RESEARCH ARTICLE
Reversible Functional Connectivity Disturbances during Transient Global Amnesia Michael Peer, MSc,1,2 Mor Nitzan, MSc,2,3 Ilan Goldberg, MD, PhD,1,2 Judith Katz,1 J. Moshe Gomori, MD,4 Tamir Ben-Hur, MD, PhD,1,2 and Shahar Arzy, MD, PhD1,2 Objective: Transient global amnesia (TGA), an abrupt occurrence of severe anterograde episodic amnesia accompanied by repetitive questioning, has been known for more than 50 years. Despite extensive research, there is no clear evidence for the underlying pathophysiological basis of TGA. Moreover, there is no neuroimaging method to evaluate TGA in real time. Methods: Here we used resting-state functional magnetic resonance imaging recorded in 12 patients during the acute phase of TGA together with connectivity and cluster analyses to detect changes in the episodic memory network in TGA. Results: Our results show a significant reduction in functional connectivity of the episodic memory network during TGA, which is more pronounced in the hyperacute phase than in the postacute phase. This disturbance is bilateral, and reversible after recovery. Although the hippocampus and its connections are significantly impaired, other parts of the episodic memory network are also impaired. Similar results were obtained for the analysis of the episodic memory network whether it was defined in a data-driven or literature-based manner. Interpretation: These results suggest that TGA is related to a functional disturbance in the episodic memory network, and supply a neuroimaging correlate of TGA during the acute phase. ANN NEUROL 2014;75:634–643
T
ransient global amnesia (TGA) is among the most enigmatic phenomena in neurology. TGA is defined as “a syndrome characterized by the rapid onset of antero- and retrograde amnesia, accompanied by temporal disorientation and iterative questioning that lasts up to 24 hours.”1 During the attack, patients are alert and communicative, and personal identity is preserved.2 Neuropsychologically, TGA is characterized by severe anterograde amnesia for episodic memory, but not semantic, procedural, or recognition memory.1,3,4 Retrograde amnesia is characterized by a temporal gradient, as older memories are more easily retrieved than newer ones, over a span that might reach several decades.2,3 TGA is characterized by a short hyperacute phase (usually 1–8 hours) in which memory is significantly impaired, disorientation is very severe, and iterative questioning is a hallmark, fol-
lowed by a postacute phase of gradual recovery of orientation and memory, which mostly lasts up to 24 hours.5,6 Despite hundreds of published cases, and extensive neurological, neuropsychological, and neuroimaging studies, the etiology of TGA remains unclear. Moreover, TGA diagnosis during the acute phase is based solely on clinical evaluation. Previous neuroimaging studies using positron emission tomography (PET) in TGA patients reported abnormalities in the hippocampus, amygdala, lentiform nucleus, and prefrontal cortex,7–9 but also in various other regions,1,6 making these studies difficult to interpret.6 Two single-case studies using task-related functional magnetic resonance imaging (fMRI) identified reversible reduced activity in a network of temporolimbic brain regions.10,11 In recent years, more and more studies based on diffusion-weighted imaging (DWI) MRI
View this article online at wileyonlinelibrary.com. DOI: 10.1002/ana.24137 Received Oct 7, 2013, and in revised form Mar 6, 2014. Accepted for publication Mar 7, 2014. Address correspondence to Dr Arzy, Neuropsychiatry Lab, Department of Neurology, Faculty of Medicine, Hadassah Hebrew University Medical School, Jerusalem, Israel. E-mail: [email protected] From the 1Department of Neurology, Hadassah Hebrew University Medical School, Jerusalem, Israel; 2Faculty of Medicine, Hadassah Hebrew University Medical School, Jerusalem, Israel; 3Racah Institute of Physics, Hebrew University of Jerusalem, Jerusalem, Israel; and 4Department of Radiology, Hadassah Hebrew University Medical Center, Jerusalem, Israel.
