” features, such as pleasure, threat, value, reward, and other features that map to multiple patterns of high-dimensional sensory and motor details (discussed in Barrett, 2017, 2022). With signal compression, array-to-array mapping becomes less and less fine-grained (Finlay & Uchiyama, 2015), until the most compressed, multimodal signal summaries are assembled in the cortical limbic areas (e.g., sgACC, pACC, EC, vAI/pOFC), at which point nearest-neighbor mapping no longer reflects the arrays across any particular sensory surface.¹⁰

Temporal Co-Occurrence Mapping

The mapping of temporal relationships between incoming chemosensory signals in the primate olfactory system is also relevant to our discussion of possible signal processing motifs in the afferent vagus nerve. Our goal in this section is to highlight features that allow for mapping the temporal co-occurrence of molecules (called “olfactants” or “odorants”) rather than to describe everything that has been discovered in the rapidly advancing research on signal processing in the olfactory system. Functionally, temporal coding (coding by temporal coincidence over a wide range of spatially distributed inputs) contributes to the ability to perceive billions of odors available in the chemosensory environment (the upper bound in humans having been estimated at close to two trillion (Gerkin & Castro, 2015; Meister, 2015). We draw on research from multiple vertebrate species (mostly mammals) because the anatomy and functioning of the olfactory system is strikingly similar across species (Manzini et al., 2022; with a few notable exceptions not relevant to this discussion, such as the existence or importance of the vomeronasal organ). Accessible reviews of olfactory anatomy can be found in Cleland & Borthakur (2020), Manzini et al. (2022), and van Hartevelt & Kringelbach (2012).

The sensory receptors for olfactant molecules are found in the olfactory epithelium that lines the nasal cavity. Receptors are part of the olfactory sensory neurons (also sometimes called olfactory receptor neurons). Each neuron is a bipolar cell equipped with a single dendrite embedded in the mucosal layer of the nasal cavity. Each dendrite expands into several microvilli called olfactory cilia containing a type of olfactory receptor. Olfactory receptors are highly diverse in molecular phenotype (~1000 types; Malnic et al., 1999; Niimura & Nei, 2005; van Hartevelt & Kringelbach, 2012). Studies using RNA sequencing indicate that the vast majority of olfactory sensory neurons express a single receptor gene at high levels, meaning that a given neuron is highly sensitive and selective for the molecules that bind to its dominant receptor (Hanchate et al., 2015; Saraiva et al., 2015; Xu et al., 2020; for additional references see Kurian et al., 2021).¹¹ Olfactory sensory neurons that express the same receptor gene are scattered across the olfactory epithelium.

The number of odor molecules that can be detected and discriminated by a mammalian nose (including a human nose) far exceeds the number of olfactory receptors available (Niimura et al., 2014) because each receptor responds to a variety of odor molecules and each molecule is coded by a combination of receptors, an arrangement called combinatorial receptor coding. Combinatorial receptor coding creates an olfactory epithelial map that discriminates odors in the chemical environment using widespread and overlapping receptor activity patterns (see Figure 2B). The correlated signal input coming from the simultaneous activation of receptors distributed across the olfactory epithelium captures the temporal co-occurrence of olfactory signals, not their spatial co-occurrence (Carey et al., 2009; Cleland & Borthakur, 2020).

There is some evidence that full-blown chemosensory signal processing begins in the olfactory epithelium (in the sensory periphery), even before signals reach the olfactory bulb in the central nervous system (e.g., Gronowitz et al., 2021; Inagaki et al., 2020; Pfister et al., 2020; Xu et al., 2020; Zak et al., 2020). Within the complex mixtures of chemicals that exist in the natural world, receptors are also impacted by the concentration as well as quality of the molecules.¹² More than one molecule might compete for the same receptor, in effect impacting olfactory receptors as agonists (stimulating receptors) or modulators (e.g., suppressing receptors or participating in synergistic effects), in much the same way that neurons inhibit, enhance or modulate signals in one another within the brain (Gronowitz et al., 2021; Inagaki et al., 2020; Pfister et al., 2020; Xu et al., 2020; Zak et al., 2020; for examples, see Kurian et al., 2021 and references therein). These effects are largely non-linear, with antagonistic interactions (i.e., inhibition) often dominating. The ability to perceive a multitude of odors may be the result of complex non-linear dynamics within the olfactory epithelium that produce patterns that are more than the sum of individual olfactant molecules, meaning that olfactory sensory neurons and their receptors may function like a complex system in an ever-changing chemical timescape (e.g., Keller, 2016).

Olfactory sensory neurons that express the same receptor gene all send their unmyelinated axons (via the olfactory nerve) into the same small, spherical structures within the olfactory bulb called glomeruli, producing a chemotopic map of the olfactory epithelium. The architecture of projections effectively removes temporal redundancy in the correlated signals at or close to their source (i.e., the sensory neurons in the olfactory epithelium; Brann & Datta, 2020; Erskine et al., 2019; Giessel & Datta, 2014), compressing the olfactory signals into more efficient summaries. Each glomerulus within the olfactory bulb is selectively tuned to a narrow subset of the olfactants detected in the olfactory epithelium, i.e., it is tuned to the molecules that bind to the primary receptor expressed in the sensory neurons that project into that glomerulus (Burton et al., 2022). Unlike the nearest-neighbor mapping found in the visual system, the glomeruli in the olfactory bulb do not map the spatial relationships among the olfactory sensory neurons in the olfactory epithelium. Instead, olfactory sensory neurons that express the same dominant receptor are scattered around the olfactory epithelium and converge into fewer glomeruli that themselves have a mosaic-like organization within the bulb, so that glomeruli with distinct odorant sensitivities are spatially interspersed (Burton et al., 2022). In effect, this anatomical arrangement generates a temporal map of olfactory receptor identity across the glomeruli of the olfactory bulb. Because many olfactory signals from sensory neurons project into far fewer glomeruli, a high-dimensional sensory array is summarized into one of lower dimensionality, reducing temporal redundancies, integrating signal intensities, (Zhang & Firestein, 2002; e.g., in humans, about 6 million receptors in the human nose converge onto an average of 5500 glomeruli, van Hartevelt & Kringelbach, 2012) and in effect, achieving signal compression. Given that olfactory sensory neurons with the same dominant receptor converge onto the same glomeruli, combinatorial coding of odorant identity is likewise found in the glomeruli, although the coding is considerably sparser due to the compression of signals via incoming axons from epithelial neurons (e.g., Chae et al., 2019; for review, see Cleland & Borthakur, 2020).¹³ This signal compression further enables coding of temporal co-occurrence.¹⁴

Chemotopic temporal coding is maintained in the bulb’s output neurons that project to the rest of the central nervous system. The apical dendrites of these output neurons (called projection neurons, or mitral and tufted cells) extend into the glomeruli, where they synapse with the axon terminals of the olfactory sensory neurons that terminate there. The excitatory signals for an individual projection neuron, therefore, derive from olfactory sensory neurons that express the same type of olfactory receptor. The inputs to the projection neurons are an example of divergent signal processing because there are more mitral and tufted cells than glomeruli (i.e., the 40,000 mitral cells receive synapses from 5500 glomeruli in the human olfactory bulb; van Hartevelt & Kringelbach, 2012). Adding divergent processing of signals after the initial convergence of signals into the glomeruli maintains the specificity of the combinatorial coding while providing functional flexibility (Brezina & Weiss, 1997). These output signals also modulate one another in complex ways, thanks to the dense interconnections among the glomeruli via different types of interneurons (bulbar or juxtaglomerular cells) as well as interneurons that connect the mitral and tufted cells (granular cells).

