A prominent finding of postmortem and molecular imaging studies on Alzheimer's disease (AD) is the accumulation of neuropathological proteins in brain regions of the default mode network (DMN). Molecular models suggest that the progression of disease proteins depends on the directionality of signaling pathways. At network level, effective connectivity (EC) reflects directionality of signaling pathways. We hypothesized a specific pattern of EC in the DMN of patients with AD, related to cognitive impairment. Metabolic connectivity mapping is a novel measure of EC identifying regions of signaling input based on neuroenergetics. We simultaneously acquired resting-state functional MRI and FDG-PET data from patients with early AD (n = 35) and healthy subjects (n = 18) on an integrated PET/MR scanner. We identified two distinct subnetworks of EC in the DMN of healthy subjects: an anterior part with bidirectional EC between hippocampus and medial prefrontal cortex and a posterior part with predominant input into medial parietal cortex. Patients had reduced input into the medial parietal system and absent input from hippocampus into medial prefrontal cortex (p < 0.05, corrected). In a multiple linear regression with unimodal imaging and EC measures (F4,25 = 5.63, p = 0.002, r2 = 0.47), we found that EC (β = 0.45, p = 0.012) was stronger associated with cognitive deficits in patients than any of the PET and fMRI measures alone. Our approach indicates specific disruptions of EC in the DMN of patients with AD and might be suitable to test molecular theories about downstream and upstream spreading of neuropathology in AD.