Introduction
The aetiology of psychosis is still poorly understood (Perala et al. Reference Perala, Suvisaari, Saarni, Kuoppasalmi, Isometsa, Pirkola, Partonen, Tuulio-Henriksson, Hintikka and Kieseppa2007). Schizophrenia (SZ) is characterized by psychotic symptoms (such as hallucinations and delusions), lack of motivation, social withdrawal and executive and memory deficits. Bipolar disorder (BD) features prominent mood swings, i.e. alternating episodes of mania and depression, and is found to include psychotic symptoms in half to two-thirds of cases (Canuso et al. Reference Canuso, Bossie, Zhu, Youssef and Dunner2008). Despite the continuous effort to provide effective treatment and management of these prevalent illnesses [lifetime prevalence of 0.5–1% for SZ (Cannon & Jones, Reference Cannon and Jones1996) and 3–4% for BD (Faravelli et al. Reference Faravelli, Guerrini Degl'Innocenti, Aiazzi, Incerpi and Pallanti1990; Szadoczky et al. Reference Szadoczky, Papp, Vitrai, Rihmer and Furedi1998; ten Have et al. Reference ten Have, Vollebergh, Bijl and Nolen2002)], they remain chronic and recurrent for most cases, entailing high rates of morbidity and mortality (Tohen et al. Reference Tohen, Waternaux and Tsuang1990; Jamison, Reference Jamison2000; Judd et al. Reference Judd, Akiskal, Schettler, Endicott, Maser, Solomon, Leon, Rice and Keller2002) and a large emotional and financial burden to society (Murray & Lopez, Reference Murray and Lopez1997; Knapp et al. Reference Knapp, Mangalore and Simon2004; Ogilvie et al. Reference Ogilvie, Morant and Goodwin2005).
Besides the overlap of symptoms between SZ and BD, emerging evidence implicates shared genetic susceptibility (Avissar & Schreiber, Reference Avissar and Schreiber2002; Knight et al. Reference Knight, Pickard, Maclean, Malloy, Soares, McRae, Condie, White, Hawkins, McGhee, van Beck, MacIntyre, Starr, Deary, Visscher, Porteous, Cannon, St Clair, Muir and Blackwood2009; Lichtenstein et al. Reference Lichtenstein, Yip, Bjork, Pawitan, Cannon, Sullivan and Hultman2009; Sims et al. Reference Sims, Hollingworth, Moskvina, Dowzell, O'Donovan, Powell, Lovestone, Brayne, Rubinsztein, Owen, Williams and Abraham2009) and like most disorders of complex (non-Mendelian) origin, they are likely to result from a combination of genetic and environmental factors [heritability being reported as high as 80% for SZ, and 93% for BD (Cardno et al. Reference Cardno, Marshall, Coid, Macdonald, Ribchester, Davies, Venturi, Jones, Lewis, Sham, Gottesman, Farmer, McGuffin, Reveley and Murray1999; Kieseppa et al. Reference Kieseppa, Partonen, Haukka, Kaprio and Lonnqvist2004)]. The involvement of multiple genes, their varied levels of penetrance and expression (epigenetics), their interaction with one another (epistasis), the influence of environment, and the highly heterogeneous clinical presentations, make the elucidation of a clear-cut genetic architecture for SZ and BD particularly challenging (Petronis et al. Reference Petronis, Gottesman, Crow, DeLisi, Klar, Macciardi, McInnis, McMahon, Paterson, Skuse and Sutherland2000; Arseneault et al. Reference Arseneault, Cannon, Witton and Murray2004; Bebbington et al. Reference Bebbington, Bhugra, Brugha, Singleton, Farrell, Jenkins, Lewis and Meltzer2004; Moore et al. Reference Moore, Zammit, Lingford-Hughes, Barnes, Jones, Burke and Lewis2007; Must et al. Reference Must, Janka and Horvath2011; Kirkbride et al. Reference Kirkbride, Errazuriz, Croudace, Morgan, Jackson, Boydell, Murray and Jones2012; Pelayo-Teran et al. Reference Pelayo-Teran, Suarez-Pinilla, Chadi and Crespo-Facorro2012).
The completion of the Human Genome Project has helped overcome some of these difficulties. It has allowed the survey of single nucleotide polymorphisms (SNPs) in thousands of subjects, some of which were found to tag DNA regions whose transmission is associated with the presence of SZ or BD (Hirschhorn & Daly, Reference Hirschhorn and Daly2005). For this, a stringent threshold for statistical significance is set given the large amount (a few thousands to a couple of million) SNPs being tested in each subject (Dudbridge & Gusnanto, Reference Dudbridge and Gusnanto2008). This method, known as genome-wide association (GWA), offers an alternative, hypothesis-free, approach to traditional pre-chosen candidate gene studies (Ansorge, Reference Ansorge2009). Starting in 2006, such recent collaborative efforts have identified GWA significant risk genes for SZ (such as ZNF804A, CACNA1C, TCF4, and NRGN) and BD (such as ZNF804A, CACNA1C, PBRM1, ANK3, and DGKH) (Lee et al. Reference Lee, Woon, Teo and Sim2012). Having survived stringent statistical testing, these findings certainly merit the emerging enthusiasm in examining their downstream physiological pathways to psychosis.
Surprisingly however, the vast majority of the above genes had not been previously implicated (and, conversely, most of previous candidate genes did not emerge as GWA significant). Thus, knowledge of how they might contribute to these illnesses is especially lacking. This may lower our confidence that they are true positives and can contribute to solving the pathophysiological puzzles of SZ/BD or inform the design of biomarkers and treatments (Prata et al. Reference Prata, Mechelli and Kapur2014).
Looking into the brain is an obvious next step in working out how these GWAS variants may elicit risk of psychosis. The utility of intermediate phenotypes at the neuronal-systems level as a means to bridge the gap between the causative risk genotype(s) and a disease has been increasingly recognized over the last decade, with an array of neuroimaging modalities (Gottesman & Gould, Reference Gottesman and Gould2003). This approach, ‘imaging genetics’, rationally predicts that a genetic variation will show higher penetrance at the brain's structure and function level than at the (more distal and complex) behavioral level (Rasetti & Weinberger, Reference Rasetti and Weinberger2011). Thus, examining the impact of risk genotypes on brain features that seem to be altered in the illness may help elucidate how they induce such susceptibility. Because deficits in working memory, attention, episodic memory, verbal fluency, and emotion processing have been consistently associated to SZ and BD (Weinberger, Reference Weinberger1999; Marwick & Hall, Reference Marwick and Hall2008; Li et al. Reference Li, Branch and DeLisi2009; MacDonald et al. Reference MacDonald, Thermenos, Barch and Seidman2009) and so have their neurocorrelates, the latter have become extensively used as neuroimaging intermediate phenotypes in genetic studies of these illnesses (Rasetti & Weinberger, Reference Rasetti and Weinberger2011).
This systematic review gathers and discusses all existing evidence of the impact of GWA-significant psychosis-risk variants on brain structure and function, across all neuroimaging modalities, being the first of its kind. Our goal was twofold: (1) primarily, to facilitate the timely integration of findings in this rapidly growing field of study – imaging genetics of psychosis GWAS variants – in the overarching research context; and, (2) second, to critically evaluate these studies in order to encourage reflection to improve future study design and reporting, for which we propose a 12-requirement checklist.
