Skip to main content
NCBI home page
iScience logoLink to iScience
. 2019 Dec 23;23(1):100800. doi: 10.1016/j.isci.2019.100800

A Unique Brain Connectome Fingerprint Predates and Predicts Response to Antidepressants

Samaneh Nemati 1,2,6, Teddy J Akiki 1,2,6, Jeremy Roscoe 1,2, Yumeng Ju 2, Christopher L Averill 1,2, Samar Fouda 1,2, Arpan Dutta 3,4, Shane McKie 3, John H Krystal 1,2, JF William Deakin 3,5, Lynnette A Averill 1,2, Chadi G Abdallah 1,2,7,∗
PMCID: PMC6992944  PMID: 31918047

Summary

More than six decades have passed since the discovery of monoaminergic antidepressants. Yet, it remains a mystery why these drugs take weeks to months to achieve therapeutic effects, although their monoaminergic actions are present rapidly after treatment. In an attempt to solve this mystery, rather than studying the acute neurochemical effects of antidepressants, here we propose focusing on the early changes in the brain functional connectome using traditional statistics and machine learning approaches. Capitalizing on three independent datasets (n = 1,261) and recent developments in data and network science, we identified a specific connectome fingerprint that predates and predicts response to monoaminergic antidepressants. The discovered fingerprint appears to generalize to antidepressants with differing mechanism of action. We also established a consensus whole-brain hierarchical connectivity architecture and provided a set of model-based features engineering approaches suitable for identifying connectomic signatures of brain function in health and disease.

Subject Areas: Drugs, Medical Imaging, Clinical Neuroscience

Graphical Abstract

graphic file with name fx1.jpg

Open in a new tab

Highlights

  • •

    Machine learning methods were used to fully investigate the brain connectome

  • •

    Network-informed features engineering approaches were proposed

  • •

    A cortical-subcortical hierarchical brain atlas was established

  • •

    A specific connectome signature was found to predict response to antidepressants

Drugs; Medical Imaging; Clinical Neuroscience

Introduction

The serendipitous discovery of slow acting antidepressants in the 1950s has generated persistent interest in identifying the biological underpinnings of depression and unraveling the mechanism of action of antidepressants (Ban, 2006). The clinical neuroscience field has since produced a wealth of knowledge related to the biological systems implicated in depression pathophysiology and to the neurochemical effects of these slow acting antidepressants, which tend to modulate monoaminergic neurotransmitters (Abdallah et al., 2018b, Coplan et al., 2014). Yet, despite more than six decades of research, it remains a mystery as to why the therapeutic behavioral effects of these drugs are only evident following weeks to months of treatment, whereas the neurochemical effects are acutely present after administration (Coplan et al., 2014). Solving this mystery may be critical to developing novel efficacious and rapid acting treatments for the large population of patients who are treatment resistant to currently available antidepressants (Trivedi et al., 2006). In recent years, accumulating evidence implicated brain circuitry and functionally connected networks in the pathology and treatment of depression (Kaiser et al., 2015). Hence, rather than focusing on the acute synaptic neurochemical effects of monoaminergic antidepressants, it may be more revealing to examine the role of early brain network changes in the mechanisms of these slow acting antidepressants.

A consistent evidence in the field is that the behavioral severity of depression significantly improves following placebo treatment, at times leading to lack of behavioral difference between placebo treatment and well-established antidepressants (Andrews, 2001). The placebo response could be due to nonspecific effects or to the milieu effects of research studies (e.g., repeated visits and assessments). Although the placebo effect is a major impediment for efficacy studies, we here used the significant improvement in depression following placebo to our advantage, as it provided an optimal control for both the test-retest of fMRI and depression measures, as well as controlling for behavioral improvement due to nonspecific and milieu effects. This allowed us to identify the biological correlates of response to the studied antidepressants, controlling for test-retest and nonspecific response. Focusing on the interaction between treatment and response, we implemented data-driven approaches to test three hypotheses.

Hypothesis 1—Better Response to Sertraline Would Be Predicted by an Increase in Prefrontal Cortex and Caudate Global Brain Connectivity

Global brain connectivity (GBC) is a measure of nodal strength within a network. Nodal strength is the most fundamental measure in a graph, as most other topology measures are affected by or are partially derived from nodal strength (Bullmore and Sporns, 2009). GBC is calculated as the average connectivity between a node (e.g., a gray matter voxel) and all other nodes within a network (e.g., all gray matter voxels) (Cole et al., 2011). In major depressive disorder (MDD), widespread reduction in GBC and other comparable nodal strength measures has been repeatedly demonstrated across samples and by several research groups, particularly in regions within the prefrontal cortex (PFC) (Abdallah et al., 2017a, Abdallah et al., 2017b, Holmes et al., 2019, Murrough et al., 2016, Scheinost et al., 2018, Wang et al., 2014). Moreover, increased GBC in bilateral caudate or lateral PFC following or during infusion with the rapid acting antidepressant ketamine was associated with enhanced response (Abdallah et al., 2017b, Abdallah et al., 2018a). Compared with healthy controls, patients with MDD were also found to have high GBC in the posterior cingulate, a critical node within the default mode (DM) network (Abdallah et al., 2017b). This DM GBC dysconnectivity is also normalized following ketamine treatment (Abdallah et al., 2017b). The GBC approach has two essential strengths: (1) nodal strength is a fundamental topology measure within a network, and (2) whole-brain analysis can be conducted, without the limitation and the bias of a priori seed selection. However, a main limitation of GBC is the inability to determine which edges (i.e., connection between two regions) are driving the abnormalities. This limitation is addressed in the current study by the use of network-restricted strength (NRS) and NRS predictive model (NRS-PM) approaches, described in hypotheses 2 and 3. Another limitation of GBC is that it does not capture network changes in opposing directions, e.g., a balanced dynamic network shift from internal to external connectivity cannot be captured using GBC values. To address this issue, we here implemented two measures: nodal internal NRS (niNRS) and nodal external NRS (neNRS).

Hypothesis 2—Better Response to Sertraline Would Be Predicted by a Reduction in Internal Default Mode Network-Restricted Strength

The DM comprises brain regions that are synchronously activated at rest and deactivated during external tasks (Andrews-Hanna et al., 2014). Although not without inconsistency (Mulders et al., 2015, Sexton et al., 2012), seed based, independent component, and meta-analysis results have suggested DM hyperconnectivity in MDD compared with controls (Alexopoulos et al., 2012, Greicius et al., 2007, Kaiser et al., 2015, Sheline et al., 2010, Wu et al., 2011). To date, connectivity studies have primarily used seed-based approaches to identify abnormalities or changes in DM of patients with MDD. Among the limitations of this approach are (1) one (or a few) seeds do not cover all the nodes of the DM and the target clusters, hence potentially lowering sensitivity and complicating interpretability of the findings, and (2) it would be difficult to fully integrate findings across studies as they are often highly dependent on the seed location (Akiki et al., 2018). Alternatively, whole-brain topology measures address some of these limitations but fail to identify the specific interactions within and between networks (a.k.a., modules, systems, or communities; e.g., DM).