C 2014 American Neurological Association 634 V
Peer et al: Functional Connectivity in TGA
protocols have pointed to hippocampal lesions, confined to the CA1 part of the hippocampus, that correlate with TGA.12–15 Such lesions were detected in 70% of TGA patients, yet they were found bilaterally in only 12% of patients, although bilateral lesions are assumed to be needed to cause memory impairment as severe as in TGA.13,16 Interestingly, lesions mostly develop over 24 to 48 hours after TGA onset, while patients are clinically intact (but not during the acute memory impairment), and disappear several weeks later.13,14 Based on the significant clinical manifestation of TGA, we hypothesized that TGA should be detected at least functionally during the attack, and more prominently in the hyperacute phase. Moreover, we hypothesized that this disturbance might affect not only the hippocampus but also other parts of the episodic memory network bilaterally. To test these hypotheses, we scanned patients with TGA on their arrival to the emergency room, using structural MRI and resting-state functional MRI (RSfMRI). RSfMRI has proven to be a useful method in the investigation of lesional and nonlesional clinical neurological disorders.17–19 RSfMRI was performed in patients in the hyperacute or postacute phase of the disorder and after recovery, and in age-matched healthy control subjects. Episodic memory networks were defined in 2 ways: by identification of a subnetwork that is highly functionally connected to the hippocampus in separate control groups (“data-driven”), and based on a meta-analysis of episodic memory studies (“literaturebased”). In addition, voxel-based morphometry (VBM) was used to detect structural changes.
Patients and Methods Patients Twelve patients (age 5 62.7 6 7.4 years; 5 male) who presented with TGA in the hyperacute or postacute phase were included in the study (Table 1). All patients met the standard clinical criteria for diagnosis of TGA2,6: anterograde amnesia witnessed by an observer with resolution of symptoms within 24 hours, cognitive impairment limited to amnesia with no clouding of consciousness or loss of personal identity, and no focal neurological or epileptic signs or a recent history of head trauma or seizures. Physical neurological examination, computed tomography scan of the brain, and electroencephalogram results were within normal limits. Neuropsychological evaluation did not reveal any other deficits. None of the patients had a history of psychiatric or neurological illness, except for 2 patients who had previous episodes of TGA (see Table 1). All patients underwent RSfMRI scan less than 14 hours after the amnestic onset. Five patients were scanned during the hyperacute phase, and another 7 were in the postacute phase at scan time (they no longer exhibited temporal disorientation or repetitive questioning, but still displayed memory deficits).5 Five of the patients (4 hyperacute, 1
May 2014
postacute) have kindly agreed to undergo a second RSfMRI scan 2 to 9 months after the TGA episode (postrecovery group). In addition, 17 age-matched healthy volunteers (age 5 62.1 6 6.9 years; 8 male) were scanned as a control group. Volunteers had no personal history of neurologic or psychiatric disorders, and had normal structural MRI results. All participants gave written informed consent, and the study was approved by the ethical committee of the Hadassah Hebrew University Medical Center.
MRI Acquisition Procedures Patients and healthy control subjects were scanned at the same site using a Trio 3T system with a 32-channel head coil (Siemens Medical Solutions, Erlangen, Germany) using the same imaging sequence. Blood oxygen level–dependent (BOLD) fMRI was performed using a whole brain, gradient-echo echo planar imaging (EPI) sequence of 160 volumes (repetition time/echo time 5 2,000/30 milliseconds; flip angle 5 900 ; field of view 5 192 3 192mm; matrix 5 64 3 64; 33 axial slices; slice thickness/gap 5 4/0mm; voxel size 5 3 3 3 3 4mm). All subjects were instructed to stay awake, keep their eyes open, and remain still. In addition, high-resolution (1 3 1 3 1mm) T1-weighted anatomical images were acquired to aid spatial normalization to standard atlas space. All patients and healthy control subjects were also scanned using a DWI protocol to ensure the lack of past or current ischemic episodes, and in search of potential hippocampal lesions.