Projection neurons carry their compressed, temporal chemotopic mapping to the rest of the brain via multiple, parallel routes (Igarashi et al., 2012).¹⁵ Each glomerulus, via its projection neurons, sends signals widely to primary olfactory cortex (or piriform cortex, between the insula and the temporal lobe, anterior and lateral to the amygdala), bypassing the thalamus, to create yet another temporal map of the olfactants detected by the olfactory epithelium.¹⁶ Each cortical neuron receives inputs from many different glomeruli, without any apparent spatial topography (Illig & Haberly, 2003; Rennaker et al., 2007; Stettler & Axel, 2009). The signal ensembles code for odors in an emergent way rather than representing individual molecular features (Gottfried, 2010; Wilson & Sullivan, 2011; Yeshurun & Sobel, 2010). Other parts of the brain receiving olfactory inputs include the hypothalamus, amygdala, hippocampal complex, and parts of the striatum and orbitofrontal cortex (Milardi et al., 2017; van Hartevelt & Kringelbach, 2012). Unlike the visual system, which has a loose hierarchy in the flow of signals from peripheral receptors in the retina to the early parts of visual cortex (e.g., DiCarlo et al., 2012; Felleman & Van Essen, 1991), the olfactory bulb projection neurons send signals broadly in parallel to these many different parts of the brain.¹⁷

Summary

We have discussed two computational architectures within the central nervous system that map incoming signals from the sensory surfaces of the body: one capturing the relationships of signals in space and another capturing the relationships of signals in time. The visual system is optimized for input array-to-target array spatial mapping, such that the spatial array of light frequencies and positions in the world maps to the sensory surface in the retina, whose signals synapse in the superior colliculus or the thalamus, then make their way to early visual cortex, creating multiple maps that preserve the spatial organization of signals (i.e., nearest-neighbors) in the retina. In the olfactory system, the spatial input array of molecular concentrations in the world does not map to the spatial array of target receptors in the olfactory epithelium, which in turn does not correspond to the spatial output array of glomeruli. Instead, the architecture is optimized for coding the temporal relationships between signals. Correspondingly, visual and olfactory systems differ in their manner of signal compression. Visual signals are compressed to some degree in the retina and in the structures where neurons of the optic tract first synapse, but a large portion of their compression occurs along an anatomical gradient within the cerebral cortex, with structures closer to the sensory periphery computing simple signal features and signals increasingly compressed as they flow from early visual areas forward. Olfactory signals, on the other hand, are primarily compressed close to the sensory periphery and broadcast those compressed signals widely.

Afferent Vagus Nerve Anatomy and Its Implications for Signal Processing

In this section, we consider new evidence about the internal molecular and anatomical structure of the nodose fibers within the afferent vagus in light of the signal processing motifs discussed in the prior section. We consider the question of if and how viscerosensory signals are integrated and/or compressed, and whether the processing is spatial, temporal, or both by virtue of the internal structure of vagal afferent neurons and the architecture of the brainstem nuclei that process those vagal inputs. We then turn to new evidence on the structural arrangement of vagal afferent fibers into fascicles, or fiber bundles enclosed in a sheath of connective tissue. We consider evidence for several types of interactions that can occur across neighboring nerve fibers running within a single vagal fascicle (including ephaptic coupling, cross-depolarization, etc.) to suggest how the fascicular organization of the vagus might afford signal integration at points along its length, from rostral to caudal. Ultimately, we propose new hypotheses about the computational capacity of the vagus nerve in terms of its potential for signal integration even before signals arrive at the brainstem.

Molecular and Structural Gradients in the Afferent Vagus Nerve and Brainstem Targets

For many years it was assumed that the nodose ganglion neurons, which originate the ascending fibers of the vagus, and the NTS, the main target of vagal afferent signals, both contained viscerotopic maps of the internal milieu consisting of fine-grained, high-dimensional signaling of sensory events in specific locations of specific organs, similar to the retinotopic maps in the earliest parts of the visual processing stream -- that is, a nearest-neighbor array-to-array map of the body corresponding to the tissues where the sensory signals originate (for the nodose ganglia; Browning & Mendelowitz, 2003; Zhuo, 1997; for the NTS; Altschuler et al., 1989; Dennison et al., 1981; Loewy, 1990). Strict viscerotopic organization in the nodose ganglion neurons and in the NTS (i.e., signals segregated by the specific location of each end-organ) has been called into question, however, by evidence arising from new methodological approaches that instead suggest that the afferent vagus is capable of both spatial and temporal co-occurrence mapping, without precise viscerotopy (see Box 1 for additional details).

Box 1. A possible rostro-caudal specificity gradient in brainstem nuclei.

The NTS has at least 10 subnuclei that are often divided into rostral, intermediate, and caudal divisions (e.g., McRitchie & Törk, 1993). Rostral NTS subnuclei receive afferent vagal fibers from more rostrally located tissues, which are relatively segregated by specific organ (tongue/epiglottis and pharynx; e.g., Altschuler et al., 1989; Kalia & Mesulam, 1980b). Intermediate and caudal NTS subnuclei receive intermingled afferent vagal fibers from more caudally located tissues, with less specificity in their spatial array (e.g., heart, lungs, GI tract; for reviews and further discussion, see Andresen & Paton, 2011; Cutsforth-Gregory & Benarroch, 2017; Neuhuber & Berthoud, 2022; Paton, 1999). Many of these studies were conducted with horseradish peroxidase as a labeling method. In contrast, a more recent viral mapping study traced connections along vagal afferent fibers and found virtually no spatial specificity in the NTS (Bassi et al., 2022). Yet another study, which used two-photon calcium imaging replicated the rostro-caudal specificity gradient when examining NTS neuron responsivity to organ stimulation (Ran et al., 2022, also described in the main text). It is possible that the inconsistent results are due to differences in methodology. $$PARABREAKHERE$$A rostro-caudal specificity gradient has also been observed, albeit with some caveats, in the efferent fibers arising from nAmb (e.g., described in Neuhuber & Berthoud, 2022). The nAmb contains a dorsal subdivision and a ventrolateral subdivision (as discussed in the section entitled “A Brief Overview of the Vagus and its Brainstem Targets”). The dorsal subdivision gives rise to efferent fibers that innervate the most rostrally located organs (e.g., pharynx, larynx, upper esophagus) with relatively high spatial segregation (e.g., Bieger & Hopkins, 1987). The ventrolateral subdivision gives rise to visceromotor efferent fibers that innervate the relatively more caudally located heart and lungs (e.g., Hopkins & Armour, 1998). These efferent fibers are relatively more intermingled with one another than the efferent fibers arising dorsally but are nonetheless themselves arranged along a loose rostro-caudal gradient with pulmonary neurons clustered rostrally and cardiac neurons arranged more caudally (Hopkins & Armour, 1998; Hsieh et al., 1998; McAllen & Spyer, 1978). Other evidence suggests that the cardiac and pulmonary vagal efferents arising from the nAmb may be further organized by cardiorespiratory function (Gatti et al., 1996; Massari et al., 1995; Veerakumar et al., 2022). $$PARABREAKHERE$$The DMV sends efferent visceromotor projections to organs along the rostro-caudal axis of the body from esophagus to colon (e.g., Huang et al., 1993; Karim et al., 1984; McLean & Hopkins, 1985a; Mussa & Verberne, 2013), but no conclusions can be drawn about organizing gradients, given the extensive species-related differences in the architecture of the visceromotor efferents to the more caudal organs in the body -- the abdominal viscera, including the GI tract and pancreas (Huang et al., 1993). For many years, researchers segmented the DMV in rodents into a series of columnar subnuclei, with each column corresponding to each of the five subdiaphragmatic vagal branches (anterior gastric, posterior gastric, hepatic, celiac, and accessory celiac; e.g., Fox & Powley, 1985; Norgren & Smith, 1988). However, these columnar subnuclei are not viscerotopically arranged, as there is not a one-to-one mapping between subdiaphragmatic vagal branches and visceral organ (e.g., Berthoud et al., 1991). Rather, most studies have found that abdominal vagal projections, particularly those to the stomach/gut, are arrayed across the entire length of the DMV rather than being clustered in close spatial proximity (Kalia & Mesulam, 1980a; Karim et al., 1981; Leslie et al., 1982; McLean & Hopkins, 1985a). Furthermore, they may be organized functionally (e.g., Tao et al., 2021) in a manner similar to the ethological maps described by Graziano (Graziano, 2016) for the motor cortex. $$PARABREAKHERE$$As we have previously noted, the DMV also receives some direct vagal afferents, in addition to sending vagal efferents to the viscera (e.g., Kim et al., 2020). There is some rostro-caudal patterning in vagal afferents to the DMV of the rat (Kalia & Sullivan, 1982) and cat (Kalia & Mesulam, 1980), such that afferent vagal terminals from the nodose ganglion were segregated to one side of the DMV at the rostral end (ipsilaterally; only the right nodose ganglion was injected with the tracer; Kalia & Sullivan, 1982), while vagal afferents were more bilaterally distributed, and with greater density, throughout the DMV at the caudal end (Kalia & Sullivan, 1982; Neuhuber & Sandoz, 1986).

