The Cognition Network:
Relations amongst Cannabis Use, Genetic Risk for Schizophrenia, and Cognitive Abilities Pia Tio University of Amsterdam Tomáš Paus University of Toronto Denny Borsboom University of Amsterdam
2 Abstract
Cannabis is the world’s most used illicit drug and its use may impair cognition. Three main issues can be identified in the literature on the relation between cannabis use and cognition: (i) the highest prevalence of cannabis use is found in adolescents, however, most research on cognition focuses on adults; (ii) not all cannabis users suffer the same degree of cognitive impairment, indicating the presence of a moderating variable; and (iii) the effects on different cognitive abilities are often studied in isolation, ignoring the relations between the different cognitive abilities. Using data from the Saguenay Youth Study, we describe these relations between cognitive abilities by applying a network model, and investigate whether this cognitive structure differs depending on the adolescents’ cannabis use. Genetic risk for schizophrenia was included as a possible moderator due to comorbidity between schizophrenia and cannabis use. We found differences in the relations amongst cognitive abilities between cannabis users and non-users. Strength of these differences depended on the adolescents’ level of genetic risk for schizophrenia. Application of network models provide new insights into the relations amongst cognitive abilities.
Keywords: cognitive abilities, cannabis use, genetic risk for schizophrenia, network model
3 Introduction
Cannabis is the most widely used illicit drug in the world, with between 125 and 227 million users (2.7-4.9%) between the ages 15 and 64 (United Nations Office on Drugs and Crime, 2014). While cannabis is known for inducing elevated mood, relaxation, and perceptual alterations, it also has several less pleasant effects. Both acute and heavy long-term use of cannabis appears to impair cognitive functions such as memory, attention, psychomotor performance, and working memory (Hall & Solowij, 1998), though in the case of chronic use these impairments appear to be more subtle (Grant, Gonzalez, Carey,
Natarajan, & Wolfson, 2003). Additionally, the longer cannabis has been used, the more pronounced cognitive impairments seem to be (Solowij, 1998). Although the relation between cannabis use and cognition seems straightforward, this is not the case. The goal of this study is to investigate the relations amongst cannabis use and cognitive abilities, taking three main issues into account with respect to the current body of literature.
First, efforts to understand the association between cannabis use and cognition are focused mostly on adults. Prevalence of cannabis use is, however, highest during adolescence (Degenhardt & Hall, 2012), a period of high vulnerability of the brain and onset of many psychiatric disorders (Paus, Keshavan, & Giedd, 2008). Given that this prevalence might even increase in the future as more and more countries legalise cannabis, studying cannabis use in adolescents is of considerable importance. Studies that investigated prospectively cannabis use in adolescents show mixed results. A preliminary report from the ALSPAC cohort (ECNP, 2014), with over 2,500 adolescents, found no relation between cannabis use and intelligence when other risky behaviours (such as other drug use) are taken into account. On the other hand, Meier et al. (2012) found in a longitudinal sample of 1,037 adolescents that cannabis use is associated with decline in different intelligence indices such as memory
4
and processing speed, even when excluding participants diagnosed with dependence of other drugs.
Second, not everyone who uses cannabis shows the same degree of cognitive deficits. Prevalence of cannabis use is higher among patients with schizophrenia than among healthy controls (Volkow, 2009), though the causal relation between the cannabis use and
schizophrenia is still a matter of debate (Gage, Zammit, & Hickman, 2013). Schizophrenia is a highly heritable psychiatric disorder (Cardno & Gottesman, 2000; Sullivan, Kendler, & Neale, 2003), with concurrent impairment of cognitive abilities (Yücel et al., 2012). A meta-analysis by Snitz, MacDonald, and Carter (2006) demonstrated that individuals with high genetic risk for schizophrenia score lower on tasks with high executive function demands than healthy controls. Given that both cannabis use and (genetic risk for) schizophrenia are associated with cognitive deficits, one would expect that cannabis using individuals with schizophrenia would fare even worse. While this might be the case for short-term cannabis use (D’Souza et al., 2005), at least certain subgroups of cannabis-using patients seem to have less cognitive impairments than non-using patients (Schnell, Koethe, Daumann, &
Gouzoulis-Mayfrank, 2009; Yücel et al., 2012). Considering all this, genetic risk for
schizophrenia is a possible moderator of the association between cannabis use and cognition. Third, although different aspects of cognition have been investigated in relation to cannabis use, cognitive abilities are often studied in isolation. The use of multiple statistical analyses such as linear regression and t-test is suitable for analysing independent skills, but it is doubtful whether this assumption is plausible for the different cognitive abilities studied at hand. In their study on cognitive impairments in individuals at high risk for schizophrenia, Egan et al. (2000) addressed this issue using factor analysis and concluded that the different cognitive abilities are independent of each other. While the use of analyses derived from reflective measurement models (such as factor analysis) is far from unusual, reflective
5
conceptualisation of psychological phenomena faces several problems of a different nature (see Schmittmann et al., 2013). For instance, an important assumption of reflective models is that there are no direct causal relations between the indicators of factors; however, is it not inconceivable that cognitive abilities such as selective attention and memory performance influence each other (and other domains) rather than being two isolated abilities.