Method
Selection
A recently published systematic review has identified a list of GWAS-supported risk genes for SZ and BD (Lee et al. Reference Lee, Woon, Teo and Sim2012). Their abbreviations were used as subject headings in the search box of PubMed (http://www.ncbi.nlm.nih.gov/pubmed), Scopus (http://www.scopus.com) and Web of Science (http://webofknowledge.com) databases along with imaging modality keyword abbreviations: ‘(ANK3 OR CACNA1C OR MHC OR TCF4 OR NRGN OR DGKH OR PBRM1 OR NCAN OR ZNF804A) AND (MRI OR fMRI OR MRS OR DTI OR PET OR EEG)’. The inclusion and exclusion criteria were tailored to obtain all existing human neuroimaging studies of these GWAS-significant risk genes published up to 31 August 2013, irrespective of publication date or subjects’ ethnicity, age group, gender and diagnosis (Fig. 1). Animal studies, literature reviews or meta-analyses and non-English written studies were excluded. The studies’ bibliographies were further surveyed for additional relevant studies. Studies were included whether their primary focus was on the SNPs identified in the GWAS (Lee et al. Reference Lee, Woon, Teo and Sim2012), or on SNPs in high linkage disequilibrium with them (r 2 ⩾ 0.80), which was checked using http://www.broadinstitute.org.
Different phases of the systematic search for imaging genetics studies on the effect of psychosis risk variants, found through genome-wide association, on brain structure and function.
Review
The recorded variables for each study were the primary gene and SNP of interest, the risk allele, subjects’ ethnicity, diagnosis, sample size, neuroimaging modality and measure, direction of association, brain search area, statistical test, software used, statistical significance thresholds, covariates (if any) and areas of reported effects (Table 2 and online Supplementary Table S1). Except when no correction was crucial [e.g. only total grey matter (GM) and/or white matter (WM) was being tested], we have described results surviving a statistical significance threshold of p < 0.05 after correction for multiple comparisons within the search brain area as statistically significant associations, and those that have not, as trends (which we avoid discussing). Trends were labelled as such in Table 2 and online Supplementary Table S1, regardless of the threshold used by the authors, in order to facilitate comparison between different studies.
Quality assessment
Each study was evaluated on a 12-quality requirement score list (detailed in Table 1). For this, information was retrieved from the main article or online Supplementary material. For each item, a score of 0–3 depended on whether there was strong (3), some (2), little (1) or no evidence (0) that it was adhered to. The sum of the 12 items’ scores, divided by the maximum sum applicable to the respective study modality, was then used as an indication of the general quality of the study (low ⩽69%, medium70–79%, high ⩾80%).
A 12-item score list to deduce the quality standard of the imaging genetics studies included in this review and the % of studies complying the highest (score 3) with each item
Results
Overview
A total of 40 studies investigating seven out of the nine identified risk genes for psychosis was revealed (Fig. 1). No studies of the MHC and PBRM1 genes nor using EEG, MRS or PET were found.
Research on the effect of CACNA1C and ZNF804A risk variants on brain structure and function was the most popular, making up 30 studies. In these, the intermediate phenotypes investigated were, in order of frequency, anatomic measures (total/regional brain volumes, cortical thickness, cortical folding and WM integrity), emotion processing, working memory, episodic memory, verbal fluency, attention, theory of mind and reward reversal learning, all previously shown to be altered in SZ or BD (Weinberger, Reference Weinberger1999; Gottesman & Gould, Reference Gottesman and Gould2003; Marwick & Hall, Reference Marwick and Hall2008; Li et al. Reference Li, Branch and DeLisi2009; MacDonald et al. Reference MacDonald, Thermenos, Barch and Seidman2009; Rasetti et al. Reference Rasetti, Mattay, Wiedholz, Kolachana, Hariri, Callicott, Meyer-Lindenberg and Weinberger2009; Rasetti & Weinberger, Reference Rasetti and Weinberger2011). In addition, three studies focused on the effect of ANK3 risk variants on working memory and brain structure (total/regional brain volumes and WM integrity), one on DGKH's effect on verbal fluency, one on NCAN's effect on cortical folding and thickness, four on NRGN's effect on episodic memory, brain volumes, emotion processing and working memory, and finally, one on TCF4's effect on brain volume and cortical thickness.
A little more than half of the studies used only healthy individuals (22 studies), and the other 17 also used affected individuals (in comparison with healthy controls): a majority with SZ (n = 7), then BD (n = 3), four including both and three including individuals at a high risk for either BD1 (n = 3) or SZ (n = 1). Structural imaging investigations (55%) were as abundant as functional (45%) ones.
ZNF804A
ZNF804A encodes the zinc-finger protein 804A – a protein so far of unknown function but with a zinc finger domain typical of DNA binding and thus possibly acting as a transcription factor. It seems to regulate gene expression of many genes involved in cell adhesion, neurite outgrowth and dendritic branching (Hill et al. Reference Hill, Jeffries, Dobson, Price and Bray2012) and of at least four long-known SZ-associated candidate genes: COMT, DRD2, PRSS16 and PDE4 (Girgenti et al. Reference Girgenti, LoTurco and Maher2012). Thus, the impact that variations in this gene may have on SZ and BD risk may be both via effects on brain structure or directly on brain function. Its rs1344706 SNP was one of the first variations found to be genome-wide-associated with SZ (O'Donovan et al. Reference O'Donovan, Craddock, Norton, Williams, Peirce, Moskvina, Nikolov, Hamshere, Carroll, Georgieva, Dwyer, Holmans, Marchini, Spencer, Howie, Leung, Hartmann, Moller, Morris, Shi, Feng, Hoffmann, Propping, Vasilescu, Maier, Rietschel, Zammit, Schumacher, Quinn, Schulze, Williams, Giegling, Iwata, Ikeda, Darvasi, Shifman, He, Duan, Sanders, Levinson, Gejman, Cichon, Nothen, Gill, Corvin, Rujescu, Kirov, Owen, Buccola, Mowry, Freedman, Amin, Black, Silverman, Byerley and Cloninger2008; Williams et al. Reference Williams, Norton, Dwyer, Moskvina, Nikolov, Carroll, Georgieva, Williams, Morris, Quinn, Giegling, Ikeda, Wood, Lencz, Hultman, Lichtenstein, Thiselton, Maher, Malhotra, Riley, Kendler, Gill, Sullivan, Sklar, Purcell, Nimgaonkar, Kirov, Holmans, Corvin, Rujescu, Craddock, Owen and O'Donovan2011b ), effect which increased with the addition of a BD sample. Its role in brain function, in healthy subjects, was first investigated by Esslinger et al. (Reference Esslinger, Walter, Kirsch, Erk, Schnell, Arnold, Haddad, Mier, Opitz von Boberfeld, Raab, Witt, Rietschel, Cichon and Meyer-Lindenberg2009). They reported that despite it having no effect on regional activation, carrying the risk allele (A) brought about a dose-dependent increase in functional connectivity of the right dorsolateral prefrontal cortex (DLPFC) with the left hippocampus, and a decrease in that with the left and (other) right DLPFC regions (Esslinger et al. Reference Esslinger, Walter, Kirsch, Erk, Schnell, Arnold, Haddad, Mier, Opitz von Boberfeld, Raab, Witt, Rietschel, Cichon and Meyer-Lindenberg2009), during working memory. The increase in prefronto-hippocampal connectivity was concluded to be task-specific since it was not seen during emotional face recognition or resting state in the same sample (Esslinger et al. Reference Esslinger, Kirsch, Haddad, Mier, Sauer, Erk, Schnell, Arnold, Witt, Rietschel, Cichon, Walter and Meyer-Lindenberg2011). On the other hand, the DLPFC inter-hemispheric dysconnectivity, was present in all three tasks. During that emotional face-recognition task, the risk allele was associated with increased connectivity between the right amygdala and the frontal, prefrontal and temporal cortices, the left hippocampus, the left amygdala and the striatum bilaterally.
