Maternal infection and immune activation

The evidence base has been strengthened by complementary approaches converging on the risk factor of interest and shedding light on biological mechanisms. For example, maternal infection at times of epidemic activity is implicated by several ecological studies, but which did not include confirmation of infection in individual cases (Brown & Derkits, Reference Brown and Derkits2010). This association has been supported by rigorous case ascertainment in prospective birth cohort studies, which demonstrated exposure to a range of pathogens as a risk factor, including influenza, toxoplasmosis gondii, herpes simplex virus, rubella, and maternal respiratory and genital infections (Brown & Derkits, Reference Brown and Derkits2010). The time of exposure is important, as serological evidence of influenza infection is associated with increased incidence in the first but not latter trimesters (Brown, Begg, et al., Reference Brown, Begg, Gravenstein, Schaefer, Wyatt, Bresnahan and Susser2004), consistent with the concept of critical windows for development. A large Finnish case-register study has suggested an interaction between familial liability and maternal exposure to infection to increase risk of SSD (Clarke, Tanskanen, Huttunen, Whittaker, & Cannon, Reference Clarke, Tanskanen, Huttunen, Whittaker and Cannon2009).

Potential mechanisms for pathogenicity include direct fetal invasion and maternal immune activation, the latter due to the inability of certain viruses to cross the placenta (Patterson, Reference Patterson2009). Altered cytokine expression has been suggested as a mediator of disease (Gilmore & Jarskog, Reference Gilmore and Jarskog1997), with evidence from case-control designs showing increased IL-8 and C-reactive protein, decreased IL-4, IL-5, and IL-13 (Allswede, Buka, Yolken, Torrey, & Cannon, Reference Allswede, Buka, Yolken, Torrey and Cannon2016; Brown, Hooton, et al., Reference Brown, Hooton, Schaefer, Zhang, Petkova, Babulas and Susser2004). The inflammatory phenotype is also associated with structural brain abnormalities such as ventriculomegaly, decreased gray matter volume, and aberrant dendritic arborisation in preclinical models and humans (Ellman et al., Reference Ellman, Deicken, Vinogradov, Kremen, Poole, Kern and Brown2010; Short et al., Reference Short, Lubach, Karasin, Olsen, Styner, Knickmeyer and Coe2010; Weir et al., Reference Weir, Forghany, Smith, Patterson, McAllister, Schumann and Bauman2015).

Maternal nutritional deprivation

Maternal nutritional deprivation has ecological evidence for increasing risk: an approximately two-fold increase in incidence was associated with the famines of the Dutch Hunger Winter (1944–1945) and the Wuhu region in China (1960–1961; Brown & Susser, Reference Brown and Susser2008). These ecological studies did not index maternal starvation in individual cases, thus making it difficult to direct focus toward specific micronutrients. Subsequent studies have suggested maternal folate, vitamin D, iron, and protein deficiency as candidates for micronutrient deficiencies associated with increased risk (Brown & Susser, Reference Brown and Susser2008). In the Dutch Hunger study, lifelong epigenetic changes were demonstrated compared to healthy siblings (Heijmans et al., Reference Heijmans, Tobi, Stein, Putter, Blauw, Susser and Lumey2008), suggesting that methylation status related to folate deficiency may be implicated in SSD risk (Kirkbride et al., Reference Kirkbride, Susser, Kundakovic, Kresovich, Davey Smith and Relton2012). Maternal starvation was also associated with reduced brain mass and DNA content in preclinical models (Gu, Shambaugh, Metzger, Unterman, & Radosevich, Reference Gu, Shambaugh, Metzger, Unterman and Radosevich1992; Hemb, Cammarota, & Nunes, Reference Hemb, Cammarota and Nunes2010).

