Full text Figures and data Side by side Abstract Editor's evaluation Introduction Results Discussion Methods Data availability References Decision letter Author response Article and author information Metrics Abstract Deep brain stimulation targeting the posterior hypothalamus (pHyp-DBS) is being investigated as a treatment for refractory aggressive behavior, but its mechanisms of action remain elusive. We conducted an integrated imaging analysis of a large multi-centre dataset, incorporating volume of activated tissue modeling, probabilistic mapping, normative connectomics, and atlas-derived transcriptomics. Ninety-one percent of the patients responded positively to treatment, with a more striking improvement recorded in the pediatric population. Probabilistic mapping revealed an optimized surgical target within the posterior-inferior-lateral region of the posterior hypothalamic area. Normative connectomic analyses identified fiber tracts and functionally connected with brain areas associated with sensorimotor function, emotional regulation, and monoamine production. Functional connectivity between the target, periaqueductal gray and key limbic areas – together with patient age – were highly predictive of treatment outcome. Transcriptomic analysis showed that genes involved in mechanisms of aggressive behavior, neuronal communication, plasticity and neuroinflammation might underlie this functional network. Editor's evaluation This study presents useful structural and functional connectivity profiles of patients receiving deep brain stimulation in the posterior hypothalamus for severe and refractory aggressive behavior. The inclusion of data from multiple centers is compelling. This study will be important for a broad readership including basic and clinical neuroscientists. https://doi.org/10.7554/eLife.84566.sa0 Decision letter Reviews on Sciety eLife's review process Introduction Aggressive behaviors are highly prevalent among psychiatric patients, presenting a major obstacle to patient care. In addition to suffering, these symptoms constitute a leading cause for institutionalization Gouveia et al., 2021a; Brentani et al., 2013; Gouveia et al., 2019. Standard treatments for aggressive behaviors involve behavioral and pharmacological therapies that mainly act on the dopaminergic and serotonergic systems (e.g. serotonin reuptake inhibitors, antipsychotics) Gouveia et al., 2021a; Brentani et al., 2013; Gouveia et al., 2019⁠. Despite their efficacy, a substantial proportion (30%) of patients fail to respond and are considered to be treatment refractory Adler et al., 2015; Gouveia et al., 2020; Gouveia et al., 2021b. For these patients, neuromodulation therapies, such as deep brain stimulation (DBS), have been investigated Gouveia et al., 2019; Gouveia et al., 2021c; Contreras Lopez et al., 2021; Benedetti-Isaac et al., 2015; Torres et al., 2020; Micieli et al., 2017; López Ríos et al., 2023. DBS is a neurosurgical therapy in which implanted electrodes are used to adjustably deliver electrical current to specific brain targets Jakobs et al., 2019. It is an established therapy for Parkinson’s Disease, dystonia, essential tremor, epilepsy Salanova et al., 2021 and obsessive-compulsive disorder Hamani et al., 2014; Li et al., 2020, showing promising results for the treatment of several other neuropsychiatric disorders, including Alzheimer’s disease Germann et al., 2021a; Lozano et al., 2016, depression, anorexia nervosa, addiction, and posttraumatic stress disorder Elias et al., 2021; Lozano et al., 2008; De Vloo et al., 2021; Elias et al., 2022; Hamani et al., 2022; Davidson et al., 2022. To date, DBS for refractory aggressive behavior has primarily targeted the posterior hypothalamic region Gouveia et al., 2019; Gouveia et al., 2021c; Contreras Lopez et al., 2021; Benedetti-Isaac et al., 2015; Torres et al., 2020; Micieli et al., 2017; López Ríos et al., 2023. The hypothalamus is a diencephalic structure with well-established roles in the control of homeostasis and motivated behaviors and is a key area in a broader neurocircuitry regulating aggressive behavior that also involves the orbitofrontal cortex, hippocampus, amygdala and periaqueductal gray Dudás, 2013; Blair, 