Introduction
Post-traumatic stress disorder (PTSD), an anxiety disorder that is characterized by intrusive vivid memories of the traumatic event, avoidance of any reminders of this trauma, and hyperarousal, has been proposed to arise as the result of two types of pathogenic neural processes, the first characterized by exaggerated responses to a stressful event and the second by inadequate mechanisms for recovery (Yehuda & LeDoux, Reference Yehuda and LeDoux2007). Consistent with this notion, many neuroimaging studies of PTSD have reported functional and structural abnormalities in regions associated with emotion and memory processing, such as the amygdala, hippocampus and anterior cingulate cortex (ACC) (Karl et al. Reference Karl, Schaefer, Malta, Dorfel, Rohleder and Werner2006; Rauch et al. Reference Rauch, Shin and Phelps2006; Francati et al. Reference Francati, Vermetten and Bremner2007).
Most of the work has focused on identifying the neural correlates of current PTSD symptomatology (e.g. Armony et al. Reference Armony, Corbo, Clement and Brunet2005; Shin et al. Reference Shin, Wright, Cannistraro, Wedig, McMullin, Martis, Macklin, Lasko, Cavanagh, Krangel, Orr, Pitman, Whalen and Rauch2005; Williams et al. Reference Williams, Kemp, Felmingham, Barton, Olivieri, Peduto, Gordon and Bryant2006; Dickie et al. Reference Dickie, Brunet, Akerib and Armony2008; Kroes et al. Reference Kroes, Whalley, Rugg and Brewin2011). However, more recently, researchers have begun to identify potential neural markers that can predict recovery from PTSD. For instance, a positive response to eye movement desensitization and reprocessing (EMDR) therapy was shown to be associated with increased grey matter (GM) density in the posterior cingulate cortex, premotor area, insula and parahippocampal gyrus at baseline (Nardo et al. Reference Nardo, Hogberg, Looi, Larsson, Hallstrom and Pagani2010), whereas GM volume in rostral ACC correlated with symptom improvement after an 8-week course of cognitive behavioural therapy (Bryant et al. Reference Bryant, Felmingham, Whitford, Kemp, Hughes, Peduto and Williams2008b). The latter group also showed that, at the functional level, greater bilateral amygdala and ventral ACC activation in response to masked fearful faces at baseline was associated with poor improvement following treatment (Bryant et al. Reference Bryant, Felmingham, Kemp, Das, Hughes, Peduto and Williams2008a). Dickie et al. (Reference Dickie, Brunet, Akerib and Armony2011) reported that memory-related activity in the hippocampus and subgenual ACC (sgACC) predicted symptom severity changes 6–9 months later.
Taken together, these studies point towards the ventromedial aspect of the prefrontal cortex (vmPFC), including the sgACC, as a potential marker for the capacity to recover from PTSD. This fits with the established role of this structure in the extinction of conditioned fear responses (Milad et al. Reference Milad, Rauch, Pitman and Quirk2006; Quirk et al. Reference Quirk, Garcia and Gonzalez-Lima2006). More direct evidence for a role of the vmPFC in the PTSD recovery process itself comes from two functional magnetic resonance imaging (fMRI) studies showing that not only does activity in this region predict recovery but also, across time, changes in its response correlate with symptom severity reduction (Felmingham et al. Reference Felmingham, Kemp, Williams, Das, Hughes, Peduto and Bryant2007; Dickie et al. Reference Dickie, Brunet, Akerib and Armony2011). Whether these functional processes are associated with concomitant structural changes remains unknown. In particular, it remains to be determined whether the anatomical differences observed in PTSD, particularly in the vmPFC, represent ‘trait-like’ individual differences in the capacity for recovery, in which case they should remain stable during the process of remission, or whether instead they are ‘state dependent’ and therefore follow, and possibly underlie, recovery. The former hypothesis would be consistent with studies showing that vmPFC thickness in healthy individuals correlates with the ability to recall fear extinction (Milad et al. Reference Milad, Quinn, Pitman, Orr, Fischl and Rauch2005), whereas the latter would be in line with studies of major depression and bipolar disorder showing structural changes in regions such as the putamen and ACC after successful treatment (Costafreda et al. Reference Costafreda, Chu, Ashburner and Fu2009; Moore et al. Reference Moore, Cortese, Glitz, Zajac-Benitez, Quiroz, Uhde, Drevets and Manji2009).