Fetal hypoxia

Fetal hypoxia is postulated as a mechanism for neuronal damage. Exposure to fetal hypoxia has been associated with an increased incidence of SSD, and the hypoxic state may manifest as a low Apgar score (Dalman, Reference Dalman2001), placental insufficiency, fetal distress, or cord entrapment (Rosso et al., Reference Rosso, Cannon, Huttunen, Huttunen, Lönnqvist and Gasperoni2000; Zornberg, Buka, & Tsuang, Reference Zornberg, Buka and Tsuang2000). Fetal hypoxia appears to potentiate SSD risk in the presence of genetic risk, which manifests as alterations in brain modeling. There is evidence for a Gene × Environment interaction as pre- and perinatal hypoxia predicts decreased gray matter and temporal lobe volumes in both adult patients and siblings of patients (Cannon et al., Reference Cannon, van Erp, Rosso, Huttunen, Lönnqvist, Pirkola and Standertskjöld-Nordenstam2002). In individuals with an established history of fetal exposure to hypoxia, genetic load for SSD has been shown to predict a smaller hippocampal volume in a dose-dependent fashion (Van Erp et al., Reference Van Erp, Saleh, Rosso, Huttunen, Lönnqvist, Pirkola and Cannon2002). Consistent with this, 42 SSD risk genes have been implicated in ischaemic/hypoxic and vascular regulation (Schmidt-Kastner, Van Os, Esquivel, Steinbusch, & Rutten, Reference Schmidt-Kastner, Van Os, Esquivel, Steinbusch and Rutten2012). Fetal hypoxia is associated with reductions in neurotrophic factors in patients (Cannon, Yolken, Buka, & Torrey, Reference Cannon, Yolken, Buka and Torrey2008), supported by preclinical models of birth asphyxia that show alterations in dendritic morphology, increased cell death, and damage secondary to oxidative stress and a predilection for impaired hippocampal function (Marriott, Rojas-Mancilla, Morales, Herrera-Marschitz, & Tasker, Reference Marriott, Rojas-Mancilla, Morales, Herrera-Marschitz and Tasker2017).

Maternal stress

Exposure to maternal psychological stress has been suggested as a risk factor for later life mental health disorder and SSD in particular (den Bergh et al., Reference den Bergh, van den Heuvel, Lahti, Braeken, de Rooij, Entringer and Schwab2017; Khashan et al., Reference Khashan, Abel, McNamee, Pedersen, Webb, Baker and Mortensen2008). The maternal stress hypothesis was suggested by Huttenen and Niskanen (Reference Huttunen and Niskanen1978), who observed an excess of deaths in offspring whose fathers’ died before birth compared to those who died in the child's first year of life. Ecological studies demonstrated increased rates of SSD in populations exposed to brief war during the first trimester (Malaspina et al., Reference Malaspina, Corcoran, Kleinhaus, Perrin, Fennig, Nahon and Harlap2008; Van Os & Selten, Reference Van Os and Selten1998). Population-based case registry studies have variably been found in support (Khashan et al., Reference Khashan, Abel, McNamee, Pedersen, Webb, Baker and Mortensen2008) and against (Abel et al., Reference Abel, Heuvelman, Jörgensen, Magnusson, Wicks, Susser and Dalman2014; Class et al., Reference Class, Abel, Khashan, Rickert, Dalman, Larsson and D'Onofrio2014) maternal stressors as a risk factor. More recently, a case control has demonstrated maternal daily life stress to be associated with male, but not female, risk of developing later life SSD (Fineberg et al., Reference Fineberg, Ellman, Schaefer, Maxwell, Shen, Chaudhury and Brown2016). Mechanisms of maternal stress transfer remain unclear, although hypercortisolaemia is a candidate pathway, particularly due to its ability to readily cross the placenta. However, this is unlikely to fully account for early gestational stress, as glucocorticoid receptors in the fetus are not yet developed (Rakers et al., Reference Rakers, Rupprecht, Dreiling, Bergmeier, Witte and Schwab2016). Additional mechanisms implicated are increased cytokine transfer across the placenta, catecholamine modulation of uterine perfusion, epigenetic mechanisms, and microbiota transfer through the birth canal (den Bergh et al., Reference den Bergh, van den Heuvel, Lahti, Braeken, de Rooij, Entringer and Schwab2017; Rakers et al., Reference Rakers, Rupprecht, Dreiling, Bergmeier, Witte and Schwab2016).

There is thus converging evidence from ecological, birth cohort, and preclinical models suggesting a considerable basis for the fetal origins of SSD. MRI offers a promising vista into envisioning mechanistic macroscopic perturbations in neural development.