2016; Miczek et al., 2007. Along the anterior-posterior axis, the hypothalamus can be divided into three regions (i.e. supraoptic or anterior, tuber cinereum or medial and supramamillary or posterior) with distinct cell types, projections and functions Dudás, 2013; Beattie et al., 1930. While the anterior region is mainly involved in thermoregulation and the control of circadian rhythms, the medial region regulates feeding and sexual behavior and plays a critical role in several endocrine and autonomic processes Dudás, 2013; Hoff, 1950; Flament-Durand, 1980. The posterior hypothalamus (pHyp) is an ergotropic area involved in the generation of sympathetic responses Hess, 1945. It provides robust projections to the midbrain and reticular formation via the hypothalamotegmental tract, thus being critical in the regulation of wakefulness and stress responses Beattie et al., 1930; Lechan and Toni, 2016; Saper and Lowell, 2014. Indeed, lesions (e.g. hamartomas and gliomas) located at the pHyp have been reported to cause apathetic and somnolent symptoms, while the selective neurosurgical ablation of a small portion of the pHyp has been successfully used to treat patients with severe treatment-resistant aggressive behavior (for a detailed review, see Gouveia et al., 2019; Sano et al., 1966; Sano and Mayanagi, 1988). More recently, DBS of the pHyp (pHyp-DBS) has also been shown to reduce aggressive symptoms in humans, with variable outcomes Gouveia et al., 2021c; Contreras Lopez et al., 2021; Benedetti-Isaac et al., 2015; Torres et al., 2020; Franzini et al., 2010; Torres et al., 2013; López Ríos et al., 2023. At present, only 21patients diagnosed with autism spectrum disorder (ASD), intellectual disability, obsessive-compulsive disorder, epilepsy and schizophrenia with ages ranging from 10 to 51 years have been reported, mostly as individual cases or small case series. In these patients, long-term improvement (up to several years of follow-up) in aggressive behaviors was in the order of 38–100% compared to baseline Gouveia et al., 2021c; Contreras Lopez et al., 2021; Benedetti-Isaac et al., 2015; Torres et al., 2020; Micieli et al., 2017; López Ríos et al., 2023. Although case studies and small series provide valuable insight into the safety and therapeutic impact of pHyp-DBS in individual patients, they do not allow the characterization of clinical phenotypes, optimal stimulation targets or brain networks underlying treatment. In this work, we gathered data from the largest international multi-centre dataset of patients treated to date with pHyp-DBS for aggressive behaviors to retrospectively investigate possible neurobiological mechanisms of action. Combining a well-documented electrode localization and volume of activated tissue (VAT) modeling pipeline (https://www.lead-dbs.org/) with probabilistic sweet-spot mapping Elias et al., 2021; Dembek et al., 2017, normative connectomics Fox, 2018; Elias et al., 2020; Germann et al., 2021b, and transcriptomics analysis Gouveia et al., 2021a; Mroczek et al., 2021 (https://alleninstitute.org/), we delineated a potentially ‘optimized’ surgical target and identified the brain networks and underlying neurobiological processes that might underpin successful pHyp-DBS. Demographic data, pre-and post-operative magnetic resonance imaging (MRI) and computed tomography (CT) scans were obtained from each participant for DBS lead localization, followed by an estimation of the VAT and the determination of the most effective region of stimulation. The VATs were further processed for connectomic analyses to investigate the structural (i.e. fiber tracts) and functional (i.e. brain areas) maps associated with symptom improvement. Using demographics and individual functional connectivity, we tested a predictive model of improvement following pHyp-DBS. Finally, we investigated genes with a spatial pattern of distribution similar to the functional connectivity map to explore associated biological processes. A graphical summary of the methodology used in this study can be found in Figure 1 and in Figure 1—figure supplement 1. Figure 1 with 1 supplement see all Download asset Open asset Illustration of the methodologies applied in this study. Preoperative MRI