The increase in functional connectivity between the right DLPFC and hippocampal areas during working memory in healthy people was replicated in another study; however, not the inter-hemispheric dysconnectivity (Paulus et al. Reference Paulus, Krach, Bedenbender, Pyka, Sommer, Krug, Knake, Nothen, Witt, Rietschel, Kircher and Jansen2013). Nevertheless, the increase was not replicated in yet another study in further healthy subjects, SZ patients and unaffected siblings of SZ patients. Instead, a significant increase in connectivity within the right DLPFC region was detected in patients (Rasetti et al. Reference Rasetti, Sambataro, Chen, Callicott, Mattay and Weinberger2011), in opposition to Esslinger et al.'s (2009) finding in healthy subjects.
In summary, rather than any effect on regional activation, the association of the risk allele with increased fronto-temporal functional connectivity in working memory has been replicated in three samples and the decreased prefrontal inter-hemispheric connectivity in two samples. Although the latter was found to be reversed in SZ patients, no genotype by diagnosis interaction test was performed to discern whether it was significantly different from what is seen in health (Rasetti et al. Reference Rasetti, Sambataro, Chen, Callicott, Mattay and Weinberger2011). The observation that a SZ risk allele could contribute to decreased prefrontal inter-hemispheric connectivity is consistent with the dysconnectivity hypothesis of SZ, which has been specially supported between the two hemispheres (Stephan et al. Reference Stephan, Baldeweg and Friston2006). Its contribution to an increase in fronto-temporal connectivity is neatly explained by a previous study, of the same group, that showed it to be abnormally persistent during working memory in SZ (Meyer-Lindenberg et al. Reference Meyer-Lindenberg, Olsen, Kohn, Brown, Egan, Weinberger and Berman2005). An effect of the risk allele in regional activation was found, rather, in working memory for faces, as an increase in the DLPFC (after controlling for performance differences) in healthy subjects (Linden et al. Reference Linden, Lancaster, Wolf, Baird, Jackson, Johnston, Donev and Thome2013), which is consistent with the commonly supported inefficient DLPFC activation during working memory in SZ (Callicott et al. Reference Callicott, Bertolino, Mattay, Langheim, Duyn, Coppola, Goldberg and Weinberger2000). The contribution of the risk allele to increased amygdala connectivity with a widespread network during facial affect recognition is an interesting finding in respect to BD more than SZ, but needs further replication.
The feature of an abnormal theory of mind system in these illnesses has also started gaining support. Behaviour-wise, two meta-analyses (Sprong et al. Reference Sprong, Schothorst, Vos, Hox and van Engeland2007; Bora et al. Reference Bora, Yucel and Pantelis2009) have shown a robust deficit and imaging-wise, a reduced activation in the theory of mind network is repeatedly seen in SZ (Russell et al. Reference Russell, Rubia, Bullmore, Soni, Suckling, Brammer, Simmons, Williams and Sharma2000; Brunet et al. Reference Brunet, Sarfati, Hardy-Bayle and Decety2003; Andreasen et al. Reference Andreasen, Calarge and O'Leary2008). Consistent with this, Walter et al. (Reference Walter, Schnell, Erk, Arnold, Kirsch, Esslinger, Mier, Schmitgen, Rietschel, Witt, Nothen, Cichon and Meyer-Lindenberg2011) found the risk allele was dose-dependently associated with decreased activation during theory of mind in bilateral dorsal medial prefrontal cortex (PFC), the left tempo-parietal cortex (TPC), left inferior parietal cortex (IPC), posterior cingulate and the left lateral PFC of healthy subjects (Walter et al. Reference Walter, Schnell, Erk, Arnold, Kirsch, Esslinger, Mier, Schmitgen, Rietschel, Witt, Nothen, Cichon and Meyer-Lindenberg2011). There was also a trend for increased functional connectivity of the left temporal parietal junction (TPJ) with several regions including the left inferior frontal gyrus (IFG), left cuneus, left caudate and right thalamus, and decreased coupling of the right DLPFC with the right precentral gyrus, medial temporal gyrus (TG) and left lingual gyrus.
With regard to anatomical measures, the risk allele was initially associated with increased total WM volume (Lencz et al. Reference Lencz, Szeszko, DeRosse, Burdick, Bromet, Bilder and Malhotra2010) in healthy subjects but reduced total GM volume in the ‘default mode network’ (which includes the angular, parahippocampal, posterior cingulate and medial orbitofrontal gyri), in healthy subjects. This was not replicated in a posterior study (in terms of both WM and GM), but a trend for an increase in GM volume in the bilateral superior temporal gyrus and insula was detected in SZ patients only (Donohoe et al. Reference Donohoe, Rose, Frodl, Morris, Spoletini, Adriano, Bernardini, Caltagirone, Bossu, Gill, Corvin and Spalletta2011). Looking a priori to this region, this study also found the same effect, and significant, in the hippocampus bilaterally, also in patients only (Donohoe et al. Reference Donohoe, Rose, Frodl, Morris, Spoletini, Adriano, Bernardini, Caltagirone, Bossu, Gill, Corvin and Spalletta2011). This appears counter-intuitive given that this allele increases risk for SZ and that cognitive ability correlates positively with GM volume (Hulshoff Pol et al. Reference Hulshoff Pol, Schnack, Mandl, Brans, van Haren, Baare, van Oel, Collins, Evans and Kahn2006), even though the authors argue that it may be relative to a subtype of SZ. Another counter-intuitive trend associated the risk allele with enlarged total WM in SZ patients and controls, with a frontal lobe increase present only in SZ (Wassink et al. Reference Wassink, Epping, Rudd, Axelsen, Ziebell, Fleming, Monson, Ho and Andreasen2012). However, as hereby designated by ‘trends’, these effects were not corrected for multiple comparisons thus caution in their interpretation is needed. In fact, no association with total/regional GM, WM and total brain volumes in an up to 10x larger sample of healthy participants (Cousijn et al. Reference Cousijn, Rijpkema, Harteveld, Harrison, Fernandez, Franke and Arias-Vasquez2012) was detected, even at an uncorrected level.