The genetics of schizophrenia

Recent insights from a major genome-wide association study (GWAS) have implicated dopaminergic, glutamatergic pathways and likely major histocompatibility complex-mediated immunologic mechanisms (Ripke et al., Reference Ripke, Neale, Corvin, Walters, Farh, Holmans and O'Donovan2014). Subsequent work highlighted the role of complement-4 protein in synaptic pruning (Sekar et al., Reference Sekar, Bialas, De Rivera, Davis, Hammond, Kamitaki and McCarroll2016).

In addition, SSD genes are expressed from the earliest periods in neurodevelopment, and it has been suggested that the genetics of schizophrenia is at least in part the genetics of brain development (Kahn et al., Reference Kahn, Sommer, Murray, Meyer-Lindenberg, Weinberger, Cannon and Insel2015). A meta-analysis of candidate gene studies demonstrated genetic variants with strongest genetic risk implicated in pathways in modulation of dopamine, glutamate, serotonin, and neuronal development and function (Gatt, Burton, Williams, & Schofield, Reference Gatt, Burton, Williams and Schofield2015). Earlier studies of candidate genes were limited by publication biases toward positive results and liberal application of significance testing (Giegling et al., Reference Giegling, Hosak, Mössner, Serretti, Bellivier, Claes and Rujescu2017). More sophisticated analysis has been undertaken in the GWAS era. Such agnostic methods have allowed identification of a number of genes including dopaminergic and glutamatergic function, neuronal calcium signalling, synaptic function, and neurodevelopment (Ripke et al., Reference Ripke, Neale, Corvin, Walters, Farh, Holmans and O'Donovan2014). Although initial work looked at common single nucleotide polymorphisms (SNPs), more recently additional work has looked at the contribution of rarer copy number variants implicating pathways with genes involved in synaptic function and scaffolding, abnormal mouse behavioral phenotypes, and overlap with pediatric developmental disorders (Marshall et al., Reference Marshall, Howrigan, Merico, Thiruvahindrapuram, Wu, Greer and Sebat2017). SNPs are common but show small effect sizes (odds ratio < 1.3), whereas copy number variants, although rare, show larger effect sizes (odd ratio = 4–70), and hence utilization of both enhances our understanding of the complex genetics of SSD risk (Birnbaum & Weinberger, Reference Birnbaum and Weinberger2017).

GWAS approaches can be leveraged to develop polygenic risk scores (PRS), which been shown to have real-world significance and predict caseness for schizophrenia in large-scale clinical data sets (Zheutlin et al., Reference Zheutlin, Dennis, Linnér, Moscati, Restrepo, Straub and Smoller2019). PRS for schizophrenia has also been shown to predict adverse neurocognitive outcomes and higher childhood internalizing and externalizing behaviors (Jansen et al., Reference Jansen, Polderman, Bolhuis, van der Ende, Jaddoe, Verhulst and Tiemeier2018; Shafee et al., Reference Shafee, Nanda, Padmanabhan, Tandon, Alliey-Rodriguez, Kalapurakkel and Robinson2018; Toulopoulou et al., Reference Toulopoulou, Zhang, Cherny, Dickinson, Berman, Straub and Weinberger2019). There is also evidence that some of the genes involved in SSD risk may be associated with early life fetal adversity. Using pathway analysis of genes derived from PRS for schizophrenia, Ursini et al. (Reference Ursini, Punzi, Chen, Marenco, Robinson, Porcelli and Weinberger2018) demonstrated that SNPs regulating oxidative and cellular stress were highly expressed in cases of abnormal placental pathology (intrauterine growth restriction and preeclampsia) indicating a crucial role for the placenta in developmental sensitivity for SSD (Ursini et al., Reference Ursini, Punzi, Chen, Marenco, Robinson, Porcelli and Weinberger2018).

Further evidence for the neurodevelopmental nature of SSD is that genetic SSD risk confers pleiotropic risk for several other childhood and adult-onset psychiatric disorders, including attention-deficit/hyperactivity disorder, autism, bipolar affective disorder, depression, and alcohol dependence (Cross-Disorder Group of the Psychiatric Genomics Consortium, 2013; Forstner et al., Reference Forstner, Hecker, Hofmann, Maaser, Reinbold, Mühleisen and Nöthen2017; Gandal et al., Reference Gandal, Haney, Parikshak, Leppa, Ramaswami, Hartl and Geschwind2018; Zheutlin et al., Reference Zheutlin, Dennis, Linnér, Moscati, Restrepo, Straub and Smoller2019)).