scans were co-registered with the postoperative MRI/CT scan, followed by normalization to standard MNI152 space (https://www.bic.mni.mcgill.ca/ServicesAtlases/ICBM152NLin2009). Individual DBS leads were manually localized in the posterior hypothalamic area (pHyp) in the patient space and normalized to MNI152 space. The estimation of the volume of activated tissue (VAT) was calculated based on individual stimulation parameters using Lead-DBS (https://www.lead-dbs.org/ See Table 1 for individual stimulation parameters). The patients’ VATs were further investigated for the analysis of the Voxel Efficacy Map (determination of the optimal stimulation site), Imaging Connectomics using Structural Connectivity Map (determining the streamlines involved in symptom improvement) and Functional Connectivity Map (determining the functionally connected areas involved in symptom improvement). For imaging Transcriptomics, we applied a Threshold Free Cluster Enhancement (TFCE) to the functional connectivity map. Functionally connected areas were averaged into the Harvard-Oxford Atlas (http://www.cma.mgh.harvard.edu/). Based on the human gene expression data from the Allen Human Brain Atlas (https://alleninstitute.org/), genes with a spatial pattern distribution similar to the TFCE map were selected for further gene ontology analysis. 3D reconstruction of the DBS leads on a 100micron resolution, 7.0 Tesla FLASH brain (https://openneuro.org/datasets/ds002179/versions/1.1.0) in MNI152 space; the pHyp label was derived from a previously published high-resolution MRI atlas of the human hypothalamic region (https://zenodo.org/record/3903588#.YHiE7pNKiF0). Results Patients included In this retrospective study, we aggregated a large dataset of 33 patients from 5 centers treated with pHyp-DBS to alleviate intractable aggressive behaviors, characterized by self-injurious and extreme aggressive behaviors towards their surroundings and others (12 females and 21 males, 24.48±10.28 years of age ranging from 10 to 52y, Figure 2A–B). The most frequent diagnoses were epilepsy (pediatric: 50%, adult: 62%) and ASD (pediatric: 34%, adult: 24%). Intellectual disability was observed in all cases. This was described as severe in 82% of patients (pediatric: 75%, adult: 85%) and moderate in 18% of patients (pediatric: 25%, adult: 15%). Table 1 presents demographic data. Clinical trials and individual cases were evaluated by the corresponding local ethics committee, and informed consent was obtained Gouveia et al., 2021c; Contreras Lopez et al., 2021; Benedetti-Isaac et al., 2015; Torres et al., 2020; Micieli et al., 2017; Torres et al., 2013; López Ríos et al., 2023; . Surgical treatment was approved on a humanitarian basis, given the chronicity and severity of symptoms and the lack of response to conservative treatment. Figure 2 Download asset Open asset Patient demographics and treatment outcome. (A) Patients were divided in three main groups according to age: pediatric population (≤17years, 11 out of 33), young adults (18–30years, 14 out of 33) and older adults (31–52years, 8 out of 33). (B) Distribution of males (21 out of 33) and females (12 out of 33) in this study. (C) Patient distribution according to the percentage of symptomatic improvement (≤20: 3 out of 33; 21–40: 1 out of 33; 41–60: 7 out of 33; 61–80: 1 out of 33; 81–100: 21 out of 33). Note that the majority of individuals presented over 30% improvement following treatment (criteria for being considered a treatment responder), and a large proportion of patients presented an improvement greater than 80%. (D) Age at surgery was significantly negatively correlated with postoperative symptomatic improvement (R=–0.61; R2=0.38; *** p<0.001). (E) There was no significant difference in the percentage of symptomatic improvement between male and female patients. Table 1 Demographics. CaseSexAge rangeImprovement (%)LateralityDBS SystemStimulation SettingsRightLeft1M31–52100BilateralMedtronic 33871.8V; 180Hz; 60msec1.8V; 180Hz; 60msec2M≤1797BilateralMedtronic 33872.2V; 200Hz; 90msec2.2V; 200Hz; 90msec3M≤1798bilateralMedtronic 33872.5V; 180Hz; 90msec2.5V; 180Hz; 90msec4M≤1789BilateralMedtronic 33875.0V; 210Hz; 90msec5.0V; 210Hz; 90msec5F≤1793BilateralMedtronic 