Hypothesis-driven structural connectivity analysis supports network over hierarchical model of brain architecture

Abstract

The brain is usually described as hierarchically organized, although an alternative network model has been proposed. To help distinguish between these two fundamentally different structure-function hypotheses, we developed an experimental circuit-tracing strategy that can be applied to any starting point in the nervous system and then systematically expanded, and applied it to a previously obscure dorsomedial corner of the nucleus accumbens identified functionally as a “hedonic hot spot.” A highly topographically organized set of connections involving expected and unexpected gray matter regions was identified that prominently features regions associated with appetite, stress, and clinical depression. These connections are arranged as a longitudinal series of circuits (closed loops). Thus, the results do not support a rigidly hierarchical model of nervous system organization but instead indicate a network model of organization. In principle, the double-coinjection circuit tracing strategy can be applied systematically to the rest of the nervous system to establish the architecture of the global structural wiring diagram, and its abstraction, the connectome.

The starting point that we chose for neural systems analysis is a 1-mm³ region of brain tissue in which injected μ-opioids increase ingestive behavior, possibly by increasing the hedonic impact of sweetness (1). The identified region, which has been called a hedonic hot spot, lies dorsomedially in the nucleus accumbens (ACB), a part of the basal ganglia's ventral striatum long thought to play a role in controlling foraging behavior and the rewarding properties of food or other natural rewards and substances of abuse (2, 3). This functional heterogeneity of the ACB suggested to us that the ACB's extrinsic axonal connectivity is also topographically organized and specialized to a greater extent than demonstrated previously (4). Thus, we used the dorsomedial ACB (ACBdm) as a circumscribed, readily identifiable anchor point for our hypothesis-driven structural connectivity analysis.

Results

We first established differential extrinsic axonal inputs and outputs of the ACBdm dorsal tip (ACBdmt) with a new double-coinjection (COIN) network tracing strategy. In the same animal, two tiny, nonoverlapping anterograde/retrograde tracer COINs—mixtures of biotinylated dextran amine/Fluorogold (BDA/FG) and Phaseolus vulgaris leucoagglutinin/cholera toxin subunit b (PHAL/CTb)—were placed stereotaxically in the medial ACB (Figs. 1 A–C and Fig. S1). The sources of extrinsic axonal inputs specifically to ACBdmt in these double-COIN topography experiments (Fig. 1A; the filled purple area is representative) were simpler that those described for the whole ACB (4, 5). Only three major cerebral hemisphere regions were retrogradely labeled with CTb (Fig. 2): (i) the cortical infralimbic area (ILA; Brodmann area 25) (Fig. 1D), where deep electrical stimulation relieves symptoms in chronically depressed patients (6); (ii) the hippocampal ventral subiculum (SUBv), which modulates hypothalamic stress responses (7) and also projects directly to ILA (8); and (iii) the amygdalar basomedial nucleus (BMA), which receives olfactory information (9) and a massive input from the SUBv (10). Only two major brainstem inputs were identified: the dorsal raphé (DR), the presumptive source of ACBdmt serotonin (11), and the interfascicular nucleus (IF), the presumptive source of ACBdmt dopamine (12) that is medial to the expected ventral tegmental area (VTA) itself (13). Thus, the ACBdmt receives direct inputs from three massively interconnected cerebral regions implicated in stress and depression, as well as restricted brainstem sites providing dopaminergic and serotonergic terminals.

Fig. 1.

