Introduction
Reduced fractional anisotropy (FA), widely reported in patients with psychotic disorder (Ellison-Wright and Bullmore, Reference Ellison-Wright and Bullmore2009), but not in individuals at higher than average genetic risk (siblings of patients) may reflect disease-related dysconnectivity or disease-related differential sensitivity to the environment (Boos et al., Reference Boos, Mandl, van Haren, Cahn, van Baal, Kahn and Hulshoff Pol2013; Domen et al., Reference Domen, Michielse, Gronenschild, Habets, Roebroeck, Schruers, van Os and Marcelis2013). The risk for psychosis, a condition with adolescent onset, has been related to environmental exposures such as pre- or postnatal birth complications, cannabis use, childhood trauma (Varese et al., Reference Varese, Smeets, Drukker, Lieverse, Lataster, Viechtbauer, Read, van Os and Bentall2012; Misiak et al., Reference Misiak, Krefft, Bielawski, Moustafa, Sąsiadek and Frydecka2017), and growing up in an urban environment (van Os and Kapur, Reference van Os Jand Kapur2009). These environmental stressors may be the trigger (Cornblatt et al., Reference Cornblatt, Lencz, Smith, Correll, Auther and Nakayama2003) or the ‘second hit’ (Maynard et al., Reference Maynard, Sikich, Lieberman and LaMantia2001), contributing to the emergence of a psychotic illness. However, the potential impact of these environmental risks on white matter (WM) connectivity (Andreasen et al., Reference Andreasen, Paradiso and O'Leary1998; Friston, Reference Friston1998) has not been the subject of detailed investigation in patients with a psychotic disorder. Cross-sectional studies did explore genetic factors, showing a moderate-to-high heritability of WM FA, ranging from 0.4 to 0.7 (Voineskos, Reference Voineskos2015). Also, several candidate genes for WM heritability have been proposed, such as neuregulin1-tyrosine kinase receptor ErbB4, involved in oligodendrocyte, myelin, and axonal development and maintenance (Wang et al., Reference Wang, Jiang, Sun, Teng, Luo, Zhu, Zang, Zhang, Yue, Qu, Lu, Hong, Huang, Blumberg and Zhang2009). Various hypotheses have been postulated, associating environmental risk factors with WM alterations. Cannabis use may induce apoptosis of oligodendrocyte progenitors, affecting WM development (Molina-Holgado et al., Reference Molina-Holgado, Vela, Arevalo-Martin, Almazan, Molina-Holgado, Borrell and Guaza2002). In the literature to date, there is evidence to suggest a negative association between cannabis use and WM volume in patients with schizophrenia (Cahn et al., Reference Cahn, Hulshoff Pol, Caspers, van Haren, Schnack and Kahn2004; Szeszko et al., Reference Szeszko, Robinson, Ashtari, Vogel, Betensky, Sevy, Ardekani, Lencz, Malhotra, McCormack, Miller, Lim, Gunduz-Bruce, Kane and Bilder2007). Results from diffusion tensor imaging (DTI) studies are less clear and have shown increased as well as decreased FA in cannabis-using v. non-using patients with schizophrenia (Peters et al., Reference Peters, Blaas and de Haan2010; James et al., Reference James, Hough, James, Winmill, Burge, Nijhawan, Matthews and Zarei2011). The same ambiguity is seen in samples of non-psychotic substance users compared with non-users, which, on the one hand, showed microstructural WM alterations in specific pathways (corpus callosum and superior longitudinal fascicules) (Baker et al., Reference Baker, Yucel, Fornito, Allen and Lubman2013) and the hippocampus (fimbriae), corpus callosum (splenium), commissural fibers (Zalesky et al., Reference Zalesky, Solowij, Yucel, Lubman, Takagi, Harding, Lorenzetti, Wang, Searle, Pantelis and Seal2012), as well as, on the other hand, absence of FA differences (Arnone et al., Reference Arnone, Barrick, Chengappa, Mackay, Clark and Abou-Saleh2008).