Borrowing from studies of complex systems (Gu et al., 2015, Guimera and Amaral, 2005), NRS approaches address these limitations and have been successfully implemented in the study of psychopathology (Akiki et al., 2018, Etkin et al., 2019, Nusslock et al., 2019, Schultz et al., 2018, Yang et al., 2018). Briefly, the network-restricted approach used in this study is capable of comprehensively assessing all identifiable nodes in a given brain network (e.g., DM), to extract internal connectivity strength within that network and also calculate the external connectivity between networks. To ensure reproducibility in future studies, we opted to use a consensus hierarchical modularity atlas recently established in 1,003 subjects with high-quality functional magnetic resonance imaging (fMRI) data (Akiki and Abdallah, 2019). Based on the hypothesized hyperconnectivity in DM, we investigated whether reduction in internal DM connectivity would predict better response to sertraline. This was followed by exploratory analysis, with appropriate correction for multiple comparisons, to determine whether response to sertraline is predicted by connectivity within or between brain networks.

Hypothesis 3—Better Response to Sertraline Would Be Predicted by a Consensus-Based NRS Predictive Model

Machine learning methods have been increasingly implemented in the study of psychopathology (Galatzer-Levy et al., 2018). Among the strengths of machine learning algorithms are the fully data-driven approach and the cross-validation component often included in these predictive models, which could address the issue of over-fitting in interpretive models (e.g., regression) and may enhance generalizability to new data (Scheinost et al., 2019, Shen et al., 2017). However, these machine learning approaches have two critical limitations in the study of psychopathology: (1) They are faced with large number of features (e.g., there are 1,799,970,000 unique edges in a cifti-based fMRI dense connectome) compared with the number of observations, which in psychiatry are often in the order of dozens to hundreds. (2) Considering the dimensionality reduction and weighting procedures involved, it is often not possible to back translate the selected/weighted features to the original space to visualize and understand the underlying neurobiology (Shen et al., 2017). Connectome-based predictive model (CPM) is a linear machine learning approach, which retains the ability to back translate findings to the original feature space (Finn et al., 2015, Shen et al., 2017). However, a limitation of CPM is the use of individual parcellated nodes, which leads to a relatively high number of features (e.g., there are 89,676 unique edges in the A424 atlas), which subsequently complicates the interpretability of the findings as the edges included in the final model are often in the order of thousands. Moreover, these nodes are based on anatomical location rather than on network affiliations. To address these limitations, we implemented an NRS-PM that reduces the connectome input features and facilitates the neurobiological interpretation of the final model, as all input features are network based. To assess the robustness of the approach and to rule out potential bias related to architecture selection, we investigated NRS-PM across all architecture levels—with and without subcortical structures, we deconstructed GBC into nodal internal and external NRS, and we determined the full connectome (FC-PM) results using a whole-brain multimodal atlas with 424 nodes (i.e., A424).

Overall, we aimed to determine whether early changes in brain functional connectivity predates and predicts response to slow acting antidepressants. To answer this question, we analyzed publicly available data from a relatively large neuroimaging clinical trial in which depressed patients (n = 202) were randomized to placebo or sertraline, a typical monoaminergic slow acting antidepressant (Trivedi et al., 2016). We then leveraged a large dataset of high-quality fMRI from healthy volunteers (n = 1,003) who participated in the Human Connectome Project (Van Essen et al., 2013) to establish a whole-brain hierarchical parcellation of intrinsic connectivity networks (ICNs). The latter allowed us to extend the cortical findings to subcortical and cerebellar brain regions. Finally, encouraged by the predictive model findings, we conducted a pilot analysis to determine whether the identified connectomic signature related to sertraline response could predate and predict response to the rapid acting antidepressant ketamine, compared with both active and inactive control (Abdallah et al., 2018a, Downey et al., 2016).

Results

The sertraline dataset was acquired from the National Institute of Mental Health Data Archive (NDA), Establishing Moderators and Biosignatures of Antidepressant Response for Clinical Care (EMBARC). All participants (n = 202) with successful resting state fMRI at baseline and week 1, who completed depression assessment at week 8, were included in the current study. The protocol and results of the EMBARC clinical trial were reported elsewhere (Pizzagalli et al., 2018, Trivedi et al., 2006). Briefly, unmedicated patients with chronic or recurrent MDD were randomized to 8 weeks of daily oral placebo or sertraline. Demographics and clinical features are described in Table 1. Hamilton rating scale for depression (HAMD) was used to determine severity. Response was defined as at least 50% improvement in HAMD at week 8, the time point at which full clinical benefit is expected following slow acting antidepressant treatment.

Table 1.

Demographics and Clinical Characteristics

Characteristics Sertraline (n = 99) Placebo (n = 103)
Demographic
 Age, mean (SD) 38.5 (14.2) 37.1 (12.2)
 Male 30.3% 35.9%
 Years of education, mean (SD) 15.2 (2.6) 15.2 (2.4)
Clinical features
 Age of onset, mean (SD) 13.6 (5.4) 14.3 (5.5)
 Chronic 55% 50%
 Baseline HAMD, mean (SD) 18.9 (4.4) 18.5 (4.3)
 W1 HAMD, mean (SD) 16.1(5.5) 15.8(5.1)
 W8 HAMD, mean (SD) 11.1(6.5) 11.8(7.3)
 W1 response 9% 7%
 W8 response 47% 35%a
Open in a new tab

W1, week 1; W8, week 8; HAMD, Hamilton Rating Scale for Depression.

a

There were no significant (p > 0.1) differences between treatment groups, except for response rate at week 8, which showed a trend (chi square = 2.8, p = 0.096).

The Antidepressant Response Is Associated with Early Increase in Caudate Global Connectivity

Guided by previous work with antidepressants (Abdallah et al., 2017b), we first conducted a whole-brain vertex-/voxel-wise GBC analysis, with false discovery rate (FDR) correction (q < 0.05), examining the interaction between treatment (sertraline versus placebo) and response (responders versus non-responders). We found that increased GBC in bilateral caudate and right rostral anterior cingulate (rACC) at week 1 were associated with better response to sertraline at week 8 compared with placebo (Figure 1A). Two clusters with reduced GBC in left Brodmann areas 4 (motor) and 5 (somatosensory) were associated with better response to sertraline (Figure 1A). The independent effects of response and treatment are provided in the Supplemental Information (Figure S1).

Figure 1.

Figure 1

Open in a new tab

Early Global Functional Connectivity Changes Predict Response to Sertraline

(A) Whole-brain interpretive analysis showed a significant interaction between treatment and response, such that increased global brain connectivity (GBC) in the bilateral caudate and right rostral anterior cingulate (red circles) and decreased GBC in sensorimotor cortices (blue circles) at week 1 predicted response to sertraline at week 8 post treatment. The color bar represents the z values (threshold at p < 0.005, with circles denoting clusters that survived FDR correction at q < 0.05).

(B) Nodal strength (nS; i.e., nodal GBC) predictive model analysis (p ≤ 0.001) revealed widespread nS increase in central executive and higher order association areas (red clusters) and decrease in nS within primary cortices (e.g., sensory, motor, and visual; blue clusters) at week 1 as predictors of enhanced response to sertraline at week 8, compared with placebo.

These results confirmed the predicted role of caudate GBC but failed to show an association between antidepressant response and increased lateral PFC GBC (hypothesis 1). We speculated that this failure may be due to the limitations (see Supplemental Information) of the vertex-/voxel-wise interpretive analysis. To rule out this possibility, we implemented a nodal strength predictive model (nS-PM), with 1,000 iterations of 10 cross-validation (CV), to identify brain regions that significantly predict a continuous measure of improvement following sertraline compared with placebo. Changes in nodal GBC (i.e., nS) at week 1 significantly predicted improvement at week 8 following sertraline compared with placebo (r = 0.23, CV = 10, iterations = 1,000, p ≤ 0.001). Enhanced antidepressant response was associated with increased nodal GBC in bilateral caudate and left lateral PFC, along with other brain regions primarily located within the central executive (CE) ICN (Figure 1B). Reduced nodal GBC in several regions within the visual (VI) and sensorimotor (SM) ICNs also predicted better antidepressant response, compared with placebo (Figure 1B).

