Functional Connectome Analyses Reveal the Human Olfactory Network Organization

Abstract

The olfactory system is uniquely heterogeneous, performing multifaceted functions (beyond basic sensory processing) across diverse, widely distributed neural substrates. While knowledge of human olfaction continues to grow, it remains unclear how the olfactory network is organized to serve this unique set of functions. Leveraging a large and high-quality resting-state functional magnetic resonance imaging (rs-fMRI) dataset of nearly 900 participants from the Human Connectome Project (HCP), we identified a human olfactory network encompassing cortical and subcortical regions across the temporal and frontal lobes. Highlighting its reliability and generalizability, the connectivity matrix of this olfactory network mapped closely onto that extracted from an independent rs-fMRI dataset. Graph theoretical analysis further explicated the organizational principles of the network. The olfactory network exhibits a modular composition of three (i.e., the sensory, limbic, and frontal) subnetworks and demonstrates strong small-world properties, high in both global integration and local segregation (i.e., circuit specialization). This network organization thus ensures the segregation of local circuits, which are nonetheless integrated via connecting hubs [i.e., amygdala (AMY) and anterior insula (INSa)], thereby enabling the specialized, yet integrative, functions of olfaction. In particular, the degree of local segregation positively predicted olfactory discrimination performance in the independent sample, which we infer as a functional advantage of the network organization. In sum, an olfactory functional network has been identified through the large HCP dataset, affording a representative template of the human olfactory functional neuroanatomy. Importantly, the topological analysis of the olfactory network provides network-level insights into the remarkable functional specialization and spatial segregation of the olfactory system.

Keywords: functional neuroanatomy; functional segregation; functional specialization; graph theory; network; olfaction.

Figure 1.

Procedures. A, Schematic illustration of analysis pipeline. (i) A total of 28 ROIs were defined. (ii) An automated procedure based on COV removed voxels contaminated with artifacts from the ROIs. (iii) Participant exclusion based on three exclusion criteria. (iv) Timeseries data extraction from the ROIs. (v) A 28 × 28 correlation matrix compiled based on pair-wise correlations across the ROIs. (vi) A binary adjacency matrix constructed with suprathreshold and subthreshold connections. (vii) Suprathreshold connections chosen to form the olfactory network. (viii) Graph-theoretical analyses performed to characterize the organization of this network. B, 3D display of ROIs before (top row) and after (bottom row) voxel removal. Insets illustrate the underlying ROIs in 3D whole-brain images with parts of dorsolateral frontal and temporal lobes removed. Omm, middle medial OFC; Opm, posterior medial OFC; Oc, central OFC; Oapc, anterior-APC OFC; Oml, medial lateral OFC; Oolfl = lateral olfactory OFC.

Figure 2.

The olfactory network. A, A weighted sparse 28 × 28 correlation matrix of group average Pearson’s rs for all suprathreshold pairs. ROIs included in the olfactory network are enclosed in the orange box, with the three identified modules (subnetworks) enclosed in the red boxes. The table lists the region names in correspondence to the ROI/node numbers. B, A transparent brain model (in sagittal and axial views) with ROIs (nodes) for the three modules coded in three respective colors. Gray nodes are ROIs not accepted into the olfactory network. C, A binary connectivity matrix reveals suprathreshold connections (shown in yellow) across the olfactory network nodes (22 parcels, enclosed in the orange box) and occipital visual cortical regions (28 parcels, enclosed in the cyan box) at three cutoff levels (top 5%, 10%, and 15%). The visual regions (Cal = calcarine gyrus; Cun = cuneus gyrus; Occ-I = occipital inferior gyrus; Occ-M = occipital middle gyrus; Occ-S = occipital superior gyrus) were strongly interconnected and relatively disconnected from the olfactory nodes.

Figure 3.

Local network metrics. A, Topology of the olfactory network. The three modules/subnetworks are indicated by the three colors of the circles. Line thickness indicates connection strength (mean correlation coefficients), and node size reflects connection density (number of connections). B, Modularity across a range of connection thresholds. Each row corresponds to one of the 28 ROIs, and columns indicate the connection threshold applied to the network while color indicates the module assignment. In general, nodes were consistently assigned to three modules identified as the limbic (red), sensory (yellow), and frontal (blue) subnetworks. At some connection thresholds, nodes were no longer connected to the network, which is indicated in gray. C, Hubness of a node as reflected by composite hub ranking and composite hub Z-scores. The three centrality indices (node degree, betweenness, and closeness centrality) are also displayed. The AMY and INSa separated from the other nodes as major hubs of the network. Changes in global efficiency of the olfactory network following a node removal were small except for the AMY and INSa nodes, which resulted in 8.5% and 7.3% reductions in global efficiency, respectively.

Figure 4.

Validation and function of the olfactory network. A, Weighted connectivity matrices of the olfactory network based on the HCP dataset and the independent dataset greatly overlapped; Spearman ρ = 0.41, p < 0.001. B, Module assignments across weighted networks of individual subjects. Each row corresponds to one of the 28 ROIs, and columns indicate individual subjects while color indicates the module assignment. The group level module assignment is provided to the far right for reference. Subjects are ordered based on the number of modules detected (left, two module subjects, n = 4; center, three module subjects, n = 21; right, four module subjects, n = 7) and beneath the module assignment matrix a key to module number is provided (purple). C, The global network metric of clustering coefficient was positively correlated with olfactory discrimination performance (d’), ρ = 0.32, *p < 0.05.