Studies on early-life stress in otherwise healthy children have shown associations between cortisol reactivity and possible alterations in hippocampal and amygdala volumes (Pagliaccio et al., Reference Pagliaccio, Luby, Bogdan, Agrawal, Gaffrey, Belden, Botteron, Harms and Barch2014). Corticosteroids may suppress the final mitosis of glial cells necessary for myelination, influencing WM microstructure (Teicher et al., Reference Teicher, Andersen, Polcari, Anderson and Navalta2002). This model is supported by studies showing reduced corpus callosum volume in paediatric inpatients with a history of abuse and neglect (Teicher et al., Reference Teicher, Dumont, Ito, Vaituzis, Giedd and Andersen2004) and reduced FA in the left inferior longitudinal fasciculus (ILF) in young adults witnessing domestic violence in childhood (Choi et al., Reference Choi, Jeong, Polcari, Rohan and Teicher2012). To date, longitudinal diffusion weighted imaging studies of patients with schizophrenia are scarce (Canu et al., Reference Canu, Agosta and Filippi2015) and apart from studies investigating the effect of medication on WM diffusion measures (Ozcelik-Eroglu et al., Reference Ozcelik-Eroglu, Ertugrul, Oguz, Has, Karahan and Yazici2014; Reis Marques et al., Reference Reis Marques, Taylor, Chaddock, Dell'acqua, Handley, Reinders, Mondelli, Bonaccorso, Diforti, Simmons, David, Murray, Pariante, Kapur and Dazzan2014), no study has examined whether traumatic experience has a differential effect on FA over time in individuals with or without (liability for) psychotic disorder (Bendall et al., Reference Bendall, Jackson, Hulbert and McGorry2008).
In a cross-sectional analysis of the baseline DTI scans of the current sample, patient-specific microstructural WM alterations were found (Domen et al., Reference Domen, Michielse, Gronenschild, Habets, Roebroeck, Schruers, van Os and Marcelis2013). These alterations remained relatively stable over a 3-year time course in contrast to a decline in mean FA in the non-affected siblings compared with healthy controls (Domen et al., Reference Domen, Peeters, Michielse, Gronenschild, Viechtbauer, Roebroeck, van Os and Marcelis2017). Consequently, both patients and siblings had decreased FA with respect to controls at follow-up. The aim of the current investigation was to study whether microstructural alterations over time were conditional on the exposure on two of the most examined environmental risk factors for schizophrenia; cannabis use and childhood trauma, for which biological plausibility exists. More specifically, we hypothesized that individuals at increased genetic risk (patients and siblings) with higher levels of exposure to these environmental risk factors would show reduced WM FA over time.
Methods
Participants
Subjects were recruited in the context of a multicenter longitudinal study (Genetic Risk and Outcome of Psychosis, G.R.O.U.P.) in the Netherlands (Korver et al., Reference Korver, Quee, Boos, Simons and de Haan2012). At baseline, 300 participants were included of which 258 underwent a DTI scan. At follow-up, approximately 3 years later (mean: 3.3 years), DTI scans were acquired from a sample of 180 participants, of which 159 provided a valid pair of DTI scans for the longitudinal analysis (see Domen et al. (Reference Domen, Michielse, Gronenschild, Habets, Roebroeck, Schruers, van Os and Marcelis2013, Reference Domen, Peeters, Michielse, Gronenschild, Viechtbauer, Roebroeck, van Os and Marcelis2017) and the Supplementary Method section for further information on inclusion and exclusion criteria, family composition and diagnostic assessments (Domen et al., Reference Domen, Michielse, Gronenschild, Habets, Roebroeck, Schruers, van Os and Marcelis2013, Reference Domen, Peeters, Michielse, Gronenschild, Viechtbauer, Roebroeck, van Os and Marcelis2017).
The standing ethics committee approved the study protocol, and all participants gave written informed consent in accordance with the committee's guidelines.
Measures
Level of psychotic symptomatology at the time of scanning was assessed with the Positive and Negative Symptom Scale (PANSS) (Kay et al., Reference Kay, Fiszbein and Opler1987).
Educational level was defined as highest accomplished level of education. Handedness was assessed using the Annett Handedness Scale (Annett, Reference Annett1970).
Antipsychotic medication
The determination of (lifetime) antipsychotic (AP) medication use at baseline and cumulative AP exposure during the 3-year follow-up period has been described in the online Supplementary Method section based on Domen et al. (Reference Domen, Michielse, Gronenschild, Habets, Roebroeck, Schruers, van Os and Marcelis2013, Reference Domen, Peeters, Michielse, Gronenschild, Viechtbauer, Roebroeck, van Os and Marcelis2017).
Substance use
Substance use was measured at both time points with the Composite International Diagnostic Interview (CIDI) sections B-J-L (WHO, 1990). As data of drug use over the last 3 years were not available, cannabis and other drug exposure was assessed as reported frequency of use during the last 12 months and lifetime use (mean number of times until baseline measurement). CIDI frequency data on alcohol (weekly consumptions), lifetime cannabis, and other drug use was available at follow-up for, respectively, 158 participants (1% missing data), 155 participants (3% missing data), and 157 participants (1% missing data). Despite the fact that the association between alcohol or other drugs and psychosis risk has been studied less than cannabis and provided a less clear picture, these substances may well contribute to some of the WM variation (Nesvåg et al., Reference Nesvåg, Frigessi, Jönsson and Agartz2007; Willi et al., Reference Willi, Barr, Gicas, Lang, Vila-Rodriguez, Su, Thornton, Leonova, Giesbrecht, Procyshyn, Rauscher, MacEwan, Honer and Panenka2017), and were therefore considered potential confounders (see Statistical analyses).
