Intrinsic neuronal network organization of a mammalian peripheral nervous system Authors: Larry W. Swanson, Joel D. Hahn, Olaf Sporns Published: 2025 ABSTRACT SignificanceThe peripheral nervous system (PNS) is usually regarded as a set of ganglia and nerves that transmit information between the central nervous system and the body’s other systems, thereby supporting the regulation and integration of physiology and behavior. Using systems science tools, we analyzed a sparse yet broad network of axonal connections between peripheral ganglia in a mammalian (rat) nervous system. The resulting model reveals a nested hierarchy of modules spatially aligned along the body axis, with two potential hubs identified as the bilateral myenteric plexus ganglia. These findings suggest that the PNS is not a series of parallel nerves and associated ganglia; instead, it presents an internally connected and organized neuronal network linking sensory and autonomic ganglia. Since the late nineteenth century (1) there has been sustained interest in the organization of axonal connections forming the neuronal network substrate of the vertebrate nervous system (NS), which controls and coordinates the body’s internal physiology and behavioral interactions with the environment through its two great topographic divisions, central nervous system (CNS) and peripheral nervous system (PNS). In tetrapods, a primary criterion for this division is developmental. In the early embryo, CNS and PNS neurons arise from essentially separate ectodermal spatial domains: CNS neurons arise from the neural plate, whereas PNS neurons arise from adjacent neural crest and nearby neuroectodermal placodes (2). In the adult, both divisions may be subdivided on histological grounds into gray matter (predominantly neuronal cell bodies and synaptic interactions) and white matter (predominantly axons). CNS gray matter forms a mosaic of spatially distinct regions, each defined by a unique set of neuron types, interconnected by a set of white matter tracts (each with a more or less spatially distinct set of axons). In the PNS, the basic cellular components are similar, but the terminology for them differs: Circumscribed gray matter regions are called ganglia, and white matter tracts are called nerves (3). Conceptually, two primary approaches have emerged for modeling the PNS: One is informed mainly by PNS topographic organization, while the other emphasizes its functional organization (3). The currently favored topographic model describes a set of sensory and autonomic ganglia and craniospinal nerves arranged from rostral to caudal and proximal to distal. In contrast, the functional modeling approach is systems-based and conceives a network of interconnected ganglia—just as the CNS corresponds to a network of interconnected regions. The tools and approaches of connectomics (the application of network science to neuroscience) offer a systematic, mathematically grounded, and relatively objective framework for analyzing and interpreting neuronal network architecture (4–7)—and for providing mechanistic insight into their structure–function organization based on dynamic structural modeling (4, 8, 9). The results of this connection matrix-based approach have been described recently for the rat intrinsic CNS neuronal network (10), with its 840 regions and some 80,000 axonal connections between them—compared to as few as 200 connections (11) reported for the mammalian CNS in an authoritative mid-twentieth century textbook (12). The results of this reevaluation suggest, for example, that the basic plan of the rat intra-CNS neuronal network has two main divisions: a) a rostral division with a pair of bilaterally symmetric modules (subsystems, subnetworks) concerned primarily with behavior control, cognition, and affect, and b) a caudal division with a single bilaterally symmetric module concerned primarily with behavior execution and reflex integration. The strategy and methodology we used for CNS network analysis is applied here to clarify the basic organization of the intrinsic PNS neuronal network in rats. However, there are substantial differences between CNS and PNS that have a bearing on the comparison and interpretation of their connectomes. For example, the terminology associated with PNS ganglia and nerves is much more standardized, and thus much less ambiguous, than the considerably more varied terminology for CNS regions and tracts (13). From a basic interpretive perspective, this is advantageous. There is, however, a more fundamental difference. CNS regions form a large yet spatially compact volume intermingled with tracts in a complex arrangement that is contained within the skull and vertebral column. In contrast, PNS regions (ganglia) form comparatively small, spatially dispersed volumes, and in comparison to CNS tracts, PNS nerves are generally more spatially dispersed, extend throughout the body, and are also separated from one another by nonneuronal tissue. Since Classical Antiquity, this spatially accessible PNS architecture has facilitated innumerable PNS dissections in humans and other mammals that have yielded a remarkably consistent basic plan at the scale of gross anatomy (13). On the other hand, in comparison to the CNS, experimental axonal pathway tracing data for the PNS are quite limited. The relative scarcity of experimental connectivity data for the intra-PNS neuronal network can be attributed to technical challenges inherent to PNS architecture. Mapping all potential intra-PNS connections for a given ganglion requires, in principle, histological analysis of nearly the entire body (excluding the CNS), due to the widespread distribution of peripheral ganglia. Further complexity arises from the small size and compact