One-hit, two-hit, and multi-hit hypotheses

SSD has been conceptualized as the endpoint of a trajectory of incremental neurodevelopmental insults (Bayer, Falkai, & Maier, Reference Bayer, Falkai and Maier1999; Waddington, Lane, Larkin, & O'Callaghan, Reference Waddington, Lane, Larkin and O'Callaghan1999). The insults involve a range of mechanisms including genetic, epigenetic, and environmental insults converging upon disorder (Birnbaum & Weinberger, Reference Birnbaum and Weinberger2017; Kahn et al., Reference Kahn, Sommer, Murray, Meyer-Lindenberg, Weinberger, Cannon and Insel2015). A wide range of risk factors implicated in pathogenesis have been described (Belbasis et al., Reference Belbasis, Köhler, Stefanis, Stubbs, van Os, Vieta and Evangelou2018; Radua et al., Reference Radua, Ramella-Cravaro, Ioannidis, Reichenberg, Phiphopthatsanee and Amir2018). Two periods of critical development have attracted the most attention: first, early neurodevelopment (i.e., the pre- and perinatal period); and second, exposures during adolescence and early adulthood, such as social stressors and substance misuse. Several lines of evidence indicate Gene × Environment interaction in additively increasing risk of SSD. For example, in a large case-control study of first-episode psychosis, the PRS for schizophrenia has been shown to interact with cannabis use and early childhood adversity and trauma (Guloksuz et al., Reference Guloksuz, Pries, Delespaul, Kenis, Luykx, Lin and van Os2019). Similarly, polygenic risk has been shown to interact with early life and obstetric complications to increase the risk of schizophrenia five-fold with results consistent across multiple samples (Ursini et al., Reference Ursini, Punzi, Chen, Marenco, Robinson, Porcelli and Weinberger2018). Thus, a wide range of risk factors, genetic and environmental, interact to converge on SSD risk (Van Os, Kenis, & Rutten, Reference Van Os, Kenis and Rutten2010; Van Os, Rutten, & Poulton, Reference Van Os, Rutten and Poulton2008). In the event of ultimate development of psychotic disorder, the biological pathways implicated have been hypothesized to converge on aberrant dopamine signaling in the ventral striatum (Howes & Kapur, Reference Howes and Kapur2009), which may be secondary to a primary glutamatergic or GABA-ergic abnormality. The fetal brain thus represents the primordium for the convergence of genetic and early environmental risk, presenting the earliest opportunity for definitive delineation of viable biomarkers.

Ventricular enlargement

Enlargement of the ventricular system and its association with SSD was demonstrated almost a century ago using pneumo-encephalographic imaging (Jacobi & Winkler, Reference Jacobi and Winkler1927), and followed up in a seminal study by Johnstone et al. utilizing computed tomography (Johnstone, Frith, Crow, Husband, & Kreel, Reference Johnstone, Frith, Crow, Husband and Kreel1976). Ventricular enlargement was observed to be present from disease onset and hence was hypothesized to predate the clinical onset of the disorder, and was also associated with a worse prognosis (for a review of early studies see, Weinberger, Wagner, & Wyatt, Reference Weinberger, Wagner and Wyatt1983). Several meta-analyses have confirmed that enlarged ventricles on MRI have effect sizes in the moderate range for chronic, first episode, and antipsychotic naive patients (Haijma et al., Reference Haijma, Van Haren, Cahn, Koolschijn, Hulshoff Pol and Kahn2013; Wright et al., Reference Wright, Rabe-Hesketh, Woodruff, David, Murray and Bullmore2000), suggesting that changes in volume predate disorder and are not simply a consequence of psychotropic medication. Quantitative analysis shows that total ventricular volumes are increased by approximately 26%, and although all components of the ventricular system appear enlarged, there appears to be a predilection for the lateral ventricles (approximately 47%) with relative sparing of the fourth ventricle (7%; Wright et al., Reference Wright, Rabe-Hesketh, Woodruff, David, Murray and Bullmore2000); these changes are less marked in patients in early stages of illness (Haijma et al., Reference Haijma, Van Haren, Cahn, Koolschijn, Hulshoff Pol and Kahn2013; Vita, De Peri, Silenzi, & Dieci, Reference Vita, De Peri, Silenzi and Dieci2006).