Early Reduction in Default Mode Connectivity Predates the Antidepressant Response

Employing validated NRS methods (Akiki et al., 2018), we constructed a general linear model examining the effects of treatment, response, and treatment-by-response on changes in internal DM NRS (i.e., week 1 minus pretreatment). We found a significant treatment-by-response interaction (F(1,195) = 5.0, p = 0.026), such that, compared with placebo, reduction in DM NRS at week 1 predicted better response to sertraline at week 8. There were no significant response or treatment effects (p > 0.1).

These results confirmed the predictions of hypothesis 2. Yet, to better characterize the DM findings and inform future studies, we conducted a follow-up analysis, with FDR correction, examining internal and external connectivity of all cortical ICNs (Figure 2A, i.e., DM, CE, dorsal salience [DS], ventral salience [VS], SM, and VI). Following correction for multiple comparisons, there was significant treatment-by-response interactive effect on the CE-SM edge (F(1,195) = 10.4, p = 0.001), reflecting increased connectivity predicts better response to sertraline compared with placebo. There was also significant treatment effect on the CE-SM edge (F(1,195) = 11.0, p = 0.001), reflecting increased connectivity at week 1 following sertraline, compared with placebo (Figure S2). There were no other significant treatment, response, or treatment-by-response effects (q > 0.05).

Figure 2.

Figure 2

Open in a new tab

Early Changes in Network-Restricted Connectivity Predict Response to Sertraline

(A) The networks nodal affiliation based on the Akiki-Abdallah cortical (AAc) hierarchical atlas at 6 modules (AAc-6) (Akiki and Abdallah, 2019), which includes the default mode (DM), central executive (CE), dorsal salience (DS), ventral salience (VS), sensorimotor (SM), and visual (VI) networks.

(B) The network-restricted strength (NRS) pentagon. Internal NRS is depicted as filled circles, whereas inter-networks external NRS is depicted as edges. * was used for p < 0.05, ** for p < 0.01, *** for p < 0.001. Interactions that survived FDR correction were denoted with squares (i.e., CE-SM edge). Filled circles and edges were colored gray for non-significant effects, blue for negative effects, and red for positive effects. The NRS interpretive analysis examining the interactive effects between treatment and response showed significant increase in connectivity between CE and SM at week 1 that predicts response to sertraline at week 8. This figure was adapted with permission from the Emerge Research Program (http://emerge.care).

The Whole-Brain Hierarchical Atlas Reveals the Subcortical Affiliation of GBC Findings

To identify the ICN affiliation of the subcortical GBC findings and to extend the cortical NRS to subcortical and cerebellar regions, we next aimed to determine the hierarchical architecture of the whole brain using subject-level clustering of functional networks from 1,003 adult healthy subjects (Van Essen et al., 2013). Using consistent definition of community across cortical, subcortical, and cerebellar structures, the whole-brain subject-level clustering and consensus community detection algorithm identified 136 significant architecture (Figures 3 and S3). The hierarchical atlas, here termed Akiki-Abdallah (AA) atlas, delineated architectures ranging from 3 to 150 modules. The cortical modules remained largely consistent with the cortex-only atlases in our previous work (Akiki and Abdallah, 2019) and in the literature (Yeo et al., 2011), as shown in Figures 2A versus 2C. The AA architecture at 24 modules (AA-24) largely matches the cortical 22 modules architecture (see Supplemental Information and Figure S4) while adding the affiliation of subcortical and cerebellar nodes. AA-50 is the architecture with the highest similarity to subject-specific modularity (Figures 3B and S3D). Notably, the ICN affiliations of the positive predictive GBC nodes findings were within the CE and global-pallidus-putamen subcortical (GPu SC) modules, whereas the negative predictive GBC nodes were within the SM and VI networks (see Figures 1B versus 3C).

Figure 3.

Figure 3

Open in a new tab

Whole-Brain Hierarchical Brain Architecture of Functional Connectivity Networks

(A) Co-classification matrix summarizing the results of the whole-brain clustering atlas in 1,003 healthy subjects, here termed Akiki-Abdallah (AA) atlas. The dendrogram represents the hierarchical organization of the nested communities. The background colors represent the network affiliation at 7 modules architecture (i.e., AA-7).

(B) Similarity plot showing the mean similarity between the partitioning in each consensus hierarchical level (i.e., from AA-3 to AA-150) and the subject-level clustering quantified by Z score of the Rand coefficient (blue line); the Rand score peaked at AA-50. The dashed red lines denote the brain architecture levels for Figures 3D and S3D.

(C and D) These maps show the whole-brain networks nodal affiliation at AA-7 (corresponds to the six main cortical networks) and AA-24 (corresponds to the peak subject-level similarity in the cortical atlas). The module abbreviations of AA-24, along with further details about the affiliation of each node, are reported in Table S1. The figures were adapted with permission from the Emerge Research Program (http://emerge.care). The hierarchical atlas maps and codes will be made publicly available at https://github.com/emergelab.

A Connectome Fingerprint Predates and Predicts Enhanced Antidepressant Response

Following ICN hierarchical parcellation, we aimed to determine whether a whole-brain connectome fingerprint predates and predicts response to sertraline, compared with placebo. As shown in Figure S5B, the whole-brain NRS-PM predicted the antidepressant response across AA-4 to AA-150 architectures (following FDR correction), with a peak at AA-58 (r = 0.27, CV = 10, iterations = 1,000, p = 0.003). Independently, the positive predictive edges peaked at AA-58 (r = 0.29, CV = 10, iterations = 1,000, p = 0.001) and the negative predictive edges peaked at AA-26 (r = 0.25, CV = 10, iterations = 1,000, p = 0.003). Compared with cortical NRS-PM, the inclusion of subcortical structure appears to have enhanced models, particularly positive predictions that were more consistent in independently predicting antidepressant response across architectures (Figure S5).

To further assess the whole-brain NRS-PM results, the AA-24 and AA-50 were visualized in Figures 4A and 4B, respectively. Predictive edges were comparable with findings with cortical atlas (see Supplemental Information and Figure S4), with richer connectomic signature and two highlights: (1) Modules within the CE and GPu SC ICNs showed reduced internal connectivity among each other but increased connectivity to the rest of the brain as predictors of enhanced antidepressant response. (2) Reduction in internal connectivity within the remaining ICNs, particularly SM and VI, also predicted better sertraline response. Modules containing the amygdala and insula also showed a shift from connection with primary cortices (e.g., sensory, motor, and visual) to increased connectivity with higher order association areas (Figure 4).

Figure 4.

Figure 4

Open in a new tab

A Unique Brain Functional Connectome Fingerprint Predates and Predicts Response to Sertraline

(A and B) Network-restricted strength (NRS) predictive models (NRS-PMs; p = 0.032 and p = 0.033, respectively) revealed a specific connectomic signature evident at week 1 post treatment, at a time of no clinically meaningful antidepressant effects, that predicts enhanced response to sertraline at week 8, compared with placebo. An overall pattern emerged consistent of reduced connectivity between modules within the central executive (CE) and globus-pallidus-putamen subcortical (GPu SC) networks, along with increased connectivity between these two networks (CE/GPu SC) and the rest of the brain as predictors of enhanced response to sertraline. The connectome fingerprint (CFP) also showed reduction in internal connectivity within the visual and sensorimotor networks, as well as reduced interference on these two networks from modules within the default mode and salience networks.