structure of most ganglia, coupled with their spatial dispersion, which complicates precise targeting. In contrast, intra-CNS connectivity can be assessed within the spatially continuous brain and spinal cord, with most studies focusing on the brain—a more accessible target aided by standardized stereotaxic atlases (for example, ref. 14). Consequently, region-level axonal connectivity data for the mammalian PNS remains far more limited than for the CNS. Our rat connectome project has progressed systematically through the NS divisions outlined in Fig. 1, beginning with the intensively studied cerebral cortical association network (15). We now extend the analysis to the network organization of bilateral intra-PNS connections (PNS2), which naturally invites comparison with the bilateral intra-CNS network (CNS2). The next logical step is to examine intra-CNS and intra-PNS connectivity in conjunction with the connections between them, collectively forming the NS neuronal network. This integrated framework can then be used to explore how the NS neuronal network interfaces with other body systems—such as the digestive, circulatory, and muscular systems—to regulate physiology and behavior. By extension, we refer to this broader integrative framework as the neurome (15). A nested hierarchy of NS gray matter parts arranged topographically for connectomics studies in adult vertebrates, applied here at lower levels to a mammal (rat). Gray matter regions, used here as a parts list for connection matrices, are a nested level below the 15 subdomains and their parent divisions at the vertebrate level; neuron types are a nested level below regions; individual neurons are a nested level below neuron types, and synapses are a nested level below individual neurons. Regions (including PNS ganglia) and neuron types are part of the inherited body plan of a species, whereas individual neurons and synapses are subject to epigenetic influences that are unique to each individual. Region numbers shown are for both sides of the bilateral NS, where each region on one side (left or right) has a corresponding region on the opposite side. Results The strategy for constructing, analyzing, and interpreting a connectome is intentionally straightforward and remains fully compatible with our previous work on intrinsic connectivity within brain and CNS divisions of the rat (see refs. 9 and 10). In the PNS connectome, gray matter regions encompass craniospinal sensory ganglia, the olfactory and vomeronasal neuroepithelia, and the paravertebral, prevertebral, and terminal autonomic ganglia (see ref. 13, and Datasets S1 and S2 for terminology and abbreviations, respectively). Creating a Benchmark Neuroinformatics Database for a Mammalian intra-PNS Connectome. Constructing an intra-PNS neuronal network model, as for a neuronal network model of any other NS division, requires selecting a nested scale of connection granularity—macro (between regions or ganglia), meso (between neuron types within a region or ganglion), or micro (between individual neurons within a neuron type), where regions or ganglia are defined as spatially compact gray matter volumes distinguished by a unique set of neuron types with a unique set of connections (3, 16, 17) (Fig. 1). The macro level in rats was chosen for our global NS connectome project because available mammalian data for both the CNS and PNS are most complete at this level, and in rats, making this species the most suitable mammalian model for global connectome analysis. However, because nested levels of scale are used, the macro level in principle also contains all interregional connections at lower levels of scale, allowing aggregation of data across macro-, meso-, and micro levels into a macroscale database. It is important to note here that the macro level, like all levels of scale, has its own set of internally consistent neurobiological features (18). Notably, macro level connections do not distinguish between connection subsets specific to neuron types (meso level), which are especially prominent in autonomic and sensory ganglia (19, 20), or between the varied axon collateralization patterns and connections of individual neurons (micro level) (1). Next, we assembled the connection database, collated primarily from the published results of experimental axonal pathway tracing studies initiated on the PNS in the late 1970s, and supplemented by authoritative reviews of the mammalian literature. The connection data were extracted manually from relevant text, tables, maps, diagrams, and photomicrographs (for a complete account of collation strategy see ref. 9). The results indicate presence, absence, and maximum weight of directed (“from–to”) axonal connections (edges in graph theory, referred to hereafter simply as connections) between all 84 regions (ganglia and sensory neuroepithelia) included in this study on the right (42 regions) and left (42 corresponding regions) sides of the adult rat PNS in our internally consistent reference nomenclature hierarchy (Dataset S1). This fully annotated PNS nomenclature is integrated with its CNS counterpart and has undergone continual revision since 1974 (21); it is now in its fourth edition (14), with updates here based on the literature review for this paper (SI Appendix, Materials and Methods). The nomenclature is part of an evolving neuroanatomical universal reference ontology of defined terms for NS parts in any species, a consideration and disambiguation problem essential for direct intraspecies and interspecies connectome comparisons (9, 13, 22, 23). The collation also employed a defined vocabulary for describing axonal connections (3). Connection reports for each possible connection in the connection matrix were expertly collated by