Using a variety of methods and analysis techniques, Zhao et al. (2022) examined the architectural arrangement of the more than 14,000 nodose ganglia neurons in mice. Seven visceral organs were injected with specially designed viruses, each with their own unique genetic projection barcodes, to retrogradely trace neuronal connections. Bioinformatic methods clustered subpopulations of vagal sensory nodose neurons tagged with Phox2b+, a genetic marker specific to (and necessary for) embryological development of viscerosensory-visceromotor reflex arcs (Dauger et al., 2003). Differentially expressed gene (DEG) transcription factors were used in both wild-type and Cre-line mice to label specific subpopulations of nodose neurons. These combined methods revealed that ascending nodose ganglion fibers from the viscera were organized according to two genetically-coded molecular gradients.

The first gradient coded the locations of nodose sensory endings along the body’s rostro-caudal axis. This gradient did not produce a strong nearest-neighbor organ-specific viscerotopic map, however. Nodose neurons were distributed very sparsely in a so-called “salt-and-pepper distribution” with minimal to no organ-specific clustering (and see Bassi et al., 2022, which showed no discernable organization of any sort within the nodose ganglia). A very roughly mapping of organ location might exist, with the sensory conditions of the more rostral organs like the heart and lungs somewhat spatially separated from the sensory conditions of more caudal organs like the colon, duodenum, and pancreas (see also Jayaprakash et al., 2023). Vagal afferent neurons innervating rostrally located organs (like heart and lungs) also maintain greater segregation from one another than do the neurons innervating the more distal organs, implying that the sensory signals informing on the metabolic and other sensory conditions of the heart, lungs and airways, pharynx, larynx and upper esophagus may be relatively more distinct (e.g., baroreceptor afferent signaling, see Andresen & Peters, 2008; or afferents in the cardiac nerve; see Jayaprakash et al., 2022), whereas the sensory signals informing on the metabolic conditions of the lower esophagus, stomach, intestines, and liver may be compressed into summaries (Jayaprakash, et al., 2022). Moreover, more caudally arising signals are more likely to influence rostral ones than vice versa simply due to the arrangement of fibers. The second “tissue layer” gradient, orthogonal to the first, coded the locations of nodose sensory endings across a gradient from the inner (lumenal or mucosal) organ layer to the outer (muscular) organ layer. Based on their findings, Zhao et al. (2022) concluded that afferent vagal fibers create a loose spatial map of sensory signals from the viscera but does not afford sufficient spatial segregation to be considered a viscerotopic map.

Zhao et al. (2022) interpreted their findings in terms of the organ and tissue locations they observed in the adult animals they studied, but we hypothesize that the gradients they observed might be best understood in terms of their developmental origin as they are laid down during embryonic development. Specifically, both of their observed molecular gradients bear a strong resemblance to the highly conserved molecular polarities that are organized from HOX genes during vertebrate embryonic development (for a discussion see Stiles, 2008 and references therein). These gradients are established when the embryo’s organs are not yet in place. By implication, the ascending fibers of the vagus nerve may be organized by the embryonic birthday of the nodose ganglion cells from which those sensory fibers arise (i.e., chronotopy) in a manner similar to what has been observed in the optic tract, rather than by the organ-related signals they carry. These gradients within the ascending vagal fibers appear to be insufficient to produce a strict nearest-neighbor array-to-array end-organ (viscerotopic) map in their termination sites just as the chronotopic organization of the optic tract is insufficient for creating retinotopic maps in the superior colliculus and thalamic nuclei. Those structures develop retinotopic maps by other means (i.e., gradient mapping), but this did not appear to be the case in the NTS. The absence of a clear viscerotopic map in the NTS was empirically confirmed by Zhao et al. (2022; also see Andresen & Paton, 2011; Bassi et al., 2022; Cutsforth-Gregory & Benarroch, 2017). Another study (Ran et al., 2022, discussed in more detail below) using in vivo calcium imaging in mice, did observe something like a rostro-caudal gradient in the NTS, however, see Box 1. One opportunity for future research might involve assessing the hypothesis that embryological chronotopy is a major driving force for the spatial location of visceral organs, the spatial organization of the nodose ganglia and their afferent vagal fibers, and the termination of those fibers in their brainstem targets like the NTS.

In addition to the spatial gradients, a third genetic gradient using yet another method was observed in the organization of nodose axons as they synapsed on their brainstem targets (Zhao et al., 2022). Using specific types of Cre mice bred for specific conditional knockouts of specific types of tissue, highly specific anterograde tracers were injected into regions of visceral organs that contained nodose ganglion sensory nerve endings, thereby tracking vagal afferent pathways for 11 different nodose neuron subtypes to their termination sites in the NTS, AP and the DMV. In addition, nerve endings in visceral organs were stimulated via different modalities (e.g., by stretch (inflation) of the lungs or stomach, or by chemical stimulation such as delivering nutrients, salt, or water to the lumen of the intestines) during calcium imaging.¹⁸ These investigations revealed that the genetically distinct nodose cell subtypes coded for modality (chemo-, mechano- or osmoreception) across different visceral organs in a many-to-many pattern reminiscent of the combinatorial receptor coding observed in the olfactory epithelium. Each stimulus modality (e.g., mechanostimulation) activated a pattern across multiple nodose cell subtypes, and each subtype was stimulated by more than one modality. For example, pulmonary stretch produced a fast but sustained signal in one cell subtype but a transient response in another cell subtype. On the other hand, the same cell subtype that showed a transient response during pulmonary stretch was also responsive to stretch stimuli in other organs as well as to chemical stimulation in the intestines. Interestingly, nearly every one of the 11 nodose neuron subtypes projected to multiple subnuclei of the NTS, and most subnuclei of the NTS, as well as other nuclei in the brainstem (i.e., DMV, AP) received convergent inputs from multiple different nodose neuron subtypes, recapitulating a combinatorial pattern that results in temporal co-occurrence maps for different types of sensory change across organs and tissues in the body. Temporally correlated signals of any specific modality (chemical, mechanical, etc.) across multiple organs are processed as a pattern of activation across nodose cell subtypes, with each pattern reaching partially overlapping targets in the brainstem. Again, this mapping represents different sensory modalities rather than segregated, spatially organized viscerotopic maps in the brainstem (for additional viral mapping evidence consistent with temporal mapping, see Bassi et al., 2022; Neuhuber & Berthoud, 2022, p. 4).

The compression of multiple modalities of viscerosensory signals from certain organs appears to occur in the NTS (Ran et al., 2022, mentioned above). In vivo calcium imaging in mice enabled real-time analysis of thousands of NTS responses to ascending vagal inputs. NTS neurons were observed to be more broadly tuned to different signal modalities coming from the same organ (e.g., the larynx or the duodenum) than were the nodose neurons that sent the ascending signals, as indicated in Zhao et al. (2022), indicating a convergence (i.e., compression) of sensory signals from nodose neurons to NTS neurons (see Box 1 for additional details).

Taken together, these findings suggest the hypothesis that afferent vagal fibers organize sensory signals from the viscera in space (possibly according to embryonic gradients rather than specific viscerotopy) and time. The ascending viscerosensory signals have a multidimensional coding architecture that enables the “massively parallel presentation of interoceptive signals in an efficient manner” (Zhao et al., 2022, p. 6) but without evidence of specific maps of specific visceral organs. By combining across signals from tissues according to rostro-caudal location, inner layer-outer layer position, and modality of sensory stimulation, vagus neurons relay signals to brainstem targets in a way that indicates particular types of sensory signals at one time within larger regions of the internal milieu (e.g., the abdomen) rather than pinpointing changes in specific locations within specific organs. Correspondingly, maps within the NTS suggest the possibility of similar gradients for space and time, with minimal evidence for any specific nearest-neighbor array-to-array mapping but some for a possible rostro-caudal gradient. Rostro-caudal gradients are a possible organizational theme present in the nodose neurons and in the brainstem, across both afferent and efferent brainstem nuclei: NTS, AP, nAmb, and potentially even the DMV (see Box 1). We now turn to the fascicular organization of vagal fibers throughout the nerve trunk to elaborate on these hypotheses further, and to understand whether a similar rostro-caudal gradient exists within the vagus nerve itself.