Direct comparison of ACB projection specificity and topography. Pathway tracer double COINs were made in different regions entirely within the ACB, allowing the direct comparison of topographic organization of both input and output projections from the two single COINS. (A) Injection site distribution for nine double-COIN experiments (18 COINs), summarized on reference atlas templates (32). PHAL/CTb tracer injections are outlined in purple, and BDA/FG COINs are outlined in orange; filled circles indicate the double-COIN experiment illustrated in B and C. (Top, Right) Each atlas level (AL) and the corresponding distance from the bregma are indicated. (B) Quadruple-labeled confocal image of a double-COIN experiment involving the medial (shell) nucleus accumbens. The background image was derived from a copy of the BDA channel and added to the composite as a separate layer (Fig. S1). For all confocal images (Figs. 1 B–F′′ and 3 B–G), grayscale channels were false-colored according to the tracer imaged; PHAL is always shown in red, CTb in magenta, BDA in green, and FG in cyan. (C) Selective presentation of one tracer from each single COIN shown in B (PHAL, red; BDA, green). (D–F) Comparison of the projections of the medial ACB, labeled by the injections shown in B. (D) Confocal image of CTb and FG neuronal labeling in ILA (AL9) (Inset, AL9, lower left). (E) Confocal image of PHAL and BDA axonal labeling in SI (AL17). (F) Dark-field photomicrograph of ACBdmt projection to LHAa (AL25). In this series of histological sections, tracers were labeled with diaminobenzidine (DAB) (PHAL) and nickel-intensified DAB (BDA) to generate high-contrast, low-power images. Because distinguishing between the brown and black reaction products at this magnification is impossible, a fluorescent companion series was prepared from adjacent sections. Green and red arrows indicate the corresponding groups of BDA- and PHAL-labeled axons illustrated at higher magnification in F′ and F′′. Note that BDA-labeled axons from the ventromedial ACB (F′; green) descend laterally to the projection from the ACBdm in tightly bundled fascicles with few boutons, suggesting virtually no input at this level. In contrast, the PHAL-labeled axons from the ACBdmt (F′′; red) generate a significant input, as indicated by the frequent branches with prominent boutons. aco, anterior commissure olfactory part; CP, caudoputamen; fx, fornix; opt, optic tract; sm, stria medullaris; VL, lateral ventricle. (Scale bars: B–C and F, 100 μm; D, F′, and F′′, 25 μm.)

Fig. 2.

Structural organization of ACBdmt-related neural circuitry. The four nodes of a closed loop (ACBdm > LHAa > PTa/PVTa > ILA > ACBdm; colored circles on left) are emphasized as a test bed for the experimental double-COIN network tracing strategy used to identify the origin, course, and termination of each pathway described in the text. Note a triple descending projection from the cerebral hemisphere (29) to LHAa: excitatory from ILA, inhibitory from ACBdmt, and disinhibitory from SIdmtr. Also note three major LHAa outputs, to (1) BSTamg and ADP, which innervate regions controlling metabolism (including ACTH and glucocorticoid responses) and feeding behavior; (2) parts of the fight-or-flight defensive behavior system, the PMd and PAG (especially the precommissural and commissural nuclei and the dorsomedial and ventrolateral columns); and (3) parts of the behavioral state control system (VTA, IF, and DR). Finally, note (4) feedback from motor output to sensory input (20). Brain part abbreviations are given in the text, and evidence for putative neurotransmitters not documented in the text (5-HT, serotonin; da, dopamine; gaba; glu, glutamate; ne, norepinephrine) is provided in SI Materials and Methods.

Only one ACBdmt axonal output was labeled with PHAL (Fig. 1A; the filled purple area is a representative injection site): a descending pathway through the medial forebrain bundle with two clear terminal fields. As expected (4), one field was highly restricted to a small oval patch in a rostrolateral sector of the substantia innominata (SI; also called the ventral pallidum) (Fig. 1E, red axons), referred to here as the ACBdmt-recipient zone (Sidmtr) (Fig. 2). Surprisingly, virtually no input was labeled to the VTA (Fig. 3), the other major ACB target. Instead, a large, clear, circumscribed terminal field was labeled in, and restricted to, a rostromedial sector of the anterior lateral hypothalamic area (LHAa) (Fig. 1 F and F′). This result was confirmed in two ways. First, all ACB COINs ventral to the ACBdmt labeled a clear projection to VTA (Fig. 1 A, C, and F′); second, very large retrograde pathway tracer implants centered in the VTA extensively labeled the ACB except in the ACBdmt, whereas another retrograde tracer placed in the LHAa of the same animal labeled the ACBdmt (Fig. 3).

Fig. 3.