(C) The full connectome predictive model (FC-PM; p = 0.005) predicted response to sertraline, but it was not possible to visually discern the underlying signature considering the large number of edges retained in the PM.

(D) Using nodal strength within the FC-PM as cutoff to retain the highest top 2.5% negative predictive edges and top 2.5% positive predictive edges showed a pattern consistent with the NRS-PM findings, but at the expense of discarding 95% of the data, and did not fully depict the shift from internal to external strength within the CE and GPu SC networks. Notes: The circular graphs in 4A, 4C, and 4D are labeled based on the Akiki-Abdallah (AA) whole-brain architecture at 24 modules (AA-24; see Figure 3D; Table S1), whereas 4B is based on AA-50 (see Figure S3D; Table S1). Modules are colored according to their AA-7 network affiliation (see Figure 3C). Edges are colored based on the initiating module using a counter-clockwise path starting at 12 o'clock. Internal edges (i.e., within module NRS) are depicted as outer circles around the corresponding module. Thickness of edges reflects their corresponding weight in the predictive model. The module abbreviations of AA-24 and AA-50, along with further details about the affiliation of each node, are reported in Table S1. The predictive models will be made publicly available at https://github.com/emergelab.

The Full Connectome Predicts Response but Yields Undiscernible Connectomic Signature

We next investigated whether a nodal based (i.e., 424x424 nodes) full connectome PM (FC-PM) would yield comparable or differing results, compared with whole-brain NRS-PM. The FC-PM significantly predicted the antidepressant response (r = 0.27, CV = 10, iterations = 1,000, p = 0.005). However, considering the large number of predictive edges, visualizing the PM is unable to discern the underlying connectomic signature (Figure 4C). Retaining the nodes with the highest degree (i.e., top 2.5% of each of positive and negative predictive edges) showed a pattern of reduced internal connectivity among nodes in the SM and VI network, but at the expense of discarding 95% of the model, and failed to reveal the internal-to-external shift within the CE network (Figure 4D).

Quantifying a Clinically Relevant Internal to External Connectivity Shift

Together, the network-restricted interpretative and predictive results supported the nS-PM GBC findings of increased connectivity in nodes within the CE and GPu SC but reduced connectivity in the SM and VI networks as predictor of response (Figures 1, 2, 3, and 4). However, the CE shift from internal to external connectivity cannot be captured by GBC measures, as the latter is an average of both internal and external connectivity. Therefore, to quantify the NRS shift, we deconstructed nS (i.e., nodal GBC) into two complementary measures: (1) nodal internal NRS (niNRS) calculated as the average connectivity between each node and all other nodes within the same ICN and (2) nodal external NRS (neNRS) calculated as the average connectivity between each node and all other nodes outside its ICN. Here, we used AA-7, which comprises the main brain networks while incorporating subcortical structures, including CE, DM, VS, DS, SC, SM, and VI networks (Figure 3C). Changes in neNRS at week 1 significantly predicted improvement at week 8 following sertraline treatment compared with placebo (r = 0.28, CV = 10, iterations = 1,000, p ≤ 0.001). Enhanced antidepressant response was associated with increased neNRS in brain regions primarily located within the CE and GPu SC networks (Figures 5A and 5B). Similarly, changes in niNRS significantly predicted the antidepressant response (r = 0.24, CV = 10, iterations = 1,000, p = 0.005). Enhanced antidepressant response was associated with reduced niNRS in brain regions primarily located within the CE, GPu SC, SM, and VI networks (Figures 5C and 5D). These results quantitatively demonstrated the internal-to-external NRS shift as predictor of antidepressant response. The cortical and whole-brain neNRS-PMs and niNRS-PMs across all architectures are provided in Supplemental Information (Figures S6 and S7). Similar to the NRS-PMs, including subcortical structures resulted in more consistent neNRS-PMs and niNRS-PMs across architectures (Figures S6 and S7).

Figure 5.

Figure 5

Open in a new tab

Quantifying the Internal to External Connectivity Shift

(A and B) A predictive model (p ≤ 0.001) using nodal external network-restricted strength (neNRS) as input, based on Akiki-Abdallah (AA) whole-brain atlas at the architecture with 7 modules (AA-7), predicted enhanced response to sertraline. The model showed increased global external connectivity in brain regions within the central executive (CE) and globus-pallidus-putamen subcortical (GPu SC) networks. There were no reductions in global external connectivity.

(C and D) A predictive model (p = 0.005) using nodal internal network-restricted strength (niNRS) as input, based on AA-7, predicted enhanced response to sertraline. The model showed reduced internal connectivity in brain regions within the CE, GPu SC, sensorimotor, and visual networks. There were no increases in internal connectivity. Red arrows point to regions that showed both reduced niNRS and increase neNRS, all of which located within the CE and GPu SC networks, consistent with the internal-to-external connectivity shift observed in previous analyses and quantitively supporting the observed pattern of a shift toward increased higher order control over primary cortices.

The Predictive Models Partially Generalize to the Rapid Acting Antidepressant Ketamine

Encouraged by the robust NRS-PM findings, we conducted a pilot follow-up analysis to investigate the generalizability of the identified whole-brain models in predicting the antidepressant response to ketamine, a well-established rapid acting antidepressant (Abdallah et al., 2018b). Here, we used data from a previously published pharmacoimaging study, which examined brain functional connectivity at a period that predates the antidepressant effects of ketamine (i.e., during infusion) (Abdallah et al., 2018a, Downey et al., 2016). In this study, 56 patients with MDD were randomized to ketamine, lanicemine, or normal saline. The antidepressant effects of ketamine often arise within hours of its administration. Thus, we examined whether connectivity changes during ketamine infusion predates and predicts response at 24h post treatment, using lanicemine and normal saline as active and inactive control arms, respectively. Ketamine and lanicemine are both N-methyl-D-aspartate (NMDA) receptor antagonists; therefore, the lanicemine arm provided control for both the milieu effect (assessments, infusion, scans, etc.) as well as the acute non-specific NMDA modulation of connectivity networks during infusion (Abdallah et al., 2018a, Downey et al., 2016).

Following FDR correction for multiple comparisons, the three whole-brain predictive models established in the sertraline study—here termed sertraline connectome fingerprint (CFP)—predicted ketamine response compared with lanicemine at AA-24-CFP (r = 0.52, n = 38, p = 0.0008), AA-50-CFP (r = 0.57, n = 38, p = 0.0002) and full connectome FC-CFP (r = 0.55, n = 38, p = 0.0003). All three sertraline CFPs failed to predict ketamine response, compared with placebo (p > 0.05).

To further characterize the ketamine versus lanicemine findings, we conducted whole-brain NRS-PM at AA-24 and AA-50, as well as FC-PM. All three models significantly predicted treatment response to ketamine, compared with lanicemine (r = 0.41 to 0.48, CV = 10, iterations = 1,000, p < 0.05). As shown in Figure 6, the identified models were largely comparable with those found in the sertraline versus placebo results.

Figure 6.

Figure 6

Open in a new tab

A Comparable Brain Functional Connectome Signature Predates and Predicts Response to Ketamine

(A and B) Network-restricted strength (NRS) predictive models (NRS-PMs; p = 0.003 and p = 0.002, respectively) revealed a comparable connectome fingerprint evident at 20 min during infusion, which predicts enhanced response to ketamine at 24 h, compared with active control. The overall pattern is consistent with findings in the sertraline models (Figure 4), but at a significantly shorter scale (i.e., at 20 min compared with at 7 days).