To determine whether cognitive abilities mutually influence each other, their underlying structure needs to be mapped. This can be done without data reduction or the introduction of latent variables, namely by using network models (Schmittmann et al., 2013). In network models, a construct is conceptualised as a network of directly related variables that influence each other over time. To construct such a network, variables are represented as nodes and an edge is drawn between two nodes when the two variables are associated (e.g. correlated) to one another. The network perspective has been applied to conceptualise
different psychological phenomena, including psychopathology (Borsboom, 2008; Borsboom & Cramer, 2013), personality (Cramer et al., 2012), and general intelligence (van der Maas et al., 2006; van Gelder, 1998).
Taking these issues into account, the present study investigated the relations amongst adolescents’ cannabis use and their cognitive abilities. We represent the relations amongst cognitive abilities using a network model, and include genetic risk for schizophrenia as a possible moderator. Given previously reported sex difference in incidence of cannabis use (Degenhardt et al., 2008), risk and age of onset for developing schizophrenia (Aleman, Kahn, & Selten, 2003; Häfner, Heiden, & Behrens, 1998), and several cognitive abilities (Halpner, 2012), all analyses will be done separately for males and females.
Methods
Data
6
Participants were 1,029 adolescents (48% male; 12-18 years old) recruited from the Saguenay Lac-Saint-Jean region in Quebec, Canada, as part of the Saguenay Youth Study (SYS, Pausova et al., 2007). While the initial focus of the SYS was to evaluate associations between prenatal exposure to maternal smoking and the adolescents’ brain and cardio-metabolic health, our study utilizes this cohort to examine the cross-sectional relationship between cannabis use, genetic risk for schizophrenia, and cognition. Details of recruitment and testing procedures are provided in Pausova et al. (2007). Their main exclusion criteria for the adolescents were: (i) premature birth (<35 weeks); (ii) alcohol abuse during pregnancy; (iii) personal history of heart disease, brain trauma, brain tumour, meningitis, or epilepsy; (iv) severe mental illness or mental retardation (IQ<70); and (v) MRI contraindications. Seven participants were excluded from all analyses because no information on cognitive abilities was available.
Cognitive abilities
Cognition abilities were assessed in French using 11 cognitive instruments: Wechsler Intelligence Scale for Children III (except for the subtest Maze; Wechsler, 1991), Woodcock-Johnson III (Woodcock, McGrew, & Mather, 2001), spelling task Orthographe d’usage (Poulin, 1982), subscales Dot Location and Stories from the Children’s Memory Scale
(Cohen, 1997), Self-Ordered Pointing Task (Petrides and Milner, 1982), a verbal fluency task based on NEPSY verbal fluency test (Korkman, Kirk, & Kemp, 1998), Stroop test (Stroop, 1935), Grooved Pegboard task, continuous and sequential tapping, Card Playing Task (Newman, Patterson, & Kosson, 1987), and Ruff 2 & 7 Selective Attention Test (Ruff & Allen, 1996). Most instruments have been described elsewhere (Kafouri et al., 2009; Schwartz et al., 2013).
7
All subtests/scales of the cognitive instruments were used, resulting in 63 variables (for an overview see Supplement 1). Scores exceeding ± 3 sd of the mean were excluded from the analysis. Data within each variable were standardized and adjusted for age (in months).
Cannabis use
Cannabis use was assessed using a life-style questionnaire. Participants indicated whether or not they had “ever” (yes) or “never” (no) used cannabis before the age of 16. Genetic risk for schizophrenia
Imputations from genome-wide single nucleotide polymorphisms were obtained to calculate a polygenic risk score for schizophrenia based on the 108 genome-wide significant loci identified by Psychiatric Genomics Consortium (Ripke et al., 2014). Adolescents were either classified as having a high (above median) or low (below median) genetic risk for schizophrenia.