Childhood trauma: Childhood trauma was assessed at baseline with the Dutch version of the Childhood Trauma Questionnaire Short Form (CTQ) (Thombs et al., Reference Thombs, Bernstein, Lobbestael and Arntz2009). The short CTQ consists of 25 items rated on a five-point Likert scale (1 = never true to 5 = very often true) inquiring about traumatic experiences in childhood. Five types of childhood maltreatment were assessed: emotional, physical and sexual abuse, and emotional and physical neglect, with five questions covering each type of trauma (Bernstein et al., Reference Bernstein, Ahluvalia, Pogge and Handelsman1997). The mean of these 25 items (range 5.0–25.0) created a general measure of childhood trauma. The CTQ data were missing for one patient.
Image acquisition
Magnetic resonance imaging scans were obtained at Maastricht University, the Netherlands, using an Allegra Magnetom MR (Siemens, Erlangen, Germany) operating at 3.0 Tesla. At both measurement points, microstructural anatomy was examined using DTI with an echo-planar-imaging sequence (field of view 230 mm × 230 mm, TR 10800 ms, TE 84 ms, voxel size 1.8 mm × 1.8 mm × 1.8 mm, b-value 1000 s/mm2, 85 slices, no overlap). As a result of an update of the scanner software during baseline acquisition, two DTI sequences were used: one with 76 directions [of which four T2-weighted (B0) and 72 diffusion-weighted (B1000)] and one with 81 directions (8 × B0 and 73 × B1000). Gradient directions were identical in both sequences. A potential association between the proportion of baseline scans and group was investigated using a Pearson χ2 test.
At follow-up, the DTI sequence comprised 81 directions (8 × B0 and 73 × B1000). Total acquisition time of the DTI sequence was 15 min.
DTI analysis
Processing of DTI data was performed using tract-based spatial statistics (TBSS) v1.2 in FSL 4.1.6 (FMRIB Analysis Group, Oxford, UK, http://www.fmrib.ox.ac.uk/analysis/research/tbss). The consecutive processing steps are described in the online Supplementary Method section (based on Domen et al., Reference Domen, Michielse, Gronenschild, Habets, Roebroeck, Schruers, van Os and Marcelis2013, Reference Domen, Peeters, Michielse, Gronenschild, Viechtbauer, Roebroeck, van Os and Marcelis2017). For the current study, three mean FA skeletons were created; for the cross-sectional analysis at baseline (n = 258: controls, siblings, patients) and at follow-up (n = 180), and for the longitudinal analysis (n = 159), one based on six groups (3 groups × 2 time points).
Statistical analyses
In order to use a multilevel (mixed-effects) model and to be able to calculate a mean FA ‘change’ (mean FA at baseline minus mean FA at follow-up) for the longitudinal model, which was not compatible with the standard protocol in TBSS, data were analyzed in R (version 3.2.0), a free software environment for statistical computing and graphics (Team, Reference Team2015). From the 38 labeled WM tracts, skeleton mean FA values per participant per time point were extracted and exported to R. Since the mean FA values per subject were based on varying number of voxels, depending on the region, we used a model in which the error variance for a particular observation was inversely weighted by the number of voxels within the corresponding region.
Cross-sectional analysis at baseline and follow-up
As of the three-level grouping structure of the data, compromising statistical independence of the observations, a multilevel (mixed-effects) model was fitted. FA was the dependent variable, group (dummy variables with the controls as the reference category, controls = 0, siblings = 1, patients = 2), and the environmental exposures (lifetime and last year cannabis exposure, childhood trauma exposure) were the independent variables and random effects (intercepts) were added for both subject and family.
The statistical basic model was: FA = β 0 + β 1 (group) + β 2 (environmental exposure) + β 3 (group × environmental exposure). This model included the a priori hypothesized confounding variables age, sex, and level of education as fixed effects. In case of significant findings, additional covariates [i.e. alcohol consumption, lifetime other drug use, and body mass index (BMI)] were separately added to the model.
Main effects of the environmental exposures (controlled for group), as well as group × environmental exposures interactions in the model of FA were examined. The environmental exposures were entered both as linear and as factored variables (i.e. representing the distribution of scores divided by its tertiles: lifetime cannabis use: no, moderate, or heavy cannabis use; childhood trauma exposure: low, moderate, or high trauma exposure), allowing visualization of dose–response. In case of significant interaction effects, stratified analyses were conducted in order to quantify whether the association between environmental exposure and FA differed between the three groups.
