Somatosensory network functional connectivity differentiates clinical pain phenotypes in diabetic neuropathy

Aims/hypothesis The aim of this work was to investigate whether different clinical pain phenotypes of diabetic polyneuropathy (DPN) are distinguished by functional connectivity at rest. Methods This was an observational, cohort study of 43 individuals with painful DPN, divided into irritable (IR, n = 10) and non-irritable (NIR, n = 33) nociceptor phenotypes using the German Research Network of Neuropathic Pain quantitative sensory testing protocol. In-situ brain MRI included 3D T1-weighted anatomical and 6 min resting-state functional MRI scans. Subgroup differences in resting-state functional connectivity in brain regions involved with somatic (thalamus, primary somatosensory cortex, motor cortex) and non-somatic (insular and anterior cingulate cortices) pain processing were examined. Multidimensional reduction of MRI datasets was performed using a machine-learning approach to classify individuals into each clinical pain phenotype. Results Individuals with the IR nociceptor phenotype had significantly greater thalamic–insular cortex (p false discovery rate [FDR] = 0.03) and reduced thalamus–somatosensory cortex functional connectivity (p-FDR = 0.03). We observed a double dissociation such that self-reported neuropathic pain score was more associated with greater thalamus–insular cortex functional connectivity (r = 0.41; p = 0.01) whereas more severe nerve function deficits were more related to lower thalamus–somatosensory cortex functional connectivity (r = −0.35; p = 0.03). Machine-learning group classification performance to identify individuals with the NIR nociceptor phenotype achieved an accuracy of 0.92 (95% CI 0.08) and sensitivity of 90%. Conclusions/interpretation This study demonstrates differences in functional connectivity in nociceptive processing brain regions between IR and NIR phenotypes in painful DPN. We also establish proof of concept for the utility of multimodal MRI as a biomarker for painful DPN by using a machine-learning approach to classify individuals into sensory phenotypes. Graphical abstract