(C) The full connectome predictive model (FC-PM) predicted response to ketamine compared with active control, but it was not possible to visually discern the underlying signature considering the large number of edges retained in the PM.

(D) Using nodal strength within the FC-PM as cutoff to retain the highest top 2.5% negative predictive edges and top 2.5% positive predictive edges showed a pattern consistent with the NRS-PM findings, but at the expense of discarding 95% of the data. Notes: The circular graphs in 6A, 6C, and 6D are labeled based on the Akiki-Abdallah (AA) whole-brain architecture at 24 modules (AA-24; see Figure 3D; Table S1), whereas 6B is based on AA-50 (see Figure S3D; Table S1). Modules are colored according to their AA-7 network affiliation (see Figure 3C). Edges are colored based on the initiating module using a counter-clockwise path starting at 12 o'clock. Internal edges (i.e., within module NRS) are depicted as outer circles around the corresponding module. Thickness of edges reflect their corresponding weight in the predictive model. The module abbreviations of AA-24 and AA-50, along with further details about the affiliation of each node are reported in Table S1. The predictive models will be made publicly available at https://github.com/emergelab.

Discussion

Using a set of traditional statistics and machine learning approaches, the study results provided strong evidence of a specific connectome fingerprint (CFP) that predates and predicts response to the slow acting antidepressant, sertraline. The study established a whole-brain hierarchical ICNs atlas and provided evidence of its relevance to the study of psychopathology and to clinical neuroscience discovery. It also presented specialized measures and tested the robustness of innovative approaches, i.e., NRS-PM, niNRS-PM, and neNRS-PM. The impact of these atlases, measures, and approaches is expected to go beyond the current findings, as similar approaches could be used to identify the CFPs of various brain functions in health and disease. Finally, the study provided pilot evidence about the potential generalizability, albeit at a much shorter timescale, of the identified CFP to the mechanisms of rapid acting antidepressants.

Partially supporting hypothesis 1, bilateral caudate clusters of increased global connectivity at week 1 post treatment predicted better response to sertraline compared with placebo at week 8 (Figure 1A). Treatment response was also associated with decreased global connectivity in the sensory motor area but increased connectivity in the rostral anterior cingulate cortex (ACC; Figure 1A). However, the latter finding lost significance in the predictive model analysis (Figure 1B). Intriguingly, although GBC is a whole-brain based measure and not network restricted, the positive predictive GBC nodes were primarily located within the CE and GPu SC modules. Similarly, the negative predictive GBC nodes were primarily within the SM and VI networks (see Figures 1B vs. 3C). Although the primary analysis failed to show increased GBC in the lateral PFC as predicted by hypothesis 1, this may be due to limitations of the used vertex-/voxel-wise approach (see below) Supplemental Information.

The interpretive NRS model confirmed the predictions of hypothesis 2, showing significant association between better response to sertraline at week 8 and earlier reduction in internal DM connectivity at week 1 post treatment. Follow-up analyses revealed increased CE-SM connectivity as early predictor of enhanced response to sertraline (Figure 2B).

Compared with placebo, the NRS-PMs significantly predicted response to sertraline, supporting hypothesis 3, while revealing dynamic shifts in the brain circuits, critically underlying the negative and positive predictions of treatment effects (Figures 4 and 5). In particular, three patterns of connectivity shifts have emerged as predictors of response to antidepressant treatment: (1) A reduction in internal connectivity among the CE and GPu SC modules, along with increased external connectivity between these modules and the rest of the brain (most evident in AA-50, but also in AA-24, and niNRS/neNRS). (2) Reduced internal connectivity among modules within the SM and VI networks. (3) In DM/VS modules containing the amygdala and insula (i.e., affective DM and inferior VS, respectively), there was reduced connectivity with perceptual and motor areas (i.e., SM and VI) but increased connectivity with higher order association regions, suggesting an early shift toward enhanced executive control Supplemental Information.

The current study focused on the changes overtime in brain functional connectivity and in psychopathology as reflected by antidepressant scores. This focused approach provided a powerful comprehensive assessment of brain connectivity in relation to treatment response. However, future complementary and follow-up studies are still required to determine whether pretreatment connectivity, as well as clinical features (e.g., early life stress or symptom clusters) would be associated with specific CFPs or would predict response to antidepressants. Notably, an outstanding recent study has investigated pretreatment connectivity and clinical features in the EMBARC cohort (Yu et al., 2019). Before treatment, it was reported that various brain networks correlated with early life stressors and that MDD was characterized by increased NRS in intrinsic networks, including DM, but reduced NRS in task-positive networks, including CE (Yu et al., 2019). Importantly, these pretreatment findings of altered DM and CE connectivity in MDD suggest that the antidepressant CFP identified in the current study may reflect an early pattern of normalization detectable biologically at week 1 but not evident behaviorally until week 8.

In the context of inconsistent patterns in the literature, we did not a priori predict the widespread role of the CE network at multiple hierarchical architectures, even though previous results from our own studies have associated increased GBC in the lateral PFC and caudate with enhanced antidepressant response (Abdallah et al., 2017b, Abdallah et al., 2018a). Better understanding of this oversight may be essential to enhance reproducibility in psychiatric neuroimaging studies. We believe that this oversight was primarily due to two common limitations of neuroimaging work. The first is that vertex-/voxel-wise traditional statistics (i.e., interpretive models, e.g., regression) are subject to overfitting owing to the lack of cross-validation and to type II error owing to the needed correction for multiple comparison. For example, an increase in vertex-wise GBC in the rostral ACC was identified in the interpretive analysis (Figure 1A) but did not hold when subjected to cross-validation procedures (Figure 1B). Moreover, the vertex-/voxel-wise interpretive model identifies only the peak effect (i.e., the vertices/voxels with the highest effect size), without the ability to fully map the behavioral brain effects (see Figures 1A vs. 1B). The second limitation is that the majority of ICN atlases were not multiscale hierarchical (i.e., lacking the information about the upstream ICN architecture) and were limited to the cortex (e.g., Yeo et al., 2011). The few ICN atlases that included subcortical structures were mostly post hoc (i.e., projecting the cortical ICNs on the subcortical regions without the use of a unified definition of ICN community; e.g., Choi et al., 2012) or were meta-analytically based missing important subcortical structures (e.g., the amygdala in Power et al., 2011). Thus, it was not commonly viewed that the caudate and frontoparietal cortex share the same upstream CE architecture. Furthermore, the biased focus of the literature on few key seeds and ICNs (e.g., DM) may have contributed to the apparent inconsistency across studies. In the current study, the full assessment of the connectome along with combined use of predictive models, hierarchical ICN architecture, and unified definition of ICN across cortical and subcortical nodes led to the robust identification of the CE ICN as critical predictor of treatment response using multiple data-driven analytical approaches.

In summary, based on the behavioral assessment of depression, there was no clinically meaningful behavioral response at week 1 (∼2 points reduction on HAMD) and there were no significant differences between the treatment arms. In contrast, the biological functional connectivity investigation at week 1 showed a robust connectomic signature that predates and predicts response to sertraline at week 8. The identified biosignature was evident using three distinct, yet overlapping and complimentary, approaches. A main impact of the findings is the identification of an antidepressant-related dynamic shift in brain networks connectivity, which is consistent with early increase in executive control weeks prior to the full therapeutic behavioral effects. Some features of the identified connectomic signature are reminiscent of findings with ketamine, and the predictive models at least partially generalized to ketamine treatment, raising the possibility that the discovered connectome fingerprint may generalize to antidepressants with differing mechanism of action.