Differential projections from the medial ACB to the lateral hypothalamic area and the ventral midbrain. (A–D) Fluorescence photomicrographs of retrograde neuronal labeling in the medial ACB (A, Inset; B) resulting from multiple tracer injections in the LHAa (C, Inset, AL25) and ventral midbrain (D; AL37) in the same animal. To specifically label terminating axons in the LHA, the retrograde tracer Fluorogold (FG) was used for injections, because it is largely resistant to uptake by undamaged fibers of passage. To maximally label all projections in the VTA and medially adjacent parts of the substantia nigra, the retrograde tracer True Blue (TB) was used, because it is avidly taken up by terminating and passing fibers. In addition, TB was delivered as desiccated “crystals” of tracer mechanically ejected from glass pipettes with tip diameters of up to 500 μm (D). The typical pattern of retrograde neuronal labeling produced by these combined injections is shown in A, where the bright-yellow FG-filled neurons are clustered in the ACBdmt, whereas all other parts of the striatum contain numerous intensely blue TB-filled neurons projecting to the ventral midbrain. The approximate location of the labeling is shown schematically in the insets in A and also in a low-power (2.5×) dark-field photograph of the same section in B. The arrows in A and B indicate features recognizable in both images. ACBdmt, nucleus accumbens dorsomedial tip; aco, anterior commissure olfactory part; ccr, rostrum of corpus collosum; CP, caudoputamen; cpd, cerebral peduncle; fr, fasciculus retroflexus; fx, fornix; MB, mammillary body; opt, optic tract; PVHpm, paraventricular nucleus of hypothalamus, posterior magnocellular part; SNr/c, substantia nigra, reticular and compact pars; V3, third ventricle. (Scale bars: A, 100 μm; B–D, 200 μm.)

We next established the global axonal outputs of the newly identified LHAa region targeted by the ACBdmt. Single COINs of each type were used for this, because retrograde labeling concentrated in ACBdmt confirmed accurate LHAa injection placement (Figs. 2 and 4 A and B). Three long projections were labeled with PHAL and confirmed with BDA (Fig. 2). Ascending LHAa axons innervate various parts of the septal region, including the BST anteromedial group (BSTamg), with outputs to regions controlling feeding behavior and metabolism, including the release of ACTH and glucocorticoids (14); the SIdmtr (a bidirectional connection due to the presence of intermixed retrograde labeling); and adjacent hypothalamic neuron populations, including the anterodorsal preoptic nucleus (ADP), a critical node in the hypothalamic visceromotor pattern generator network (15). A dorsally directed hypothalamo > thalamic pathway innervates restricted adjacent parts of the anterior paratenial and paraventricular nuclei (PTa/PVTa) (Fig. 4C), rostral mediodorsal nucleus (MDr), and lateral habenula (LH) (Fig. 3D). In addition, a descending medial forebrain bundle pathway innervates several regions forming critical components of the fight-or-flight defensive behavior system (16)—the dorsal premammillary nucleus (bilaterally) (Fig. 4E) and several periaqueductal gray (PAG) components (Fig. 2)—and multiple components of the behavior state control system, including the VTA, IF, and DR.

Fig. 4.

Axonal projections from the ACBdmt-recipient part of the LHAa. Single COINs were stereotaxically placed in this region (A, Inset, AL25; compare with Fig. 1F) and confirmed by the distribution of retrograde neuronal labeling in the ACB (B; AL13). Large injections like the BDA-FG experiment shown in A were acceptable, because the ACBdmt projection has a discrete origin and circumscribed terminal region in the LHA. All injections showing this pattern also generated dense axonal projections to the thalamic paratenial nucleus (C; AL23), lateral habenula (D; AL32), and bilateral dorsal premammillary nucleus (E; AL33), and lighter projections listed in the text. The arrows in B indicate the medial border of the ACB. The arrows in C and E indicate the midline. aco, anterior commissure olfactory part; CP, caudoputamen; fx, fornix; isl, island of Calleja; LH, lateral habenula; MH, medial habenula; opt, optic tract; PMd, dorsal premammillary nucleus; PT, paratenial nucleus; PVT, thalamic paraventricular nucleus; sm, stria medullaris; SO, supraoptic nucleus; V3m, third ventricle mammillary recess; V3t, third ventricle thalamic part; VL, lateral ventricle. (Scale bars: 100 μm.)

