Late-life depression (LLD) is a severe mental illness with a high risk of suicide. Characterized by heterogeneity of depressive symptoms and cognitive impairment, it is thought to arise from aberrant functioning within various brain networks, particularly with networks related to executive control, reward, self-referential cognition, as well as attentional processes involving switching and control (Menon, 2011; Szymkowicz et al., 2023). Among these networks associated with LLD, substantial evidence reported a significant correlation between the executive control network, particularly the dorsolateral prefrontal cortex (DLPFC), and deficits in executive functioning (Szymkowicz et al., 2023). Together with the DLPFC, the ventral lateral prefrontal cortex (VLPFC) has been demonstrated to connect to the caudate, thereby contributing to the brain reward system (Leh et al., 2007). Furthermore, a recent study using repetitive transcranial magnetic stimulation (rTMS) on VLPFC in healthy volunteers found attenuated activity in representative hubs of the subcortical affective system, which are the amygdala and insula (He et al., 2023). Alternatively, rTMS targeting both dorsal and ventral subregions of the lateral prefrontal cortex (LPFC) in LLD has been associated with reduced depression scores and improved remission rates (Kaster et al., 2018; Levkovitz et al., 2009). These findings suggest that delivering rTMS non-invasively to the LPFC holds considerable promise in treating depression.

Targeting the LPFC, rTMS is currently a clinical standard for treating patients with treatment-resistant major depression. However, its efficiency has been shown to have limited probability due to the suboptimal localization of LPFC targets (Cash et al., 2021b, Cash et al., 2021a; Luber et al., 2017). Clinically, the DLPFC target location is typically determined using the empirical 5-cm rule or the international 10/20 electroencephalogram scalp-based system (Cash et al., 2021a, Cash et al., 2021b). This approach omits the individual variations in head size, leading to the misplacement of stimulation targets, not only located in the prefrontal but also the motor and premotor cortices, subsequently reducing remission rates among patients undergoing TMS treatments (Ahdab et al., 2010; Herbsman et al., 2009). Furthermore, this scalp-based method was less effective than imaging-guided methods when tested in healthy volunteers (Sack et al., 2009). Of these imaging-guided TMS, efficacy was reported to improve progressively with targeting derived from group-based functional magnetic resonance imaging (fMRI), individual anatomical MRI, and individualized fMRI (Sack et al., 2009). The fMRI is a high-resolution and non-invasive brain-mapping method for localizing DLPFC during cognitive tasks like the Sternberg working memory task (Balderston et al., 2020a) or through resting state connectivity analysis (Fox et al., 2013). Through the systematic comparison, the individualized fMRI method demonstrated the highest effect size than the remaining imaging-guided targeting methods (Sack et al., 2009), suggesting a positive association between targeting precision for each individual and the treatment responses to TMS.

Beyond studying healthy volunteers, recent research conducted a randomized trial on depression patients, demonstrating significant spatial discrepancies in DLPFC targets between rTMS responders and non-responders (Rosen et al., 2021). Similarly, Cash et al. performed a retrospective study presenting a significant reduction in depressive scores correlated with the proximity of stimulated targets to individualized targets, which were identified using resting-state fMRI to localize the strongest temporal correlation with the subgenual anterior cingulate cortex (sgACC) (Cash et al., 2021a, Cash et al., 2021b). This approach has gained attention in depression treatment; however, substantial inter-individual variability in connectivity between ACC and DLPFC, along with the illness duration, weakens the reliability of DLPFC targeting via resting-state fMRI (Sheng et al., 2022). In contrast, task-based fMRI has been used to localize DLPFC in anxiety, eliminating seed-selection ambiguity (Balderston et al., 2020b). Despite this, there is limited research on individual DLPFC localization through task-based fMRI for depression treatment. Therefore, it is imperative to establish the reliability of this individual localization in depression patients, particularly among diverse depression subgroups.

Convergent evidence since 2012 has highlighted potential targets beyond the DLPFC for treating emotional functions, including dorsomedial prefrontal cortex, frontopolar cortex, ventromedial prefrontal cortex (VMPFC), and VLPFC (Downar and Daskalakis, 2013). Notably, VLPFC facilitation plays a pivotal role in mediating the VMPFC activity for attenuating amygdala activity and down-regulating negative emotions (He et al., 2023). Among LLD patients, its connectivity with caudate distinguished patients without suicidal risk from those with suicidal thoughts or past suicide attempts (Shao et al., 2021). This finding suggests the VLPFC as a promising target for LLD patients with high suicide risk, where currently no target consensus exists. To date, personalized DLPFC-targeting using resting-state fMRI is proliferated, but VLPFC localization remains under-explored for LLD patients. In contrast, task-based fMRI, incorporating numerical Stroop and face/shape matching tasks, has been designed to raise the brain activity of the LPFC (both dorsal and ventral) in older adults (Huang et al., 2012; Lin et al., 2019) and patients with late-life depression (Lin et al., 2019). Literature suggests that the brain functionality in LLD populations is heterogeneous (Sun et al., 2023), and the reliability of LPFC localization using the numerical Stroop and face/shape matching tasks remains to be investigated. While the numerical Stroop has not been explicitly examined in depression, a previous study demonstrated significant differences in VLPFC activation between subgroups of mid-life depression using the face/shape matching task (Townsend et al., 2010). This finding suggests that VLPFC localization in the face/shape matching task may be less reliable and influenced by anxiety severity. Therefore, we hypothesized that (1) task-based fMRI using the numerical Stroop task provides more reliable LPFC localization in LLD patients than the face/shape matching task, and (2) the reliability of individual-level LPFC localization varies across the heterogeneous LLD subgroups. This work aimed to systematically evaluate the reliability of these fMRI tasks for localizing the DLPFC and VLPFC among LLD patients with varying degrees of suicidality, considering that suicide's devastating consequence in this demographic.