To examine whether childhood trauma and lifetime cannabis use contributed independent effects, planned sensitivity analyses were performed with both environmental exposures in the model.
To examine whether the scanner software update at baseline affected the results, the interaction analyses at baseline were repeated in subgroups stratified by the number of scan directions: 76 (n = 191) v. 81 (n = 67) directions.
Longitudinal analysis
A mean FA ‘change’ (delta, Δ) per participant per region (n = 159) was calculated by subtracting mean FA at baseline from mean FA at follow-up. The same analyses were carried out as described in the cross-sectional part, now using ΔFA as the dependent variable and group and the environmental exposures (lifetime and last year cannabis exposure, childhood trauma exposure) as the independent variables.
In addition, to control for a potential effect of depression on ΔFA, the main longitudinal analysis was repeated with exclusion of the subgroup of participants in the control (n = 11) and sibling group (n = 15) with a history of a depressive disorder.
Since our previous study showed a small effect of last 3-year and lifetime AP use on ΔFA (Domen et al., Reference Domen, Peeters, Michielse, Gronenschild, Viechtbauer, Roebroeck, van Os and Marcelis2017), planned sensitivity analyses were performed to rule out potential AP effects, using patient subgroups [with low, moderate, and high AP exposure (for the number of participants per subgroup see online Supplementary Table)]. Thus, the group × environmental exposure interactions in the model of ΔFA were examined in three AP subgroups to ascertain whether a potential effect remained significant in the respective subgroups.
Results
Demographics
The patients represented a relatively stable population (not in need of inpatient care or intensive treatment), as reflected by the low PANSS scores and the number of patients that fulfilled the remission criteria (Table 1). The gender distribution in the samples was skewed, showing more male patients and male siblings as heavy cannabis users and more male patients and female controls exposed to high levels of childhood trauma at follow-up (Table 2). The mean current dosage of AP medication in terms of standard haloperidol equivalents was 5.5 milligrams (mg) (s.d. = 4.6) at baseline and 4.7 mg (s.d. = 5.1) at follow-up (over the last 3 years). The proportion of baseline scans with 76 directions did not differ between the groups (84% in controls, 82% in siblings, and 71% in patients: χ2 = 3.02, df = 2, p = 0.22).
Demographic characteristics of the participants [baseline (n = 258) and longitudinal analysis (n = 159)]
PANSS, Positive and Negative Syndrome Scale; dis, disorder; NOS, not otherwise specified.
Means ± s.d., range are reported.
a Range: from −100 (entirely left-handed) to +100 (entirely right-handed).
b Range: 0 = no education to 8 = university degree.
c Cumulative exposure, lifetime until baseline in haloperidol equivalents.
d Cumulative exposure, last 3 years in haloperidol equivalents.
e History of major depressive disorder, no current episodes at baseline or in last 3 years.
f Mean number of times; lifetime.
g Mean number of times; last 12 months.
h Weekly consumptions; lifetime.
i Weekly consumptions; last 12 month.
Within-group distribution of the environmental exposures
The number of subjects per group (and proportion of total group) for the baseline (n = 258) and longitudinal analysis (follow-up, n = 159) per environmental stress factor. The difference between groups in proportion of high, medium, and low cannabis use was analyzed with the Pearson's (χ2).
m/f ratio, male/female ratio.
Cross-sectional analysis of FA and environmental risk factors at baseline
There were no significant associations between cannabis exposure and FA (lifetime: B = 0.002, p = 0.24, last year: B = 1.0 × 10−5, p = 0.43) or between childhood trauma exposure and FA (B = −0.001, p = 0.60). In addition, no significant interactions were found between cannabis exposure and group in the model of FA (lifetime: χ2 = 1.3, df = 2, p = 0.52, last year: χ2 = 0.1, df = 2, p = 0.93) and between childhood trauma and group in the model of FA (χ2 = 2.9, df = 2, p = 0.24) (all analyses with linear environmental variables). The results did not change after stratification by number of scan directions (76 directions: cannabis exposure; lifetime: χ2 = 1.4, df = 2, p = 0.49, last year: χ2 = 0.2, df = 2, p = 0.89, childhood trauma exposure: χ2 = 2.3, df = 2, p = 0.31; and 81 directions: cannabis exposure; lifetime: χ2 = 1.2, df = 2, p = 0.56, last year: χ2 = 1.5, df = 2, p = 0.48, childhood trauma exposure: χ2 = 0.2, df = 2, p = 0.93).