In terms of WM density, the risk allele has been associated at trend level with an increase in the hippocampus bilaterally in both the Chinese SZ patient and control groups (Wei et al. Reference Wei, Kang, Diao, Shan, Li, Zheng, Guo, Liu, Zhang and Zhao2012). This finding may explain the earlier finding of increased prefronto-temporal functional connectivity by Esslinger et al. (Reference Esslinger, Walter, Kirsch, Erk, Schnell, Arnold, Haddad, Mier, Opitz von Boberfeld, Raab, Witt, Rietschel, Cichon and Meyer-Lindenberg2009), but replication is needed. In the left prefrontal lobe, this effect was still seen in patients, but reversed in healthy subjects, which was reflected in a significant genotype by diagnosis interaction but difficult to interpret by itself. Decreased cortical thickness in the anterior cingulate cortex (ACC), posterior cingulate cortex (PCC) and superior temporal gyrus (STG) in healthy subjects was also predictably associated with the risk allele (Voineskos et al. Reference Voineskos, Lerch, Felsky, Tiwari, Rajji, Miranda, Lobaugh, Pollock, Mulsant and Kennedy2011), but this was later disputed by Bergmann et al. (Reference Bergmann, Haukvik, Brown, Rimol, Hartberg, Athanasiu, Melle, Djurovic, Andreassen, Dale and Agartz2013) who found no effect in a more than double healthy sample (Bergmann et al. Reference Bergmann, Haukvik, Brown, Rimol, Hartberg, Athanasiu, Melle, Djurovic, Andreassen, Dale and Agartz2013). With regard to WM integrity, studies carried out have mostly reported no association of the present risk allele, in either patients and healthy participants (Voineskos et al. Reference Voineskos, Lerch, Felsky, Tiwari, Rajji, Miranda, Lobaugh, Pollock, Mulsant and Kennedy2011; Sprooten et al. Reference Sprooten, McIntosh, Lawrie, Hall, Sussmann, Dahmen, Konrad, Bastin and Winterer2012; Wei et al. Reference Wei, Kang, Diao, Guidon, Wu, Zheng, Li, Guo, Hu, Zhang, Liu and Zhao2013), with one study reporting whole-brain trends for decreased WM integrity in the bilateral parietal lobes and left central gyrus in SZ patients and increases in the right temporal lobe of healthy participants (Kuswanto et al. Reference Kuswanto, Woon, Zheng, Qiu, Sitoh, Chan, Liu, Williams, Ong and Sim2012). This was reflected in significant genotype by diagnosis interactions, whereby, not surprisingly, having two copies of the risk allele and SZ consistently (in the four areas) showed lower WM integrity.
CACNA1C
CACNA1C had been implicated via GWA in BD (Ferreira et al. Reference Ferreira, O'Donovan, Meng, Jones, Ruderfer, Jones, Fan, Kirov, Perlis, Green, Smoller, Grozeva, Stone, Nikolov, Chambert, Hamshere, Nimgaonkar, Moskvina, Thase, Caesar, Sachs, Franklin, Gordon-Smith, Ardlie, Gabriel, Fraser, Blumenstiel, Defelice, Breen, Gill, Morris, Elkin, Muir, McGhee, Williamson, MacIntyre, MacLean, St, Robinson, Van Beck, Pereira, Kandaswamy, McQuillin, Collier, Bass, Young, Lawrence, Ferrier, Anjorin, Farmer, Curtis, Scolnick, McGuffin, Daly, Corvin, Holmans, Blackwood, Gurling, Owen, Purcell, Sklar and Craddock2008), SZ (Green et al. Reference Green, Grozeva, Jones, Jones, Kirov, Caesar, Gordon-Smith, Fraser, Forty, Russell, Hamshere, Moskvina, Nikolov, Farmer, McGuffin, Holmans, Owen, O'Donovan and Craddock2010a ; Nyegaard et al. Reference Nyegaard, Overgaard, Su, Hamilton, Kwintkiewicz, Hsieh, Nayak, Conti, Conover and Giudice2010) and major depression (Green et al. Reference Green, Grozeva, Jones, Jones, Kirov, Caesar, Gordon-Smith, Fraser, Forty, Russell, Hamshere, Moskvina, Nikolov, Farmer, McGuffin, Holmans, Owen, O'Donovan and Craddock2010b ; Curtis et al. Reference Curtis, Vine, McQuillin, Bass, Pereira, Kandaswamy, Lawrence, Anjorin, Choudhury, Datta, Puri, Krasucki, Pimm, Thirumalai, Quested and Gurling2011). The type of L-type calcium channels it codes for regulates neuronal plasticity and may have direct effects on transcription of genes involved in neuronal signaling and excitability (Gomez-Ospina et al. Reference Gomez-Ospina, Tsuruta, Barreto-Chang, Hu and Dolmetsch2006). Kempton et al. (Reference Kempton, Ruberto, Vassos, Tatarelli, Girardi, Collier and Frangou2009) were the first to investigate the plausible impact of CACNA1C rs1006737 risk variant on brain anatomy; but they reported that the risk variant brought about a trend for a dose-dependent increase in total GM volume in healthy subjects(Kempton et al. Reference Kempton, Ruberto, Vassos, Tatarelli, Girardi, Collier and Frangou2009), rather than an expected decrease. This was, however, later tentatively refuted by a trend for an association with decreased total GM volume, also in a gene–dose effect (Franke et al. Reference Franke, Vasquez, Veltman, Brunner, Rijpkema and Fernandez2010), and in terms of both total brain volume and thickness, no effects were found in another study in (BD and SZ) patients or healthy participants (Tesli et al. Reference Tesli, Egeland, Sonderby, Haukvik, Bettella, Hibar, Thompson, Rimol, Melle, Agartz, Djurovic and Andreassen2013a ). Throughout the brain, the direction of effect was very region-specific. The risk allele was associated with decreased GM volume in the left putamen, but in patients only, with an opposite trend in controls (Perrier et al. Reference Perrier, Pompei, Ruberto, Vassos, Collier and Frangou2011), but it was also associated with increased GM volume in the right amygdala and right hypothalamus in both BD patients and controls (Perrier et al. Reference Perrier, Pompei, Ruberto, Vassos, Collier and Frangou2011), and, in a separate study, in bilateral ventral, rostral and dorsolateral PFC, ACC and temporal cortex (TC) (with trends with the same direction in the insular, parietal, and occipital cortices in healthy subjects only) (Wang et al. Reference Wang, McIntosh, He, Gelernter and Blumberg2011). These were later disputed by an at least similarly-sized study which found no association with the amygdala or hippocampus volumes in BD patients or healthy controls (Soeiro-de-Souza et al. Reference Soeiro-de-Souza, Otaduy, Dias, Bio, Machado-Vieira and Moreno2012). Three additional SNPs (rs2051992, rs2239050 and rs7959938) within 100 kb of the rs1006737 SNP have also shown significant associations with increased brainstem WM volume (Franke et al. Reference Franke, Vasquez, Veltman, Brunner, Rijpkema and Fernandez2010), an area, however, with little research and implication in psychotic illnesses so far.
In terms of brain function, the same risk allele showed a trend for increased activation in bilateral hippocampi during emotional memory and in the PFC during working memory (Bigos et al. Reference Bigos, Mattay, Callicott, Straub, Vakkalanka, Kolachana, Hyde, Lipska, Kleinman and Weinberger2010). The effect in the hippocampus is consistent with it also having been reported in BD during an emotional task (Whalley et al. Reference Whalley, McKirdy, Romaniuk, Sussmann, Johnstone, Wan, McIntosh, Lawrie and Hall2009). The increased effect in the PFC, given that performance level was controlled for, could again be interpreted as lower efficiency, which is compatible with this being a feature of SZ. However, the latter was contested by another study, which surprisingly, found the opposite effect direction (Paulus et al. Reference Paulus, Bedenbender, Krach, Pyka, Krug, Sommer, Mette, Nothen, Witt, Rietschel, Kircher and Jansen2014). The latter study also found increased functional coupling of that region (right DLPFC) with bilateral hippocampal formations (HFs) (again dose-dependently), which interestingly mimics what was found for ZNF804A rs1344706 risk allele, suggesting perhaps a common downstream pathway for both risk variants.