To confirm and extend these foundational ACBdmt-related connections, we used a second double-COIN strategy with injections placed in opposite corners of a hypothesized four-node circuit (Fig. 2, colored circles). In the first configuration, PHAL/CTb was iontophoresed into the ILA, and BDA/FG was iontophoresed into the LHAa (Fig. 5G and Fig. S2 A and B). These experiments confirmed an ILA > ACBdmt corticostriatal projection with PHAL anterograde tracing (Fig. 1D) and an overlapping ACBdmt > LHAa striatohypothalamic projection with FG retrograde tracing (Figs. 4B and 5E). The experiments also extended earlier results by providing direct structural support* for a closed forebrain chain involving the ACBdmt, LHAa, PTa/PVTa, and ILA (Fig. 5H). The ILA and LHAa COINs are in opposite or alternate nodes of a potential four-link circuit, and when the two COINs are accurately targeted in a single animal, overlapping anterograde-retrograde labeling was observed in the uninjected nodes, the ACBdmt (Fig. 4F) and PTa/PVTa (Fig. S2C). Therefore, the evidence thus far suggests (see Fig. 2) that the ACBdmt is part of a closed cortico > striato > hypothalamo > thalamo > cortical loop; that this loop receives direct input from subdivisions of hippocampus, amygdala, and brainstem aminergic groups; and that this circuitry provides output to hypothalamic stress and ingestive behavior systems (from the ventral subiculum and BSTamg), the fight-or-flight system (from the LHAa), and widespread parts of the hypothalamus, PAG, and visceral sensorimotor system [from the ILA (17), confirmed here with PHAL].

Fig. 5.

All major ACBdmt axonal projections contribute to a single thalamo > cortico > striatal closed chain (circuit). Shown are analyses of circuit connections based on placement of matched pairs of COINs (A and B) (Fig. S2 A and B) and the resulting interactions between labeled tracers (C–F) (Fig. S2C). The general strategy for circuit tracing involves hypothesizing specific structural relationships between a limited number of known projections. This four-node loop (G) can then be tested with COIN pairs (red circles) in each of the two possible sets of indirectly connected, or “opposite,” nodes (X/Y or A/B) to detect the presence or absence of interactions (blue circles), indicated by overlapping anterograde and retrograde labeling in both “adjacent” nodes. If opposite nodes are connected, then both X/Y COINs will contribute to interactions in each of the A/B nodes, whereas the complementary COIN pair in A/B will label corresponding interactions in nodes X/Y. Because an ILA > ACBdm > LHAa > PTa/PVTa > ILA circuit was hypothesized (H), the X/Y double-COIN experiment targeted PTa/PVTa (A, Inset, AL23) and ACBdm (B; AL13), predicting contacts in the LHAa (C; AL25) and ILA (D; AL9). The complementary A/B pair corresponds to COINs in ILA and LHAa (Fig. S2 A and B) and shows robust interactions not only in the PTa (Fig. S2C) and ACBdmt (E; AL13), but also in SI/ventral pallidum (F; AL16). In A–F, the yellow letters in the upper right refer to corresponding positions in the generic circuit (G). Panels showing tracer interactions also include the relationship to adjacent nodes and, by extension, the direction of tracer transport, indicated by arrows. In all panels, PHAL is shown in red, CTb in magenta, BDA in green, and FG in cyan. CP, caudoputamen; VL, lateral ventricle; sm, stria medullaris; V3t, third ventricle thalamic part. (Scale bars: A and B, 100 μm; C, 5 μm; D–F, 25 μm.)

As a control, the injection sites in these double-COIN circuit-tracing experiments were “rotated” 90 degrees, so that double COINs were now aimed for ACBdmt and PTa/PVTa (Fig. 5 A, B, and H). As hypothesized, mixed anterograde-retrograde labeling was observed in the uninjected LHAa and ILA (Fig. 5 C and D), independently confirming results of the alternate COIN injection scheme for a closed forebrain chain at the level of confocal microscopy. But more interestingly, the rotated double-COIN experiment revealed that the closed chain also incorporates a “classical” (18) cortico > striato > pallido > thalamo > cortical closed chain involving the SIdmtr (Fig. 2). That is, overlapping anterograde-retrograde labeling was observed in the SIdmtr (Fig. 5D), labeled by the same COINs that generated the specific interactions in both the LHAa and ILA (Fig. 5 C and D). The resulting configuration, a bifurcated circuit or closed chain with one parallel node (Fig. 5H, Right), requires further structural characterization (e.g., with double COINS in the SIdmtr and LHAa), as well as careful functional characterization within the context of the network outlined in Fig. 2.