Cross-sectional analysis of FA and environmental risk factors at follow-up
No significant associations between cannabis exposure (linear variable) and FA (lifetime: B = 1.0 × 10−4, p = 0.95, last year: B = −3.0 × 10−6, p = 0.88) or between childhood trauma exposure (linear variable) and FA (B = −0.001, p = 0.67) were found at follow-up.
Cannabis
No significant interaction was found between last year cannabis use and group in the model of FA (χ2 = 0.8, df = 2, p = 0.66). A significant interaction was found between, respectively, (lifetime) cannabis exposure (χ2 = 9.3, df = 2, p = 0.01) and group in the model of FA. This interaction remained significant after controlling for childhood trauma (χ2 = 9.6, df = 2, p = 0.008), BMI (χ2 = 9.7, df = 2, p = 0.008), alcohol use (χ2 = 9.8, df = 2, p = 0.008), and other drug use (χ2 = 7.6, df = 2, p = 0.02) (all linear variables).
Stratified analyses showed a mean FA decrease in the heavy cannabis using patients, which was significantly different from the controls and the siblings. An association in opposite direction was found in cannabis-using siblings, but only for the moderate cannabis exposure level and not for the highest exposure level (see Table 3).
Mean FA as a function of group status and environmental exposure at follow-up
N, number of participants; P v. C, patients v. controls; P v. S, patients v. siblings; S v. C, siblings v. controls.
Results from the interaction: environmental exposure × group in the model of FA. The B’s represent the stratified group effect sizes. Group differences are displayed with χ2, and the p value (<0.05*).
a Reference level.
Childhood trauma
A significant interaction between childhood trauma exposure and group in the model of FA (χ2 = 6.1, df = 2, p = 0.05) was found. The interaction remained significant after controlling for alcohol use (χ2 = 8.8, df = 2, p = 0.01), inconclusive for lifetime cannabis use (χ2 = 4.9, df = 2, p = 0.09) and BMI (χ2 = 5.5, df = 2, p = 0.06), however not significant anymore with addition of other drug use (χ2 = 4.4, df = 2, p = 0.11) (all linear variables).
Stratified analyses showed a mean FA decrease in patients exposed to a high childhood trauma level, which was significantly different from the siblings and inconclusive from the controls (see Table 3).
Longitudinal analysis of ΔFA and environmental risk factors
In the whole group, there was no significant association between, respectively, lifetime cannabis exposure (B = −3.0 × 10−5, p = 0.16), last year cannabis exposure (B = −1.0 × 10−5, p = 0.45), or childhood trauma exposure (all linear variables) and ΔFA (B = −7.0 × 10−4, p = 0.13).
Cannabis
There was a significant interaction between (lifetime) cannabis exposure and group in the model of ΔFA (χ2 = 6.2, df = 2, p = 0.04). This interaction remained significant after controlling for childhood trauma (χ2 = 5.9, df = 2, p = 0.05), BMI (χ2 = 5.8, df = 2, p = 0.05), alcohol use (χ2 = 5.9, df = 2, p = 0.05), other drug use (χ2 = 7.4, df = 2, p = 0.02), and scan type (χ2 = 6.5, df = 2, p = 0.04). After exclusion of the 26 participants with a history of depression (16% reduction in sample size), the interaction became inconclusive (χ2 = 5.4, df = 2, p = 0.07) (all linear variables).
Stratified analyses showed a significant effect in patients, indicating a decrease in FA over time with increasing cannabis exposure in patients, resulting in a significant group difference between patients and controls and between patients and siblings. Compared with patients with no cannabis use, patients with heavy cannabis consumption had significantly more FA decrease over time. This was not the case for the moderate v. no cannabis use comparison (Table 4, Fig. 1).
The association between, respectively, cannabis (a) and childhood trauma (b) (dummy variables) and ΔFA, stratified per group. The effect of high cannabis exposure v. no cannabis exposure and high trauma exposure v. low trauma exposure on mean whole-brain ΔFA was significantly different for patients compared with controls (cannabis; χ2 = 3.7, p = 0.05, trauma; χ2 = 10.0, p = 0.002) and for patients compared with siblings (cannabis; χ2 = 4.3, p = 0.04, trauma; χ2 = 7.2, p = 0.007) (*p < 0.05).
Mean ΔFA as a function of group status and environmental exposure
N, number of participants; P v. C, patients v. controls; P v. S, patients v. siblings; S v. C, siblings v. controls.
Results from the interaction: environmental exposure × group in the model of ΔFA. The B’s represent the stratified group effect sizes. Group differences are displayed with χ2, and the p value (<0.05*).
a Reference level.