In episodic memory, healthy risk allele carriers showed, compared to their counterparts, decreased activation during recall in the bilateral hippocampus, ACC, ventral striatum, medial and STG and superior frontal gyrus (SFG), and also decreased coupling between the left and right hippocampus (Erk et al. Reference Erk, Meyer-Lindenberg, Schnell, Opitz von Boberfeld, Esslinger, Kirsch, Grimm, Arnold, Haddad, Witt, Cichon, Nothen, Rietschel and Walter2010), which also seems to cause deficient recall in mice (Canals et al. Reference Canals, Beyerlein, Merkle and Logothetis2009). In semantic verbal fluency, these subjects showed increased activation in the left precuneus and, again putatively reflecting inefficiency, in the left IFG (Krug et al. Reference Krug, Nieratschker, Markov, Krach, Jansen, Zerres, Eggermann, Stocker, Shah, Treutlein, Muhleisen and Kircher2010). In an attentional network task, the association was with decreased activation in the right inferior parietal lobule (IPL) during orienting and in the medial frontal gyrus (MFG), during attention (Thimm et al. Reference Thimm, Kircher, Kellermann, Markov, Krach, Jansen, Zerres, Eggermann, Stocker, Shah, Nothen, Rietschel, Witt, Mathiak and Krug2011).
In negative facial affect recognition, BD patients, unaffected siblings of BD patients and healthy participants together, showed an association between the risk allele and increased activation in the right amygdala, while a decreased activation in the right ventral lateral PFC was seen in patients only (Jogia et al. Reference Jogia, Ruberto, Lelli-Chiesa, Vassos, Maieru, Tatarelli, Girardi, Collier and Frangou2011). This BD-specific combination of effects seems consistent with the replicated finding of ventral lateral PFC regulating (toning down) amygdala activity (Yurgelun-Todd et al. Reference Yurgelun-Todd, Gruber, Kanayama, Killgore, Baird and Young2000; Hariri et al. Reference Hariri, Tessitore, Mattay, Fera and Weinberger2002). Similarly, a main effect of increased activation was also shown across BD, SZ and controls, while significant only in the former, but in the left amygdala during a similar paradigm, not right (Tesli et al. Reference Tesli, Skatun, Ousdal, Brown, Thoresen, Agartz, Melle, Djurovic, Jensen and Andreassen2013b ). Again in the right amygdala, but in a reward reversal-learning task, the risk allele was associated with increased activation (Wessa et al. Reference Wessa, Linke, Witt, Nieratschker, Esslinger, Kirsch, Grimm, Hennerici, Gass, King and Rietschel2010). Further studies are needed to address this laterality inconsistency.
ANK3
ANK3, which codes for Ankyrin G, a protein important in voltage gating in neurotransmission, had been associated with both BD (Ferreira et al. Reference Ferreira, O'Donovan, Meng, Jones, Ruderfer, Jones, Fan, Kirov, Perlis, Green, Smoller, Grozeva, Stone, Nikolov, Chambert, Hamshere, Nimgaonkar, Moskvina, Thase, Caesar, Sachs, Franklin, Gordon-Smith, Ardlie, Gabriel, Fraser, Blumenstiel, Defelice, Breen, Gill, Morris, Elkin, Muir, McGhee, Williamson, MacIntyre, MacLean, St, Robinson, Van Beck, Pereira, Kandaswamy, McQuillin, Collier, Bass, Young, Lawrence, Ferrier, Anjorin, Farmer, Curtis, Scolnick, McGuffin, Daly, Corvin, Holmans, Blackwood, Gurling, Owen, Purcell, Sklar and Craddock2008) and SZ (Athanasiu et al. Reference Athanasiu, Mattingsdal, Kahler, Brown, Gustafsson, Agartz, Giegling, Muglia, Cichon, Rietschel, Pietilainen, Peltonen, Bramon, Collier, Clair, Sigurdsson, Petursson, Rujescu, Melle, Steen, Djurovic and Andreassen2010). Its rs9804190 risk variant has since been associated only once with brain function, showing increased activation in the left IFG and left MFG during working memory by Roussos et al. (Reference Roussos, Katsel, Davis, Bitsios, Giakoumaki, Jogia, Rozsnyai, Collier, Frangou, Siever and Haroutunian2012), with performance differences controlled for. Inefficient activation being associated itself with SZ (Callicott et al. Reference Callicott, Bertolino, Mattay, Langheim, Duyn, Coppola, Goldberg and Weinberger2000), this is a plausible finding. No effect was found on WM integrity when looking in the anterior limb of the internal capsule (ALIC), uncinate fasciculus (UF) and corpus callosum in healthy subjects (Linke et al. Reference Linke, Witt, King, Nieratschker, Poupon, Gass, Hennerici, Rietschel and Wessa2012). However, the latter study also investigated another SNP in ANK3, rs10994336 and found the risk allele to be associated with decreased WM integrity in bilateral ALIC. No effect was, however, seen for either SNP on brain volumes and cortical thickness in a sample of SZ and BD patients and healthy subjects (Tesli et al. Reference Tesli, Egeland, Sonderby, Haukvik, Bettella, Hibar, Thompson, Rimol, Melle, Agartz, Djurovic and Andreassen2013a ).
NRGN
NRGN codes for a post-synaptic protein kinase substrate and had been associated both with SZ (Stefansson et al. Reference Stefansson, Ophoff, Steinberg, Andreassen, Cichon, Rujescu, Werge, Pietilainen, Mors, Mortensen, Sigurdsson, Gustafsson, Nyegaard, Tuulio-Henriksson, Ingason, Hansen, Suvisaari, Lonnqvist, Paunio, Borglum, Hartmann, Fink-Jensen, Nordentoft, Hougaard, Norgaard-Pedersen, Bottcher, Olesen, Breuer, Moller, Giegling, Rasmussen, Timm, Mattheisen, Bitter, Rethelyi, Magnusdottir, Sigmundsson, Olason, Masson, Gulcher, Haraldsson, Fossdal, Thorgeirsson, Thorsteinsdottir, Ruggeri, Tosato, Franke, Strengman, Kiemeney, Melle, Djurovic, Abramova, Kaleda, Sanjuan, de Frutos, Bramon, Vassos, Fraser, Ettinger, Picchioni, Walker, Toulopoulou, Need, Ge, Yoon, Shianna, Freimer, Cantor, Murray, Kong, Golimbet, Carracedo, Arango, Costas, Jonsson, Terenius, Agartz, Petursson, Nothen, Rietschel, Matthews, Muglia, Peltonen, St Clair, Goldstein, Stefansson and Collier2009) and BD (Williams et al. Reference Williams, Craddock, Russo, Hamshere, Moskvina, Dwyer, Smith, Green, Grozeva, Holmans, Owen and O'Donovan2011a ; Steinberg et al. Reference Steinberg, de Jong, Mattheisen, Costas, Demontis, Jamain, Pietilainen, Lin, Papiol, Huttenlocher, Sigurdsson, Vassos, Giegling, Breuer, Fraser, Walker, Melle, Djurovic, Agartz, Tuulio-Henriksson, Suvisaari, Lonnqvist, Paunio, Olsen, Hansen, Ingason, Pirinen, Strengman, Hougaard, Orntoft, Didriksen, Hollegaard, Nordentoft, Abramova, Kaleda, Arrojo, Sanjuan, Arango, Etain, Bellivier, Meary, Schurhoff, Szoke, Ribolsi, Magni, Siracusano, Sperling, Rossner, Christiansen, Kiemeney, Franke, van den Berg, Veldink, Curran, Bolton, Poot, Staal, Rehnstrom, Kilpinen, Freitag, Meyer, Magnusson, Saemundsen, Martsenkovsky, Bikshaieva, Martsenkovska, Vashchenko, Raleva, Paketchieva, Stefanovski, Durmishi, Pejovic Milovancevic, Lecic Tosevski, Silagadze, Naneishvili, Mikeladze, Surguladze, Vincent, Farmer, Mitchell, Wright, Schofield, Fullerton, Montgomery, Martin, Rubino, van Winkel, Kenis, De Hert, Rethelyi, Bitter, Terenius, Jonsson, Bakker, van Os, Jablensky, Leboyer, Bramon, Powell, Murray, Corvin, Gill, Morris, O'Neill, Kendler, Riley, Craddock, Owen, O'Donovan, Thorsteinsdottir, Kong, Ehrenreich, Carracedo, Golimbet, Andreassen, Borglum, Mors, Mortensen, Werge, Ophoff, Nothen, Rietschel, Cichon, Ruggeri, Tosato, Palotie, St Clair, Rujescu, Collier, Stefansson and Stefansson2014). For episodic memory and during memory encoding, the NRGN rs12807809 risk variant has shown an association with increased (putatively more inefficient, given performance was controlled for) activation in the left lingual gyrus and during memory retrieval, decreased deactivation in the left precentral gyrus (PCG), cingulate gyrus (CG) and left insula (Krug et al. Reference Krug, Krach, Jansen, Nieratschker, Witt, Shah, Nothen, Rietschel and Kircher2013), but not the hippocampus as expected. In contextual fear processing, however, the left hippocampus was found to be deactivated during late acquisition in healthy risk allele homozygotes (Pohlack et al. Reference Pohlack, Nees, Ruttorf, Witt, Nieratschker, Rietschel and Flor2011). In spatial memory, decreased load-independent activation in the left SFG has been associated with the risk allele in health (Rose et al. Reference Rose, Morris, Fahey, Robertson, Greene, O'Doherty, Newell, Garavan, McGrath, Bokde, Tropea, Gill, Corvin and Donohoe2012).