33872.0V; 200Hz; 90msec2.0V; 200Hz; 90msec6F≤1785BilateralMedtronic 33874.0V; 180Hz; 90msec4.0V; 180Hz; 90msec7F≤17100BilateralMedtronic 33873.5V; 180Hz; 90msec3.5V; 180Hz; 90msec8F31–52100BilateralMedtronic 33875.0V; 200Hz; 100msec5.0V; 200Hz; 100msec9F18–30100BilateralMedtronic 33872.3V; 200Hz; 120msec2.3V; 200Hz; 120msec10M≤1785BilateralMedtronic 33872.0V; 130Hz; 60msec2.0V; 130Hz; 130msec11M18–30100BilateralMedtronic 33873.0V; 180Hz; 90msec3.0V; 180Hz; 90msec12M≤1789BilateralMedtronic 33875.5V; 185Hz; 130msec5.5V; 185Hz; 130msec13F18–3047BilateralMedtronic 33877.0V; 250Hz; 120msec7.0V; 250Hz; 120msec14M18–30100BilateralMedtronic 33875.0V; 210Hz; 130msec5.0V; 210Hz; 130msec15F18–3090BilateralMedtronic 33873.5V; 180Hz; 90msec3.5V; 180Hz; 90msec16M18–30100BilateralMedtronic 33875.0V; 200Hz; 100msec5.0V; 200Hz; 100msec17F≤17100BilateralMedtronic 33873.5V; 180Hz; 90msec3.5V; 180Hz; 90msec18M≤1797BilateralMedtronic 33874.5V; 180Hz; 90msec4.5V; 180Hz; 90msec19M31–5288BilateralMedtronic 33893.0V; 180Hz; 90msec3.0V; 180Hz; 90msec20M≤1791BilateralMedtronic 33893.2V; 180Hz; 90msec3.2V; 180Hz; 90msec21F18–3018BilateralMedtronic 33893.8V; 180Hz; 90msec3.8V; 180Hz; 90msec22M31–5289BilateralMedtronic 33893.5V; 180Hz; 90msec3.5V; 180Hz; 90msec23M31–522BilateralMedtronic 33894.5V; 150Hz; 247msec4.5V; 150Hz; 247msec24M18–3057UnilateralMedtronic 33890.3V; 150Hz; 450msecNot applicable25M31–522UnilateralMedtronic 33890.9V; 150Hz; 450msecNot applicable26F31–5265BilateralMedtronic 33890.1V; 60Hz; 180msec0.1V; 60Hz; 300msec27M18–3059BilateralMedtronic 33890.7V; 150Hz; 330msec0.5V; 150Hz; 450msec28F31–5248UnilateralMedtronic 33890.1V; 150Hz; 450msecNot applicable29M18–30100BilateralMedtronic 33872.0V; 180Hz; 120msec2.0V; 180Hz; 120msec30M18–3050BilateralBoston, Vercise1.0mA; 185Hz; 90msec1.0mA; 185Hz; 90msec31F18–3050BilateralBoston, Vercise1.2mA; 113Hz; 120msec1.2mA; 113Hz; 120msec32M18–3036BilateralBoston, Vercise1.0mA; 170Hz; 70msec1.0mA; 170Hz; 70msec33M18–3058BilateralBoston, Vercise3mA; 185Hz; 60msec3mA; 185Hz; 60msecTo preserve patients' anonymization, the diagnoses observed in this group are presented as the following list, from more to less frequent. Epilepsy, autism spectrum disorder, tuberous sclerosis, congenital rubella, intermittent explosive disorder, agenesia of the corpus callosum, schizophrenia, obsessive-compulsive disorder, West syndrome, Landau-Kleffner syndrome, Cri-du-chat syndrome, Lennox-Gastaut syndrome, Sotos syndrome, meningoencephalitis, perinatal hypoxia, periventricular leucomalacia, microcephaly, arteriovenous malformation. In all centers, patients were evaluated by a multidisciplinary healthcare team that reviewed all clinical and medication history. Patients were only considered for surgery when a consensus was reached that they indeed presented severe medication-resistant aggressive behavior (i.e. persistent severe symptomatology despite using multiple medications at well-established doses and duration). Whole-brain T1-weighted MRI was acquired preoperatively for surgical planning (1.5T MRI used in 14 cases at 3 centers, and 3T MRI used in 19 cases at 2 centers). Postoperative brain MRI and/or CT were obtained for confirmation of electrode localization. Aggressive behavior was assessed using standard questionnaires conducted by a neuropsychologist and answered by the parents/caregivers in all centers (i.e. Overt Aggressive Behaviour [OAS] used in 3 centers; Modified Overt Aggressive Behaviour [MOAS] used in 1 center; Inventory for Client and Agency Planning [ICAP] used in 1 center). Treatment response is reported as the percentage of improvement at the last follow-up relative to baseline (preoperative). Patients presenting more than 30% improvement were considered to be treatment responders. pHyp-DBS was bilaterally implanted in 30 patients and unilaterally implanted in 3 using either Medtronic Activa (3387 leads in 19 cases and 3389 leads in 10 cases) or Boston Vercise DBS Systems (4 cases). Average stimulation parameters for the right hemisphere were 2.45V±1.76V (range 0.3–4.5V), 166.25Hz±18.87Hz (range 150–185Hz) and 219.25µsec ± 172.46µsec (range 60 to 450µsec). For the left hemisphere, average stimulation parameters were 3.17V±1.26V (range 2–4.5V), 