Structural changes in the cerebral parenchyma

Structural anomalies in antipsychotic naive patients include a reductions of global brain parenchymal volume by approximately 2%, total gray matter by 3.8%, white matter by 2.4%, and limbic system substructures (hippocampus 4.3%, caudate 5.9%, thalamus 9.6%; Haijma et al., Reference Haijma, Van Haren, Cahn, Koolschijn, Hulshoff Pol and Kahn2013). Hippocampal volumes are unchanged between first episode and chronic patients, suggesting that volume reduction predates disease onset (Adriano, Caltagirone, & Spalletta, Reference Adriano, Caltagirone and Spalletta2012).

In addition to volume changes, higher dimensional transformations occur in brain structures. Analysis of regional variations in surface representations of adult subcortical structures have shown significant differences in the medial thalamic nuclei of patients with SSD (Kang et al., Reference Kang, Kim, Kim, Choi, Jang, Jung and Kwon2008). Shape changes are also present in adult basal ganglia in patients with SSD and their unaffected siblings (Mamah et al., Reference Mamah, Harms, Wang, Barch, Thompson, Kim and Csernansky2008).

Widespread cortical thinning in temporal, frontal, and parietal regions is seen in first-episode patients, and drug-naive patients (Crespo-Facorro et al., Reference Crespo-Facorro, Roiz-Santiáñez, Párez-Iglesias, Rodriguez-Sanchez, Mata, Tordesillas-Gutierrez and Vázquez-Barquero2011; Narr et al., Reference Narr, Toga, Szeszko, Thompson, Woods, Robinson and Bilder2005; Schultz et al., Reference Schultz, Koch, Wagner, Roebel, Schachtzabel, Gaser and Schlösser2010) Cortical gray matter reductions have been identified in several regions in adolescents at genetic high risk for SSD, which include the mediotemporal, limbic system (hippocampus, parahippocampus, and anterior cingulate), temporal, parietal, and prefrontal regions (Brent, Thermenos, Keshavan, & Seidman, Reference Brent, Thermenos, Keshavan and Seidman2013). Longitudinal analysis pre- and postonset of psychosis revealed larger gray matter volume reductions associated with progression to disease onset, particularly affecting the prefrontal cortex and deep gray matter structures (Mcintosh et al., Reference Mcintosh, Owens, Moorhead, Whalley, Stanfield, Hall and Lawrie2011; Pantelis et al., Reference Pantelis, Velakoulis, McGorry, Wood, Suckling, Phillips and McGuire2003).

Altered connectivity—Structural

Perturbations in structural connectivity using diffusion tensor imaging and gyrification patterns are well established, although specific findings and regions implicated are heterogeneous across studies particularly in first-episode and high-risk states (Fitzsimmons, Kubicki, & Shenton, Reference Fitzsimmons, Kubicki and Shenton2013; White & Gottesman, Reference White and Gottesman2012). A lower global fractional anisotropy has been shown to be heritable and associated with schizophrenia liability (Bohlken et al., Reference Bohlken, Brouwer, Mandl, Van Den Heuvel, Hedman, De Hert and Hulshoff Pol2016). There is, however, also evidence that distinct subgroups may exhibit distinct patterns of alterations in white matter integrity: cannabis using patients have been shown to have distinct corpus callosal changes not seen in patients who do not use, which may indicate an environmental effect (Rigucci et al., Reference Rigucci, Marques, Di Forti, Taylor, Dell'Acqua, Mondelli and Dazzan2015).