A “Split and Merge” Fascicular Organization within the Afferent Vagus Nerve

The afferent fibers of the vagus nerve that ascend through the nodose ganglia are arranged into fascicles, each of which is surrounded by its own connective tissue sheath called a perineurium. In a recent study in swine (which have similar vagal anatomy to humans; Pelot et al., 2020; Settell et al., 2020), afferent vagal fibers tended to segregate into organ-specific fascicles as they leave the tissues of the body, with progressively more merging of fibers from different organs into the same fascicles as signals ascended rostrally (Jayaprakash et al., 2023). A recent study in human vagus observed considerable merging and splitting along the mid-cervical vagus, beginning rostrally at the nodose and terminating caudally at the level of the clavicle — vagal fibers running within one fascicle split off from that fascicle and merged with vagal fibers in another fascicle, approximately every ~560 μm, along the entire length of the cervical vagus (Upadhye et al., 2022). This “split and merge” architecture occurred regardless of fascicle diameter, or the diameter of fiber types carried within, suggesting a remarkable degree of fiber re-organization within fascicles as viscerosensory signals ascend to the nodose. Because Upadhye et al. (2022) examined only a 5 cm section from the middle of the cervical vagus, it will be important for future studies to examine whether the “split and merge” pattern holds across a longer swath of the vagus nerve (e.g., subdiaphragmatic vagus). The fact that vagal fibers carrying ascending viscerosensory signals assume new spatial configurations relative to one another within fascicles as they merge and split, at least in the cervical and thoracic vagus suggests possible signal integration as the axons arise toward the nodose ganglion. This possibility is made even more likely by the observation that axons, including those in peripheral sensory nerve bundles, can dynamically modify the signals they carry in a variety of ways, a point to which we now turn.

Possible Ephaptic Coupling within Fascicles of the Vagus

The high frequency of fascicular splits and merges along the vagus may provide opportunities for axonal ‘crosstalk’ or ephaptic signaling, whereby spatially contiguous axon fibers modulate one another’s activity (Katz & Schmitt, 1940; Ramon & Moore, 1978), an idea first suggested by Damasio and Carvalho (Carvalho & Damasio, 2021; Damasio & Carvalho, 2013). In brief, neuronal action potentials generate extracellular local field potentials that can influence the excitability of nearby axons that are not in direct contact by synapses or gap junctions (e.g., Binczak et al., 2001; Clark & Plonsey, 1970), and these effects are thought to be especially likely between nearby unmyelinated fibers (e.g., Hartline, 2008). Importantly, a large proportion of vagal axons are unmyelinated (85% of fibers in the cervical vagus and nearly all (99%) in the abdominal vagus; Havton et al., 2021), providing considerable possibilities for ephaptic effects.

The hypothesis of ephaptic coupling in the vagus remains, to our knowledge, untested, but ephaptic interactions have been demonstrated in the mammalian olfactory nerve, which is composed of unmyelinated, densely packed axons, where all axons within a fascicle influenced one another, as did axons in neighboring fascicles (Bokil et al., 2001). Ephaptic interactions have also been observed between axons in the myelinated dorsal root and sciatic nerve fibers in intact rat nerve tissues (i.e., in the absence of any injury that disrupts myelin; Bolzoni & Jankowska, 2019). The majority of evidence for ephaptic signaling, however, comes from the study of neurons in cortical areas in rodents (e.g., Anastassiou et al., 2011; Blot & Barbour, 2014; Han et al., 2018, 2020; Qiu et al., 2015; Taylor & Dudek, 1982; reviewed in Anastassiou & Koch, 2015), studies of unmyelinated nerves of various invertebrates (Arvanitaki, 1942; Katz & Schmitt, 1940; Ramon & Moore, 1978), and mathematical models showing that adjacent myelinated axons within the same fiber bundle produce local phase-locking between action potentials (Arvanitaki, 1942; Binczak et al., 2001; Capllonch-Juan et al., 2017; Capllonch-Juan & Sepulveda, 2020; Marrazzi & de No, 1944; Schmidt & Knösche, 2022) or ephaptic coupling between unmyelinated fibers (e.g., Barr & Plonsey, 1992; Clark & Plonsey, 1970).

Some models have shown that greater fiber density within a fascicle increases the strength of ephaptic coupling, although greater heterogeneity of axon diameters within a fascicle can reduce ephaptic interactions (Capllonch-Juan & Sepulveda, 2020; Schmidt & Knösche, 2022). Such findings are potentially relevant for testing hypotheses about ephaptic coupling in the ascending vagus given the extensive variability in fascicular size and composition observed in both light- and electron-microscopic assessments of well-preserved human cervical and abdominal vagus samples (from 27 middle-aged donors undergoing organ harvesting at brain death; Havton et al., 2021). Specifically, many vagal fascicles contain a heterogeneous mix of axon diameters and myelination, with unmyelinated afferents distributed throughout, whereas other fascicles contain axons segregated by diameter and myelination (Jayaprakash et al., 2023), suggesting the possibility that ephaptic coupling effects might be uneven or patterned along the length of the ascending vagus. Greater ephaptic coupling might occur within fascicles that contain a more homogeneous distribution of fiber diameters, especially when many of those fibers are unmyelinated. In addition, ephaptic signaling is plausible even between nodose fibers that make up the more caudally-located ascending axons (from organ to ganglion), since ephaptic signaling also has been documented to occur between dendrites as well as axons (Han et al., 2018; Yip & Heiman, 2018). More generally, ephaptic coupling may contribute to synchronizing the independently generated cardiorespiratory rhythms of the heart and lungs, peristalsis in the gastrointenstinal (GI) tract, longer-scale circadian and ultradian rhythms, and so on (in addition to the usual sources of synchronization, such as descending visceromotor, peptidergic and hormonal influences).

Together, this evidence suggests that ephaptic coupling is a viable hypothesis for signal processing within the vagus nerve. If ephaphtic coupling is indeed a source of non-synaptic communication in vagal fascicles, then axons carrying signals from different tissues, organs, or stimulus modalities may effectively modulate one another as they ascend the vagal trunk. Furthermore, it is unclear whether ephaptic coupling could mitigate or enhance the rostro-caudal gradient of signals that we have previously described. Going forward, an imporant piece of any empirical work will be considering how ephaptic coupling contributes to these gradients.

Possible Afferent and Efferent Signal Interaction within the Fascicles of the Vagus

As we noted in the first section of this paper, the vagus is often referred to as a “mixed” nerve because it carries both afferent (viscerosensory) fibers from the periphery to the brain (80%) and efferent (parasympathetic motor) fibers with signals moving the other direction (20%). In swine (e.g., Jayaprakash et al., 2023), ascending sensory fibers and descending motor fibers can be found within the same fascicles with varying degrees of myelination, suggesting opportunities for afferent signals to be influenced by efferent signals even before they reach the nodose ganglion or their brainstem targets, and correspondingly, for afferent signals to influence efferent signals before they reach their target tissues.¹⁹ In other words, there may be sensorimotor signal integration, not just in the brain, but also along portions of the vagus nerve. Such a possibility is consistent with evidence of signal integration in other parts of the peripheral nervous system, such as the enteric nervous system (Fung & Vanden Berghe, 2020), where interactions between the vagal mechanosensory nerve endings in the GI tract and the myenteric ganglia (e.g., Neuhuber, 1987) suggest peripheral integration of vagal afferent signals even before those signals ascend in the vagus. If ascending signals have been modulated within the fascicles of the vagus by descending signals, then they may not, strictly speaking, be exclusively sensory by the time they reach the brainstem. Any signal integration in the vagus nerve, if it exists, would contribute to the extensive signal integration already known to occur in the brainstem nuclei (see Box 2; e.g., Ran et al., 2022), midbrain (e.g., Llewellyn-Smith & Verberne, 2011; Saper, 2002), and forebrain.

Box 2. Afferent/Efferent Integration in the Brainstem.