L.W.S. The number of connection reports ranged from 0 to 7 per connection (mean: 1.01, mode: 1). When doubled to account for both sides of the PNS, the maximum becomes 14, with a mean of 2.02, and mode of 2. This information was recorded in a conditionally formatted Excel worksheet (AxiomeM 3.32, designed by J.D.H.; Dataset S3) with 18 columns of data and metadata per report; a comparison of collation results for the same connection matrix by two experts is reported in ref. 24. Basic Attributes of the PNS2 Neuronal Network (Table 1). No statistically significant differences were identified across left–right, male–female, or strain comparisons during systematic collation. Consequently, all ipsilateral and contralateral connections were assigned to one side of the PNS, and the same dataset was applied bilaterally. This region-level analysis reflects the genetically hard-wired, species-specific neuronal network of adult Rattus norvegicus domestica (15), rather than the epigenetically modifiable micro level connectivity of individual neurons (25). The full dataset comprises 7,164 connection reports covering 6,972 possible PNS2 connections, collated from 59 publications spanning 26 journals, books, and theses since 1978. Notably, only 8.5% of the reports originated from the Journal of Comparative Neurology (compared to 50.1% for our CNS2 dataset). Seventeen distinct axonal pathway tracing methods were employed (SI Appendix, Materials and Methods). Our collation identified 98 existing (present) connections and 6,040 absent connections, yielding a sparse network connection density of 1.6%. No data were available for 12.0% of possible connections, resulting in a connectome fill ratio of 88.0%. Assuming uniform sampling, we estimate the PNS2 network contains about 111 connections. Due to the lack of quantitative data, existing connections were assigned ordinal weights (Fig. 2, First column; Datasets S4 and S5, Layer E). Bilateral intra-PNS connectomes for adult rats (PNS2). Directed and weighted monosynaptic connection matrices with gray matter region (ganglion) sequence in a module arrangement derived from multiresolution consensus cluster (MRCC) analysis (fourth column). Collated data are represented by descriptive terms with ordinal weight values (first column); these values were binned for analysis (second column): “present” was assigned a “moderate” value, “axons-of-passage” (referring to PNS fibers-of-passage) was converted to “weak,” and both “unclear” and “no data” were treated as “absent.” The binned matrix was then transformed into log-weighted values for computational analysis (third column). MRCC analysis of the log-weighted data produced a coclassification matrix (fourth column), which quantifies how consistently each ganglion pair affiliates with the same module across all clustering resolutions. Coclassification values range from 0 (never coclassified) to 1 (always coclassified). The matrix is also visualized as a nested hierarchy tree (dendrogram, fifth column; see Fig. 3 for higher resolution). Two independent MRCC runs, each sampling 100,000 partitions, yielded identical results. Sixteen isolated ganglion pairs with no intra-PNS connections were excluded from MRCC analysis. All modules in the hierarchy passed statistical testing at α = 0.05. For ganglion (row and column) and additional details, see high-resolution Datasets S3 and S4, where connection data and module organization are easy to explore. For network analysis, no data entries were binned with absent and unclear values (Fig. 2, Second column; Datasets S4 and S5, Layer F). The ordinal scale was then converted to a more realistic scale spanning five orders of magnitude in connection weight, consistent with rat neuroanatomical data (9) (Fig. 2, Third column; Datasets S4 and S5, layer G). Sixteen ganglion pairs (32 of 84 total ganglia) lacked reported intrinsic PNS connections and were excluded from further network analysis (10, 26, 27). This high proportion of isolated (“orphan”) nodes contrasts with no orphans in the CNS2 network (Table 1). The ganglia that are isolated from other ganglia in the PNS network presumably form only extrinsic connections—with CNS and/or other body systems—and have no known intra-PNS connectivity. Accordingly, a 52 × 52 connection matrix was used for PNS2 network analysis. Table 1.Summary and comparison of major connection metrics for the bilateral adult PNS and CNSConnectionsDivisionRegions (nodes)*possible†reported‡projected§Connection density¶Outputs/ region#Isolated regions||1st order modules**End modules**Hubs††Connection reports‡‡ PNS 846,97298~1111.6%~1.332/8471427,164 CNS 840704,76074,700~81,43511.6%~95.90/8403160461,145,524*Regions (nodes): total number of nodes per bilateral division—CNS gray matter regions or PNS ganglia (Fig. 1).†Possible connections: calculated as N2 − N, where N is the number of nodes (regions/ganglia); excludes self-connections.‡Reported connections: sum of connections collated from the primary literature.§Projected connections: projected sum of connections extrapolated from dataset completeness (PNS2: 87.1%, CNS2: 91.7%). Includes axons-of-passage reports (assigned a conservative connection strength of weak, see text for details).¶Connection density (%): ratio of connections reported as present (as shown) to those reported as absent (PNS2: 6,040; CNS2: 571,784), expressed as a percentage. This value is the same as the ratio of projected to possible connections.#Mean degree (out/in): average number of output (outdegree) or input (indegree) connections per node. These are equal in a fully mapped directed network.||Isolated regions: nodes with no connections within the subnetwork—structurally and functionally isolated; called “orphans.”