Optimal deep brain stimulation sites and networks for stimulation of the fornix in Alzheimer’s disease

Deep brain stimulation (DBS) to the fornix is an investigational treatment for patients with mild Alzheimer’s Disease. Outcomes from randomized clinical trials have shown that cognitive function improved in some patients but deteriorated in others. This could be explained by variance in electrode placement leading to differential engagement of neural circuits. To investigate this, we performed a post-hoc analysis on a multi-center cohort of 46 patients with DBS to the fornix (NCT00658125, NCT01608061). Using normative structural and functional connectivity data, we found that stimulation of the circuit of Papez and stria terminalis robustly associated with cognitive improvement (R = 0.53, p < 0.001). On a local level, the optimal stimulation site resided at the direct interface between these structures (R = 0.48, p < 0.001). Finally, modulating specific distributed brain networks related to memory accounted for optimal outcomes (R = 0.48, p < 0.001). Findings were robust to multiple cross-validation designs and may define an optimal network target that could refine DBS surgery and programming.


Inclusion Criteria Exclusion Criteria
7. The subject is currently taking a stable dose of cholinesterase inhibitor (AChEI) medication for at least 60 days. 7. Current alcohol or substance abuse as defined by Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, Text Revision (DSM-IV-TR).
9. Contraindications for MRI scanning, including implanted metallic devices (e.g., non-MRI-safe cardiac pacemaker or neurostimulator; some artificial joints metal pins; surgical clips; or other implanted metal parts), or claustrophobia or discomfort in confined spaces.
10. Abnormal lab results that, in the opinion of the investigator and/or enrollment review committee, would preclude participation in the study.
11. Abnormal cardiovascular or neurovascular disorder that, in the opinion of the investigator and/or enrollment review committee, would preclude participation in the study.
12. Unstable doses of any medication prescribed for the treatment of memory loss or Alzheimer's disease.
13. Currently prescribed any non-AD medications that, in the opinion of the investigator and/or enrollment committee, would preclude participation in the study.
14. Is unable or unwilling to comply with protocol followup requirements.
15. Has a life expectancy of < 1 year.
16. Is actively enrolled in another concurrent clinical trial.
Supplementary Table 3. Demographic and clinical data of the patients included. Clinical outcomes measured by Alzheimer's Disease Assessment Scale 11 -cognitive subscale (ADAS-cog 11). Absolute change calculated subtracting 12-month ADAS-cog 11 value from Baseline ADAS-cog 11. Relative change calculated by dividing absolute changes by pre-operatory score and multiplying by 100. Resting-state fMRI data from 1,000 health subjects (average 1.7 runs per subject)

Final model parameters of DBS fiber filtering
During the training phase of model optimization in the DBS fiber filtering analysis, a variety of parameters were tested with the aim to create a tract-set that was i) robustly predictive during cross-validation within the training set (leave-one-out and several k-fold strategies were interactively tested) and ii) was not robust to permutations of improvement data (also see fig. 4). The following set of parameters were finally selected and used to cross-predict outcomes in the test-cohort.
Supplementary This setting was used since it follows the same logic as in sweetspot and network mapping approaches also used here.

Supplementary Methods
Narrative section of methods / predictive models: In all three models, each patient contributed their relative improvement of ADAS-cog-11 scores (before surgery, one year after surgery).
Beyond that, each model (i) tracts, ii) sweetspots and iii) functional networks) was run independently from one another.
-i) For tracts, each patient contributed the peak E-field amplitude that each tract of the normative connectome was modulated by. -ii) For sweetspots, each patient contributed the modeled electric field in MNI space (represented as a NIfTI volume). -iii) For functional networks, each patient contributed a (normative) rs-fMRI map seeding from the individual patient ("connectivity fingerprints").
Then, the three models created a i) combination of tracts ii) optimal target (sweetspot), and iii) functional network profile associated with optimal clinical improvements.
-i) For tracts, this was achieved by rank correlating the modulation amplitude imposed on each tract with clinical improvements across the set of patients. This led to an R-value for each tract, denoting how well its modulation correlated with clinical improvements (the concept was introduced in Irmen et al. Finally, data was cross-validated within the three models: -i) For tracts, this was achieved by rank correlating the impacts of the E-Fields of an unseen patient on all tracts and their R-values This led to a fiberscore denoting how specifically an unseen E-Field modulated tracts associated with optimal outcomes (the concept was introduced in Horn et al. 2022 PNAS). -ii) For sweetspots, this was achieved by spatially correlating the E-Fields of an unseen patient with the R-map model. This led to a sweetspot score denoting correlation coefficients of agreement between the actual stimulation field and an "optimal" stimulation field (represented by the R-map; the concept was introduced in Horn et al. 2022 PNAS). -iii) For functional networks, this was achieved by spatially correlating the functional connectivity fingerprints with the R-map model. This led to a network score denoting correlation coefficients of agreement between the actual network profile and an optimal network profile (represented by the R-map; the concept was introduced in Horn et al. 2017 Annals of Neurology).

Data Availability
Anonymized derivatives of stimulation data used for the described analyses are openly available on OSF (https://osf.io/bckuf). The resulting tract atlas, sweet spot and fMRI network pattern are openly available within Lead-DBS software (www.lead-dbs.org).

Code availability
All code used to analyze the dataset is openly available within Lead-DBS/-Connectome software (https://github.com/leaddbs/leaddbs). Code to reproduce main results and figures is openly available on OSF (https://osf.io/bckuf).