The DMV both receives direct vagal afferent fibers and sends visceromotor efferent fibers via the vagus (e.g., Kim et al., 2020; Neuhuber & Sandoz, 1986), making this brainstem nucleus a site for sensorimotor integration on its own. AP is also a site for sensorimotor integration. It receives a direct viscerosensory input from vagal neurons and expresses receptors for multiple peptides and hormones in the periphery (that arrive via fenestrated capillaries, which line the cerebral ventricles; see Cottrell & Ferguson, 2004; Price et al., 2008); these chemicals are considered motor signals. More broadly, the architectural connections among the DMV, AP, NTS, and nAMB facilitate integration between interoceptive and visceromotor signals from the vagus nerve as well as chemical motor signals circulating in the blood. The NTS, which receives direct vagal afferent signals, also contains DMV dendrites (Miselis & Shapiro, 1983; Shapiro & Miselis, 1985a) allowing for transmission of interoceptive signals to the DMV (in addition to the direct viscerosensory signals it receives from the vagus). The NTS also receives axons from AP, which as we just noted, carry interoceptive signals modulated by chemical motor signals. By virtue of these AP connections, the NTS can be considered a site of sensorimotor integration. In addition, AP signals influence the DMV, not directly (as it does not robustly connect to the DMV directly), but via its axons that terminate near or at the location of DMV dendrites in the NTS (e.g., Cunningham et al., 1994; Shapiro & Miselis, 1985a, 1985b).

A familiar rostro-caudal gradient characterizes fiber mixing within fasciculi, such that ascending and descending fibers are relatively more segregated in rostral fascicles but converge more extensively within caudal fascicles (Jayaprakash et al., 2023). Correspondingly, the potential for efferent modulation of vagal afferent signaling may also vary along this gradient, since there is less mixing of afferent and efferent fibers within fascicles of the more rostral vagus targeting the heart, lungs, and airways, but extensive mixing of afferent and efferent vagal fibers in more caudal vagus targeting the liver, stomach, and intestines. This pattern parallels the viscerosensory and sensorimotor convergence observed in brainstem nuclei, including the NTS, AP and even the DMV and nAmb (see Boxes 1 and 2). Moreover, the organizational arrangement of fascicles in the vagus also would suggest that intermixing of sensorimotor signals arising within the caudal vagus would necessarily influence more rostral signals.

Possible Asynaptic Chemical Modulation of Signaling by Glial Cells Within the Nodose Ganglia

Yet another possible opportunity for signal processing within the vagus nerve comes from evidence of asynaptic chemical modulation of the nodose ganglion cells by specialized glia called satellite glial cells. Satellite glial cells support the metabolic function and transmission of sensory neurons in peripheral ganglia. A small number of satellite glial cells surround each cell body and as well as the neuron’s axons, and neuron-specific communities of satellite glial cells communicate very rapidly with one another, quickly altering neuronal signal transmission (Hanani & Spray, 2020). Most of what is known about how satellite glial cells impact sensory neurons has come from studies of their role in chronic pain after different kinds of peripheral nerve injury (e.g., Dublin & Hanani, 2007; Feldman-Goriachnik & Hanani, 2021). In mouse models of systemic inflammatory pain, for example, satellite glial cells in the nodose ganglia upregulate the expression of a gene that creates membrane channels, thereby allowing for the release of ATP and a resulting increased sensitivity to ATP, which contributes to increased intercellular communication and hyperexcitability of neurons in the periphery (Feldman-Goriachnik et al., 2015).

Communities of neuron-specific satellite glial cells also surround each nodose ganglion cell (Hanani, 2010; Hanani & Spray, 2020). Cell bodies in the nodose ganglion appear to influence one another via non-synaptic means (using neurotransmitter diffusion to facilitate asynaptic cross-depolarization, which refers to the increased spike probability in the soma of some cells when the axons of neighboring cells fire repeatedly). This suggests the possibility of yet another source of signal processing, in this case within the nodose itself (Oh & Weinreich, 2002). If the satellite glial cells of the nodose ganglia function similarly to satellite glia around other peripheral sensory ganglia, then these cells also could broaden the possibilities for signal integration by supporting and potentially even augmenting non-synaptic chemical communication occurring between cell bodies in the nodose,²⁰ although the extent to which this sensory integration would target spatial, temporal, or both types of coding remains to be seen.

Summary

The signals that flow up the individual fibers of nodose ganglion neurons result in two spatial gradients, one rostral to caudal (with respect to the spatial positioning of organs along the long axis of the entire body) and one from inner to outer layer of an organ, plus one sensory modality gradient that maps the temporal co-occurrences in the same modality of signals across organs and tissues. Interestingly and perhaps importantly, some of the signal processing within the vagus is likely occurring within individual nodose neurons. Based on the totality of this evidence, we hypothesized that viscerosensory signals are processed and, in some cases, compressed both spatially and temporally by virtue of the internal structure of vagal afferent neurons and the architecture of the brainstem nuclei that receive vagal afferents. Multimodal compression across signals and tissues appears to occur in the NTS, and not in the nodose ganglion neurons themselves (see Box 1). One hypothesis for future research is the idea that this two-dimensional (rostro-caudal and inner-outer layer) spatial organization of viscerosensory signals may arise from chronotopic polarities that set up gradient-matching -- genetically-expressed guidance molecules that allow peripheral processes of the nodose to find their tissue targets as the nervous system is assembling itself during embryological development well before visceral organs have even formed. The degree of rostro-caudal organization of vagal afferents in the NTS, perhaps by gradient matching, would also be interesting and important to explore.

Furthermore, we hypothesized that some degree of sensorimotor integration, another type of signal processing, may be possible within the vagus nerve itself, blurring the line between afferent and efferent signals that are propagated along the nerve fibers to and from the sensory organs and brainstem targets. Various studies of asynaptic interactions (i.e., ephaptic signaling, cross-depolarization, and satellite glial hyperexcitation in nodose cell bodies) across adjacent peripheral nerve fibers suggest that this hypothesis is at least physiologically plausible, although the extent to which these interactions occur in the vagus nerve itself is a matter for future empirical study. In the next and final section, we consider several biological and psychological implications that would arise if these hypotheses were supported by future work on the computational capacity of the vagus nerve. We also address current barriers needing resolution and measurement tools required to test these hypotheses, which are aimed at better understanding signal processing in the vagus (see Box 3).

Box 3. Methodological Considerations for Measuring the Vagus Nerve.

Vagal activity following metabolically significant events has been recorded using denoised compound action potentials, which represent the cumulative firing activity of multiple vagal fibers (Chang et al., 2020; Zanos, 2019). These measures might provide a future avenue for empirically testing hypotheses about the existence of visceromotor motifs and their relation to temporal coding. For example, compound action potentials recorded from the vagus nerve reflect events that were observed to correspond to changes in blood glucose levels (Zanos, 2019) and cytokine levels (Steinberg et al., 2016; Zanos et al., 2018). Blood glucose level is a highly relevant metabolic signal and likely to be represented across multiple organs of the body, suggesting the possibility that compound action potentials communicate multi-organ, temporally-concordant changes in blood glucose levels. If so, this may reflect the vagus nerve compressing temporal redundancies in the afferent signals that are important for the sensing and control of blood glucose. However, compound action potentials are notoriously prone to sources of electrical noise and measuring them in an animal requires anesthesia, which can confound the recording and interpretation of waveforms (e.g., Silverman et al., 2018). Recent work also has demonstrated the feasibility of single-unit recordings in the vagus nerve of awake humans (guided by microneurography) from both afferent and efferent fibers (Ottaviani et al., 2020). Once the technical demands of these newer methods have been reduced and some limitations addressed (building off existing work by Silverman et al., 2018 and Zanos, 2019), many of the hypotheses proposed here potentially could be tested, including whether the afferent and efferent signals in the vagus are modulating one another within fascicles.

Implications and Future Directions

Signal Processing Gradients in the Ascending Vagus: Implications for Allostasis

Before we dive into the implications of the material we’ve discussed so far, it is necessary to provide a bit of background about signal processing throughout the brain. The framework that we outline below belongs to a larger, mathematically formalized, neuroscience-inspired account referred to as predictive processing (e.g., Clark, 2013; Keller & Mrsic-Flogel, 2018; Rao & Ballard, 1999). We start with a brief overview of predictive control of skeletomotor movements to motivate our discussion of allostasis, which is the predictive control of the visceromotor movements of the body.