Statistical analyses Correlation matrices
Within the network model, the underlying structure of a set of variables can be studied in different ways. A straightforward measure of relationships is to use correlations. Partial correlations represent the unique variance between two variables by removing the influence of all other variables, and are especially useful when considering moderation effects. However, due to extracting multiple variables from the same cognitive instrument (see Supplement 1), multicollinearity amongst the cognitive abilities variables was expected. In the presence of (high) multicollinearity, partial correlation estimates are less precise than if the variables were less correlated with each other. Therefore, Pearson’s correlations were calculated for all four combinations of cannabis use (ever/never) and genetic risk for
8
schizophrenia (low/high). Non-normal variables were transformed using the nonparanormal transformation function with the R package huge (Zhao, Liu, & Roeder, 2012).
Differences in the underlying structure of cognitive abilities were quantified by testing whether two or more correlation matrices differ significantly from each other. If the samples on which these correlation matrices are based belong to the same population, we expect that the correlation matrices, apart from sampling bias, are equal to each other. In other words, we expect that the difference between the matrices (i.e., subtracting one matrix from the other) is (close to) zero. If the samples on which the correlation matrices are based belong to different populations, the absolute difference between the matrices is expected to be greater than zero.
In this study, the difference between the correlation matrices was qualified using the Frobenius norm (Golub & van Loan, 1996), which is the summed quadrated difference for each pair of variables. To test whether this Difference Value (DV) is close to zero, an
empirical p-value was calculated. All participants were randomly re-assigned membership to one of two samples 10,000 times and the DV was calculated, thus creating a reference distribution for the DV. Under the null hypothesis (i.e., assuming that the two samples are from the same population) our original DV (i.e., the one we are interested in) will not deviate much from the distribution mean. If the two samples are not from the same population, however, then our original DV will deviate from the distribution mean. The empirical p-value was defined as the number of absolute permuted DVs that are equal to or more extreme than the original DV, divided by the total amount of permuted DVs (i.e., 10,000). Two correlation matrices were considered significantly different if the empirical p-value is < .05. Comparing more than two correlation matrices required the summing of DV’s for each pair of matrices.
Correlation matrices were visualized as networks using the R package qgraph (Epskamp, Cramer, Waldorp, Schmittmann, & Borsboom, 2012). Here, the nodes represent
9
the 63 different cognitive abilities variables and the connecting edges reflect the strength of the relationship (i.e., Pearson’s correlation) between a pair of variables. All networks were corrected for multiple comparisons using the False Discovery Rate (FDR; Benjamini & Hochberg, 1995).
Another way the relations amongst cannabis use, genetic risk for schizophrenia, and cognitive abilities were examined was by additionally calculating the correlations between the cognitive abilities variables on one hand and the variables “Cannabis use” and “Genetic risk for schizophrenia” on the other hand. The strength of these correlations provided information on which cognitive abilities variables are associated with cannabis use and genetic risk for schizophrenia.
Network properties
The underlying structure of variables can also be studied using network properties, which are typically designed to assess the centrality of a node in the network. Node degree is a measure of centrality defined as the total number of direct neighbours (nodes) a given node has. The shortest path length between two nodes is the number of nodes that lie on the shortest path that connects these nodes, and the average shortest path length is a measure of how well connected a network is.
Differences in the network properties degree and shortest path length were examined between the following groups: i) ever/never cannabis use; ii) low/high genetic risk for schizophrenia; and iii) ever/never cannabis use within low and high genetic risk for
schizophrenia. Degree and shortest path length for each (pair of) node(s) were calculated as described by Opsahl, Agneessens, and Skvoretz (2010), and their distributions were
compared using Kolmogorov-Smirnov test. All analyses were performed using the FDR corrected networks. Two distributions were considered different for p-values < .05.
10 Results
Participant characteristics can be found in Table 1. Females and males did not differ in age, cannabis use, and median genetic risk score for schizophrenia (p>.05). Both male and female cannabis users showed impaired cognitive abilities compared to none-users (Males: F(63, 411) = 2.3188, p<.001; Females: F(63,450) = 2.5919, p < .001). No main effect of genetic risk for schizophrenia nor an interaction effect between cannabis use and genetic risk for schizophrenia on cognitive abilities was found (p>.05).
*****************
Table 1 about here
*****************
Correlation matrices
Relations amongst cannabis use, genetic risk for schizophrenia, and cognitive abilities were examined by investigating the correlation matrices of cognitive abilities variables. The hypothesis that the correlation matrices for males and females are the same could not be rejected (p = .08; Supplement 2). Similar results were found when comparing all 4 correlation matrices (ever/never cannabis use X low/high genetic risk for schizophrenia) for either males (p = .78) or females (p = .42).