**See Figs. 3 and 4; and Dataset S4, Layer D for graphical representations of these metrics.††Top-ranked hubs: number of bilateral hub pairs; for PNS this is the right–left pair of myenteric plexus ganglia.‡‡Connection reports: total number of reports entered (collated) for all possible connections; this number includes reports for which data were available, as well as reports indicating that no data were found. Bilateral counts are doubled from unilateral data.This table compares key network measures for the bilateral PNS and CNS in the adult rat. CNS data are adapted from ref. 10. All values reflect bilateral networks, with each side assumed to be symmetrical. Summary and comparison of major connection metrics for the bilateral adult PNS and CNS *Regions (nodes): total number of nodes per bilateral division—CNS gray matter regions or PNS ganglia (Fig. 1). †Possible connections: calculated as N2 − N, where N is the number of nodes (regions/ganglia); excludes self-connections. ‡Reported connections: sum of connections collated from the primary literature. §Projected connections: projected sum of connections extrapolated from dataset completeness (PNS2: 87.1%, CNS2: 91.7%). Includes axons-of-passage reports (assigned a conservative connection strength of weak, see text for details). ¶Connection density (%): ratio of connections reported as present (as shown) to those reported as absent (PNS2: 6,040; CNS2: 571,784), expressed as a percentage. This value is the same as the ratio of projected to possible connections. #Mean degree (out/in): average number of output (outdegree) or input (indegree) connections per node. These are equal in a fully mapped directed network. ||Isolated regions: nodes with no connections within the subnetwork—structurally and functionally isolated; called “orphans.” **See Figs. 3 and 4; and Dataset S4, Layer D for graphical representations of these metrics. ††Top-ranked hubs: number of bilateral hub pairs; for PNS this is the right–left pair of myenteric plexus ganglia. ‡‡Connection reports: total number of reports entered (collated) for all possible connections; this number includes reports for which data were available, as well as reports indicating that no data were found. Bilateral counts are doubled from unilateral data. This table compares key network measures for the bilateral PNS and CNS in the adult rat. CNS data are adapted from ref. 10. All values reflect bilateral networks, with each side assumed to be symmetrical. Binned data indicate that, on average, each PNS ganglion projects to and receives input from 1.1 other ganglia. Given 84 PNS ganglia (42 bilateral pairs), each could theoretically form 83 interganglionic output or input connections. However, observed output connections ranged from 0 to 9, input connections from 0 to 8; and combined input–output totals from 0 to 13 (out of 166 possible). Finally, a 7-point ordinal validity metric (1 = lowest, 7 = highest) was applied to each axonal tracing method used in the connection reports (28, 29) (Dataset S5, Layer I). This metric incorporates three criteria: reach (favoring monosynaptic over polyneuronal), sensitivity, and potential involvement of fibers-of-passage (axons or dendrites). Six validity scores are given, for present or absent connections within one side, between sides, and for both sides. In addition, for connections reported to be absent these scores are given for two datasets: the dataset used for PNS2 network analysis and the complete dataset. The validity scores for both datasets differ slightly for connections reported as absent because 16 (out of 42) PNS ganglia were excluded from the PNS2 network analysis dataset due to their being unconnected (isolated) from the rest of the network. For these, the first number given relates to the data used for PNS2 network analysis (indicated by “network” in parenthesis), and the second number relates to the complete dataset (indicated by “complete” in parentheses). Accordingly, the following average validity scores were determined. For connections reported to exist: ipsilateral (within one side) = 4.89, contralateral (between sides) = 3.67, both sides = 4.82. For connections reported to not exist: ipsilateral = 3.54 (network)/3.26 (complete), contralateral = 3.65 (network)/3.31 (complete), both sides = 3.60 (network)/3.29 (complete). Structure–Function Module Organization Revealed by MRCC. Following construction of the connection database, we investigated local differentiation within the PNS neuronal network using cluster analysis based on connection weight and the hypothesis that densely interconnected ganglia form modules (subsystems or subnetworks) with distinct structure–function attributes. To this end, we applied unsupervised MRCC analysis, which samples all levels of clustering resolution and identifies variably sized modules arranged in a matrix and also visualized as a nested hierarchy (dendrogram) (30, 31). Together, these representations offer a compact summary of all modules, covering multiple spatial scales, and their interrelationships (Fig. 2, Right two columns; Datasets S4 and S5, Layer H; for definitions of dendrogram parts see ref. 10). The MRCC output constitutes a predictive model—or testable hypothesis—of neuronal network organization. The MRCC-defined PNS2 hierarchy displays 22 modules organized across six hierarchical levels, including at the second level (the level below the first or root level corresponding to the unpartitioned network), seven first-order modules and 14 end modules (Dataset S5, Layers B and D). As an interpretive framework, this structural hierarchy was provisionally and conservatively annotated (Fig. 3; Dataset S5, Layer D) with general functional correlates at the region level, based on documentation from the connection