To address this question directly, we analysed the GM structure (cortical thickness) of highly symptomatic PTSD patients who were beginning to undergo treatment. To identify potential predictors of the recovery process, the GM thickness values from vertices across the whole cortical surface at baseline (Time 1) were correlated with changes in symptom severity 6–9 months later (Time 2), when about half of the participants were no longer symptomatic. To understand the longitudinal progression of these potential GM markers of the recovery process, some participants were scanned again at Time 2. This allowed us to directly explore the time course of structural integrity in the regions identified as predictors of recovery and its relationship to symptom severity reduction. Specifically, our main objectives were: (1) to identify brain regions in which cortical thickness predicted symptom severity reduction 6 to 9 months later, and (2) to assess whether there was a change in structural integrity in these areas as a function of recovery.
Method
Subjects
Thirty participants (age range 20–60 years) suffering from DSM-IV-TR PTSD were recruited from Montreal area clinics (Traumatys Clinic and Charles LeMoyne Hospital). To be considered for the study, individuals had to have a primary diagnosis of PTSD, no history of neurological, learning or psychotic disorders and no medical contra-indications for MRI scanning. In addition, inclusion criteria required participants to score >45 in the Clinician-Administered PTSD Scale (CAPS; Blake et al. Reference Blake, Weathers, Nagy, Kaloupek, Gusman, Charney and Keane1995). Clinical and demographic characteristics of the final sample of participants are given in Table 1. Participants were scanned after a first clinical assessment (Time 1; delay: mean = 9.3, s.d. = 9.0 days) and 6–61 weeks following trauma exposure. A second clinical assessment (Time 2) took place 22–53 weeks later (mean = 33.8, s.d. = 6.3). Twenty-five of these 30 participants completed a second MRI scan shortly after this second clinical assessment (delay: mean = 11.8, s.d. = 12.5 days). Missing data from five participants were due to pregnancy (n = 2), severe back pain (n = 1) and claustrophobia (n = 1), in addition to technical issues (n = 1).
Table 1. Demographic and clinical data (n = 30)
PDI, Peritraumatic Distress Inventory; PDEQ, Peritraumatic Dissociative Experience Questionnaire; MVA, motor vehicle accident; CAPS, Clinician-Administered PTSD Scale; BDI, Beck Depression Inventory; s.d., standard deviation; n.s., not significant; n.a., not statistically assessed (due to lack of statistical power, as factor was only present in 1 or 2 participants).
Continuous variables and longitudinal measures given as mean (s.d.), categorical variables as n (%).
Symptom improvement is defined as CAPS1 – CAPS2.
a Tested those who suffered MVA against all others.
b Missing Time 2 data from one subject.
Clinical assessments were conducted by a clinical psychologist (V.A.) and consisted of the CAPS, the Mini International Neuropsychiatric Interview (MINI; Sheehan et al. Reference Sheehan, Lecrubier, Sheehan, Amorim, Janavs, Weiller, Hergueta, Baker and Dunbar1998), the Beck Depression Inventory (BDI; Beck et al. Reference Beck, Ward, Mendelson, Mock and Erbaugh1961), the Peritraumatic Distress Inventory (PDI; Brunet et al. Reference Brunet, Weiss, Metzler, Best, Neylan, Rogers, Fagan and Marmar2001) and the Peritraumatic Dissociative Experience Questionnaire (PDEQ; Marmar et al. Reference Marmar, Weiss, Schlenger, Fairbank, Jordan, Kulka and Hough1994).
All participants provided informed consent and received financial compensation for their time and travel expenses. All procedures were approved by the ethical review boards of the Douglas Mental Health University Institute, the Faculty of Medicine of McGill University and the Charles LeMoyne University Hospital. Most patients had also been referred by these Montreal clinics for psychotherapy with a local therapist; however, detailed information about the type of psychotherapy provided (i.e. cognitive behaviour, interpersonal, etc.) or the degree of compliance with prescribed medication was not available to the authors of this study.