Future studies can capitalize on the established set of approaches that combine the strengths of consensus hierarchical architecture of the brain, along with neuroscience-informed features engineering and machine learning to fully assess the functional connectome and successfully identify psychopathology-relevant reproducible fingerprint. These data-driven approaches are likely to have impact beyond answering the research question of the current study, by allowing the full assessment of functional connectivity in relation to behavior without the reliance on a limited number of seeds or networks. Together, the study findings enriched our understanding of the neurobiology of depression and revealed a previously unknown connectomic signature that may serve as a treatment target, as a biological indicator of response to optimize treatment regimen, or as a surrogate for early stages in drug development.

Limitation of the Study

The relatively small sample in the ketamine study is considered a limitation. Thus, the generalizability finding of the identified monoaminergic CFP to glutamatergic rapid acting antidepressants should be considered pilot evidence. In particular, using an active control, we found that ketamine induced a CFP highly comparable with the one identified with sertraline, which predated and predicted the rapid acting antidepressant effects (Figures 4 and 6). However, this was not the case when normal saline was used as control. It is plausible that the direct NMDA antagonism effects on pyramidal neurons are not fully required for the antidepressant properties of ketamine; thus, removing these non-specific effects using an active comparator may have facilitated the identification of the unique connectivity signature. This hypothesis would be consistent with accumulating evidence underscoring the role of the ketamine metabolite, (2R-6R)-hydroxynorketamine, which does not exert direct NMDA antagonistic effects (Riggs et al., 2019, Zanos et al., 2016, Zanos et al., 2019). Yet, it remains important to highlight the pilot nature of this follow-up analysis and the need for replication in a larger sample prior to making any firm conclusions.

Methods

All methods can be found in the accompanying Transparent Methods supplemental file.

Acknowledgments

The authors would like to thank the subjects who participated in these studies for their invaluable contribution. Data used in the preparation of this manuscript were obtained and analyzed from the controlled access datasets distributed from the NIMH-supported National Database for Clinical Trials (NDCT). NDCT is a collaborative informatics system created by the National Institute of Mental Health to provide a national resource to support and accelerate discovery related to clinical trial research in mental health. Dataset identifier(s): STU 092010-151; Establishing Moderators and Biosignatures of Antidepressant Response for Clinical Care (EMBARC). Data were provided in part by the Human Connectome Project, WU-Minn Consortium (Principal Investigators: David Van Essen and Kamil Ugurbil; 1U54MH091657) funded by the 16 NIH Institutes and Centers that support the NIH Blueprint for Neuroscience Research; and by the McDonnell Center for Systems Neuroscience at Washington University. This manuscript reflects the views of the authors and may not reflect the opinions or views of the NIMH or of the Submitters submitting original data to NDCT.

Funding support was provided by NIMH (K23MH101498), VA Career Development Award (L.A.A.), the VA National Center for PTSD, and AstraZeneca. We are grateful for the expert assistance of the staff of the Clinical Research Facilities in Manchester and Oxford in which the ketamine study was conducted. P1vital managed the original ketamine study on behalf of AstraZeneca. The content of this report is solely the responsibility of the authors and does not necessarily represent the official views of the sponsors, the Department of Veterans Affairs, NIH, or the US Government.

Author Contributions

Conceptualization, C.G.A., T.J.A., and S.N.; Methodology, C.G.A. and T.J.A.; Formal Analysis, C.G.A., T.J.A., and S.N.; Investigation, S.N., T.J.A., J.R., C.L.A., S.F., A.D., S.M., J.F.W.D., and C.G.A.; Writing – Original Draft, C.G.A., T.J.A., S.N., Y.J., and C.L.A.; Writing – Review/Editing, all authors; Funding Acquisition, C.G.A., L.A.A., J.H.K., and J.F.W.D.; Resources, C.G.A. and L.A.A.; Supervision, C.G.A., L.A.A., J.H.K., and J.F.W.D.

Declaration of Interests

Dr. Abdallah has served as a consultant and speaker and/or on advisory boards for Genentech, Janssen, Lundbeck, and FSV7 and editor of Chronic Stress for Sage Publications, Inc. and filed a patent for using mTORC1 inhibitors to augment the effects of antidepressants (filed on Aug 20, 2018). Dr. Krystal is a consultant for AbbVie, Inc., Amgen, Astellas Pharma Global Development, Inc., AstraZeneca Pharmaceuticals, Biomedisyn Corporation, Bristol-Myers Squibb, Eli Lilly and Company, Euthymics Bioscience, Inc., Neurovance, Inc., FORUM Pharmaceuticals, Janssen Research & Development, Lundbeck Research USA, Novartis Pharma AG, Otsuka America Pharmaceutical, Inc., Sage Therapeutics, Inc., Sunovion Pharmaceuticals, Inc., and Takeda Industries; is on the Scientific Advisory Board for Lohocla Research Corporation, Mnemosyne Pharmaceuticals, Inc., Naurex, Inc., and Pfizer; is a stockholder in Biohaven Pharmaceuticals; holds stock options in Mnemosyne Pharmaceuticals, Inc.; holds patents for Dopamine and Noradrenergic Reuptake Inhibitors in Treatment of Schizophrenia, U.S. Patent No. 5,447,948 (issued Sep 5, 1995), and Glutamate Modulating Agents in the Treatment of Mental Disorders, U.S. Patent No. 8,778,979 (issued Jul 15, 2014); and filed a patent for Intranasal Administration of Ketamine to Treat Depression – U.S. Application No. 14/197,767 (filed on Mar 5, 2014); U.S. application or Patent Cooperation Treaty international application No. 14/306,382 (filed on Jun 17, 2014). Filed a patent for using mTORC1 inhibitors to augment the effects of antidepressants (filed on Aug 20, 2018). Dr. Deakin currently advises or carries out research funded by Autifony, Sunovion, Lundbeck, AstraZeneca, and Servier. All other co-authors declare no conflict of interest.

Published: January 24, 2020

Footnotes

Supplemental Information can be found online at https://doi.org/10.1016/j.isci.2019.100800.

Data and Code Availability

The study data are available through the Human Connectome Project (https://www.humanconnectome.org) and the National Institute of Mental Health (NIMH) Data Archive (NDA; https://nda.nih.gov/). The developed Akiki-Abdallah hierarchical modularity atlas, the network-restricted strength function, and the predictive model codes will be made publicly available at https://github.com/emergelab along with the predictive models established in the current paper.