For many years, it was assumed that the brain created a somatotopic array of muscles and appendages as elemental building blocks for skeletomotor movements. Primary motor cortex, for example, was thought to contain such a map, with feet represented at the top of the precentral sulcus and the tongue at the bottom. More recent discoveries vigorously challenge this idea, however, and instead suggest that primary motor cortex, as well as premotor cortices, appear to be mapping movement ensembles that are fundamentally rooted in the spatial and temporal statistics of ethologically meaningful behavior (Coffman et al., 2011; Gordon et al., 2023; Graziano, 2016, 2023 and references therein; Levinthal & Strick, 2020). The sensory features of the environment are integrated as part of these maps (e.g., Rizzolatti et al., 2013) in a way that accounts for the current sensory state of the animal’s body (Dum et al., 2016; Herzfeld & Shadmehr, 2014; Levinthal & Strick, 2012, 2020).

The mechanical requirements for executing a specific set of muscle movements are never perfectly predictable, making stored patterns of precise neuromuscular activity triggered from fixed, preprogrammed circuits ill-suited to the task of skeletomotor control. It is now increasingly accepted that the specifics of skeletomotor movements are constructed, as needed, from these maps of ethologically meaningful behavior in a complex combinatorial fashion (e.g., Flash & Bizzi, 2016; Mussa-Ivaldi & Bizzi, 2000; for a discussion and additional references, see Barrett & Finlay, 2018). Movements are assembled (probabilistically and inferentially) in a signal processing hierarchy that spans the cerebral cortex to the ventral horn of the spinal cord.

Pre-motor and primary motor cortices assemble an ensemble of signals for ethologically meaningful behavior as abstract features; these are typically referred to as a “plan” or a “goal”, to indicate their abstraction away from motor particulars (Graziano, 2016; Rizzolatti et al., 2013). However, we have described them as an action concept – integrated, compressed multimodal signals that plan action in a particular context (Barrett, 2017; Barrett & Finlay, 2018). These plans are low-dimensional summaries of sensorimotor signals (see section “Signal Processing Motifs in the Nervous System”) corresponding to the statistical relationships in spatial and temporal patterns of activity that have been compressed across multiple modalities. These summaries correspond to motor control signals or motor reference signals in the systems control literature. To be implemented as actual movements, these summaries must be decompressed (probabilistically and inferentially) to recruit ever more specific neural assemblies as the signals cascade through the midbrain and brainstem. Ultimately, they are particularized at greater spatial and temporal specificity in modules of the ventral horn of the spinal cord that implement the signals for a specific pattern of muscle fiber activity and joint angles (Flash & Bizzi, 2016; Mussa-Ivaldi & Bizzi, 2000). This cascade of assemblies at ever-greater particularization proceeds in a generative way that is more flexible and functional than what could be accomplished with pre-set motor programs. Flexibility and robustness derive not only from which action is executed, but also how any given action is achieved via a specific pattern of muscle contractions and joint angles in the spinal cord (for further discussion and references, see Barrett & Finlay, 2018). Even the ‘reflexes’ that are present in spinal cord circuits and pattern generators in the brainstem are flexibly modulated in a context-dependent way via this architecture (e.g., Bhattacharyya et al., 2012; Pearson & Gordon, 2014).

Action concepts are thought to control motor movements by prediction rather than reaction, embodying the inferred causal relationships between potential future actions and their expected sensory consequences (e.g., Lochmann & Deneve, 2011; Shadmehr et al., 2010; Wolpert et al., 2013).²¹ They are the brain’s best guess, generalized from past experience, as to which actions will be most functional in a given context and how those actions can be most efficiently implemented in that context. The cascade of signals that are decompressed from cortex to spinal cord are effectively inferences that predict forward in time and space, to anticipate how the motor system’s state will change as a function of the motor commands. These cascading signals also infer the expected sensory consequences of those motor movements. In this way of understanding skeletomotor control, incoming signals from the sensory surfaces of the body do not trigger motor responses anew, but rather are modulatory signals that serve upcoming motor control. They either confirm the sensory prediction signals (and the associated skeletomotor plans from which those sensory predictions arose) or correct the prediction signals. It is hypothesized that incoming signals from the sensory periphery are compared to prediction signals at every synapse (Deneve, 2008) along the various processing gradients within the brain.²² Within each gradient, the ascending signals closer to the sensory periphery are higher in dimensionality with a specificity that is closer to the signals from the receptor cells in those sensory surfaces; these signals become increasingly compressed into summaries of lower dimensionality as they flow to the limbic portions of the cerebral cortex (either as they ascend from the brainstem or flow from primary sensory areas along the main architectural gradient of the cortex; see section “Signal Processing Motifs in the Nervous System”).

There are many reasons to hypothesize that visceromotor control works in a similarly complex, combinatorial, and predictive manner (including evidence from evolution and embryological development, as well as overlapping architecture throughout the brain discussed below). In the visceromotor domain, this predictive control is called allostasis -- anticipating the body’s metabolic needs and preparing to meet those needs before they arise (e.g., Sterling, 2012). Allostasis is one of the brain’s core functions (for discussion of converging evidence, references with additional details and complementary views; see the following and references therein: Barrett, 2017; Barrett & Simmons, 2015; Chanes & Barrett, 2016; Hutchinson & Barrett, 2019; Katsumi et al., 2022, 2023; Kleckner et al., 2017; Schulkin & Sterling, 2019; Sterling & Laughlin, 2015). By implication, the brain appears to be mapping allostatic ensembles that are fundamentally rooted in the spatial and temporal statistics of ethologically meaningful behavior, as occurs in the skeletomotor domain, rather than attempting to separately control individual visceral organs or tissues. Consistent with this hypothesis, primary motor cortex and premotor cortex, which implement functional skeletomotor ensembles also implement maps of the viscera (e.g., Dum et al., 2016; Levinthal & Strick, 2012, 2020) and some of the functional movement ensembles described in primary motor cortex involve visceromotor components (see Graziano, 2016). In addition, a key premotor cortical area, the aMCC, is a primary site for visceromotor control via its connection to sgACC and pACC via the cingulum bundle (e.g., Vogt, 2016; Vogt et al., 2003). Such findings suggest that the cingulate cortex is involved in coordinating visceromotor movements and the skeletomotor movements that they support.

In this processing context, the gradients in the nodose ganglion and its brainstem targets seem to underscore the idea that allostasis may be coordinating tissue changes across the body prior to further processing and integration beyond the brainstem. Such coordination is consistent with Peter Sterling’s notion of allostasis as “stability through change,” whereby a physiological system need not achieve stability by regulating back to specific set points or reducing metabolic costs per se, but rather by coordinating its various parts to maximize metabolic efficiency (e.g., Sterling, 2004; 2012). A corresponding hypothesis is that poor coordination in the face of changing energetic circumstances may be an indicator of dysfunction that could be associated with physical illness or mood symptoms.

Even at the level of the brainstem nuclei where prediction signals descending from the forebrain and midbrain meet sensory signals ascending from the nodose that inform on the metabolic conditions of the body’s internal milieu, there seems be relatively little spatial specificity like that achieved with nearest-neighbor array-to-array mapping, as in vision and audition. One possibility is that high-dimensional viscerotopic specificity, if required for allostasis, can be computed on the fly. Alternatively, the rostro-caudal gradient in spatial specificity in the nodose ganglia (e.g., Jayaprakash et al., 2023; Williams et al., 2016; Zhao et al., 2022) suggests that such specificity may not be required to accomplish all aspects of allostasis – knowing when and approximately where a sensory change is occurring may be sufficient for some regions of the periphery (e.g., the GI tract), whereas other organs may evidence greater specificity (e.g., heart and lungs).