In Figure 1, the correlation matrices for males and females are visualised as networks. Interestingly, most of the cognitive variables are connected to each other to form a so called giant component: a large, constant fraction of the total network that is not connected to (i.e. correlated with) the rest of the network. Variables from the instruments Dot Location, Newman’s Card Sorting, Self-ordered Pointing, and Grooved Pegboard only cluster together with variables from the same cognition tool. The exception is Ruff 2-&-7: While the error
11
and accuracy variables are solely connected to themselves, the two speed variables can be found in the middle of the highly connected giant component.
*****************
Figure 1 about here
*****************
Table 2 provides the global descriptive properties of the networks in Figure 1. The male cognition network contains more edges, has a higher average degree, and a shorter average shortest path length than the female cognition network. The female cognition network contains more clusters than the male cognition network. Visualisation and
descriptive properties of the four correlation matrices (ever/never cannabis use X low/high genetic risk for schizophrenia) for males and females can be found in Supplement 3. For both sexes, the cannabis networks have less edges, lower average degree, and longer average shorter path length than network of non-users. Also similar for both sexes, networks of adolescents with high genetic risk for schizophrenia have more edges, higher degree, and longer average shortest path lengths than low genetic risk for schizophrenia networks.
With 63 variables, a total of 1953 unique correlations were calculated1. Of these correlations, only 10.5-16.1% exceeded the FDR threshold, indicating a non-zero edge strength. Because of this edge sparseness, the Frobenius norm might not be powerful enough to detect any existing differences. Therefore all analyses were run again, replacing the Frobenius norm with the Strength norm. This norm equals the differences in the sum of absolute correlation values, and is thought to have a higher power for detecting differences in sparse networks (Claudia van Borkulo, personal communication, May 22nd 2015).
Application of the Strength norm, however, did not alter the findings (p>.05).
1 Of the 63 by 63 correlation matrix, the upper and lower triangle provide the same information. Removing these
12
*****************
Table 2 about here
*****************
Additionally, the correlations between the cognitive abilities variables on one hand and the variables “cannabis use” and “genetic risk for schizophrenia” on the other hand were calculated. Figure 2 illustrates the visualisation of the cognition network that was enriched with two additional nodes: cannabis use and genetic risk for schizophrenia. As can be seen, “Cannabis use” (grey) is negatively correlated to “Newman’s Card Perseveration” (green) in males, while it does not correlate with any nodes in females. In the networks of neither males nor females “Genetic risk for schizophrenia” (white) is correlated to any other node.
*****************
Figure 2 about here
*****************
Network properties
The relations amongst variables can also be studied using the network properties degree and shortest path length. Figure 3 represents the degree distributions for males and females. Male cannabis users with low genetic risk for schizophrenia and female cannabis users with high genetic risk for schizophrenia differ in their degree distribution compared to their non-using peers (D = 0.29, p = .01 and D = .25, p = .03 respectively). As Figure 3 illustrates, the degree distribution of these cannabis users (Males: blue; Females: black) is shifted to the left compared to their non-using peers (Males: green; Females: red), indicating that their cognition networks contain more nodes with a lower degree. Other contrasts between cannabis users and non-users were not significant (p>.05).
13
*****************
Figure 3 about here
*****************
Figure 4 represents the shortest path length distributions for males and females. For both males and females, regardless of their genetic risk for schizophrenia, cannabis users and non-users differed in their shortest path length (p<.001). As Figure 4 illustrates, the shortest path length distribution of cannabis users (black and blue) is shifted to the right, indicating that those cognition networks are characterised by longer shortest path lengths than non-users. An exception to this are the male cannabis users with low genetic risk for
schizophrenia (blue); compared with non-users (green), their cognition networks are characterised by shorter shortest path lengths.
*****************
Figure 4 about here
*****************
Discussion
The present study investigated the relations amongst cannabis use, genetic risk for schizophrenia, and cognitive abilities in a large community-based sample of adolescents. The relations amongst the cognitive abilities were represented as i) correlations and ii) network properties. No evidence was found in favour of correlations between cannabis use, genetic risk for schizophrenia, and cognitive abilities, indicating that cannabis users and non-users show similar correlations amongst their cognitive abilities. Sole exception is a negative correlation found in males between their cannabis use and perseveration skill. When investigating the network properties degree and shortest path length, however, differences between cannabis users and non-users were found in both males and females. Strength of
14
these differences were dependent on the adolescents’ level of genetic risk for schizophrenia, suggesting a moderating role of genetic risk for schizophrenia in the relations amongst cannabis use and cognitive abilities.