report references in Dataset S3. This annotation strategy mirrors our previous analyses of intrinsic connectivity in the forebrain (28), midbrain (29), rostral sector (8), rhombicbrain/intermediate sector (26), brain (9), spinal cord/caudal sector (27), and CNS (10). Structure–function interpretation of the MRCC module hierarchy for PNS2. General functional associations are annotated with concise labels; more specific interpretations may be developed by examining the full ganglion composition of each module (see Dataset S4 for annotated hierarchy and connection matrix context). In MRCC hierarchies, the order of child branches and the sequence of ganglia (nodes) within end modules are arbitrary and may be rearranged for clarity without affecting network topology. Hierarchy levels (red) indicate sequential branch points, with level 1 corresponding to the root at the top (the entire matrix with no partitioning) and level 6 corresponding to the final branches at the bottom (in M3 and M4). For region abbreviations, see Dataset S1 or Dataset S2; abbreviations are prefixed with “1” or “2,” where 1 indicates one side of the NS (left or right), and 2 refers to the opposite side. Several features of the MRCC hierarchy for PNS2 are notable (Dataset S5, Layers B and D). First, the full connection matrix partitions immediately into seven first-order modules, each showing strong resistance to further subdivision, as represented by branch length in the dendrogram. Second, these seven first-order modules comprise one bilaterally symmetric module (M5) and three bilaterally symmetric module pairs (M1–M2, M3–M4, and M6–M7). Third, module pair M1 and M2 show no branching; M3 and M4 branch minimally and late, with very short second- and third-order branches; and M6 and M7 exhibit longer second-order end branches, indicating greater internal differentiation. Fourth, the unpaired first-order module M5 contains a bilaterally symmetric pair of relatively short second-order end modules. This modular architecture contrasts sharply with the CNS2 hierarchy (Table 1), which includes only three first-order modules but 160 end modules—highlighting the relative simplicity and sparsity of intra-PNS network organization compared to the CNS. Bottom–up PNS module analysis drives testable hypothesis formulation. Bottom–up analysis of the MRCC hierarchy offers the most detailed resolution of modular organization within the neuronal network. As such it is particularly useful for generating strong testable hypotheses, exemplified here for PNS2, whose 14 end modules are experimentally tractable ganglion sets presumed to share functional relationships—there are on average four relatively robustly interconnected ganglia per terminal module (range: 2 to 7 ganglia). Thus, structure–function annotation of the PNS2 MRCC hierarchy began with terminal modules and proceeded systematically up the hierarchy (Dataset S5, Layers C, D), guided by three practical hypotheses (28). First, module members with poorly understood functional attributes share overall functional attributes of the entire module. Second, if a parent module has two or more children with apparently distinct functional properties, then the parent module integrates those two or more functional properties. And third, if a parent module has two children, one with clear functional properties and the other without, then the parent module displays the known and unknown functions of its children. Top–down module analysis reveals general organizing principles. Top–down analysis of an MRCC neuronal network hierarchy provides a high-level perspective that aids comprehension and interpretation of its nested modules, offering broad insights into the basic plan or wiring diagram, in contrast to bottom–up analysis that offers narrower insights, reflected in greater functional specificity. Because the PNS2 network is quite sparse and its module hierarchy quite shallow—compared, for example, to the CNS2 network and hierarchy (10)—top–down interpretation of its organization yields comparatively richer insight into overall network architecture. As shown in Fig. 3, its seven first-order modules form four basic groups: three bilaterally symmetric pairs (M1–M2, M3–M4, and M6–M7) and one that is unpaired and bilaterally symmetric (M5). Functionally, the four groups are associated with accessory olfactory structures (M1–M2); sensory and autonomic innervation of the head, neck, and upper limbs (M3–M4); vagal sensory and cardiac parasympathetic innervation (M5); and spinal sensory and autonomic innervation of the trunk and lower limbs (M6–M7). A solution of seven first-order modules is the highest number for this metric that we have encountered for the intrinsic connectivity of major rat NS divisions (bilateral). The next highest is the cerebral cortex that displays six first-order modules (24), both pons and medulla display five (26), CNS displays three (10) (Table 1), and forebrain (28) and cerebellum (26) display two first-order modules. Spatial distribution of PNS2 modules. Ganglia associated with a module could be distributed across the PNS2 network in a dispersed (“checkerboard”) pattern, in a spatially aggregated set, or in some variation combining elements of both. When the four groups of ganglia assigned to the seven first-order modules (Fig. 3) are plotted on a flatmap of the rat NS (32), it is readily apparent from visual inspection that they form spatially segregated, partly overlapping components of the PNS2 neuronal network that are arranged rostrocaudally (from M1 to M7) and are aligned with major CNS divisions (Fig. 4). Spatial distribution of PNS2 ganglia plotted on a flatmap of NS gray matter parts. Ganglia are color-coded (Legend) to match the module assignments shown