171.67Hz±18.93Hz (range 150–185Hz) and 142.33µsec ± 95.48µsec (range 60 to 257µsec). Individual stimulation parameters are shown in Table 1. After treatment, 91% (30 out of 33) of the patients were considered to be responders (>30% decrease in validated scores compared to baseline) Adler et al., 2015; Gouveia et al., 2020; Gouveia et al., 2021b; Benedetti-Isaac et al., 2015. The average percentage of improvement was 75.25 ± 29.59% (Figure 2C). Younger patients were found to have a more pronounced benefit, with the pediatric population (patients ≤ 17 years of age) exhibiting greater symptom improvement compared to the adult population (93% vs 66%, Figure 2D). No differences were observed between males (75.57% ± 31.13%) and females (74.68% ± 28.00%; Figure 2E). Probabilistic sweet-spot mapping To provide insight into the relationship between stimulation location and response to pHyp-DBS treatment, probabilistic maps of efficacious stimulation were generated using previously described methods Elias et al., 2021; Dembek et al., 2017. Briefly, preoperative MRI scans were co-registered to individual postoperative MRI/CT scans, normalized to standard MNI152 space for estimation of the VAT. Left-sided VATs were flipped at the sagittal plane, weighted by the corresponding percentage of improvement and the mean improvement of overlapping VATs was calculated at each voxel. These average maps were then thresholded for voxel-wise significance using a Wilcoxon signed rank test (p<0.05). Additionally, to exclude outlier voxels, only those detected in>10% of all maps were included. Finally, we performed nonparametric permutation testing, randomly assigning each clinical score to a VAT, as previously described Dembek et al., 2017; Eisenstein et al., 2014; Dembek et al., 2019. This analysis revealed greater symptom alleviation related to the stimulation of a more posterior-inferior-lateral region of the posterior hypothalamic area (Figure 3). This area encompassed 684 voxels associated with a behavioral improvement greater than 90%. The centroid of this highly efficacious area can be found at MNI152 space at the coordinates x=7.5, y=-15, z=−6.5 (https://www.bic.mni.mcgill.ca/ServicesAtlases/ICBM152NLin2009) and in the Talairach-Tournoux space (http://www.talairach.org/) at coordinates x=6.5, y=-16, z=−1.5. The permutation test showed that this pattern indeed reflects the specific relationship of individual VATs with the individual outcome (ppermute <0.01). Figure 3 with 2 supplements see all Download asset Open asset Probabilistic Sweet-spot Mapping. (A) The area of stimulation associated with greater symptomatic improvement (red) was located in the more posterior-inferior-lateral region of the posterior hypothalamic area (from left to right: sagittal, coronal and axial views). (B) The extent of the volumes of activated tissue (VATs) responsible for eliciting at least 50% improvement is shown in successive coronal MRI slices. All results are illustrated on slices of a 100micron resolution, 7.0 Tesla FLASH brain (https://openneuro.org/datasets/ds002179/versions/1.1.0) in MNI152 space (https://www.bic.mni.mcgill.ca/ServicesAtlases/ICBM152NLin2009). The posterior hypothalamic nucleus (pHyp n.) label (shown in beige) was derived from a previously published high-resolution MRI atlas of the human hypothalamic region (https://zenodo.org/record/3903588#.YHiE7pNKiF0). Normative connectomics analyses - structural and functional connectivity mapping To investigate white matter tracts and brain networks associated with symptom improvement, normative structural and functional connectivity mapping were performed Fox, 2018; Elias et al., 2020; Germann et al., 2021b. Structural connectivity mapping employed diffusion MRI-based tractography data sourced from Human Connectome Project subjects to identify streamlines that intersected individual VATs. Functional connectivity mapping used Genomics Superstruct Project-derived resting-state functional MRI information to generate voxel-wise correlation maps that reflected each VTA’s brain-wide functional connectivity. Subsequent statistical analyses were conducted to determine which streamlines and functional connectivity patterns were associated with positive