The hypothesis that visceromotor control is, like skeletomotor control, combinatorial and predictive is also consistent with empirical evidence that signals from visceral organs array into reliable and repeatable physiological “motifs” (Feng & Narayanan, 2019; Hoemann et al., 2020), just as there are signals from voluntary (skeletomotor) movements and behavioral “motifs” (Brown et al., 2013; Datta et al., 2019; Grover & Tavare, 2009; Luxem et al., 2022; Robie et al., 2017). A motif is a repeating ensemble of observed signals that can be statistically summarized by a pattern of lower dimensionality, such as the high-dimensional multitude of muscular movements and vasculature control required for a person to raise their arms above their head. We hypothesize that different physiological or behavioral motifs begin in the brain as a compressed, multimodal action concept (where here, we mean ‘action’ in both the skeletomotor and visceromotor senses). Our lab has recently documented individual-specific, recurrent patterns of signal change across multiple physiological measures in humans; these patterns reflect the visceromotor (here, cardiovascular and hemodynamic) changes that support mental experiences and behavior in daily life (Hoemann et al., 2020; see also Feng & Narayanan, 2019).

One very well-studied multimodal physiological pattern relies specifically on signaling in the vagus: respiratory sinus arrhythmia (RSA). RSA is the variability in heart rate that is linked to respiration via cardiac vagal activity and is a result of both afferent and efferent signaling in the vagus nerve (as well as influence from the sympathetic nervous system; Grossman & Taylor, 2007; for a review, see Berntson et al., 1993). Heart rate increases during inhalation and decreases during exhalation, resulting in shortened and lengthened interbeat intervals, respectively. It is difficult to disentangle the afferent and efferent elements responsible for RSA (discussed in Berntson et al., 1993). This difficulty appears to be a more general feature given the architectures of the vagus nerve and the brainstem nuclei with which it communicates (see Box 2). Specifically, RSA may represent an emergent²³ phenomenon arising from combinatorial coding that we have described, where converging interactions from multiple physiological sources (both the afferent and efferent vagus nerve, as well as indirect influence from sympathetic nerves) are reflected in a single physiological signal that yields a measurement capturing both cardiovascular and pulmonary sensorimotor dynamics. While speculative, such a hypothesis is consistent with the possibility of signal compression and combinatorial coding by temporal coincidence that the vagus architecture possibly affords, along with the afferent and efferent signal integration that occurs within brainstem nuclei (and possibly even within vagal fasciculi). Moreover, signals arising in the nodose ganglia, being lower in dimensionality and possibly entailing more signal compression than afferent spinal cord signals for skeletomotor control, might at some point be revealed to contribute to functional “visceromotor concepts” in the primary visceromotor control areas of the cerebral cortex (e.g., the anterior cingulate, the ventral anterior insula/posterior orbitofrontal cortex, etc.) in the same way that the abstract functional ensembles in primary motor cortex are action concepts. The existence of such concepts orchestrating particular physiological “motifs” might present a partial answer to the question of how gradients of viscerosensory signals can be processed in the service of visceromotor control in the absence of viscerotopic specificity.

A Methodological Note

The research on visceromotor and behavioral motifs discussed above was conducted with measurements from human and non-human animals while they freely moved in real-world contexts or laboratory situations constructed to be similar to real-world contexts. This methodological insight is important for future research on vagal signal processing functions. Standard laboratory procedures, by contrast, intentionally remove signal variation in the world that is inherently present in naturalistic ecological contexts; laboratory settings are both spatially constrained and temporally artificial. The result is impoverished signals and signal patterns inside the brain and body because the metabolic and movement possibilities are severely limited by design. The consequence of stripping away this multidimensional context is that there may be insufficient signal variance from which to discover and map the hypothesized signal motifs and their associated concepts. Evidence consistent with this concern comes from the studies that discovered action concepts in early motor cortices described earlier. These studies used extended trains of microstimulation in cortical areas to evoke movements, on the timescale of meaningful behavior, rather than the short bursts of stimulation that had been traditionally used and that resulted in a somatotopic map (e.g., Graziano, 2016, 2023). Other methodological considerations for testing signal processing hypotheses in the vagus can be found in Box 3.

Temporal Mapping in the Vagus and Event Segmentation: Implications for Memory

The possibility of temporal coding in the vagus nerve can be extended further to hypothesize that physiological and behavioral motifs (i.e., temporally co-occurring events in the sensory periphery) may play a role in structuring learning, memory and prediction (maybe even establishing integrated skeletomotor and visceromotor concepts in early development) via their impact on the hippocampus. Ascending vagal signals reach the hippocampus and the rest of the forebrain via multiple pathways from its brainstem targets, but for present purposes we focus on inputs that project to the medial septal nuclei in the basal forebrain (Castle et al., 2005; Takeuchi et al., 2021).²⁴ The vagus projects to the medial septum via a variety of multi-synaptic pathways, including prominent projections from the NTS (as well as brainstem monoaminergic nuclei and the hypothalamus; Takeuchi et al., 2021). Cholinergic, GABAergic, and glutamatergic neurons from the medial septum then project to various aspects of the hippocampal complex, with projections specifically to the entorhinal cortex, the dentate, and cornu ammonis regions (CA3 and CA1).²⁵

The hypothesis that vagal afferents are important to hippocampal function is consistent with considerable functional evidence. For example, direct vagal nerve stimulation in rodents modulates the firing of neurons within the hippocampus (Broncel et al., 2017, 2018, 2021) and, correspondingly, has been shown to modulate neurotransmitter levels in hippocampal neurons (such as norepinephrine, serotonin, dopamine and GABA; e.g., Bocian et al., 2023; Broncel et al., 2019a, 2021; Raedt et al., 2011; Roosevelt et al., 2006). Mechanical distention of the stomach, intestinal infusion of nutrients, and electrical stimulation of the stomach all result in GI vagal signals that increase hippocampal neuron firing (Min, Tuor, & Chelikani, 2011; Min, Tuor, Koopmans, et al., 2011; Wang et al., 2006). Respiratory signals, potentially carried by the afferent vagus, serve as an “oscillatory pacemaker” (along with coordinated signals from the olfactory bulb) in freely moving rodents, modulating hippocampal and medial prefrontal cortex activity and their dynamics in a way that supports signal segregation and integration (e.g., Karalis & Sirota, 2022). Electrical stimulation of the cervical vagus has been shown to enhance memory consolidation (e.g., Clark et al., 1998, 1999; Ghacibeh et al., 2006; Ura et al., 2013) and facilitate neurogenesis in the hippocampus (e.g., Biggio et al., 2009; Follesa et al., 2007; Noble et al., 2011), and endogenous stimulation of the subdiaphragmatic vagus (via the GI tract) appears to be necessary for several hippocampal-dependent memory phenomena, including spatial working memory and contextual episodic memory for objects (Suarez et al., 2018). These effects specifically involve vagal inputs to the medial NTS, which project to the medial septum, which in turn innervates glutamatergic neurons in the dorsal hippocampal complex in rodents (equivalent to the posterior hippocampal complex in primates).

Viscerosensory afferent signals appear to influence hippocampus function (mediated by the medial septum) via endogenous theta oscillations (3–12 Hz signals) arising in the vagus itself. Medial septal inputs are known to be an important source of theta oscillations in the hippocampus (for discussions, see Butler et al., 2016; Tsanov, 2018), but recent evidence from studies of vagal nerve stimulation suggests that those oscillations arise in the vagus nerve itself. Vagal nerve stimulation has been shown to induce slower wave theta oscillations (3–6 Hz) in the hippocampal formation (Broncel et al., 2017, 2018, 2019a, 2019b, 2021, 2022; for corroborating evidence from pharmacological manipulations paired with large-scale electrical recordings in freely moving rodents, see Karalis & Sirota, 2022).²⁶ Theta oscillations from respiration in combination with theta oscillations from the olfactory bulb (another system known for temporal co-occurrence mapping) are particularly impactful (see Karalis & Sirota, 2022; for a recent summary of how breathing regulates the strength and synchronization of neural oscillations, see Brændholt et al., 2023).