in Figs. 2 and 3, and detailed in Dataset S4. Modules 1 to 7 (M1–M7) exhibit a general rostral-to-caudal arrangement, progressing from left to right across the flatmap. The CNS portion of the flatmap (Fig. 1) is derived from the embryonic neural plate and its transformation into the neural tube. In contrast, peripheral ganglia and sensory neuroepithelia originate from adjacent neural crest and nearby ectodermal placodes. The CNS portion functions as a fate map of the neural plate and is adapted from ref. 32. For ganglion abbreviations, see Dataset S1 or Dataset S2. Connections between PNS2 Modules. The results of our MRCC analysis visualized as a dendrogram show all identified modules in a nested hierarchy (Fig. 2, Right column; Fig. 3 and Dataset S5, Layer D); however, this representation does not show the existence of weaker connections between modules, which are clear in the MRCC matrix itself (Fig. 2, Second column; Datasets S4 and S5, Layers C–J). One way to quantify how directly ganglion pairs are linked, and the extent of intermodular interactions, is to plot the shortest path length for all ganglion pairs in a step-based distance matrix (Fig. 5; Dataset S5, Layer J), with paths computed from the weighted connection matrix and then reported in terms of the number of steps. Here, one step is equivalent to a direct connection between two ganglia, two steps are equivalent to a path with two connections between three ganglia, and so on—some ganglion pairs having no intra-PNS path between them. The relatively sparse PNS2 neuronal network, even after excluding ganglia with no reported connections (Table 1), has many ganglion pairs with no path between them (red cells in Fig. 5)—no signals can pass between them in this network—and in fact, modules 1 and 2 (M1 and M2; accessory olfactory) are completely isolated from the rest of the network. In contrast, there are at least potential interactions between the rest of the modules in the network, although some may be very weak, requiring up to nine steps—the maximal shortest path, defined as network diameter (Fig. 5; Dataset S5, Layer J). How this intra-PNS network fits within the larger context of the NS as a whole, and NS interactions with the rest of the body, remains a key question for future investigation. A distance matrix for the PNS2 neuronal network, excluding orphans. This matrix displays the shortest path lengths, computed from the weighted connection matrix, between all ganglion pairs in the PNS2 network, expressed in steps—the number of connections in the path between the ganglion of origin and the ganglion of termination. First-order modules M1–M7 are outlined in yellow along the main diagonal (upper left to lower right; see Dataset S4, Layer A). Red indicates that no path exists between the indicated ganglion pair; thus, modules 1 and 2 are completely isolated from the rest of the network. The maximal shortest path (network diameter) in this network comprises nine steps. Global Neuronal Network Attributes of the PNS2 Connectome. After considering local differentiations of the PNS2 neuronal network as reflected in a nested hierarchy of interconnected modules, we computed and analyzed four basic measures of global network organization based on measures of ganglion/node centrality (5–7). Centrality. This fundamental property captures the relative importance of nodes (here ganglia) within the network. Four standard centrality measures (degree, strength, betweenness, and closeness) were computed for all ganglia in the PNS2 neuronal network (Fig. 6 A–D). For example, the data suggest that the indegree and outdegree (the number of input and output connections) (Fig. 6E) is roughly similar for the thoracic spinal nerve ganglia (GSNth1-13); whereas the outdegree is considerably greater than the indegree for the myenteric plexus ganglia (GMYP), suggesting a predominantly “sender” role. In contrast, the superior cervical ganglion (GSC) may function primarily as a “receiver” node because it has considerably more reported input connections than output connections within the PNS2 neuronal network. Centrality rankings of PNS2 ganglia. Ganglia in the PNS2 network are ranked according to four standard centrality measures: degree (A), strength (B), betweenness (C), and closeness (D). In directed networks, degree centrality can be further divided into indegree and outdegree, for (respectively) incoming and outgoing connections (E). Degree centrality reflects the number of direct connections associated with a node; strength centrality captures the total weight of those connections; betweenness centrality quantifies how often a node lies on the shortest paths between other nodes; and closeness centrality measures the average shortest path length from a node to all others. Regions are ranked by degree centrality (A) across all plots. Purple bars in A–D highlight ganglia in the top 20th percentile for all four metrics. An aggregate centrality score (range: 0 to 4) was calculated for each ganglion based on the number of metrics in which it ranked in the top 20th percentile. Only one side of each bilaterally symmetric ganglion pair is shown. For abbreviations, see Dataset S1 or Dataset S2. Hubs. These are relatively highly connected nodes and are thus often importantly involved in regulating information flow through a network. Here, hub identification was based on aggregated rankings across the four node centrality measures (degree, strength, betweenness, and closeness), after ranking each region on each metric (Fig. 6). An aggregate “hub score” was determined for each ganglion by summing the number of centrality metrics for which each ganglion appeared in the top 20% (15). Using this criterion, there are only two ganglia that meet the hub