Childhood trauma
A significant interaction between trauma exposure and group in the model of ΔFA (χ2 = 11.3, df = 2, p = 0.003) was found. The interaction remained significant after controlling for lifetime cannabis use (χ2 = 12.3, df = 2, p = 0.002), BMI (χ2 = 10.5, df = 2, p = 0.005), alcohol use (χ2 = 12.0, df = 2, p = 0.003), other drug use (χ2 = 14.6, df = 2, p = 0.0007), and scan type (χ2 = 11.5, df = 2, p = 0.003). The interaction was inconclusive after exclusion of the 26 participants with a history of depression (χ2 = 5.6, df = 2, p = 0.06) (all linear variables).
Stratified analysis revealed a significant negative association between childhood trauma and ΔFA in patients, resulting in a significant patient–control and patient–sibling difference. Compared with low trauma exposure, patients exposed to high trauma levels had significantly more FA decrease over time. This was not the case for moderate compared with low trauma exposure (Table 4, Fig. 1).
Sensitivity analyses in AP medication subgroups
In the AP medication subgroup analyses (with smaller N), interactions between both environmental risk factors and group in the model of ΔFA remained largely significant, most prominent in patients with the lowest AP exposure levels (lifetime and 3-year interval). Wald tests showed that the significant negative associations between the environmental factor and ΔFA in the low AP subgroups were significantly different for the patient–control and the patient–sibling comparison (see online Supplementary Table S1).
Discussion
Despite the absence of cross-sectional associations between environmental risk factors and WM FA at baseline, there were significant interactions between group and cannabis and childhood trauma exposure in models of FA at follow-up and FA change over a 3-year time period. Patients exposed to the highest levels of cannabis or childhood trauma had a greater FA decrease over time compared with controls and siblings.
Cannabis exposure and WM alterations
Higher levels of (lifetime) cannabis exposure in patients with psychotic disorder were not associated with FA alterations at baseline, but with significant FA reduction over a 3-year period compared with siblings and healthy controls. This is the first longitudinal study examining the effect of cannabis on WM FA in relation to (familial risk for) psychotic disorder, suggesting that the extent of WM alterations is conditional on the level of cannabis exposure in patients with the disorder. Especially, cannabis use before or at disease onset may cause an additional reduction in WM FA, given the absence of an interaction between group and last year cannabis use in the model of ΔFA. This finding complies with other structural imaging findings of more severe WM deficits in young adults exposed to cannabis prior to the age of 16 (Cookey et al., Reference Cookey, Bernier and Tibbo2014). It strengthens the evidence, from both preclinical human and animal models, for a neurotoxic effect of cannabis in adolescence (Rubino and Parolaro, Reference Rubino and Parolaro2014); a sensitive age period for the neuronal maturation of the endocannabinoid system, possibly resulting in disrupted network connectivity of various brain areas.
To date, three longitudinal DTI studies examined substance abuse in non-psychotic populations. A lower rate of change in FA was found in adolescent cannabis users (n = 23) in five clusters of fronto-parietal association fibers over a 2-year interval (Becker et al., Reference Becker, Collins, Lim, Muetzel and Luciana2015). Moreover, a significant FA decrease over time in the left ILF in adolescents with cannabis use disorder (n = 19), associated with more cannabis exposure, compared with healthy controls (Epstein and Kumra, Reference Epstein and Kumra2015). A third study showed poorer WM integrity after an 18-month follow-up, although mainly predicted by alcohol, not marijuana, in seven fronto-parietal tracts in adolescent substance users (n = 41) (Bava et al., Reference Bava, Jacobus, Thayer and Tapert2013).
In previous cross-sectional studies, positive as well as negative associations between FA and cannabis use in patients with schizophrenia have been reported. James et al. (Reference James, Hough, James, Winmill, Burge, Nijhawan, Matthews and Zarei2011) revealed associations between early cannabis use and decreased FA in, e.g. the internal capsule, corona radiata, superior and ILF, in patients with adolescent-onset schizophrenia (James et al., Reference James, Hough, James, Winmill, Burge, Nijhawan, Matthews and Zarei2011). In contrast, Peters et al. (Reference Peters, de Haan, Vlieger, Majoie, den Heeten and Linszen2009) found FA increases in the bilateral uncinate fasciculus, anterior internal capsule, and frontal WM in patients with recent-onset schizophrenia who had started using cannabis before the age of 17 years, compared with a similar group with no history of cannabis use (Peters et al., Reference Peters, de Haan, Vlieger, Majoie, den Heeten and Linszen2009). Reduced FA has also been found in the splenium of the corpus callosum in non-cannabis using patients with schizophrenia compared with patients with schizophrenia and early-onset cannabis use (before age 15 years) (Dekker et al., Reference Dekker, Schmitz, Peters, van Amelsvoort, Linszen and de Haan2010). It has been shown that cannabis use may have different effects at various neurodevelopmental stages of life (Jakabek et al., Reference Jakabek, Yucel, Lorenzetti and Solowij2016). In the present study, absence of a significant cross-sectional association between FA and lifetime cannabis exposure at baseline in contrast to significant associations at follow-up between, respectively, high and moderate cannabis exposure and FA in patients and siblings compared with controls may imply an age-related effect. The effect of cannabis on brain WM may only be visible after many years, depending on the WM developmental curvature of the specific subject and the timing of the measurement.