In terms of brain anatomy, one study has reported an association with decreased GM, but not WM, volume in the left anterior cingulate cortex in SZ patients (Ohi et al. Reference Ohi, Hashimoto, Yasuda, Nemoto, Ohnishi, Fukumoto, Yamamori, Umeda-Yano, Okada, Iwase, Kazui and Takeda2012), but all associations with regional GM or WM volumes in healthy subjects were negative (Pohlack et al. Reference Pohlack, Nees, Ruttorf, Witt, Nieratschker, Rietschel and Flor2011; Rose et al. Reference Rose, Morris, Fahey, Robertson, Greene, O'Doherty, Newell, Garavan, McGrath, Bokde, Tropea, Gill, Corvin and Donohoe2012; Wirgenes et al. Reference Wirgenes, Sonderby, Haukvik, Mattingsdal, Tesli, Athanasiu, Sundet, Rossberg, Dale, Brown, Agartz, Melle, Djurovic and Andreassen2012).
DGKH
DGKH code for diacylglycerol kinase, a key protein in the lithium-sensitive phosphatidyl inositol pathway and may be important in the activity of protein kinase C (PKC), which is involved in the phosphatidyl inositol and Wingless (Wnt) signalLing pathways (Berridge, Reference Berridge1989). The DGKH rs1012053 risk variant associated with BD (Baum et al. Reference Baum, Akula, Cabanero, Cardona, Corona, Klemens, Schulze, Cichon, Rietschel, Nothen, Georgi, Schumacher, Schwarz, Abou Jamra, Hofels, Propping, Satagopan, Detera-Wadleigh, Hardy and McMahon2008), along with two other risk variants of the DGKH gene (of rs9315885 and rs1170191), have been plausibly associated with increased (putatively more inefficient, given performance was controlled for) verbal fluency-related activation in the PFC, precuneus and parahippocampus in individuals at high familial risk of BD, but not in healthy participants; where the effect was, surprisingly, significantly reversed (Whalley et al. Reference Whalley, Papmeyer, Romaniuk, Johnstone, Hall, Lawrie, Sussmann and McIntosh2012).
NCAN
NCAN encodes neurocan, an extracellular matrix glycoprotein, with putative roles in cell adhesion and migration and had been associated with BD (Cichon et al. Reference Cichon, Muhleisen, Degenhardt, Mattheisen, Miro, Strohmaier, Steffens, Meesters, Herms, Weingarten, Priebe, Haenisch, Alexander, Vollmer, Breuer, Schmal, Tessmann, Moebus, Wichmann, Schreiber, Muller-Myhsok, Lucae, Jamain, Leboyer, Bellivier, Etain, Henry, Kahn, Heath, Hamshere, O'Donovan, Owen, Craddock, Schwarz, Vedder, Kammerer-Ciernioch, Reif, Sasse, Bauer, Hautzinger, Wright, Mitchell, Schofield, Montgomery, Medland, Gordon, Martin, Gustafsson, Andreassen, Djurovic, Sigurdsson, Steinberg, Stefansson, Stefansson, Kapur-Pojskic, Oruc, Rivas, Mayoral, Chuchalin, Babadjanova, Tiganov, Pantelejeva, Abramova, Grigoroiu-Serbanescu, Diaconu, Czerski, Hauser, Zimmer, Lathrop, Schulze, Wienker, Schumacher, Maier, Propping, Rietschel and Nothen2011). Subsequently, the NCAN rs1064395 risk variant has been associated with increased cortical folding (a putative marker of increased neuronal efficiency) in the lateral occipital cortex and the PFC in SZ patients, but not in healthy subjects (Schultz et al. Reference Schultz, Muhleisen, Nenadic, Koch, Wagner, Schachtzabel, Siedek, Nothen, Rietschel, Deufel, Kiehntopf, Cichon, Reichenbach, Sauer and Schlosser2014). The same study reported no association with cortical thickness in patients, nor controls.