treatment outcomes. A variety of streamline bundles were identified by this analysis as being clinically relevant, such that VATs touching said streamlines corresponded to better outcomes than VATs that did not. These streamlines bundles may be divided into three main categories of function: I. Somatosensation (Medial Lemniscus and Spinothalamic Tract); II. Regulation of emotions (Amygdalofugal Pathway, Anterior Limb of the Internal Capsule, Medial Forebrain Bundle Nieuwenhuys et al., 2007; Nieuwenhuys et al., 1982); III. Motor Connections (Superior Cerebellar Peduncle, Rubrospinal tract, Frontopontine tract, Central Tegmental Tract, Medial-longitudinal Fasciculus, Motor Projections; Figure 4A–B, Figure 4—figure supplement 1). Interestingly, when overlapping the voxel efficacy and the structural connectivity maps (Figure 4C), we observed that voxels associated with higher efficacy (in red) were more closely related to these fiber tracts. Figure 4 with 1 supplement see all Download asset Open asset Structural connectivity mapping. (A) Magnetic resonance imaging (MRI) in the sagittal plane showing the fiber density of streamlines connected to the volumes of activated tissue (VATs) associated with significantly greater symptomatic improvement. (B) 3D reconstruction of the streamlines associated with significantly greater improvement illustrated on the MNI152 brain (https://www.bic.mni.mcgill.ca/ServicesAtlases/ICBM152NLin2009); the posterior hypothalamic nucleus label (in red) was derived from a previously published high-resolution MRI atlas of the human hypothalamic region (https://zenodo.org/record/3903588#.YHiE7pNKiF0). (C) MRI showing the relation between VATs responsible for eliciting at least 50% improvement and the fiber density map (from top to bottom: sagittal, coronal and axial views). The results presented in A and C are illustrated on a 100micron resolution, 7.0 Tesla FLASH brain (https://openneuro.org/datasets/ds002179/versions/1.1.0) in MNI152 space. Abbreviations: AFP: Amygdalofugal Pathway; ALIC: Anterior Limb of the Internal Capsule; CTT: Central-Tegmental Tract; FPT: Frontopontine Tract; MFB: Medial Forebrain Bundle; ML: Medial Lemniscus; MLF: Medial-Longitudinal Fasciculus; MP: Motor Projections; RBT: Rubrospinal Tract; SCP: Superior Cerebellar Peduncle; STT: Spino-Thalamic Tract. The functional connectivity analysis showed that the extent of VAT connectedness to several areas was significantly associated with clinical benefits. These areas are related to the production of monoamines (i.e. dorsal and medial raphe nuclei [serotonin]; substantia nigra [dopamine]; Figure 5) and are known to be components of the neurocircuitry of aggressive behavior (e.g. amygdala; nucleus accumbens, rostral anterior cingulate cortex; bed nucleus of the stria terminalis; hypothalamus; dorsal anterior cingulate cortex, insula, periaqueductal grey; Figure 5) Gouveia et al., 2021a; Gouveia et al., 2019; Gouveia et al., 2020; Gouveia et al., 2021b. Functional connectivity mapping after Threshold-Free Cluster Enhancement (TFCE) analysis FDR corrected at q<0.0001 is presented in Figure 5—figure supplement 1. Figure 5 with 1 supplement see all Download asset Open asset Functional connectivity mapping. Magnetic resonance imaging (MRI) in the axial plane showing areas found to have a positive (warm colors) or a negative (cold colors) correlation between clinical outcome and functional connectivity. Results are illustrated on a 100micron resolution, 7.0 Tesla FLASH brain in MNI152 space (https://openneuro.org/datasets/ds002179/versions/1.1.0). Abbreviations: ACC: Anterior Cingulate Cortex; BNST: Bed Nucleus of Stria Teminalis; LH: Lateral Hypothalamus; n.Acc: Nucleus Accumbens; OFC: Orbitofrontal Cortex; PAG: Periaqueductal Grey matter; Pe: Periventricular Hypothalamus; PVN: Paraventricular Hypothalamus; SN: Substantia Nigra; STN: Subthalamic Nucleus; VMH: Ventromedial Hypothalamus; ZI: Zona Incerta. Estimation of clinical outcome To investigate whether individual functional connectivity to particular hubs within the neurocircuitry of aggressive behavior could be used to estimate symptom improvement following pHyp-DBS, additive linear models were created. For this, we extracted