MRI acquisition and cortical thickness analysis
Scanning was conducted at the Montreal Neurological Institute (MNI) using a 1.5-T Siemens Sonata whole-body system equipped with a standard head coil. A T1-weighted anatomical volume was collected using a gradient echo pulse sequence [repetition time (TR) = 22 ms, echo time (TE) = 9.2 ms, flip angle = 30°, voxel size 1 mm3 isotropic]. Participants also completed an fMRI scan of memory encoding, reported in Dickie et al. (Reference Dickie, Brunet, Akerib and Armony2008, Reference Dickie, Brunet, Akerib and Armony2011).
Cortical thickness was calculated using a processing pipeline developed at MNI (CIVET; http://www.bic.mni.mcgill.ca/∼alan/tools.html;Zijdenbos et al. Reference Zijdenbos, Forghani and Evans2002; Ad-Dab'bagh et al. Reference Ad-Dab'bagh, Lyttelton, Muehlboeck, Lepage, Einarson, Mok, Ivanov, Vincent, Lerch, Fombonne, Evans and Corbetta2006). Native images were corrected for non-uniformity (Sled et al. Reference Sled, Zijdenbos and Evans1998) and then registered to the stereotaxic space of Talairach & Tournoux (Reference Talairach and Tournoux1988) using the International Consortium for Brain Mapping (ICBM) nonlinear sixth-generation template (Collins et al. Reference Collins, Neelin, Peters and Evans1994; Grabner et al. Reference Grabner, Janke, Budge, Smith, Pruessner and Collins2006). Registered images were classified into GM, white matter (WM) and cerebrospinal fluid (CSF) (Zijdenbos et al. Reference Zijdenbos, Forghani and Evans2002). Next, the Constrained Laplacian Anatomic Segmentation using Proximity (CLASP; Kim et al. Reference Kim, Singh, Lee, Lerch, Ad-Dab'bagh, MacDonald, Lee, Kim and Evans2005) algorithm was applied to each hemisphere independently. This algorithm iteratively warps an 81920 polygonal surface mesh (encompassing 40962 vertices) to fit the boundary between GM and WM (the inner surface), and then expands outward along a Laplacian map to find the outer boundary of the cortex (the GM/CSF boundary). A mid-surface map was created using the mid-points of linked inner and outer surfaces. These surface maps were then nonlinearly aligned to an iterative group registration template (Lyttelton et al. Reference Lyttelton, Boucher, Robbins and Evans2007) using a two-dimensional registration procedure (Robbins et al. Reference Robbins, Evans, Collins and Whitesides2004). Cortical thickness was measured as the distance between corresponding vertices of the GM/WM and the GM/CSF boundary surfaces (Lerch & Evans, Reference Lerch and Evans2005). An inverse linear registration matrix was applied so that values presented would correspond to measurements in native space. In addition, cortical surfaces from the second assessment (Time 2) were aligned to the surfaces acquired during the first assessment (Time 1). Cortical thickness and cortical surface area maps were blurred using a surface-based 20-mm full-width at half-maximum (FWHM) diffusion smoothing kernel (Lerch & Evans, Reference Lerch and Evans2005). Diffusion smoothing, unlike the volumetric blurring used in other methods (e.g. voxel-based morphometry, VBM), follows the surface manifold and thus respects anatomical boundaries.
Statistical analyses and figure generation were performed using the SurfStat Matlab toolbox (www.math.mcgill.ca/keith/surfstat/). Symptom improvement, defined as the difference between CAPS scores at the two testing sessions (CAPS1 – CAPS2), was regressed against cortical thickness and surface area values at each vertex. Participants' age, number of weeks between clinical testing sessions and total brain volumes were included as covariates in regression models. Multiple comparisons across the cortical surface were accounted for using random field theory for non-isotropic images on a cluster level threshold of p < 0.05 family-wise error rate (Worsley et al. Reference Worsley, Andermann, Koulis, MacDonald and Evans1999). The thickness values from vertices of interest were extracted for post-hoc analysis and plotting. To test potential confounding of depressive symptom change, the changes in BDI scores were introduced independently as additional covariates into the regression model at each peak of interest. The same approach was applied to investigate potential mediating effects of participants' sex, and time since trauma exposure.