TCF4
Last, TCF4 encodes a transcription factor (Liu et al. Reference Liu, Ray, Yang, Luntz-Leybman and Chiu1998) with a role in neurodevelopment and had been associated with SZ (Stefansson et al. Reference Stefansson, Ophoff, Steinberg, Andreassen, Cichon, Rujescu, Werge, Pietilainen, Mors, Mortensen, Sigurdsson, Gustafsson, Nyegaard, Tuulio-Henriksson, Ingason, Hansen, Suvisaari, Lonnqvist, Paunio, Borglum, Hartmann, Fink-Jensen, Nordentoft, Hougaard, Norgaard-Pedersen, Bottcher, Olesen, Breuer, Moller, Giegling, Rasmussen, Timm, Mattheisen, Bitter, Rethelyi, Magnusdottir, Sigmundsson, Olason, Masson, Gulcher, Haraldsson, Fossdal, Thorgeirsson, Thorsteinsdottir, Ruggeri, Tosato, Franke, Strengman, Kiemeney, Melle, Djurovic, Abramova, Kaleda, Sanjuan, de Frutos, Bramon, Vassos, Fraser, Ettinger, Picchioni, Walker, Toulopoulou, Need, Ge, Yoon, Shianna, Freimer, Cantor, Murray, Kong, Golimbet, Carracedo, Arango, Costas, Jonsson, Terenius, Agartz, Petursson, Nothen, Rietschel, Matthews, Muglia, Peltonen, St Clair, Goldstein, Stefansson and Collier2009; Steinberg et al. Reference Steinberg, de Jong, Irish Schizophrenia Genomics, Andreassen, Werge, Borglum, Mors, Mortensen, Gustafsson, Costas, Pietilainen, Demontis, Papiol, Huttenlocher, Mattheisen, Breuer, Vassos, Giegling, Fraser, Walker, Tuulio-Henriksson, Suvisaari, Lonnqvist, Paunio, Agartz, Melle, Djurovic, Strengman, Jurgens, Glenthoj, Terenius, Hougaard, Orntoft, Wiuf, Didriksen, Hollegaard, Nordentoft, van Winkel, Kenis, Abramova, Kaleda, Arrojo, Sanjuan, Arango, Sperling, Rossner, Ribolsi, Magni, Siracusano, Christiansen, Kiemeney, Veldink, van den Berg, Ingason, Muglia, Murray, Nothen, Sigurdsson, Petursson, Thorsteinsdottir, Kong, Rubino, De Hert, Rethelyi, Bitter, Jonsson, Golimbet, Carracedo, Ehrenreich, Craddock, Owen, O'Donovan, Ruggeri, Tosato, Peltonen, Ophoff, Collier, St Clair, Rietschel, Cichon, Stefansson, Rujescu and Stefansson2011). Decreased expression in blood had been associated with the psychotic state (Kurian et al. Reference Kurian, Le-Niculescu, Patel, Bertram, Davis, Dike, Yehyawi, Lysaker, Dustin, Caligiuri, Lohr, Lahiri, Nurnberger, Faraone, Geyer, Tsuang, Schork, Salomon and Niculescu2011), whereas up-regulation had been found in the cerebellar cortex of SZ patients (Mudge et al. Reference Mudge, Miller, Khrebtukova, Lindquist, May, Huntley, Luo, Zhang, van Velkinburgh, Farmer, Lewis, Beavis, Schilkey, Virk, Black, Myers, Mader, Langley, Utsey, Kim, Roberts, Khalsa, Garcia, Ambriz-Griffith, Harlan, Czika, Martin, Wolfinger, Perrone-Bizzozero, Schroth and Kingsmore2008). The presence of the TCF4 rs9960767 risk variant has since shown a trend for a correlation with increased hippocampal volume and reduced ventricular volume in BD and SZ patients and healthy subjects (Wirgenes et al. Reference Wirgenes, Sonderby, Haukvik, Mattingsdal, Tesli, Athanasiu, Sundet, Rossberg, Dale, Brown, Agartz, Melle, Djurovic and Andreassen2012), which is against plausible expectation given that these phenotypes are inversely associated with these illnesses (Nelson et al. Reference Nelson, Saykin, Flashman and Riordan1998; Wright et al. Reference Wright, Rabe-Hesketh, Woodruff, David, Murray and Bullmore2000). In the same study, additional 59 SNPs in TCF4 were also tested for their possible association with brain volume measures: the rs12966547 G and rs4309482 A variants showed a trend for increased ventricular volumes, which is, this time, in line with expectations (Wright et al. Reference Wright, Rabe-Hesketh, Woodruff, David, Murray and Bullmore2000).
Quality assessment
There were 32 studies scoring ‘high’, seven scoring ‘medium’ and one scoring ‘low’ (Table 2, online Supplementary Table S1), according to our criteria. Total and item-by-item scoring of each study is available in online Supplementary Table S2. Overview of compliance of studies with each criteria is added in Table 1.
A summary of all imaging genetics studies investigating the impact of GWA-supported psychosis risk variants on brain structure and function. (Further details such as sample ethnicity, the statistical software, statistical tests, statistical significance threshold, covariates, performance control and regions of interest used in each study are included in Supplementary Table 1.)
ACC; Anterior cingulate cortex; ALIC, anterior limb of the internal capsule; BD, bipolar disorder; CG, cingulate gyrus; CSF, cerebrospinal fluid; DLPFC, dorsolateral prefrontal cortex; FA, fractional anisotropy; GM, grey matter; HF, hippocampal formation; HR, high risk; IFG, inferior frontal gyrus; IPL, inferior parietal lobule; L, left; MD, mean diffusivity; MFG, medial frontal gyrus; MGP, middle globus pallidus; MOG, medial occipital gyrus; MTG, medial temporal gyrus; NA, not applicable; NS, not significant; PCC, posterior cingulate cortex; NOS, not otherwise specified; NR, not reported; PCG, precentral gyrus; PFC, prefrontal cortex; R, right; RD, radial diffusivity; SFG, superior frontal gyrus; STG, superior temporal gyrus; SZ, schizophrenia; TC, temporal cortex; TPC, tempo-parietal cortex; TPJ, tempo-parietal junction; US, unaffected siblings; WM, white matter.
Discussion
Overview
The majority of the published studies have demonstrated a positive association of psychosis risk variants with neuroimaging phenotypes of SZ and BD. Such is consistent with the finding that the cognitive deficits under study, and indeed their functional and structural neurocorrelates, have a heritable component and are associated with these illnesses. Nevertheless, one cannot exclude the possibility that the effect of the genetic risk variant on these intermediate phenotypes is independent (pleiotropic) of its effect on the clinical phenotype risk (Gottesman & Gould, Reference Gottesman and Gould2003). In addition, reasonable uncertainty lies in what molecular and cellular-level role these genes and risk SNPs hold, in e.g. regulation in gene expression, mRNA processing and translation, etc. Also to keep in mind, as in any genetic association study, the identified GWA risk variants are not necessarily the real ‘culprit’ variants, and may be being passed on in high linkage disequilibrium with causative ones. One way of addressing this may be conducting hypothesis-driven GWAS where additional markers are applied to detect the SNPs that are associated with the disease at the highest significance level.
In the reviewed studies, the genotype's impact on brain activation rarely surfaced at the behavioral level, even in the same subjects. This putatively reflects the greater penetrance of genetic variation at a neurobiological level, confirming the usefulness of neuroimaging intermediate phenotypes. Also useful was the inclusion of patient populations. Although the use of healthy candidates avoids the potential contamination of illness-related factors such as symptoms, co-morbidities and medication, a patient population is key to providing a bigger picture, specially because the effect of the risk allele in healthy subjects cannot be absolutely generalized to patients: a detrimental genotype or brain phenotype in healthy controls may not be so in patients (or vice versa), where the context of other risk genetic and environmental effects may augment or limit it.
Inconsistencies between findings were mainly observed in brain structure abnormalities, more specifically for ZNF804A, possibly due to it simply being the most studied gene, and to false positives coming from insufficient correction for multiple comparisons. Ultimately, confidence in statistically significant association findings relies on their replication. However, as most of these studies are the first of their kind, replications are still infrequent.
Methodology
Heterogeneity in the neuroimaging methods, especially in the statistical analysis, may explain some of the heterogeneity in results. Only half of the studies reviewed chose significance thresholds corrected for multiple testing, most of them applied at voxel-level and some at cluster-level. This lack of a gold-standard approach impairs comparison between findings. Indeed, a recent review of 241 fMRI studies showed that 223 unique analysis strategies were used so that almost no single strategy occurred more than once (Carp, Reference Carp2012) – this can make study results vary widely (Ioannidis, Reference Ioannidis2005). A lack of a complete set of reported results (i.e. T, F or Z scores, effect sizes, and exact p values per area) was also unfortunately common. This represents wasted effort and a serious disadvantage to future research. In particular, effect sizes (Friston, Reference Friston2012) while able to provide an invaluable measure of the magnitude of the genetic effect (i.e. its penetrance) and generally disclosed in genetic association studies of complex phenotypes, were reported in little more than 1/3 of the studies herein reviewed.
