In this study of relapse and recurrence in late-life depression, we found robust differences in network connectivity between relapse, stable remission, and heathy comparisons, differences that centered heavily on the DMN. The remitted LLD group as a whole exhibited lower within and between network connectivity of the DMN, visual, and SMN than healthy comparisons, with the exception of greater connectivity between DMN and ECN. Stable remission and relapse differed primarily by lower connectivity between DMN and SN and SMN in the relapse group. Overall, the connectivity of participants that went on to relapse was more similar to healthy comparisons than was the connectivity of the stable remitted participants.
Consistent with our first hypothesis, the remitted LLD group as a whole showed significant differences in network connectivity from healthy comparisons, reflecting a new homeostatic setpoint associated with remission. The DMN featured prominently in these differences, consistent with previous working implicating aberrant DMN connectivity in both adult and geriatric depression2,21. The finding of greater connectivity between the DMN and ECN in the remitted LLD group partially replicates findings in younger adults that connectivity between specific DMN-ECN regions was elevated in relapsing participants (though not in stable remitted participants)22,23. There is also evidence that DMN-ECN connectivity is higher in acutely depressed individuals21, while our previous review identified a positive association between DMN-ECN connectivity and antidepressant treatment response24. The coherence of the DMN and ECN appears to be a key feature of depression, though the precise role remains unclear. The SMN and visual network also showed lower within-network connectivity in LLD, which is consistent with findings from a large Chinese consortium16,17, though at least one small study has reported the opposite effect in the visual network25. Overall, remitted LLD participants show robust differences from healthy comparisons, supporting that stable remission from depression is not simply a “return to normal,” but is associated with a new configuration of the neural landscape.
The neural landscape also differed between stabled remitted and relapsed participants. DMN-SN connectivity was lower in relapsed participants than in stable remitted, and was also associated with time to relapse. This is consistent with a recent study in midlife depression that reported relapse was associated with lower connectivity between the DMN and portions of the SN26, but stands in contrast to another a recent finding (albeit in only 9 relapsed participants) showing higher FC between the right anterior insula (a SN hub) and the subcallosal cingulate (a DMN hub) predicted depression recurrence20. Interestingly, lower DMN-SN connectivity has also been reported in acutely depessed individuals27. Lower SMN connectivity to the DMN and ECN was also associated with relapse which is the first finding to our knowledge relating SMN connectivity to relapse and requires further exploration.
Contrary to our third hypothesis, participants who would go on to relapse displayed a functional connectivity profile more similar to healthy comparisons than the stable remitted participants did. This finding may indicate that significant reconfigurations of the neural landscape in context of successful antidepressant treatment are required to result in stable remission. There is wide-spread evidence for this reconfiguration across the DMN, ECN, and SN28–32. This type of dynamic network reconfiguration (as opposed to a return to baseline) is consistent with work in learning reversal, showing that new circuits that will override the learned behavior, rather than a reversal of the learned circuitry33. As a corollary, a partial return to baseline (i.e., failure to establish a new stable setpoint) may represent an unstable equilibrium and thus susceptibility to relapse, especially in the presence of other persistent alterations. This may be evidenced in our study by relapsed participants showing greater similarity of functional network organization to healthy comparisons than stable remitted participants.
Our final hypothesis—that the key networks of the LLD neural landscape are the DMN, ECN, and SN—appears to be mostly true. The DMN showed widespread and persistent differences between the groups, underscoring its crucial role in the neural underpinnings of late-life depression. Importantly, these differences did not lie just within the DMN, but often in the interaction between the DMN and other networks. The ECN and SN did not differ as robustly as the DMN, though their connectivity with the DMN served as important differentiators between healthy comparisons and LLD, and between stable remission and relapse, respectively. Connectivity of the SMN and visual networks also differed frequently between groups, adding to a growing body of evidence that these unimodal networks are also involved in depression17.
The prospective design and 4 month inclusion cutoff following remission are significant strengths of our study, providing ecological validity as evidenced by the excellent agreement with previously observed relapse rates of 43% within two years34. This study has several limitations. Our ability to assess the temporal stability of neural networks is hampered by analyzing only baseline neuroimaging. We do not have pretreatment imaging data in the LLD group. Sample size is moderate, although above average for neuroimaging of clinical populations and sufficiently powered for our analytic method that leverages omnibus testing to reduce dimensionality. Remitted participants did not receive uniform antidepressant treatment; while a general treatment algorithm was followed, medications were individually tailored. However, this resulted in a high rate of remission (73% across all three sites) and provides better generalization/clinical translation. The delineation between the stable remitted and relapsed groups was subject to right-censoring; while all participants had ≥ 8 months of follow up at the time of analysis, most participants had a longer duration (up to 2 years). We were insufficiently powered to analyze a uniform cutoff of relapse within 8 months, which would result in only 13 relapsed participants. Since relapse tends to occur sooner rather than later (e.g., only 29% of relapsed participants with two years of data relapsed after one year) and 85% of the participants had ≥ 1 year of follow up, we believe our inclusive approach represents a better approximation to the “true” relapse group that we would observe with 2 years of longitudinal data for all participants. In this study we chose to define the neural landscape in terms of within and between network connectivity using the canonical seven Yeo networks. While this choice is well-justified, there are models/evidence that support other meaningful organizational levels of inquiry (e.g., subnetworks or specific regions, especially subcortical regions). Restricting our analysis to 7 networks allowed for cortex-wide coverage while also minimizing the number of comparisons. This investigation is focused on resting state fMRI; future investigations will tests the effect of task-based fMRI or structural connectivity (e.g., white matter hyperintensities, diffusion imagining measures).
In conclusion, we identified robust differences in the functional connectome between healthy comparison and remitted participants in late-life, as well as differences between participants who relapsed and those who remained depression-free. These differences were apparent at the network level and were robust to parcellation scheme. These findings, consistent with our proposed disruption of neural homeostasis model of recurrence and relapse in late-depression, may be used to identify depressed older adults at higher risk of relapse and to adequately tailor preventative interventions to prevent the burden of additional depressive episodes.