fMRI Preprocessing Preprocessing was conducted using SPM8 (www.fil.ion.ucl.ac. uk/spm), DPARSF,20 and MATLAB (MathWorks, Natick, MA) software. The first 5 volumes were discarded to ensure magnetization equilibrium. All functional time-series were slice-time– corrected, and motion-corrected to the mean functional image using a trilinear interpolation with 6 degrees of freedom, coregistered with the anatomical image, normalized to standard anatomical space (Montreal Neurological Institute EPI template, resampling to 3mm cubic voxels), and spatially smoothed (4mm full-width half-maximum [FWHM], isotropic). Additional preprocessing steps included the removal of linear trends to correct for signal drift and filtering with a 0.01 to 0.15Hz band-pass filter to reduce non-neuronal contributions to BOLD fluctuations. In line with recent concerns regarding the effect of subjects’ motion on functional connectivity characteristics,21–23 we performed multiple regression of 24 motion parameters24: 6 rigid-body head motion parameter values—x, y, and z translations and rotations—their value at the previous time point, and the 12 corresponding squared values. In addition, motion “spikes” were also included as regressors (identified by framewise displacement of 0.5mm), in addition to global mean, white matter, and cerebrospinal fluid signals.25 The percent of motion spikes was matched between experimental groups by addition of arbitrary spike regressors at random time points.
Extraction of Regionwise Resting-State Time Series To measure functional connectivity we first defined a whole brain network using the Automatic Anatomical Labeling (AAL)
635
636
65
64
52
70
77
54
62
68
72
64
3
4
5
6
7
8
9
10
11
12
M
M
F
M
F
M
M
F
F
F
F
—
—
—
—
—
—
—
1
Physical activity —
Stressful event
Physical activity —
Stressful event
—
At work
—
At work
Stressful event
Physical activity —
Stressful event
2
Several months
1 day
1 day
1 hour
Several months
None
Several days
1 week
Several months
1 hour
1 day
1 day
5
5
5
5
5
5
Unknown
2
Unknown
Unknown
3
3
—
—
—
—
—
—
—
—
1
—
—
—
7
3
4
10
10
0.5
2
5
6
7
12
12
14
5
7
11
12
4
12
2
5
6
5
8
7
2
3
1
2
3.5
10
23
21
21
27
24
Postacute
Postacute
Postacute
Postacute
Postacute
Postacute
Postacute
Hyperacute
Hyperacute
Hyperacute
Hyperacute
Hyperacute
No
No
No
Yes
No
No
No
No
Yes
Yes
Yes
Yes
Past TGA Length of Time Hippocampal TGA Scan Time Scan Time TGA Phase Scan at Time of after Events, Retrograde Orientation at Lesion in Duration, from from Scan Recovery No. Amnesia Time of Scana DWI h Onset, h End, h
Orientation in time was measured using questions from the Mini-Mental State Examination, and scored on a scale of 1 to 5. DWI 5 diffusion-weighted imaging; F 5 female; M 5 male; TGA 5 transient global amnesia.
a
74
2
Stressful event
66
1
F
Age, yr Sex Onset Circumstances
Patient
TABLE 1. Patient Demographic and Clinical Details
ANNALS
of Neurology
Volume 75, No. 5
Peer et al: Functional Connectivity in TGA
atlas, which defines 45 brain regions in each cerebral hemisphere.26 To ensure that voxels in each region were indeed a part of the cerebral cortex, we used the new-segment algorithm of SPM8, which identifies different tissue types on the T1 anatomical image of each subject, and used the resulting gray matter image to create a mask of the gray matter (gray matter segmentation intensity > 0.01). We used this mask to ensure that only gray matter voxels were used for averaging each region. To avoid using voxels that are affected by signal dropout,27 we fitted a Gaussian model to the voxel intensity graph across the image and used a threshold of a 5 0.01 to create a mask including only high-intensity voxels and excluding voxels with low functional signal. Fitting was performed on images before the smoothing, filtering, and covariates regression steps, which change the image intensity (average goodness of fit [adjusted R2] 2 0.8). Regions containing 0.3). This disturbance was found in both hemispheres and statistical analysis did not reveal any difference between right and left hemispheres (all p values, >0.2). These connectivity differences were specific to the episodic memory network, as no significant differences were found between groups in other functional networks (motor, language, p 5 0.33, 0.16, respectively), and in a network comprised of stress-related regions (p 5 0.31).30–33 Distribution of Functional Connectivity Disturbance of the Episodic Memory Network in TGA To further investigate the connectivity disturbance in TGA and its specificity to hippocampal connections, we used a hierarchical clustering analysis of the healthy control subjects, which divides the episodic memory network into clustered subnetworks based on their connectivity 639
ANNALS
of Neurology