Age, gender, handedness, drug use including alcohol consumption, smoking and medication, IQ, or years of education may affect brain structure and function or engagement with the task. Consideration of these factors is important. If they have an expected effect on the neuroimaging phenotype, (1) they may deem genotype effects under-detected, by increasing the error variance in the phenotype, and (2) if they are also correlated with the genotype (or diagnosis), they may confound its effects. Restricting or matching of groups or introducing such variables in the model could help compensate for the former, albeit not the latter (Miller & Chapman, Reference Miller and Chapman2001). Rather, to exclude the latter scenario, the individual association of those potential confounders with the imaging phenotype in the reported areas can be tested. Fortunately, genotype is guarded against any association with age and (unless in sexual chromosomes) gender; even though these may be (even if not confounders) mediating or interacting variables. An effort to exclude closely genetically related participants goes also much unreported. However, genetic similarities between related subjects confound correlations between genotype of choice and phenotype, unless a genetic cluster analysis is implemented. Performance bias (difference between subjects in activation dependent on ability to perform the task or degree of attention) may also be a crucial confounder in functional studies. This can be compensated by matching subjects for level of performance, covarying for performance, or restricting analysis to images of correct responses. This was well taken into account in all studies. One power consideration is the common attempt to increase group size by merging the smallest homozygote group with the heterozygotes. As the mode of inheritance of these risk variants is still unknown, this strategy may lead to false negative effects (e.g. if allele is recessive or co-dominant, its weight is diluted when its homozygotes are merged with the heterozygotes). Also in terms of sample size, studies had at least 10 subjects per group but the threshold of 20, calculated to be necessary for minimum reliability in neuroimaging (Thirion et al. Reference Thirion, Pinel, Meriaux, Roche, Dehaene and Poline2007), was reached in less that one-tenth of the studies. This problematic as low-powered studies, coupled with publication bias towards significant results, are predominant in neuroscience and decrease the likelihood that true effects are found but also that those found are true–strategies to prevent this have been suggested elsewhere (Button et al. Reference Button, Ioannidis, Mokrysz, Nosek, Flint, Robinson and Munafo2013; Cumming, Reference Cumming2014).
Further approaches
Given that a combination of polymorphisms is probably behind the genetic etiology of SZ and BD (Gottesman et al. Reference Gottesman, Laursen, Bertelsen and Mortensen2010), one could assess the amount of composite SNP variation contributing to neuroimaging intermediate phenotypes. This has been calculated for SZ and BD risk using genome-wide complex trait analysis (GCTA) (Yang et al. Reference Yang, Lee, Goddard and Visscher2011). Such knowledge of how much heritability of brain phenotypes derives from common SNP data, would add conceptual strength to SNP associations. Moreover, the presence of specific risk SNP alleles can be summed into a composite polygenic risk score for each individual, weighted by the respective ORs (after being ranked according to p values, in an independent GWA analysis) (Dudbridge, Reference Dudbridge2013). This GWA-based score has being correlated with disease risk (Dudbridge, Reference Dudbridge2013) and, as above, could also be correlated with neuroimaging measures to allow us to assess the impact of a set of risk alleles for SZ or BD on brain phenotypes – a powerful leap from the usual single SNP approach (McIntosh et al. Reference McIntosh, Gow, Luciano, Davies, Liewald, Harris, Corley, Hall, Starr, Porteous, Tenesa, Visscher and Deary2013), that could increase effect sizes, even if not clarifying each SNP's individual role.
A more complete insight into these genes mode of action, via epistasis, environment and epigenetics research would help explain some of the discrepancies in the literature, but we found no such studies with the GWA risk variations herein discussed. The effect of gene-gene (epistasis) and gene–environment interactions in the brain are, nevertheless, starting to be assessed in relation to psychosis risk (Prata et al. Reference Prata, Mechelli, Fu, Picchioni, Toulopoulou, Bramon, Walshe, Murray, Collier and McGuire2009; Haukvik et al. Reference Haukvik, Saetre, McNeil, Bjerkan, Andreassen, Werge, Jonsson and Agartz2010; Nicodemus et al. Reference Nicodemus, Callicott, Higier, Luna, Nixon, Lipska, Vakkalanka, Giegling, Rujescu, St Clair, Muglia, Shugart and Weinberger2010a , Reference Nicodemus, Law, Radulescu, Luna, Kolachana, Vakkalanka, Rujescu, Giegling, Straub, McGee, Gold, Dean, Muglia, Callicott, Tan and Weinberger b ). Neuronal gene expression research however is difficult to perform due to the invasiveness in obtaining cellular materials – yet, we might be able to circumvent this soon by growing neurons in vitro using pluripotent stem cell technology (Brennand & Gage, Reference Brennand and Gage2011).
The association of risk variants with novel brain regions for SZ/BD, e.g. the reported effect of CANA1C risk SNPs on brainstem volume, reflects the need to investigate previously less-considered regions (Franke et al. Reference Franke, Vasquez, Veltman, Brunner, Rijpkema and Fernandez2010). The brainstem is a central structure for vital physiological processes and may reveal a novel mechanism implicated in SZ/BD. In fact, functional studies should broaden their scope by integrating tasks and modalities. Electrophysiological neurocorrelates of sensory motor gating and eye-tracking are also robustly found to be altered in SZ and BD (Calkins & Iacono, Reference Calkins and Iacono2000; Braff et al. Reference Braff, Geyer and Swerdlow2001), but were not used in neither study herein reviewed. Future studies on the impact of the GWA risk genes on these phenotypes accessible with EEG, would help expand on the genetic dissection of these illnesses.
Conclusion
Investigating the effects of GWAS supported risk variants on brain structure and function is a promising strategy for understanding the neuronal basis of psychiatric illness susceptibility. This is especially timely as GWAS studies of SZ and BD, despite the initial excitement, have had limited success in elucidating the underlying biological mechanisms. This review gathers evidence that most of GWAS risk variants seem to affect phenotypes previously implicated in SZ and BD such as WM integrity (ANK3 and ZNF804A), volume (CACNA1C and ZNF804A) and density (ZNF804A), GM volume (CACNA1C, NRGN, TCF4 and ZNF804A), ventricular volume (TCF4), cortical folding (NCAN) and thickness (ZNF804A), regional activation during executive tasks (ANK3, CACNA1C, DGKH, NRGN and ZNF804A) and functional connectivity during executive tasks (CACNA1C and ZNF804A), facial affect recognition (CACNA1C and ZNF804A) and theory of mind (ZNF804A). However, further efforts should be made to test the reliability of these early findings in replication studies, with greater emphasis placed on the need for larger samples and clinical samples, and gold-standard design and reporting. Also timely would be the use of other imaging modalities (preferably in the same sample), research on interactions of genetic with environmental/epigenetic factors on brain structure and function, and development of statistical tools that allow for the integration of effects of different genetic risk variations.
Supplementary material
For supplementary material accompanying this paper visit http://dx.doi.org/10.1017/S0033291715000537.
Acknowledgements
This research was supported by a UK National Institute for Health Research fellowship (NIHR-PDF-2010-03-047) to D.P.
Declaration of Interest
None.