The opposite direction of effect, i.e. the FA increases in relatives with increasing lifetime cannabis exposure may suggest a delayed maturation, an imaging artifact or even a protective effect of a small amount of cannabis. A cannabinoid neuroprotective effect on brain matter with improvement in WM efficiency (Westlye et al., Reference Westlye, Walhovd, Dale, Bjornerud, Due-Tonnessen, Engvig, Grydeland, Tamnes, Ostby and Fjell2010) has been proposed by recent in vitro studies (Sarne and Mechoulam, Reference Sarne and Mechoulam2005).
The negative association between especially the heavy cannabis using patients with psychotic disorder and FA may fit with the hypothesis that heavy cannabis use at a young age may have altered the normal trajectory of WM brain maturation. Interference on the extensive pruning and myelination processes in an already vulnerable adolescent brain (Lubman et al., Reference Lubman, Cheetham and Yucel2015) may have caused additional reduction in WM FA later on in life. Whether this has clinical implications or influences on long-term prognosis needs further investigation.
Childhood trauma exposure and WM alterations
The present study examined the association between childhood trauma and FA in individuals with (risk for) psychotic disorder. At baseline, neither a significant interaction between childhood trauma exposure and group in the model of FA was found, nor a main effect of childhood trauma in any of the groups. However, at follow-up, a significant (dose–response) negative association between childhood trauma exposure and group in the model of FA was found in patients with psychotic disorder. In other words, higher exposure to childhood trauma is associated with lower whole-brain mean FA later in life. As with cannabis, differences in baseline and follow-up results may be explained by the timing of genetic and environmental influences (and their interactions) impacting cerebral plasticity during the life span.
The follow-up findings are in line with several studies that show reduced FA in non-psychotic traumatized subjects (Choi et al., Reference Choi, Jeong, Polcari, Rohan and Teicher2012; Daniels et al., Reference Daniels, Lamke, Gaebler, Walter and Scheel2013) in stress-processing-related areas, such as the corpus callosum (Jackowski et al., Reference Jackowski, Douglas-Palumberi, Jackowski, Win, Schultz, Staib, Krystal and Kaufman2008; Paul et al., Reference Paul, Henry, Grieve, Guilmette, Niaura, Bryant, Bruce, Williams, Richard, Cohen and Gordon2008) and the cingulum bundle (Wang et al., Reference Wang, Zhang, Tan, Yin, Chen, Wang, Zhang, Wang, Guo, Tang and Li2010; Zhang et al., Reference Zhang, Zhang, Li, Li, Li, Ma, Hou, Zhang, Zhang, Wang, Duan and Lu2011). Thus, cross-sectional studies indicate that the FA reductions in traumatized populations may show overlap with the WM abnormalities in schizophrenia (Kubicki et al., Reference Kubicki, McCarley, Westin, Park, Maier, Kikinis, Jolesz and Shenton2007; Ellison-Wright and Bullmore, Reference Ellison-Wright and Bullmore2009), suggesting that (part of) the WM tract alterations may be non-specific, contributing to different phenotypes.
The current study did also find a greater FA decline over a 3-year period in the patients with the highest level of childhood trauma exposure with respect to siblings and healthy controls. This significant FA decline over time associated with higher levels of childhood trauma may fit with the literature describing the neurotoxic impact of childhood trauma on WM development at a sensitive age period (Heim and Binder, Reference Heim and Binder2012).
The patient-specific finding with regard to childhood trauma may refer to a complicated interplay between trauma-related factors and psychosis-related factors to account for the more pronounced WM alterations, predominantly in the group with the highest trauma exposure. Although a causal relationship between childhood trauma and WM decreases cannot be determined from these data, experiencing severe childhood trauma may cause an additive effect on already disrupted WM development. Alternatively, illness-related factors such as disadvantageous life style and health issues (e.g. reduced physical activity, social deprivation, smoking) (von Hausswolff-Juhlin et al., Reference von Hausswolff-Juhlin, Bjartveit, Lindstrom and Jones2009) may have contributed to increased cerebral vulnerability in patients, either or not in interaction with the environmental exposures under investigation.