Gray matter changes, including gyrification and diminishment of cortical thickness, can also be shown to demonstrate network properties (Palaniyappan, Das, Winmill, Hough, & James, Reference Palaniyappan, Das, Winmill, Hough and James2019; Palanyiyappan et al., Reference Palaniyappan, Marques, Taylor, Mondelli, Reinders, Bonaccorso and Dazzan2016; Palaniyappan, Park, Balain, Dangi, & Liddle, Reference Palaniyappan, Park, Balain, Dangi and Liddle2015), indicating coordinated pathogenic and adaptive responses to schizophrenia rather than isolated focal lesions. Recently progressive gray matter changes have been postulated to be indicative of a compensatory, maladaptive response in schizophrenia (Palaniyappan et al., Reference Palaniyappan, Marques, Taylor, Mondelli, Reinders, Bonaccorso and Dazzan2016).

Altered connectivity—Functional

Conceptions of SSD have moved from identification of anatomic lesions toward a more distributed model of schizophrenia as a disorder of dysconnectivity (Friston, Reference Friston1998; Friston, Brown, Siemerkus, & Stephan, Reference Friston, Brown, Siemerkus and Stephan2016; McGuire & Frith, Reference McGuire and Frith1996). A recent formulation of the dysconnectivity hypothesis (Friston et al., Reference Friston, Brown, Siemerkus and Stephan2016) posits that psychosis is best explained as a systems-level decrease in functional integration of specialized disparate regions of the brain. Functional MRI imaged via blood oxygen level dependent response demonstrates perturbations in executive function, working memory, verbal learning, attentional processing, and emotional salience. These findings implicate the prefrontal, parietal, and mediotemporal regions, including the hippocampus and amygdala (Gur & Gur, Reference Gur and Gur2010). Resting-state analysis provides evidence of a generalized connectivity deficit with reductions in local connectivity and modular structure (Fornito, Zalesky, Pantelis, & Bullmore, Reference Fornito, Zalesky, Pantelis and Bullmore2012; Sheffield & Barch, Reference Sheffield and Barch2016). Several resting-state studies have shown evidence for reduced functional connectivity key brain networks (O'Neill, Mechelli, & Bhattacharyya, Reference O'Neill, Mechelli and Bhattacharyya2018; Orliac et al., Reference Orliac, Delamillieure, Delcroix, Naveau, Brazo, Razafimandimby and Joliot2017; Pettersson-Yeo, Allen, Benetti, McGuire, & Mechelli, Reference Pettersson-Yeo, Allen, Benetti, McGuire and Mechelli2011; Zalesky, Fornito, & Bullmore, Reference Zalesky, Fornito and Bullmore2010); however, this is not universal. It has been observed that alterations in functional connectivity are mixed (Fornito et al., Reference Fornito, Harrison, Goodby, Dean, Ooi, Nathan and Bullmore2013; Fornito & Bullmore, Reference Fornito and Bullmore2015), where functional hypoconnectivity may indicate an underlying deficit, a pattern of hyperconnectivity may indicate a compensatory response that may be adaptive associated with improved prognosis or symptom scores (Andreou et al., Reference Andreou, Nolte, Leicht, Polomac, Hanganu-Opatz, Lambert and Mulert2015; Anticevic et al., Reference Anticevic, Hu, Xiao, Hu, Li, Bi and Gong2015) or maladaptive associated with worse psychopathology (Whitfield-Gabrieli et al., Reference Whitfield-Gabrieli, Thermenos, Milanovic, Tsuang, Faraone, McCarley and Seidman2009). Taken together, although there is good evidence for alterations on functional connectivity between patients and controls a robust resting-state functional MRI biomarker consistent between samples has yet to be identified. A further complication arises in that perturbations in functional and structural connectivity are not uniformly aligned (Fornito & Bullmore, Reference Fornito and Bullmore2015).