FIGURE 3: Hierarchical clustering of the episodic memory network. (A) Connectivity and subnetwork clusters in the data-driven (left) and literature-based (right) episodic memory networks. Each row/column represents a specific brain region. Strength of functional connectivity between the corresponding regions is represented by color code. Hierarchical clustering revealed 6 main clusters: frontocingulate (FC, purple), medial-temporal (MTL, turquoise), deep structures (DS, yellow), medial-occipital (MO, blue), inferiortemporal (IT, red) and inferior-frontal triangular (TR, orange). (B) Intra- and intercluster connectivity of TGA (transient global amnesia) patients (hyper- and postacute phases) and healthy control subjects in the data-driven (left) and literature-based (right) episodic memory networks. Line width represents connectivity strength between clusters, and cluster size represents intracluster connectivity. Inter- and intracluster connectivity values that significantly differ between control subjects and TGA patients are marked by solid lines, and values that do not significantly differ are marked by dashed lines. Note that functional disturbances are apparent in large parts of the episodic memory network. inf 5 inferior; L 5 left; med 5 medial; mid 5 middle; post 5 posterior; R 5 right; sup 5 superior.
pattern, in a data-driven fashion. The clustering assignments were similar across the 4 experimental groups in the data-driven and literature-based networks, revealing 5 major clusters: orbitofrontal cingulate, medial temporal, deep structures, inferior temporal, and occipital (Fig 3A; in the literature-based network an additional cluster of inferior frontal triangular gyri was found, which is anticorrelated to the rest of the network). Comparison of intercluster connectivity in control subjects and in TGA patients (hyper- and postacute combined) revealed significant connectivity disturbances between the medial temporal cluster and other clusters, but also between parts of the network that do not involve the hippocampus (see Fig 3B). In addition to intercluster connectivity disturbances, significant disturbances were 640
also apparent within the medial temporal, frontocingulate, medial occipital, inferior temporal, and deep-structures clusters (all p values
Reversible Functional Connectivity Disturbances during Transient Global Amnesia Michael Peer, MSc,1,2 Mor Nitzan, MSc,2,3 Ilan Goldberg, MD, PhD,1,2 Judith Katz,1 J. Moshe Gomori, MD,4 Tamir Ben-Hur, MD, PhD,1,2 and Shahar Arzy, MD, PhD1,2 Objective: Transient global amnesia (TGA), an abrupt occurrence of severe anterograde episodic amnesia accompanied by repetitive questioning, has been known for more than 50 years. Despite extensive research, there is no clear evidence for the underlying pathophysiological basis of TGA. Moreover, there is no neuroimaging method to evaluate TGA in real time. Methods: Here we used resting-state functional magnetic resonance imaging recorded in 12 patients during the acute phase of TGA together with connectivity and cluster analyses to detect changes in the episodic memory network in TGA. Results: Our results show a significant reduction in functional connectivity of the episodic memory network during TGA, which is more pronounced in the hyperacute phase than in the postacute phase. This disturbance is bilateral, and reversible after recovery. Although the hippocampus and its connections are significantly impaired, other parts of the episodic memory network are also impaired. Similar results were obtained for the analysis of the episodic memory network whether it was defined in a data-driven or literature-based manner. Interpretation: These results suggest that TGA is related to a functional disturbance in the episodic memory network, and supply a neuroimaging correlate of TGA during the acute phase. ANN NEUROL 2014;75:634–643
T
ransient global amnesia (TGA) is among the most enigmatic phenomena in neurology. TGA is defined as “a syndrome characterized by the rapid onset of antero- and retrograde amnesia, accompanied by temporal disorientation and iterative questioning that lasts up to 24 hours.”1 During the attack, patients are alert and communicative, and personal identity is preserved.2 Neuropsychologically, TGA is characterized by severe anterograde amnesia for episodic memory, but not semantic, procedural, or recognition memory.1,3,4 Retrograde amnesia is characterized by a temporal gradient, as older memories are more easily retrieved than newer ones, over a span that might reach several decades.2,3 TGA is characterized by a short hyperacute phase (usually 1–8 hours) in which memory is significantly impaired, disorientation is very severe, and iterative questioning is a hallmark, fol-