Supplemental Information

Document S1. Transparent Methods and Figures S1–S7
mmc1.pdf (5MB, pdf)
Table S1. Cortical and Whole-Brain Network Affiliation of the Akiki-Abdallah Atlas, Related to Figures 3, 4, 6, and Transparent Methods
mmc2.xlsx (52.5KB, xlsx)

References

  1. Abdallah C.G., Averill C.L., Salas R., Averill L.A., Baldwin P.R., Krystal J.H., Mathew S.J., Mathalon D.H. Prefrontal connectivity and glutamate transmission: relevance to depression pathophysiology and ketamine treatment. Biol. Psychiatry Cogn. Neurosci. Neuroimaging. 2017;2:566–574. doi: 10.1016/j.bpsc.2017.04.006. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Abdallah C.G., Averill L.A., Collins K.A., Geha P., Schwartz J., Averill C., DeWilde K.E., Wong E., Anticevic A., Tang C.Y. Ketamine treatment and global brain connectivity in major depression. Neuropsychopharmacology. 2017;42:1210–1219. doi: 10.1038/npp.2016.186. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Abdallah C.G., Dutta A., Averill C.L., McKie S., Akiki T.J., Averill L.A., William Deakin J. Ketamine, but not the NMDAR antagonist lanicemine, increases prefrontal global connectivity in depressed patients. Chronic Stress (Thousand Oaks) 2018;2 doi: 10.1177/2470547018796102. 2470547018796102. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Abdallah C.G., Sanacora G., Duman R.S., Krystal J.H. The neurobiology of depression, ketamine and rapid-acting antidepressants: is it glutamate inhibition or activation? Pharmacol. Ther. 2018;190:148–158. doi: 10.1016/j.pharmthera.2018.05.010. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Akiki T.J., Abdallah C.G. Determining the hierarchical architecture of the human brain using subject-level clustering of functional networks. Sci. Rep. 2019;9:19290. doi: 10.1038/s41598-019-55738-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Akiki T.J., Averill C.L., Wrocklage K.M., Scott J.C., Averill L.A., Schweinsburg B., Alexander-Bloch A., Martini B., Southwick S.M., Krystal J.H., Abdallah C.G. Default mode network abnormalities in posttraumatic stress disorder: a novel network-restricted topology approach. Neuroimage. 2018;176:489–498. doi: 10.1016/j.neuroimage.2018.05.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Alexopoulos G.S., Hoptman M.J., Kanellopoulos D., Murphy C.F., Lim K.O., Gunning F.M. Functional connectivity in the cognitive control network and the default mode network in late-life depression. J. Affect. Disord. 2012;139:56–65. doi: 10.1016/j.jad.2011.12.002. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Andrews G. Placebo response in depression: bane of research, boon to therapy. Br. J. Psychiatry. 2001;178:192–194. doi: 10.1192/bjp.178.3.192. [DOI] [PubMed] [Google Scholar]
  9. Andrews-Hanna J.R., Smallwood J., Spreng R.N. The default network and self-generated thought: component processes, dynamic control, and clinical relevance. Ann. N Y Acad. Sci. 2014;1316:29–52. doi: 10.1111/nyas.12360. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Ban T.A. The role of serendipity in drug discovery. Dialogues Clin. Neurosci. 2006;8:335–344. doi: 10.31887/DCNS.2006.8.3/tban. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Bullmore E., Sporns O. Complex brain networks: graph theoretical analysis of structural and functional systems. Nat. Rev. Neurosci. 2009;10:186–198. doi: 10.1038/nrn2575. [DOI] [PubMed] [Google Scholar]
  12. Choi E.Y., Yeo B.T., Buckner R.L. The organization of the human striatum estimated by intrinsic functional connectivity. J. Neurophysiol. 2012;108:2242–2263. doi: 10.1152/jn.00270.2012. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Cole M.W., Anticevic A., Repovs G., Barch D. Variable global dysconnectivity and individual differences in schizophrenia. Biol. Psychiatry. 2011;70:43–50. doi: 10.1016/j.biopsych.2011.02.010. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Coplan J.D., Gopinath S., Abdallah C.G., Berry B.R. A neurobiological hypothesis of treatment-resistant depression - mechanisms for selective serotonin reuptake inhibitor non-efficacy. Front. Behav. Neurosci. 2014;8:189. doi: 10.3389/fnbeh.2014.00189. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Downey D., Dutta A., McKie S., Dawson G.R., Dourish C.T., Craig K., Smith M.A., McCarthy D.J., Harmer C.J., Goodwin G.M. Comparing the actions of lanicemine and ketamine in depression: key role of the anterior cingulate. Eur. Neuropsychopharmacol. 2016;26:994–1003. doi: 10.1016/j.euroneuro.2016.03.006. [DOI] [PubMed] [Google Scholar]
  16. Etkin A., Maron-Katz A., Wu W., Fonzo G.A., Huemer J., Vertes P.E., Patenaude B., Richiardi J., Goodkind M.S., Keller C.J. Using fMRI connectivity to define a treatment-resistant form of post-traumatic stress disorder. Sci. Transl. Med. 2019;11:eaal3236. doi: 10.1126/scitranslmed.aal3236. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Finn E.S., Shen X., Scheinost D., Rosenberg M.D., Huang J., Chun M.M., Papademetris X., Constable R.T. Functional connectome fingerprinting: identifying individuals using patterns of brain connectivity. Nat. Neurosci. 2015;18:1664–1671. doi: 10.1038/nn.4135. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Galatzer-Levy I.R., Ruggles K., Chen Z. Data science in the research domain criteria era: relevance of machine learning to the study of stress pathology, recovery, and resilience. Chronic Stress. 2018;2 doi: 10.1177/2470547017747553. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Greicius M.D., Flores B.H., Menon V., Glover G.H., Solvason H.B., Kenna H., Reiss A.L., Schatzberg A.F. Resting-state functional connectivity in major depression: abnormally increased contributions from subgenual cingulate cortex and thalamus. Biol. Psychiatry. 2007;62:429–437. doi: 10.1016/j.biopsych.2006.09.020. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Gu S., Satterthwaite T.D., Medaglia J.D., Yang M., Gur R.E., Gur R.C., Bassett D.S. Emergence of system roles in normative neurodevelopment. Proc. Natl. Acad. Sci. U S A. 2015;112:13681–13686. doi: 10.1073/pnas.1502829112. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Guimera R., Amaral L.A. Cartography of complex networks: modules and universal roles. J. Stat. Mech. 2005;2005 doi: 10.1088/1742-5468/2005/02/P02001. nihpa35573. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Holmes S.E., Scheinost D., Finnema S.J., Naganawa M., Davis M.T., DellaGioia N., Nabulsi N., Matuskey D., Angarita G.A., Pietrzak R.H. Lower synaptic density is associated with depression severity and network alterations. Nat. Commun. 2019;10:1529. doi: 10.1038/s41467-019-09562-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Kaiser R.H., Andrews-Hanna J.R., Wager T.D., Pizzagalli D.A. Large-scale network dysfunction in major depressive disorder: a meta-analysis of resting-state functional connectivity. JAMA psychiatry. 2015;72:603–611. doi: 10.1001/jamapsychiatry.2015.0071. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Mulders P.C., van Eijndhoven P.F., Schene A.H., Beckmann C.F., Tendolkar I. Resting-state functional connectivity in major depressive disorder: a review. Neurosci. Biobehav. Rev. 2015;56:330–344. doi: 10.1016/j.neubiorev.2015.07.014. [DOI] [PubMed] [Google Scholar]