Studying the relations amongst cognitive abilities without imposing any constricting assumptions leads to several interesting observations. While cognitive variables from the same cognitive instrument, supposedly indicative of the same cognitive ability, tend to highly correlate with each other, only a handful do so in isolation. These include instruments that are assumed to measure visuospatial memory, perseveration, working memory, and fine motor skills. Indices of auditory/verbal memory, intelligence, academic achievement, resistance to interference, and cognitive flexibility on the other hand correlate strongly with each other. Remarkable enough, visuospatial and auditory/verbal memory variables are not correlated with each other. This findings might reflect the domain-specific storage and rehearsal processes that, next to more domain-general executive-attention processes, play a role in memory (Kane et al., 2004). In a similar manner, fine motor skills as measured by the Grooved Pegboard are not in any way related to the indices of the Tapping task (which supposedly measures a similar skill). Lastly, selective attention was split into an isolated accuracy cluster and a highly connected speed cluster, with no association between the two. This raises the question whether our current view on cognition truthfully reflects the relations amongst cognitive abilities.
One possibility to acquire more information on the relations amongst cognitive abilities is by eliminating the multicollinearity found amongst the variables of several
cognitive instruments. This can be achieve by either data reductive methods such as Principal Component Analysis (PCA) or the introduction of latent variables (see Epskamp, Rhemtulla, & Borsboom, manuscript in prep, for the rationale of combining latent variables and network models). Removal of multicollinearity not only reduces noise but also makes it possible to
15
look at the unique relations amongst cognitive abilities using partial correlations. Partial correlations have the additional benefit of providing a data generating mechanism that allows for simulation of the relations amongst cognitive abilities. Implementing these and other techniques moves us from exploring to modelling the relations amongst cognitive abilities.
As in the study by Meier et al. (2012), our results show that cannabis use in
adolescents is associated with variations in their cognition abilities. These variations are not only expressed as impaired cognitive abilities, but are also present in the relations amongst the different cognitive abilities. A logical follow-up is to investigate whether all relations amongst cognitive abilities are equally different or whether the differences are limited to a few specific relations. Thus the next step in unravelling the associations amongst cannabis use and cognitive abilities is to compare pairwise correlations and network properties of individual nodes (using permutation and using Kolmogorov-Smirnov tests respectively) between cannabis users and non-users.
In contrast to Snitz et al. (2006), whose meta-analysis showed that individuals with high genetic risk for schizophrenia have impaired executive functioning, our findings do not indicate that elevated genetic risk for schizophrenia is associated with impaired cognitive abilities. Possible reasons for this discrepancy include the use of different target populations (high-risk adult versus community-based adolescent) and different conceptualisation of cognition (assuming independence versus incorporating dependence between individual cognitive abilities). Nonetheless, our results suggest a sex-dependent moderating effect of genetic risk for schizophrenia on the network properties of cognitive abilities. In males, differences in degree between cannabis users and non-users were only found in adolescents with low genetic risk for schizophrenia. Additionally, cannabis users with high genetic risk for schizophrenia had longer shorter path lengths than their non-users peers. The opposite association was observed for cannabis users with low genetic risk for schizophrenia. In
16
females, differences in degree distribution between cannabis users and non-users were present in adolescents with high genetic risk for schizophrenia. Longer shortest path lengths were found in female cannabis users compared to non-users, regardless of their genetic risk for schizophrenia. Further knowledge regarding the association between cognitive
performance and the relations amongst cognitive abilities is necessary to make valid interpretation of these moderation effects.
It must be emphasised that, due to correlational nature of this study, no causal inferences can be made concerning the effects of cannabis use on cognition. Additionally, this study cannot provide evidence that the associations found amongst cannabis use, genetic risk for schizophrenia, and cognitive abilities are specific for either of the three. Cannabis use, for example, co-occurs often with alcohol use (Agosti, Nunes, & Levin, 2002), which by itself is associated with impaired cognitive functioning in adolescents (Brown, Tapert,
Granholm, & Delis, 2000). As such cannabis use might best be seen as a proxy for a certain type of (risky) behaviour associated with impaired cognitive functioning.
Given that, to our knowledge, this is the first study investigating relations amongst cognitive abilities represented as networks, no benchmark exists that can be used to interpret our cognition network properties-related findings. Network models have, however, been applied to investigate structural differences of the brain. Evaluating brain networks based on EEG, MEG, fMRI, and MRI data, Bullmore & Sporns (2009) conclude that the archetypical brain structure is characterised by short path lengths and degree distributions that are
compatible with the existence of hubs. Deviations from this brain structure can be found in patients with Alzheimer’s Disease (Stam, Jones, Nolte, Breakspear, & Scheltens, 2007), epilepsy (Vlooswijk et al., 2011), and schizophrenia (Bassett et al., 2008; Liu et al., 2008; Lynall et al., 2010), and include longer path lengths and reduced probability of high-degree hubs. The overlap in abnormal network structure of patients’ brains on one hand and of
17
cannabis users’ cognitive abilities on the other, suggests that degradation of brain and impairment of cognition might be guided by similar processes. Application of network models can thus contribute to the research on how cognitive (and behavioural) structures relate to those of the brain.