lowed by a postacute phase of gradual recovery of orientation and memory, which mostly lasts up to 24 hours.5,6 Despite hundreds of published cases, and extensive neurological, neuropsychological, and neuroimaging studies, the etiology of TGA remains unclear. Moreover, TGA diagnosis during the acute phase is based solely on clinical evaluation. Previous neuroimaging studies using positron emission tomography (PET) in TGA patients reported abnormalities in the hippocampus, amygdala, lentiform nucleus, and prefrontal cortex,7–9 but also in various other regions,1,6 making these studies difficult to interpret.6 Two single-case studies using task-related functional magnetic resonance imaging (fMRI) identified reversible reduced activity in a network of temporolimbic brain regions.10,11 In recent years, more and more studies based on diffusion-weighted imaging (DWI) MRI
View this article online at wileyonlinelibrary.com. DOI: 10.1002/ana.24137 Received Oct 7, 2013, and in revised form Mar 6, 2014. Accepted for publication Mar 7, 2014. Address correspondence to Dr Arzy, Neuropsychiatry Lab, Department of Neurology, Faculty of Medicine, Hadassah Hebrew University Medical School, Jerusalem, Israel. E-mail: [email protected] From the 1Department of Neurology, Hadassah Hebrew University Medical School, Jerusalem, Israel; 2Faculty of Medicine, Hadassah Hebrew University Medical School, Jerusalem, Israel; 3Racah Institute of Physics, Hebrew University of Jerusalem, Jerusalem, Israel; and 4Department of Radiology, Hadassah Hebrew University Medical Center, Jerusalem, Israel.
C 2014 American Neurological Association 634 V
Peer et al: Functional Connectivity in TGA
protocols have pointed to hippocampal lesions, confined to the CA1 part of the hippocampus, that correlate with TGA.12–15 Such lesions were detected in 70% of TGA patients, yet they were found bilaterally in only 12% of patients, although bilateral lesions are assumed to be needed to cause memory impairment as severe as in TGA.13,16 Interestingly, lesions mostly develop over 24 to 48 hours after TGA onset, while patients are clinically intact (but not during the acute memory impairment), and disappear several weeks later.13,14 Based on the significant clinical manifestation of TGA, we hypothesized that TGA should be detected at least functionally during the attack, and more prominently in the hyperacute phase. Moreover, we hypothesized that this disturbance might affect not only the hippocampus but also other parts of the episodic memory network bilaterally. To test these hypotheses, we scanned patients with TGA on their arrival to the emergency room, using structural MRI and resting-state functional MRI (RSfMRI). RSfMRI has proven to be a useful method in the investigation of lesional and nonlesional clinical neurological disorders.17–19 RSfMRI was performed in patients in the hyperacute or postacute phase of the disorder and after recovery, and in age-matched healthy control subjects. Episodic memory networks were defined in 2 ways: by identification of a subnetwork that is highly functionally connected to the hippocampus in separate control groups (“data-driven”), and based on a meta-analysis of episodic memory studies (“literaturebased”). In addition, voxel-based morphometry (VBM) was used to detect structural changes.
Patients and Methods Patients Twelve patients (age 5 62.7 6 7.4 years; 5 male) who presented with TGA in the hyperacute or postacute phase were included in the study (Table 1). All patients met the standard clinical criteria for diagnosis of TGA2,6: anterograde amnesia witnessed by an observer with resolution of symptoms within 24 hours, cognitive impairment limited to amnesia with no clouding of consciousness or loss of personal identity, and no focal neurological or epileptic signs or a recent history of head trauma or seizures. Physical neurological examination, computed tomography scan of the brain, and electroencephalogram results were within normal limits. Neuropsychological evaluation did not reveal any other deficits. None of the patients had a history of psychiatric or neurological illness, except for 2 patients who had previous episodes of TGA (see Table 1). All patients underwent RSfMRI scan less than 14 hours after the amnestic onset. Five patients were scanned during the hyperacute phase, and another 7 were in the postacute phase at scan time (they no longer exhibited temporal disorientation or repetitive questioning, but still displayed memory deficits).5 Five of the patients (4 hyperacute, 1
May 2014
postacute) have kindly agreed to undergo a second RSfMRI scan 2 to 9 months after the TGA episode (postrecovery group). In addition, 17 age-matched healthy volunteers (age 5 62.1 6 6.9 years; 8 male) were scanned as a control group. Volunteers had no personal history of neurologic or psychiatric disorders, and had normal structural MRI results. All participants gave written informed consent, and the study was approved by the ethical committee of the Hadassah Hebrew University Medical Center.