Methodological considerations
Apart from the strength of this study – a relatively large longitudinal imaging design with a gene–environment approach – some limitations need to be addressed.
The sample size of some longitudinal subgroup analyses was only modest or small, resulting in loss of power (and thus weak statistical effects) and increased likelihood of false-negative results. Taken together with a rather skewed gender distribution in our sample, gender-specific sub-analyses were not considered feasible although it is known that several WM tracts show gender-specific FA differences (Menzler et al., Reference Menzler, Belke, Wehrmann, Krakow, Lengler, Jansen, Hamer, Oertel, Rosenow and Knake2011; Kanaan et al., Reference Kanaan, Chaddock, Allin, Picchioni, Daly, Shergill and McGuire2014).
A full understanding of the biological and clinical relevance of the reported small changes in FA is hampered. Nevertheless, one can imagine that disproportionally higher changes may arise due to stronger regional effects, either or not in combination with higher mean trauma levels. Future studies with larger sample sizes may provide more precise estimates of regional FA effect sizes associated with these environmental exposures.
Conform the various results across studies on the association between AP use and WM alterations (Szeszko et al., Reference Szeszko, Robinson, Sevy, Kumra, Rupp, Betensky, Lencz, Ashtari, Kane, Malhotra, Gunduz-Bruce, Napolitano and Bilder2008; Ozcelik-Eroglu et al., Reference Ozcelik-Eroglu, Ertugrul, Oguz, Has, Karahan and Yazici2014; Reis Marques et al., Reference Reis Marques, Taylor, Chaddock, Dell'acqua, Handley, Reinders, Mondelli, Bonaccorso, Diforti, Simmons, David, Murray, Pariante, Kapur and Dazzan2014), and the small effect of last 3-year and lifetime AP use on ΔFA found in our previous analyses (Domen et al., Reference Domen, Peeters, Michielse, Gronenschild, Viechtbauer, Roebroeck, van Os and Marcelis2017), the current results suggest a minor confounding effect of AP use as not all the G × E interactions remained significant in different AP subgroups. However, the results of these sensitivity analyses must be viewed with caution given the sizable lack of power (patient–AP subgroups comprised one-third of the sample).
It is unlikely that the two DTI sequences used at baseline would have contributed to a systematic bias, as the proportions of the two sequences were almost equal between the groups. In addition, stratified analyses (by number of scan directions) and adjustment for scanning sequence did not change the results, fitting the suggestion that the variation in tensor estimation is negligible with more than 30 diffusion directions (Jones, Reference Jones2004). Extracting mean FA values from the TBSS skeleton has the disadvantage of only examining the central portion of the WM tract, but will procure that WM was indeed examined. This is in line with more recent cannabis – diffusion studies that took a distance from a voxel-based comparison approach (Jakabek et al., Reference Jakabek, Yucel, Lorenzetti and Solowij2016).
Lastly, FA is a rather non-specific diffusion measure, containing information on myelination, fiber organization, and number of axons, and therefore not completely synonymous to ‘WM integrity’, so that the current findings must be interpreted with caution (O'Donnell and Pasternak, Reference O'Donnell and Pasternak2015). Nonetheless, a whole-brain, hypothesis-generating approach was chosen, as studies investigating the influence of environmental risk factors on WM alterations in patients with psychotic disorder are scarce and ambiguous.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291718001320
Acknowledgements
The authors thank Truda Driesen and Inge Crolla for their coordinating roles in the data collection, as well as the G.R.O.U.P. investigators: Richard Bruggeman, Wiepke Cahn, Lieuwe de Haan, René S. Kahn, Carin Meijer, Inez Myin-Germeys, Jim van Os, and Durk Wiersma.
Financial support
This work was supported by the Dutch organization for scientific research NWO [Genetic Risk and Outcome of Psychosis (G.R.O.U.P)] and the European Community's Seventh Framework Programme under Grant Agreement No. HEALTH-F2-2009-241909 (European Network of National Schizophrenia Networks Studying Gene-Environment Interactions Consortium).
Both funding sources had no further role in study design; in the collection, analysis, and interpretation of data; in the writing of the report; and in the decision to submit the paper for publication.
Conflict of interest
Jim van Os is or has been, in the last 3 years, an unrestricted research grant holder with, or has received financial compensation as an independent symposium speaker from Lundbeck and Janssen. Machteld Marcelis has received, in the last 3 years, financial compensation as an independent symposium speaker from Lundbeck and Janssen. All other authors report no biomedical financial interests or potential conflicts of interest.
Ethical standards
The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.