criteria: the right–left pair of myenteric plexus ganglia (1GMYP, 2GMYP), which are regarded here as terminal autonomic ganglia in the digestive tract. Rich club. This feature is defined as a core set of individually highly connected (high degree) nodes or ganglia that are mutually highly interconnected—exceeding levels predicted by random network models and implying a role in information sharing, facilitating communication, and/or network integration (33). Rich club organization has previously been found in many neuronal networks from different species, including some, but not all, subdivisions of the rat NS (10). No statistically significant sets of ganglia forming rich clubs were identified in the rat PNS2 neuronal network. Small world. Networks with highly clustered nodes connected by short paths exhibit small-world organization (34, 35), which may indicate efficient communication in terms of energy cost (36). Small-world attributes calculated for the relatively sparse PNS2 neuronal network suggest that they exhibit among the highest small-world coefficients observed for the 22 bilateral subconnectomes forming smaller components within the CNS2 network (Fig. 7). Thus, the small-world coefficient for PNS2 is very similar to that for the bilateral endbrain and brain networks, and is only outranked by the bilateral forebrain, rostral sector, and CNS networks (Fig. 7B). At the other extreme, the bilateral intermediate sector/rhombicbrain, thalamic, cerebellar, medullary, and spinal cord subnetworks display no small-world attributes (Fig. 7 A and B). As noted previously (9), in every case including the PNS, adding commissural connections to unilateral connections increases clustering but has variable effects on path length and therefore on small-world index (SWI). It must be noted, however, that the large number of node pairs for which no path exists across the network (Fig. 5) indicates that communication is restricted to specific subsets of network elements. This tempers the interpretation of the PNS2 SWI as an index of global information integration. Small-world attributes of rat PNS and CNS divisions. (A) A comparison of small-world attributes across PNS and major CNS divisions (the latter from our previously reported subconnectomes—refs. 8, 9, 24, 26–29, 31, and (37–40) as summarized in ref. 10. Small-world networks are characterized by high clustering (dense local connectivity) and short path lengths (efficient global communication). Clustering is computed as the mean of all weighted-directed clustering coefficients, and path length is the mean of weighted shortest paths between all region-pairs for which finite-length paths can be found. Both metrics are scaled against medians from 1,000 degree-preserving randomized networks. The SWI is calculated as the ratio of scaled clustering to scaled path length (35). SWI > 1 indicates small-world topology. (B) SWI values from (A) are used to rank-order PNS and CNS divisions. Bar colors correspond to the circle colors in (A). Prefixes 1 and 2 denote unilateral and bilateral networks, respectively; suffixes f and m indicate female and male datasets. For abbreviations, see Fig. 1 and: PCB, pons-cerebellum; PMY, pons-medulla. Adapted from ref. 10, which is licensed under CC BY-NC-ND 4.0. Features of Crossed and Bidirectional Intra-PNS Connections. Because the rat PNS neuronal network’s right and left halves are bilaterally symmetric with respect to axonal connections between ganglia, we compared ipsilateral connections and connections crossing from one side to the other (contralateral, crossed connections) and compared the results with those for the CNS (10) (Table 2), as done previously for the cerebral cortex (24), brain (9), and CNS. Major findings include the following: a) contralateral intra-PNS connections are exceedingly rare, accounting for only 0.17% of all possible contralateral connections and 6.1% of reported intra-PNS connections, in contrast to intra-CNS connections that are substantially more prevalent, comprising 6.6% of all possible contralateral connections, and 31.5% of all reported intra-CNS connections; b) only three heterotopic contralateral connections (between nonhomologous ganglia on opposite sides) are reported within the PNS; no homotopic contralateral connections (between mirror image ganglia) were observed; c) Approximately 6.1% of reported ipsilateral connections in the PNS have a matching heterotopic contralateral counterpart; and d) all three reported contralateral connections are paired with a corresponding ipsilateral connection. Qualitative differences between the intra-PNS and intra-CNS networks are pronounced (Table 2). Table 2.Comparison of intra-PNS and intra-CNS network metrics for ipsilateral and crossed (contralateral) connectionsintra-PNSintra-CNSData metricNumberPercentNumberPercentd1 Ipsilateral connections462.67%25,58314.54%d2 Contralateral connections30.17%11,7076.64%d3 Homotopic contralateral connections0017842.38%d4 Heterotopic contralateral connections30.17%11,5896.59%d5 Ipsilateral connections with matching heterotopic contralateral connections30.17%10,7486.11%d6 Ipsilateral connections without matching heterotopic contralateral connections432.5%14,8358.43%d7 Heterotopic contralateral connections without matching ipsilateral connections008410.48%This table compares the number and percentage of existing connections originating from one side of the PNS and CNS. Values for both sides are double those shown to reflect assumed bilateral symmetry. All values refer to the existence of connections, not their weight. Percentages represent the proportion of possible connections within each of the six data categories (d1 to d6), based on a unilateral CNS network of 420 gray matter regions and a