MRI Acquisition Procedures Patients and healthy control subjects were scanned at the same site using a Trio 3T system with a 32-channel head coil (Siemens Medical Solutions, Erlangen, Germany) using the same imaging sequence. Blood oxygen level–dependent (BOLD) fMRI was performed using a whole brain, gradient-echo echo planar imaging (EPI) sequence of 160 volumes (repetition time/echo time 5 2,000/30 milliseconds; flip angle 5 900 ; field of view 5 192 3 192mm; matrix 5 64 3 64; 33 axial slices; slice thickness/gap 5 4/0mm; voxel size 5 3 3 3 3 4mm). All subjects were instructed to stay awake, keep their eyes open, and remain still. In addition, high-resolution (1 3 1 3 1mm) T1-weighted anatomical images were acquired to aid spatial normalization to standard atlas space. All patients and healthy control subjects were also scanned using a DWI protocol to ensure the lack of past or current ischemic episodes, and in search of potential hippocampal lesions.
fMRI Preprocessing Preprocessing was conducted using SPM8 (www.fil.ion.ucl.ac. uk/spm), DPARSF,20 and MATLAB (MathWorks, Natick, MA) software. The first 5 volumes were discarded to ensure magnetization equilibrium. All functional time-series were slice-time– corrected, and motion-corrected to the mean functional image using a trilinear interpolation with 6 degrees of freedom, coregistered with the anatomical image, normalized to standard anatomical space (Montreal Neurological Institute EPI template, resampling to 3mm cubic voxels), and spatially smoothed (4mm full-width half-maximum [FWHM], isotropic). Additional preprocessing steps included the removal of linear trends to correct for signal drift and filtering with a 0.01 to 0.15Hz band-pass filter to reduce non-neuronal contributions to BOLD fluctuations. In line with recent concerns regarding the effect of subjects’ motion on functional connectivity characteristics,21–23 we performed multiple regression of 24 motion parameters24: 6 rigid-body head motion parameter values—x, y, and z translations and rotations—their value at the previous time point, and the 12 corresponding squared values. In addition, motion “spikes” were also included as regressors (identified by framewise displacement of 0.5mm), in addition to global mean, white matter, and cerebrospinal fluid signals.25 The percent of motion spikes was matched between experimental groups by addition of arbitrary spike regressors at random time points.
Extraction of Regionwise Resting-State Time Series To measure functional connectivity we first defined a whole brain network using the Automatic Anatomical Labeling (AAL)
635
636
65
64
52
70
77
54
62
68
72
64
3
4
5
6
7
8
9
10
11
12
M
M
F
M
F
M
M
F
F
F
F
—
—
—
—
—
—
—
1
Physical activity —
Stressful event
Physical activity —
Stressful event
—
At work
—
At work
Stressful event
Physical activity —
Stressful event
2
Several months
1 day
1 day
1 hour
Several months
None
Several days
1 week
Several months
1 hour
1 day
1 day
5
5
5
5
5
5
Unknown
2
Unknown
Unknown
3
3
—
—
—
—
—
—
—
—
1
—
—
—
7
3
4
10
10
0.5
2
5
6
7
12
12
14
5
7
11
12
4
12
2
5
6
5
8
7
2
3
1
2
3.5
10
23
21
21
27
24
Postacute
Postacute
Postacute
Postacute
Postacute
Postacute
Postacute
Hyperacute
Hyperacute
Hyperacute
Hyperacute
Hyperacute
No
No
No
Yes
No
No
No
No
Yes
Yes
Yes
Yes
Past TGA Length of Time Hippocampal TGA Scan Time Scan Time TGA Phase Scan at Time of after Events, Retrograde Orientation at Lesion in Duration, from from Scan Recovery No. Amnesia Time of Scana DWI h Onset, h End, h
Orientation in time was measured using questions from the Mini-Mental State Examination, and scored on a scale of 1 to 5. DWI 5 diffusion-weighted imaging; F 5 female; M 5 male; TGA 5 transient global amnesia.