Results
Clinical outcomes
Clinical and demographic characteristics of the group, and their relationship to symptom improvement (the difference between CAPS scores at the two testing sessions: CAPS1 – CAPS2), are shown in Table 1. A total of 90% of patients experienced a reduction in symptom severity between our initial assessment (Time 1) and the 6–9 month follow-up (Time 2), with 56% of them no longer meeting diagnostic criteria for PTSD at Time 2. PTSD symptom severity was correlated with BDI scores at both times (r = 0.67, p < 0.001 and r = 0.55, p = 0.002 respectively). PDEQ scores correlated with CAPS scores at Time 1 (r = 0.46, p = 0.01) but not at Time 2 (r = 0.10, p = 0.6). As expected, reduction in PTSD symptoms was correlated with a reduction in depressive symptoms (r = 0.48, p = 0.008).
Cortical thickness
PTSD symptoms severity improvement correlated with Time 1 cortical thickness in the right sgACC [peak: x, y, z = (3, 19, −10), z (peak) = 3.72, cluster: 299 vertices, p = 0.01 corrected, see Fig. 1], after controlling for the effects of age, weeks between clinical testing sessions and total brain volume. No regions showed the opposite relationship (i.e. a negative correlation between symptom changes and cortical thickness) after controlling for multiple comparisons across the cortical surface. To investigate the longitudinal nature of this relationship, the mean cortical thickness values from all vertices within this cluster at Times 1 and 2 were extracted for the 25 participants who underwent both scans. Symptom improvement was correlated with cortical thickness within this sgACC cluster at both Time 1 (r = 0.56, p = 0.003) and Time 2 (r = 0.50, p = 0.01). There were no significant differences in cortical thickness in this region between both scanning times [t(24) = 0.9, p = 0.37], nor a relationship between change in cortical thickness across time and symptom improvement (r = 0.12, p > 0.5). Lack of significant effects was unlikely to be due insufficient number of subjects, as power analysis indicated that achieving significance at a liberal threshold of p = 0.05, uncorrected for multiple comparisons, would require a sample size of at least 240 and 540 for the former and latter comparisons, respectively.
Fig. 1. (a) Statistical parametric map depicting the positive correlation between right anterior cingulate cortical (ACC) thickness at the time of our initial assessment (Time 1) and post-traumatic stress disorder (PTSD) symptom improvement, indexed by change in the Clinician-Administered PTSD Scale (CAPS), between Time 1 and the follow-up assessment 6–9 months later (Time 2), controlling for participants' age, number of weeks between clinical assessments and total brain volume. The cluster is plotted on the average of all participants' realigned mid-surfaces. This was the only significant cluster on the cortical surface after correcting for multiple comparisons using random Gaussian field theory. (b) Cortical thickness values from all vertices within this cluster at Time 1 and Time 2 were extracted and averaged for each participant, then plotted against symptom improvement (CAPS1 – CAPS2). Lighter circles represent those participants for whom Time 2 magnetic resonance imaging (MRI) data were not available. The line represents the significant positive correlation between cortical thickness and symptom improvement. (c) Scatterplots of the relationship between CAPS scores (left) and cortical thickness for the ACC (right) at Time 1 and Time 2. Broken lines represent the 45° line representing no change between testing times. These plots illustrate that most participants experienced a reduction in symptom severity but that there were no significant changes in cortical thickness in the ACC cluster.
Potential mediating effects of other clinical and demographic factors (sex, time since trauma and change in BDI) were also tested independently against sgACC thickness at Time 1, Time 2 and their difference (all models included CAPS1 – CAPS2, along with total brain volume, age and time between scanning sessions added as covariates). No significant correlations with our potential mediators were observed for any of the analyses, even without applying Bonferroni correction for multiple testing (highest t = 1.29, p = 0.21 uncorrected). Additionally, potential modulation of the relationship between symptom improvement and sgACC cortical thickness by any of the above-mentioned factors was tested for through the independent addition of interaction terms to the original (Time 1) model (keeping total brain volume, age and time between testing sessions as covariates). No significant interactions were found (highest t = 1.13, p = 0.27).