In conclusion, differences between cannabis users and non-users are found in the relations amongst their cognitive abilities, but only when these relations are represented as a network of directly related variables that mutually influence each other. In line with this concept of mutualism (van der Maas et al., 2006), reducing the functioning of a few specific cognitive abilities (e.g. due to cannabis use) might not only contribute to cognitive
impairments but might also interfere with the development of the relations amongst cognitive abilities. Key issues for future include investigating what age is most vulnerable for such interruptions, whether this degradation is selective for some cognitive abilities, and if so, whether this disruptive pattern differs over people.
Acknowledgements
We thank Manon Bernard for managing the online database, and Angelita Wong and Jolanda Kossakowski for reviewing earlier drafts of this manuscript. This work was supported by the Saguenay Youth Study project.
18 References
Agosti, V., Nunes, E., & Levin, F. (2002). Rates of psychiatric comorbidity among U.S. residents with lifetime cannabis dependence. The American Journal of Drug and Alcohol Abuse, 28(4), 643–652. http://doi.org/10.1081/ADA-120015873
Aleman, A., Kahn, R. S., & Selten, J.-P. (2003). Sex differences in the risk of schizophrenia: evidence from meta-analysis. Archives of General Psychiatry, 60(6), 565–71.
http://doi.org/10.1001/archpsyc.60.6.565
Bassett, D. S., Bullmore, E., Verchinski, B. A., Mattay, V. S., Weinberger, D. R., & Meyer-Lindenberg, A. (2008). Hierarchical organization of human cortical networks in health and schizophrenia. The Journal of Neuroscience : The Official Journal of the Society for Neuroscience, 28(37), 9239–48. http://doi.org/10.1523/JNEUROSCI.1929-08.2008
Benjamini, Y., & Hochberg, Y. (1995). Controlling the false discovery rate: a practical and powerful approach to multiple testing. Journal of the Royal Statistical Society. Series B, 57(1), 289-300.
Borsboom, D. (2008). Psychometric perspectives on diagnostic systems. Journal of Clinical Psychology, 64(9), 1089–1108.
Borsboom, D., & Cramer, A. O. J. (2013). Network analysis: an integrative approach to the structure of psychopathology. Annual Review of Clinical Psychology, 9(March), 91–121. http://doi.org/10.1146/annurev-clinpsy-050212-185608
19
Brown, S. a, Tapert, S. F., Granholm, E., & Delis, D. C. (2000). Neurocognitive functioning of adolescents: effects of protracted alcohol use. Alcoholism, Clinical and Experimental Research, 24(2), 164–171. http://doi.org/10.1111/j.1530-0277.2000.tb04586.x
Bullmore, E., & Sporns, O. (2009). Complex brain networks: graph theoretical analysis of structural and functional systems. Nature Reviews. Neuroscience, 10(3), 186–98. http://doi.org/10.1038/nrn2575
Cardno, A. G., & Gottesman, I. I. (2000). Twin studies of schizophrenia: From bow-and-arrow concordances to star wars Mx and functional genomics. American Journal of Medical Genetics - Seminars in Medical Genetics, 97(1), 12–17.
http://doi.org/10.1002/(SICI)1096-8628(200021)97:1<12::AID-AJMG3>3.0.CO;2-U
Cramer, A. O. J., van der Sluis, S., Noordhof, A., Wichers, M., Geschwind, N., Aggen, S. H., … Borsboom, D. (2012). Dimensions of normal personality as networks in search of equilibrium: You can’t like parties if you don't like people. European Journal of Personality, 26(4), 414–431. http://doi.org/10.1002/per.1866
D’Souza, D. C., Abi-Saab, W. M., Madonick, S., Forselius-Bielen, K., Doersch, A., Braley, G., … Krystal, J. H. (2005). Delta-9-tetrahydrocannabinol effects in schizophrenia: Implications for cognition, psychosis, and addiction. Biological Psychiatry, 57(6), 594– 608. http://doi.org/10.1016/j.biopsych.2004.12.006
Degenhardt, L., Chiu, W. T., Sampson, N., Kessler, R. C., Anthony, J. C., Angermeyer, M., … Wells, J. E. (2008). Toward a global view of alcohol, tobacco, cannabis, and cocaine use: Findings from the WHO world mental health surveys. PLoS Medicine, 5(7), 1053– 1067. http://doi.org/10.1371/journal.pmed.0050141
20
Degenhardt, L., & Hall, W. (2012). Extent of illicit drug use and dependence, and their contribution to the global burden of disease. The Lancet, 379(9810), 55–70. http://doi.org/10.1016/S0140-6736(11)61138-0
Egan, M. F., Goldberg, T. E., Gscheidle, T., WEirich, M., Bigelow, L. B., & Weinberger, D. R. (2000). Relative risk of attntion deficits in siblings of patients with schizophrenia. Ameican Journal of Psychiatry, 157(8), 1309–1316.