  25. Murrough J.W., Abdallah C.G., Anticevic A., Collins K.A., Geha P., Averill L.A., Schwartz J., DeWilde K.E., Averill C., Jia-Wei Yang G. Reduced global functional connectivity of the medial prefrontal cortex in major depressive disorder. Hum. Brain Mapp. 2016;37:3214–3223. doi: 10.1002/hbm.23235. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Nusslock R., Brody G.H., Armstrong C.C., Carroll A.L., Sweet L.H., Yu T., Barton A.W., Hallowell E.S., Chen E., Higgins J.P. Higher peripheral inflammatory signaling associated with lower resting-state functional brain connectivity in emotion regulation and central executive networks. Biol. Psychiatry. 2019;86:153–162. doi: 10.1016/j.biopsych.2019.03.968. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Pizzagalli D.A., Webb C.A., Dillon D.G., Tenke C.E., Kayser J., Goer F., Fava M., McGrath P., Weissman M., Parsey R. Pretreatment rostral anterior cingulate cortex Theta activity in relation to symptom improvement in depression: a randomized clinical trial. JAMA psychiatry. 2018;75:547–554. doi: 10.1001/jamapsychiatry.2018.0252. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Power J.D., Cohen A.L., Nelson S.M., Wig G.S., Barnes K.A., Church J.A., Vogel A.C., Laumann T.O., Miezin F.M., Schlaggar B.L., Petersen S.E. Functional network organization of the human brain. Neuron. 2011;72:665–678. doi: 10.1016/j.neuron.2011.09.006. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Riggs L.M., Aracava Y., Zanos P., Fischell J., Albuquerque E.X., Pereira E.F.R., Thompson S.M., Gould T.D. (2R,6R)-hydroxynorketamine rapidly potentiates hippocampal glutamatergic transmission through a synapse-specific presynaptic mechanism. Neuropsychopharmacology. 2019;45:426–436. doi: 10.1038/s41386-019-0443-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Scheinost D., Holmes S.E., DellaGioia N., Schleifer C., Matuskey D., Abdallah C.G., Hampson M., Krystal J.H., Anticevic A., Esterlis I. Multimodal investigation of network level effects using intrinsic functional connectivity, anatomical covariance, and structure-to-function correlations in unmedicated major depressive disorder. Neuropsychopharmacology. 2018;43:1119–1127. doi: 10.1038/npp.2017.229. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Scheinost D., Noble S., Horien C., Greene A.S., Lake E.M., Salehi M., Gao S., Shen X., O'Connor D., Barron D.S. Ten simple rules for predictive modeling of individual differences in neuroimaging. Neuroimage. 2019;193:35–45. doi: 10.1016/j.neuroimage.2019.02.057. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Schultz D.H., Ito T., Solomyak L.I., Chen R.H., Mill R.D., Anticevic A., Cole M.W. Global connectivity of the fronto-parietal cognitive control network is related to depression symptoms in the general population. Netw. Neurosci. 2018;3:107–123. doi: 10.1162/netn_a_00056. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Sexton C.E., Allan C.L., Le Masurier M., McDermott L.M., Kalu U.G., Herrmann L.L., Maurer M., Bradley K.M., Mackay C.E., Ebmeier K.P. Magnetic resonance imaging in late-life depression: multimodal examination of network disruption. Arch. Gen. Psychiatry. 2012;69:680–689. doi: 10.1001/archgenpsychiatry.2011.1862. [DOI] [PubMed] [Google Scholar]
  34. Sheline Y.I., Price J.L., Yan Z., Mintun M.A. Resting-state functional MRI in depression unmasks increased connectivity between networks via the dorsal nexus. Proc. Natl. Acad. Sci. U S A. 2010;107:11020–11025. doi: 10.1073/pnas.1000446107. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Shen X., Finn E.S., Scheinost D., Rosenberg M.D., Chun M.M., Papademetris X., Constable R.T. Using connectome-based predictive modeling to predict individual behavior from brain connectivity. Nat. Protoc. 2017;12:506–518. doi: 10.1038/nprot.2016.178. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Trivedi M.H., Fava M., Wisniewski S.R., Thase M.E., Quitkin F., Warden D., Ritz L., Nierenberg A.A., Lebowitz B.D., Biggs M.M. Medication augmentation after the failure of SSRIs for depression. N. Engl. J. Med. 2006;354:1243–1252. doi: 10.1056/NEJMoa052964. [DOI] [PubMed] [Google Scholar]
  37. Trivedi M.H., McGrath P.J., Fava M., Parsey R.V., Kurian B.T., Phillips M.L., Oquendo M.A., Bruder G., Pizzagalli D., Toups M. Establishing moderators and biosignatures of antidepressant response in clinical care (EMBARC): Rationale and design. J. Psychiatr. Res. 2016;78:11–23. doi: 10.1016/j.jpsychires.2016.03.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Van Essen D.C., Smith S.M., Barch D.M., Behrens T.E., Yacoub E., Ugurbil K., WU-Minn HCP Consortium The WU-Minn human connectome Project: an overview. Neuroimage. 2013;80:62–79. doi: 10.1016/j.neuroimage.2013.05.041. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Wang L., Dai Z., Peng H., Tan L., Ding Y., He Z., Zhang Y., Xia M., Li Z., Li W. Overlapping and segregated resting-state functional connectivity in patients with major depressive disorder with and without childhood neglect. Hum. Brain Mapp. 2014;35:1154–1166. doi: 10.1002/hbm.22241. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Wu M., Andreescu C., Butters M.A., Tamburo R., Reynolds C.F., 3rd, Aizenstein H. Default-mode network connectivity and white matter burden in late-life depression. Psychiatry Res. 2011;194:39–46. doi: 10.1016/j.pscychresns.2011.04.003. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Yang Z., Gu S., Honnorat N., Linn K.A., Shinohara R.T., Aselcioglu I., Bruce S., Oathes D.J., Davatzikos C., Satterthwaite T.D. Network changes associated with transdiagnostic depressive symptom improvement following cognitive behavioral therapy in MDD and PTSD. Mol. Psychiatry. 2018;23:2314–2323. doi: 10.1038/s41380-018-0201-7. [DOI] [PubMed] [Google Scholar]
  42. Yeo B.T., Krienen F.M., Sepulcre J., Sabuncu M.R., Lashkari D., Hollinshead M., Roffman J.L., Smoller J.W., Zollei L., Polimeni J.R. The organization of the human cerebral cortex estimated by intrinsic functional connectivity. J. Neurophysiol. 2011;106:1125–1165. doi: 10.1152/jn.00338.2011. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Yu M., Linn K.A., Shinohara R.T., Oathes D.J., Cook P.A., Duprat R., Moore T.M., Oquendo M.A., Phillips M.L., McInnis M. Childhood trauma history is linked to abnormal brain connectivity in major depression. Proc. Natl. Acad. Sci. U S A. 2019;116:8582–8590. doi: 10.1073/pnas.1900801116. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Zanos P., Highland J.N., Stewart B.W., Georgiou P., Jenne C.E., Lovett J., Morris P.J., Thomas C.J., Moaddel R., Zarate C.A., Jr., Gould T.D. (2R,6R)-hydroxynorketamine exerts mGlu2 receptor-dependent antidepressant actions. Proc. Natl. Acad. Sci. U S A. 2019;116:6441–6450. doi: 10.1073/pnas.1819540116. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Zanos P., Moaddel R., Morris P.J., Georgiou P., Fischell J., Elmer G.I., Alkondon M., Yuan P., Pribut H.J., Singh N.S. NMDAR inhibition-independent antidepressant actions of ketamine metabolites. Nature. 2016;533:481–486. doi: 10.1038/nature17998. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Document S1. Transparent Methods and Figures S1–S7
mmc1.pdf (5MB, pdf)
Table S1. Cortical and Whole-Brain Network Affiliation of the Akiki-Abdallah Atlas, Related to Figures 3, 4, 6, and Transparent Methods
mmc2.xlsx (52.5KB, xlsx)

Data Availability Statement

The study data are available through the Human Connectome Project (https://www.humanconnectome.org) and the National Institute of Mental Health (NIMH) Data Archive (NDA; https://nda.nih.gov/). The developed Akiki-Abdallah hierarchical modularity atlas, the network-restricted strength function, and the predictive model codes will be made publicly available at https://github.com/emergelab along with the predictive models established in the current paper.

Articles from iScience are provided here courtesy of Elsevier

RESOURCES