The function of theta oscillations in the hippocampus is widely debated, but one prominent view suggests that they synchronize signals to result in segmented “events” (i.e., creating temporal windows, e.g., Buzsáki, 2006; Sirota & Buzsáki, 2005) that chunk or compress continuous streams of signals into temporally meaningful segments (e.g., Gupta et al., 2012). These segments allow neurons to fire in a phase-locked manner, creating time-compressed “cognitive maps” (Buzsáki & Llinás, 2017; Buzsáki & Tingley, 2018). There is evidence that theta rhythm-dependent event segmentation plays a necessary role in memory formation and consolidation (e.g., Kota et al., 2020) and disruption of septally-mediated theta oscillations impairs learning and memory (Takeuchi et al., 2021 and references therein)

A guiding hypothesis here is that theta oscillations originating in the vagus (specifically, temporal coding in the vagus) could allow the hippocampus to segment and sequence the temporal co-occurrences in signals transduced from the body’s sensory surfaces; the resulting events would correspond to statistical regularities in the brain’s periphery (Barron et al., 2020; Buzsáki & Tingley, 2018; Lisman & Redish, 2009; Pezzulo et al., 2017). Both the hippocampus (Schapiro et al., 2012, 2017; Sherman & Turk-Browne, 2020) and the cerebral cortex are generating and exchanging prediction and prediction error signals in a non-hierarchical fashion (Katsumi et al., 2022, 2023). From the perspective of the cortex, hippocampal events, which are constructed using temporal signal statistics of the body in the world, may function as prediction error signals. These error signals reweight the cortical prediction signals, which are constructed using spatial, nearest-neighbor statistics of the surrounding environment (Kumaran et al., 2016). This hippocampal re-weighting of cortical prediction signals would ensure that all prediction signals – visceromotor, skeletomotor and sensory–- are calibrated to the current metabolic state of the body in an event-consistent manner (i.e., weighted for the current and predicted conditions of the body’s sensory conditions; Kumaran et al., 2016). If correct, this hypothesis would suggest a formative role for the vagus nerve in predictive processing accounts for a variety of psychological phenomena including emotion and categorization (e.g., Barrett, 2017), attention and perception (see Hutchinson & Barrett, 2019 and Kleckner et al., 2017, and references therein), and social development (e.g., Atzil et al., 2018), just to name a few. Future research is required, of course, to fully test the hypothesis that event boundaries are drawn by the temporal coding of sensory changes in the body’s internal milieu by the vagus nerve.

Sensorimotor Integration in the Vagus and the Potential for Self-Fulfilling Prophecy: Implications for Disorders of Prediction

Within the predictive processing framework described earlier, the main function of ascending viscerosensory signals in the vagus is not to represent the sensory conditions of the body per se, but to constrain and correct the descending visceromotor commands that are implementing allostasis in the service of reducing uncertainty and enhancing metabolic efficiency. Unexpected viscerosensory changes from the internal milieu of the body, via ascending nodose axons, are errors of prediction. For example, in a recent study, exerting optogenetic control over the beating of the heart allowed researchers to control cardiac rhythms in freely moving mice (Hsueh et al., 2023).²⁶ From the brain’s perspective, the ascending sensory signals from these optogenetic manipulations function as prediction errors. Using genetic labeling, it was observed that these unexpected signals produced an increase in gene expression that is a marker of neural activation in various brain regions, including the NTS, posterior insular cortex (which is primary interoceptive cortex), as well as agranular anterior insular cortex (also called posterior orbitofrontal cortex) and the anterior cingulate cortex, both of which are key visceromotor regulation regions in the cerebral cortex (Hsueh et al., 2023; Öngür & Price, 2000; Vogt et al., 2003; for additional discussion, see Kleckner et al., 2017).

Unexpected changes in viscerosensory signaling computed as interoceptive prediction errors, like all sensory prediction errors, are teaching signals in the nervous system. Prediction errors informing on the actual energetic state of the body have the potential to adjust incoming visceromotor predictions in the moment and, if encoded and consolidated as memory, will modify future predictions, including visceromotor predictions and the anticipated interoceptive consequences (i.e., interoceptive prediction signals; see footnote 22, also see Barrett, 2017; Barrett & Simmons, 2015). One implication of possible sensorimotor integration in the fascicles of the vagus nerve, and certainly of sensorimotor integration in its brainstem targets, is that ascending sensory signals have the potential to be tuned by the brain’s descending visceromotor signals, allowing for the possibility that errors of prediction can be either magnified or minimized before they reach their midbrain and forebrain targets. This dynamic points to the potential for an elegantly orchestrated self-fulfilling prophecy embodied within the architecture of the nervous system.

In addition, ascending viscerosensory signals potentially have the capacity to modulate descending visceromotor signals outside the CNS, allowing for visceromotor changes that the brain did not issue. Even if the ascending signals improve the efficacy of descending control signals at their ultimate tissue targets, the result would be viscerosensory consequences that the brain did not predict and does not expect. This dynamic points to the potential for increasing the magnitude of eventual interoceptive prediction errors or increasing the noise in (i.e., decreasing the precision of) those errors, both of which would heighten the metabolic demand on the brain, by making allostasis more effortful and expensive to achieve. This result could materialize even when the ascending signals in the vagus or in their brainstem targets are enacting allostasis by correcting descending signals that arose from poorly calibrated prediction signals in the forebrain.

These hypotheses suggest the possibility that early sensorimotor integration might contribute to mental and physical illness in a way that has not yet been fully appreciated. For example, the brain’s processing of ascending interoceptive signals has been implicated in mood disorders such as depression and anxiety disorders. Symptoms of depression, such as insensitivity to context (e.g., Rottenberg et al., 2005), have been hypothesized to arise from metabolic problems and associated allostatic disruptions that make it difficult for the brain to update its prediction signals in the presence of prediction errors (for discussion and references, see Barrett et al., 2016; Shaffer et al., 2022). Symptoms of anxiety, in contrast, might arise when overly precise visceromotor predictions are routinely generated, resulting in noisy (less informative) prediction error signals (Paulus et al., 2019; Paulus & Stein, 2010).

If supported by empirical evidence, these hypotheses suggest other even broader implications for the role of the vagus in both sensing and control of metabolism, and would speak to our recent proposal that many disorders, both those traditionally considered physical (e.g., metabolic syndrome) and mental (e.g., major depression), can be reinterpreted as, at their core, disorders of metabolic inefficiency (Kleckner et al., 2017; Shaffer et al., 2022). Accordingly, the mood disruptions observed in both mood disorders, metabolic disorders such as diabetes and cardiovascular disease, and even disorders of the immune system (which have considerable metabolic cost; e.g., Bucks et al., 2008; Chrysohoou et al., 2018; Gavard et al., 1993; Kiriella et al., 2021) can be thought of as arising from disruptions in allostatic control, and the interoceptive contributions to those disruptions, which ultimately lead to prolonged energetic inefficiencies and the symptoms they produce. Evidence for this comes from the innumerable metabolic symptoms seen in depression, as well as aberrant functional and structural patterns in key limbic cortices that are the neural backbone of allostasis (Kleckner et al., 2017; Shaffer et al., 2022). This same architecture makes up the central autonomic network (CAN) that interfaces with the NTS and other brainstem nuclei to integrate viscerosensory signals from the vagus and other peripheral nerves in the service of physiological regulation (e.g., Benarroch, 1993; Valenza et al., 2019). The same architecture also contributes to the “pain connectome” (Kucyi & Davis, 2015) that implements varieties of pain (nociceptive and neuropathic).

Conclusions

We have reviewed recent empirical evidence to support the viability of several hypotheses regarding signal processing within the vagus nerve. The evidence revealed three signal gradients within the vagus that may afford spatial and temporal processing of viscerosensory signals, perhaps with relatively greater spatial specificity of afferent signals at the rostral end of the body’s rostro-caudal axis. The evidence also suggests that the sensory and motor fibers of the vagus are not structurally separate and may not be as functionally separate as first supposed. These hypotheses await further study, possibly using innovative methods for studying signaling in the vagus (see Box 3). If supported by empirical evidence, these hypotheses would suggest more fundamental implications for the vagus in both allostatic and psychological phenomena such as learning, memory, and mood, as well as novel implications for the vagus’s role in disorders of prediction.

Highlights.

• Recent genetic and structural evidence on the ascending vagus nerve suggest that it integrates and compresses viscerosensory signals as they ascend brain.

• The nodose ganglion does not contain a precise viscerotopic map but instead maps viscerosensory signals along two spatial gradients and one temporal gradient.

• Evidence from both central and peripheral neurons suggests that asynaptic signal processing within the vagus nerve is physiologically plausible.

• The anatomy of the vagus nerve suggests some degree of sensorimotor integration may be possible.