Epskamp, S., Cramer, A. O. J., Waldorp, L. J., Schmittmann, V. D., & Borsboom, D. (2012). qgraph: Network visualizations of relationships in psychometric data. Journal Of Statistical Software, 48(4), 1–18.
Gage, S. H., Zammit, S., & Hickman, M. (2013). Stronger evidence is needed before accepting that cannabis plays an important role in the aetiology of schizophrenia in the population. F1000 Medicine Reports, 5(2). http://doi.org/10.3410/M5-2
Grant, I., Gonzalez, R., Carey, C. L., Natarajan, L., & Wolfson, T. (2003). Non-acute (residual) neurocognitive effects of cannabis use: a meta-analytic study. Journal of the International Neuropsychological Society : JINS, 9(5), 679–689.
http://doi.org/10.1017/S1355617703950016
Häfner, H., Heiden, W. an der, & Behrens, S. (1998). Causes and consequences of the gender difference in age at onset of schizophrenia. Schizophrenia Bulletin, 24(1), 99-113.
Kafouri, S., Leonard, G., Perron, M., Richer, L., Séguin, J. R., Veillette, S., … Paus, T. (2009). Maternal cigarette smoking during pregnancy and cognitive performance in adolescence. International Journal of Epidemiology, 38(1), 158–172.
21
Liu, Y., Liang, M., Zhou, Y., He, Y., Hao, Y., Song, M., … Jiang, T. (2008). Disrupted small-world networks in schizophrenia. Brain : A Journal of Neurology, 131, 945–61. http://doi.org/10.1093/brain/awn018
Lynall, M.-E., Bassett, D. S., Kerwin, R., McKenna, P. J., Kitzbichler, M., Muller, U., & Bullmore, E. (2010). Functional connectivity and brain networks in schizophrenia. The Journal of Neuroscience : The Official Journal of the Society for Neuroscience, 30(28), 9477–87. http://doi.org/10.1523/JNEUROSCI.0333-10.2010
Meier, M. H., Caspi, a., Ambler, a., Harrington, H., Houts, R., Keefe, R. S. E., … Moffitt, T. E. (2012). From the Cover: PNAS Plus: Persistent cannabis users show
neuropsychological decline from childhood to midlife. Proceedings of the National Academy of Sciences, 109(40), E2657–E2664. http://doi.org/10.1073/pnas.1206820109
Opsahl, T., Agneessens, F., & Skvoretz, J. (2010). Node centrality in weighted networks: Generalizing degree and shortest paths. Social Networks, 32(3), 245–251.
http://doi.org/10.1016/j.socnet.2010.03.006
Paus, T., Keshavan, M., & Giedd, J. N. (2008). Why do many psychiatric disorders emerge during adolescence? Nature Reviews. Neuroscience, 9(12), 947–957.
http://doi.org/10.1038/nrn2513
Pausova, Z., Paus, T., Abrahamowicz, M., Almerigi, J., Arbour, N., Bernard, M., … Watkins, K. (2007). Genes, maternal smoking, and the offspring brain and body during
adolescence: Design of the Saguenay Youth Study. Human Brain Mapping, 28(6), 502– 518. http://doi.org/10.1002/hbm.20402
22
Ripke, S., Neale, B. M., Corvin, A., Walters, J. T. R., Farh, K.-H., Holmans, P. a., … O’Donovan, M. C. (2014). Biological insights from 108 schizophrenia-associated genetic loci. Nature, 511, 421–427. http://doi.org/10.1038/nature13595
Schmittmann, V. D., Cramer, A. O. J., Waldorp, L. J., Epskamp, S., Kievit, R. a., & Borsboom, D. (2013). Deconstructing the construct: A network perspective on psychological phenomena. New Ideas in Psychology, 31(1), 43–53.
http://doi.org/10.1016/j.newideapsych.2011.02.007
Schnell, T., Koethe, D., Daumann, J., & Gouzoulis-Mayfrank, E. (2009). The role of
cannabis in cognitive functioning of patients with schizophrenia. Psychopharmacology, 205(1), 45–52. http://doi.org/10.1007/s00213-009-1512-9
Schwartz, D. H., Leonard, G., Perron, M., Richer, L., Syme, C., Veillette, S., … Paus, T. (2013). Visceral fat is associated with lower executive functioning in adolescents. International Journal of Obesity (2005), 37(10), 1336–1343.