individual connectivity values from the peak within each brain area where functional connectivity with the VATs was found to be significantly related to outcomes at the group level (as described in the Normative Connectomics Analyses, Figure 6A). The best-performing parsimonious model, which incorporated patient age as well as individual VAT functional connectivity, revealed the Periaqueductal Grey Matter (PAG) to be the only structure that, together with age, significantly predicted more than half of the variance in individual symptom improvement (R=0.72, R2=0.52, p=1.86e-05, Figure 6B, Table 2), retaining significance during leave-one-out cross-validation (LOOCV) (R=0.65, R2=0.42, p=4.406e-5; Figure 6C). Additionally, we investigated if functional connectivity of the VATs with any 2 brain areas, in addition to patient age, would improve the estimation of clinical outcome (Supplementary file 1). Indeed, connectivity with PAG and limbic structures, namely the amygdala (R=0.75, R2=0.57, p=4.20e-07), anterior cingulate cortex (rostral: R=0.76, R2=0.58, p=2.75e-07; dorsal: R=0.75, R2=0.57, p=3.96e-07), bed nucleus of the stria terminallis (R=0.76, R2=0.58, p=2.75e07), left nucleus accumbens (R=0.76, R2=0.58, p=8.51e-07), right orbitofrontal cortex (R=0.76, R2=0.58, p=2.75e-07) and right fusiform gyrus (R=0.75, R2=0.56, p=4.67e-07) was superior in predicting outcome. Figure 6 Download asset Open asset Estimation of clinical outcome. (A) Location of the peak extracted for each area found to have significant functional connectivity with the volume of activated tissue Illustrated in the coronal plane in MNI152 standard-space (http://www.bic.mni.mcgill.ca/ServicesAtlases/HomePage). (B) A model using age and individual VAT connectivity to the periaqueductal gray significantly estimated individual symptom improvement in the whole dataset and (C) retained significance during leave-one-out cross-validation (LOOCV). Table 2 Estimation of clinical outcome based on functional connectivity map and patient age. Functionally connected brain areaPeak coordinateRR2p-valuePeriaqueductal Grey Matterx=-1 y=-30 z=-100.7250.5251.86e-06Vermisx=1 y=-49 z=-120.7020.4935.21e-06Medial Raphe nucleusx=0 y=-25 z=-150.6890.4759.11e-06Right Subst. Nigra, Subthalamic n., Zona Incertax=16 y=-14 z=-70.6810.4641.28e-05Left Subst. Nigra, Subthalamic n., Zona Incertax=-14 y=-16 z=-70.6730.4531.77e-05Left Claustrumx=-30 y=14 z=-20.6720.4511.88e-05Left Amygdalax=-25 y=-8 z=-270.6720.4511.88e-05Right Fusiform Gyrusx=37 y=-10 z=-340.6680.4472.14e-05Left Putamenx=-33 y=-4 z=20.6560.4303.41e-05Left Dorsal Anterior Cingulate Cortexx=-2 y=25 z=220.6540.4283.65e-05Right Superior Parietal Lobulex=23 y=-62 z=630.6520.4253.95e-05Left Transition Orbitofrontal Cortex- Insulax=-24 y=11 z=-180.6450.4165.06e-05Right Nucleus Accumbensx=11 y=8 z=-70.6450.4165.07e-05Right Amygdalax=-22 y=-6 z=-260.6450.4165.14e-05Left Anterior Insulax=-39 y=18 z=-10.6420.4135.59e-05Right Rostral Anterior Cingulate Cortexx=9 y=40 z=-50.6380.4076.57e-05Right Bed Nucleus Of The Stria Terminallisx=7 y=8 z=-50.6380.4076.57e-05Left Bed Nucleus Of The Stria Terminallisx=-6 y=8 z=-50.6380.4076.57e-05Left Nucleus Acuumbensx=-11 y=8 z=-70.6380.4076.57e-05Right Transition Orbitofrontal Cortex- Insulax=25 y=10 z=-150.6380.4076.57e-05Right Hypothalamusx=5 y=-3 z=-110.6310.3988.25e-05Left Fusiform Gyrusx=-34 y=-12 z=-330.6290.3968.72e-05Left Hypothalamusx=-4 y=-3 z=-110.6210.3861.13e-04 Imaging transcriptomics – Gene set analysis Finally, to investigate neural phenotypes and possible neurobiological mechanisms of treatments, we performed imaging transcriptomics analysis using the abagen toolbox (https://abagen.readthedocs.io/en/stable/index.html) and the human gene expression data from the Allen Human Brain Atlas (https://alleninstitute.org/). We investigated genes with a spatial distribution of expression that resembled the pattern of brain regions with clinically relevant (following TFCE correction) functional connectivity to the stimulation locus (qFDRcor <0.0001). This type of analysis shows genes whose spatial pattern correlates with brain changes that are also found in post-mortem and large-scale GWAS studies of specific patient populations Arnatkeviciute et