Discussion
In this study we sought to identify the structural neural correlates of potential for recovery from PTSD. We found that future recovery was associated with larger cortical thickness in the sgACC. To investigate the potential change in this area as a function of symptom reduction, we measured cortical thickness in a subgroup of participants at the time of the second clinical evaluation, 6–9 months later, when about half of the participants were no longer symptomatic. The relationship between recovery and cortical thickness was also observed at this later time point. Importantly, there was no relationship between changes in cortical thickness and changes in symptom severity between these two time points.
A large body of work has highlighted structural abnormalities in the vmPFC, including the pregenual and subgenual ACC, in PTSD, using manual tracing (Rauch et al. Reference Rauch, Shin, Segal, Pitman, Carson, McMullin, Whalen and Makris2003; Woodward et al. Reference Woodward, Kaloupek, Streeter, Martinez, Schaer and Eliez2006), VBM (Yamasue et al. Reference Yamasue, Kasai, Iwanami, Ohtani, Yamada, Abe, Kuroki, Fukuda, Tochigi, Furukawa, Sadamatsu, Sasaki, Aoki, Ohtomo, Asukai and Kato2003; Corbo et al. Reference Corbo, Clement, Armony, Pruessner and Brunet2005; Chen et al. Reference Chen, Xia, Li, Liu, He, Zhang, Yan, Zhang and Hu2006; Kasai et al. Reference Kasai, Yamasue, Gilbertson, Shenton, Rauch and Pitman2008) and cortical thickness (Woodward et al. Reference Woodward, Schaer, Kaloupek, Cediel and Eliez2009). Furthermore, hypoactivity in this region is one of the most commonly reported functional alterations observed in this population (Francati et al. Reference Francati, Vermetten and Bremner2007). More specifically, this region may play an important role in the recovery process, as suggested by some functional (Felmingham et al. Reference Felmingham, Kemp, Williams, Das, Hughes, Peduto and Bryant2007; Dickie et al. Reference Dickie, Brunet, Akerib and Armony2011) and structural (Bryant et al. Reference Bryant, Felmingham, Whitford, Kemp, Hughes, Peduto and Williams2008b) neuroimaging studies. The correlation between sgACC thickness and future symptom reduction observed in this study provides additional support to this hypothesis.
The observed association between anterior cingulate structure and potential for recovery from PTSD may be related to the process of extinction learning, a proposed crucial component of recovery from PTSD (Milad et al. Reference Milad, Rauch, Pitman and Quirk2006; Rauch et al. Reference Rauch, Shin and Phelps2006). This proposed link between PTSD and a failure in extinction learning has received important empirical support in the past few years. For instance, it has been shown that Vietnam war veterans with PTSD differ from trauma-matched controls and also from their identical twin brothers in the GM integrity of the ACC (Kasai et al. Reference Kasai, Yamasue, Gilbertson, Shenton, Rauch and Pitman2008) and in their ability to recall fear extinction (Milad et al. Reference Milad, Orr, Lasko, Chang, Rauch and Pitman2008). This reduced ability to recall fear extinction was replicated in a separate population of civilian PTSD patients and shown to be related to differences in function of the amygdala, the vmPFC and the dorsal ACC during extinction learning (Milad et al. Reference Milad, Pitman, Ellis, Gold, Shin, Lasko, Zeidan, Handwerger, Orr and Rauch2009). In healthy individuals, reduced cortical thickness in the vmPFC was associated with a reduced ability to successfully recall fear extinction (Milad et al. Reference Milad, Quinn, Pitman, Orr, Fischl and Rauch2005).
Our second, novel finding, namely that cortical thickness in the sgACC remained an index of recovery both before and after treatment, not changing as a function of symptom reduction, brings additional strong support to the proposed relationship between this brain structure, extinction and PTSD recovery. Indeed, our results suggest that GM integrity in this region may be a stable, possible trait-like, marker for the capacity to recover from trauma. According to this view, the reduced cortical thickness in the ACC we observed in PTSD patients who were less likely to recover from the disorder may have been indicative of a reduced capacity for extinction learning.