unilateral PNS network of 42 ganglia. Ipsilateral self-connections are excluded. The following relationships must hold across the data: d1 = d5 + d6 (applies to both numbers and percentages), d2 = d3 + d4 (applies to numbers only), d4 = d5 + d7 (applies to both numbers and percentages). CNS data are adapted from ref. 10, which also included connection data for major CNS divisions. Comparison of intra-PNS and intra-CNS network metrics for ipsilateral and crossed (contralateral) connections This table compares the number and percentage of existing connections originating from one side of the PNS and CNS. Values for both sides are double those shown to reflect assumed bilateral symmetry. All values refer to the existence of connections, not their weight. Percentages represent the proportion of possible connections within each of the six data categories (d1 to d6), based on a unilateral CNS network of 420 gray matter regions and a unilateral PNS network of 42 ganglia. Ipsilateral self-connections are excluded. The following relationships must hold across the data: d1 = d5 + d6 (applies to both numbers and percentages), d2 = d3 + d4 (applies to numbers only), d4 = d5 + d7 (applies to both numbers and percentages). CNS data are adapted from ref. 10, which also included connection data for major CNS divisions. Out of the 6,972 possible connections between the 84 ganglia (42 bilateral pairs) in PNS2, there are 3,486 possible node pairings. Of these possible node pairings, the evidence suggests that only 82 exist, and they form a total of 98 connections, of which 66 are unidirectional (A→B) and 16 are bidirectional (A↔B). Thus, 19.5% of connected node pairs maintain bidirectional connections, compared to 25.7% for CNS2. Expressed differently, 32.6% of PNS2 connections participate in bidirectionally connected node pairs, compared to 41% for CNS2 (10). Discussion Six principal findings emerge from our network analysis of bilateral intra-PNS macroconnectivity in rats. First, there is evidence for a network of axonal connections among PNS2 ganglia. Second, this network is relatively sparse, comprising an estimated 111 connections, out of 6,972 possible between 84 PNS ganglia (equating to a connection density of 1.6%). Third, 38.1% of the ganglia are isolated—lacking any known intra-PNS2 connections. Fourth, the MRCC hierarchy of local differentiation in the network has seven first-order modules. Fifth, only two hubs were identified: the bilateral myenteric plexus ganglia that control digestive tract functionality. Sixth, just 6.1% of connections are contralateral, while 19.5% of connected ganglion pairs exhibit bidirectional connectivity. These findings contrast sharply with those for the CNS2 neuronal network (10), where connection density = 11.6%, isolated nodes = 0.7%, first-order modules = 3; hubs = 46, contralateral connections = 32%, and 25.7% of connected region-pairs exhibit bidirectional connectivity. Currently, a broad neuronal network linking PNS ganglia is not widely acknowledged in vertebrate neuroanatomy (41–44), and based on our analysis the potential participation of more than a third of the 84 ganglia in this network remains to be determined. Nevertheless, our analysis suggests that a sparse, widely distributed intra-PNS network may exist, involving cranial and spinal sensory ganglia as well as paravertebral, prevertebral, and terminal autonomic ganglia (Table 1; Dataset S5, layer D). The potential existence of a broad intra-PNS neuronal network raises two key issues, one technical and the other conceptual. Technically, the limited number of systematic experimental connectional studies of PNS ganglia—especially as compared to CNS regions—suggests that the intra-PNS network may be more extensive than currently documented. Promising advances in tissue clearing, transparent embedding, sectioning, and block-face imaging (including immunofluorescence) offer new avenues for tracing nerve fibers throughout the body (45, 46), particularly in adult mouse, and evolving online neuroinformatics resources such as “SPARC—bridging the body and the brain” (47) offer tools for archiving, accessing, and analyzing the large amounts of data being generated. However, the possibility of false positive results due to the labeling of fibers-of-passage—nerve fibers that traverse ganglia without synapsing—remains a critical concern in pathway tracing experiments (9). Conceptually, the functional significance of an intra-PNS neuronal network remains unclear. One helpful strategy is to lesion computationally either individual nodes or node sets within the PNS2 subnetwork and then assess the impact on network topology, as we have done previously for the forebrain (28), rostral sector (8), and brain (9) subnetworks of the entire NS network. However, because the heaviest projections of PNS ganglia are to the CNS (in the case of sensory ganglia) and to other body systems (in the case of autonomic motor ganglia), a more fruitful approach may be to examine intra-PNS network function within this broader integrative context. Accordingly, the next logical steps in our systematic rat connectome analysis are to examine CNS and PNS together as a unified NS connectome and neuronal network, and finally to extend this framework to include interactions between the NS and other body systems—as the neurome. Materials and Methods The connectome database was systematically and expertly collated from the neuroanatomical literature (ref. 9). Cluster analysis was conducted using the MRCC method (30, 31). Module identification, visualization, and analysis, were performed as detailed in the text. Supplementary Material Appendix 01 (PDF) Dataset S01 (PDF) Dataset S02 (XLSX) Dataset S03 (XLSX) Dataset S04 (XLSX) Dataset S05 (PDF)