Two more features of this circuitry are informative. First, the massive ACBdm > LH projection is critical, because the LH then projects (19) to three behavioral state system components (20, 21) with ascending forebrain projections (VTA, IF, and DR) (Fig. 2). Thus, the LHAa > LH projection may be part of a feedback loop to the behavioral state system representing in part a negative reward/motivational value (22), whereas the LHAa > PTa/PVTa projection is part of a feedback loop to the cerebral cortex. Second, the viscerosensory nucleus of the solitary tract (NTS) projects directly to the behavioral state system (23, 24), the PTA/PVTa, and the ILA (Fig. 2), the latter two of which are noradrenergic (25, 26). Aside from the pharmacological evidence of a central noradrenergic influence on stress, depression, and appetite (27), this is interesting because the ILA projects directly to the NTS, providing the substrate for a possible very long cortico > NTS feedback loop (Fig. 2).

Discussion

Our findings establish the overall structural organization of a unique neural network that specifically involves the ACBdmt, is relatively simple by limbic forebrain standards, and thus is relatively amenable to experimental functional analysis, in part because network nodes typically are quite localized and restricted to subdivisions of well-known brain regions. Unexpectedly, the ACBdmt projects heavily to the LHAa but not to the VTA, and its dopaminergic input arises not from the VTA, but apparently from the medially adjacent IF instead. Furthermore, although inputs to the ACBdmt from the ILA, BMA, and SUBv have been demonstrated previously, other reported accumbens inputs specifically avoid the ACBdmt (4, 5). Finally, in the context of the ACBdm as a hedonic hot spot, other relevant SUBv outputs (Fig. 2) are to medial hypothalamic regions involved in stress and ingestive behavioral responses, both directly and through the ventral lateral septal nucleus and BSTamg (10, 14, 28), as well as to the BMA (10).

Our double-COIN strategy directly suggests that the ACBdmt is part of a large, previously unappreciated bifurcated feedback circuit in the forebrain (Fig. 5H). One core feature is the inclusion of all four components of a classical cortical > basal ganglia > thalamo > cortical circuit, confirming that this motif applies generally to cerebral functions as diverse as eye movement control (18) and affective tone (29). Another core feature is the circuit bifurcation, which allows alternate, parallel routes between the ACBdmt and PTa/PVTa, one through the SIdmtr (ventral pallidum) and the other through the LHAa. The thalamic PTa/PVTa is critically placed because it receives GABAergic input from the SIdmtr and presumably glutamatergic input from the LHAa (30) (SI Materials and Methods), as well as noradrenergic inputs from the viscerosensory NTS and a major input from the circadian rhythm generator suprachiasmatic nucleus (25, 31).

The double-COIN strategy also labels all other known axonal inputs and outputs of injected gray matter regions, indicating that circuitry involving the ACBdm extends through the nervous system as a longitudinal series of circuits (Fig. 2). Because “higher” and “lower” are difficult to define objectively in this organization, a network model of organization (e.g., the Internet) is clearly favored over a hierarchical model (e.g., the US Army) (20). Further evidence to clarify this distinction can be obtained by simply expanding systematically the double-COIN analysis started here to eventually include all gray matter regions of the nervous system. Double-COIN analysis is in fact a hypothesis-driven, internally controlled, experimental strategy for mapping systematically the nervous system's global wiring diagram, which can be abstracted to a formal connection matrix known as the connectome. In principle, this can be done at successively greater levels of resolution, ranging from the macroconnection level of gray matter regions (as done here) to the mesoconnection level of neuron type and the microconnection level of individual neurons with all of their synaptic relationships. For example, connection analysis at the level of neuron type is required to determine whether the ACB connections to GABAergic neurons in the SI and glutamatergic neurons in the LHA arise from a single neuron population with branched axons or from two ACB populations, one to the SI and the other to the LHA. The same strategy is then applied systematically to adjacent nodes in the circuitry.

Materials and Methods

All coinjection experiments involved two central iontophoretic injections, each containing one anterograde and one retrograde tract tracer (four different tracers in all) combined in solution and delivered via the same pipette. All tracers, either individually or in various combinations, were immunolabeled and visualized with either colorimetric or fluorescence methods. Fluorescent signals were imaged with confocal scanning laser microscopy configured in a multitrack format and acquired as separate channels in a single file. Individual channels were consistently false-colored by a tracer. Fig. 1B is a composite image of six panels, each a projection of nine slices for each of the four tracers. Adobe Photoshop CS3 was used to assemble the final image and to adjust the brightness and contrast of all images. The background image in Fig. 1B was created by duplicating the individual channels of the BDA tracer labeling and applying a second, extreme adjustment of brightness and contrast. Summary and schematic diagrams were created using Adobe Illustrator CS3. Additional details are provided in SI Materials and Methods.