a
74
2
Stressful event
66
1
F
Age, yr Sex Onset Circumstances
Patient
TABLE 1. Patient Demographic and Clinical Details
ANNALS
of Neurology
Volume 75, No. 5
Peer et al: Functional Connectivity in TGA
atlas, which defines 45 brain regions in each cerebral hemisphere.26 To ensure that voxels in each region were indeed a part of the cerebral cortex, we used the new-segment algorithm of SPM8, which identifies different tissue types on the T1 anatomical image of each subject, and used the resulting gray matter image to create a mask of the gray matter (gray matter segmentation intensity > 0.01). We used this mask to ensure that only gray matter voxels were used for averaging each region. To avoid using voxels that are affected by signal dropout,27 we fitted a Gaussian model to the voxel intensity graph across the image and used a threshold of a 5 0.01 to create a mask including only high-intensity voxels and excluding voxels with low functional signal. Fitting was performed on images before the smoothing, filtering, and covariates regression steps, which change the image intensity (average goodness of fit [adjusted R2] 2 0.8). Regions containing 0.3). This disturbance was found in both hemispheres and statistical analysis did not reveal any difference between right and left hemispheres (all p values, >0.2). These connectivity differences were specific to the episodic memory network, as no significant differences were found between groups in other functional networks (motor, language, p 5 0.33, 0.16, respectively), and in a network comprised of stress-related regions (p 5 0.31).30–33 Distribution of Functional Connectivity Disturbance of the Episodic Memory Network in TGA To further investigate the connectivity disturbance in TGA and its specificity to hippocampal connections, we used a hierarchical clustering analysis of the healthy control subjects, which divides the episodic memory network into clustered subnetworks based on their connectivity 639
ANNALS
of Neurology
FIGURE 3: Hierarchical clustering of the episodic memory network. (A) Connectivity and subnetwork clusters in the data-driven (left) and literature-based (right) episodic memory networks. Each row/column represents a specific brain region. Strength of functional connectivity between the corresponding regions is represented by color code. Hierarchical clustering revealed 6 main clusters: frontocingulate (FC, purple), medial-temporal (MTL, turquoise), deep structures (DS, yellow), medial-occipital (MO, blue), inferiortemporal (IT, red) and inferior-frontal triangular (TR, orange). (B) Intra- and intercluster connectivity of TGA (transient global amnesia) patients (hyper- and postacute phases) and healthy control subjects in the data-driven (left) and literature-based (right) episodic memory networks. Line width represents connectivity strength between clusters, and cluster size represents intracluster connectivity. Inter- and intracluster connectivity values that significantly differ between control subjects and TGA patients are marked by solid lines, and values that do not significantly differ are marked by dashed lines. Note that functional disturbances are apparent in large parts of the episodic memory network. inf 5 inferior; L 5 left; med 5 medial; mid 5 middle; post 5 posterior; R 5 right; sup 5 superior.
pattern, in a data-driven fashion. The clustering assignments were similar across the 4 experimental groups in the data-driven and literature-based networks, revealing 5 major clusters: orbitofrontal cingulate, medial temporal, deep structures, inferior temporal, and occipital (Fig 3A; in the literature-based network an additional cluster of inferior frontal triangular gyri was found, which is anticorrelated to the rest of the network). Comparison of intercluster connectivity in control subjects and in TGA patients (hyper- and postacute combined) revealed significant connectivity disturbances between the medial temporal cluster and other clusters, but also between parts of the network that do not involve the hippocampus (see Fig 3B). In addition to intercluster connectivity disturbances, significant disturbances were 640
also apparent within the medial temporal, frontocingulate, medial occipital, inferior temporal, and deep-structures clusters (all p values