However, as all participants were scanned after trauma exposure, when they were already diagnosed as having PTSD (indeed, that was our key inclusion criteria), we have no direct evidence that the differences in ACC were pre-morbid. Indeed, recent evidence that Vietnam veterans with PTSD have less ACC GM volume than their monozygotic twin brothers suggests that ACC GM abnormalities are acquired as a consequence of trauma and/or PTSD (Kasai et al. Reference Kasai, Yamasue, Gilbertson, Shenton, Rauch and Pitman2008). Furthermore, individuals' coping strategies during the immediate aftermath of a traumatic event have been shown to be crucial for the likelihood of developing PTSD (Clohessy & Ehlers, Reference Clohessy and Ehlers1999; Bryant et al. Reference Bryant, Marosszeky, Crooks, Baguley and Gurka2000; Schnider et al. Reference Schnider, Elhai and Gray2007), and changes in brain structure can be observed after only a few weeks of training in healthy volunteers (Johansen-Berg, Reference Johansen-Berg2011) and following short-term pharmacological treatment in psychiatric patients, including in the sgACC (Costafreda et al. Reference Costafreda, Chu, Ashburner and Fu2009; Moore et al. Reference Moore, Cortese, Glitz, Zajac-Benitez, Quiroz, Uhde, Drevets and Manji2009). Thus, it is possible that the structural changes observed in our study took place in the period between trauma exposure and the first scan, although the lack of a correlation between sgACC thickness and current CAPS scores or time since trauma seems at odds with this interpretation. Nonetheless, further studies are necessary to help resolve this important issue.
Limitations
As stated earlier, we were not involved in the treatment of these patients, which precluded our ability to monitor treatment compliance or even the theoretical orientation of psychotherapy (e.g. cognitive behavioural, interpersonal, etc.) that was available to our participants. Thus, the results of the present study are not interpretable in terms of treatment response to any particular regimen, but instead as an ecologically valid investigation of the recovery process. Although the results reported here were significant at a fairly stringent criterion of statistical significance (see Method), our sample size was relatively small. Thus, it will be important to replicate these findings in the future in other, larger samples. Our sample size also precluded us from investigating, across the whole cortical surface, potential sex differences, which have been shown to influence the likelihood symptom remission and relapse (Felmingham & Bryant, Reference Felmingham and Bryant2012). Likewise, we could not fully explore the potential modulatory effects of time since trauma or medication on cortical thickness.
Importantly, we only measured one aspect of GM integrity, namely cortical thickness (Fischl & Dale, Reference Fischl and Dale2000; Hutton et al. Reference Hutton, De Vita, Ashburner, Deichmann and Turner2008; Johansen-Berg, Reference Johansen-Berg2011). Thus, we cannot rule out the existence of other anatomical changes associated with PTSD symptom reduction that could not be detected with the technique used here. This remains a distinct possibility, especially given the previously reported functional changes in this region during recovery (Felmingham et al. Reference Felmingham, Kemp, Williams, Das, Hughes, Peduto and Bryant2007; Dickie et al. Reference Dickie, Brunet, Akerib and Armony2011). Future studies using other measures and analysis methods should help to link these functional changes to the underlying brain structure.
Conclusions
This study confirmed and extended previous work into structure predictors of recovery from PTSD. As reported previously using VBM (Bryant et al. Reference Bryant, Felmingham, Whitford, Kemp, Hughes, Peduto and Williams2008b), symptom improvement was predicted by increased GM integrity (in this case cortical thickness) in the sgACC. This relationship remained in a second measurement of cortical thickness, 6–9 months later. Importantly, no changes were observed in cortical thickness in this region during the course of recovery, suggesting that it represents a stable marker of recovery potential from this disorder.
Acknowledgements
This project was funded by grants from Fonds de la recherche en santé du Québec and the Canadian Psychiatric Research Foundation to J.L.A. and A.B.; E.W.D. was supported by a scholarship from the Natural Sciences and Engineering Research Council of Canada. We are grateful to Dr L. Gaston and the staff at Traumatys Clinic, and also the nurses, especially F. Lauriault, at Charles LeMoyne Hospital for their invaluable help in recruiting subjects. We also thank O. Lyttelton, L. Buchy and B. Bernhardt for their help with cortical thickness analysis software.
Declaration of Interest
None.