http://doi.org/10.1038/ijo.2013.104
Snitz, B. E., MacDonald, A. W., & Carter, C. S. (2006). Cognitive deficits in unaffected first-degree relatives of schizophrenia patients: A meta-analytic review of putative
endophenotypes. Schizophrenia Bulletin, 32(1), 179–194. http://doi.org/10.1093/schbul/sbi048
Stam, C. J., Jones, B. F., Nolte, G., Breakspear, M., & Scheltens, P. (2007). Small-world networks and functional connectivity in Alzheimer’s disease. Cerebral Cortex (New York, N.Y. : 1991), 17(1), 92–9. http://doi.org/10.1093/cercor/bhj127
23
Sullivan, P. F., Kendler, K. S., & Neale, M. C. (2003). Schizophrenia as a Complex Trait. Archives of General Psychiatry, 60, 1187–1192.
http://doi.org/10.1001/archpsyc.60.12.1187
Van der Maas, H. L. J., Dolan, C. V, Grasman, R. P. P. P., Wicherts, J. M., Huizenga, H. M., & Raijmakers, M. E. J. (2006). A dynamical model of general intelligence: the positive manifold of intelligence by mutualism. Psychological Review, 113(4), 842–861.
http://doi.org/10.1037/0033-295X.113.4.842
Van Gelder, T. (1998). The dynamical hypothesis in cognitive science. The Behavioral and Brain Sciences, 21(5), 615–628; discussion 629–665.
http://doi.org/10.1017/S0140525X98001733
Vlooswijk, M. C. G., Vaessen, M. J., Jansen, J. F. A., de Krom, M. C. F. T. M., Majoie, H. J. M., Hofman, P. A. M., … Backes, W. H. (2011). Loss of network efficiency associated with cognitive decline in chronic epilepsy. Neurology, 77(10), 938–44.
http://doi.org/10.1212/WNL.0b013e31822cfc2f
Volkow, N. D. (2009). Substance use disorders in schizophrenia--clinical implications of comorbidity. Schizophrenia Bulletin, 35(3), 469–472.
http://doi.org/10.1093/schbul/sbp016
Yücel, M., Bora, E., Lubman, D. I., Solowij, N., Brewer, W. J., Cotton, S. M., … Pantelis, C. (2012). The impact of cannabis use on cognitive functioning in patients with
schizophrenia: A meta-analysis of existing findings and new data in a first-episode sample. Schizophrenia Bulletin, 38(2), 316–330. http://doi.org/10.1093/schbul/sbq079
24
Zhao, T., Liu, H., & Roeder, K. (2012). The huge package for high-dimensional undirected graph estimation in r. The Journal of Machine …, 13, 1059–1062.
25 Figure captions
Figure 1. Visualisation of the cognition correlation matrices as networks. The cognition variables are represented as nodes and connected by an edge is the correlation exceeds the False Discovery Rate threshold. Green (red) edges indicate a positive (negative) correlation. The thicker the line, the higher the correlation. Colour of nodes represent the cognitive instrument where the variable is from. Supplement 2 gives the definitions of the abbreviations.
Figure 2. Visualisation of the cognition correlation matrices as networks, enriched with variables “Cannabis use” and “Genetic risk for schizophrenia”. The variables are represented as nodes and connected by an edge is the correlation exceeds the False Discovery Rate threshold. Green (red) edges indicate a positive (negative) correlation. The thicker the line, the higher the correlation. Colour of nodes represent the cognition instrument where the variable is from. Supplement 2 gives the definitions of the abbreviations.
Figure 3. The degree distributions of the FDR corrected male and female cognition networks.
Figure 4. The shortest path length distributions of the FDR corrected male and female cognition networks.
26 Table 1
Participant demographics
Males (n = 493) Females (n = 529) p-value
Age in months (sd) 179.90 (21.45) 181.62 (22.48) 0.24†
% Cannabis ever-users 30.63 33.65 0.33‡
Median genetic risk score for schizophrenia
-0.058 -0.075 0.53*
†Mann-Whitney U test; ‡Chi-square test of independence; *Mood’s median test.
Table 2
Descriptive network properties for the FDR corrected cognition networks as visualised in Figure 1. Males (n = 493) Females (n = 529) Number of nodes 63 63 Number of edges 314 206 Number of clusters 8 9 Average degree 4.81 3.32
Average shortest path length 3.73 4.96