Altered metabolic and neurotransmitter profiles

A well-established neurochemical finding is altered dopaminergic signaling, with evidence for elevation in striatal presynaptic dopamine synthesis capacity across psychotic states and prior to disease onset (Howes et al., Reference Howes, Bose, Turkheimer, Valli, Egerton, Valmaggia and McGuire2011, Reference Howes, Kambeitz, Kim, Stahl, Slifstein, Abi-Dargham and Kapur2012; Jauhar et al., Reference Jauhar, Nour, Veronese, Rogdaki, Bonoldi, Azis and Howes2017) although this cannot directly be interrogated by MRI methods. Elevations in glutamatergic indices in basal ganglia and the thalamus, which appear to decline with treatment, have been demonstrated using magnetic resonance spectroscopy (MRS; Marsman et al., Reference Marsman, Van Den Heuvel, Klomp, Kahn, Luijten and Hulshoff Pol2013; Merritt, Egerton, Kempton, Taylor, & McGuire, Reference Merritt, Egerton, Kempton, Taylor and McGuire2016; Poels et al., Reference Poels, Kegeles, Kantrowitz, Slifstein, Javitt, Lieberman and Girgis2014), whereas recent meta-analyses have not found alterations in GABA-ergic indices (Egerton, Modinos, Ferrera, & McGuire, Reference Egerton, Modinos, Ferrera and McGuire2017; Schür et al., Reference Schür, Draisma, Wijnen, Boks, Koevoets, Joëls and Vinkers2016).

In sum, several robust alterations have been identified: ventricular enlargement, whole brain atrophy, reduced cortical thickness, and volume reduction in deep gray matter structures, particularly the hippocampus. Many abnormalities predate disease onset and are not fully explained by progression of disease or treatment effects. These abnormalities are subtle, and are differences at the group rather than the individual level, leading to difficulty in using MRI approaches as a diagnostic test, but may improve with optimized postprocessing techniques and machine learning methodology (McGuire et al., Reference McGuire, Sato, Mechelli, Jackowski, Bressan and Zugman2015).

How early do brain abnormalities manifest?

Structural analysis of the CNS in childhood SSD (onset before age 18) demonstrates a more profound reduction in gray matter volume and white matter integrity compared to adult-onset SSD (Douaud et al., Reference Douaud, MacKay, Andersson, James, Quested, Ray and James2009; Ordóñez, Sastry, & Gogtay, Reference Ordóñez, Sastry and Gogtay2015). For all patients, it remains to be determined how early these abnormalities manifest and become detectable with imaging techniques. It is currently challenging to fully elucidate this ontogeny without neuroimaging-based longitudinal cohorts with SSD as the outcome of interest.

There is sufficient evidence to suggest that viable biomarkers exist throughout all stages of childhood. Neurocognitive differences appear early, and it is well established that objective assessments of neuromotor development appear within the first 2 years of life (Jones, Murray, Jones, Rodgers, & Marmot, Reference Jones, Murray, Jones, Rodgers and Marmot1994; Walker, Savole, & Davis, Reference Walker, Savole and Davis1994). Neurocognitive indices of impairment become progressively identifiable in those who go onto develop psychotic disorders (Mollon, David, Zammit, Lewis, & Reichenberg, Reference Mollon, David, Zammit, Lewis and Reichenberg2018). SSD is associated with MPA across multiple studies in independent populations (Radua et al., Reference Radua, Ramella-Cravaro, Ioannidis, Reichenberg, Phiphopthatsanee and Amir2018; Xu, Chan, & Compton, Reference Xu, Chan and Compton2011); MPA are indicative of early in utero environmental or genetically driven insults due to their neuro-ectodermal origin.

One approach to identify biomarkers is to employ imaging techniques in high-risk cohorts. One such cohort has performed neonatal scans on the offspring of mothers with SSD. Several Gender × Group differences between the risk group and controls were identified: males had increased intracranial and ventricular volumes (Gilmore et al., Reference Gilmore, Schmitt, Knickmeyer, Smith, Lin, Styner and Neale2010), functional networks had decreased global efficiency and increased connection distance (Shi et al., Reference Shi, Yap, Gao, Lin, Gilmore and Shen2012), whereas females had decreased right lateral occipital cortical thickness (Li et al., Reference Li, Wang, Shi, Lyall, Ahn, Peng and Shen2016). Subsequent studies demonstrated that these abnormalities evolved over the first 2 years of life in this group, while associated with an impaired neurocognitive profile (Murphy et al., Reference Murphy, Short, Cornea, Goldman, Li, Shen and Gilmore2018). This was a small cohort of neonates at risk (n = 26), and hence only 10% of these children would be expected to develop SSD. Further work is needed to replicate and extend these findings in larger, representative samples.

Thus there are converging lines of evidence to suggest that considering the in utero condition is of particular importance in understanding SSD, particularly in at-risk individuals, and is complementary to investigations in postnatal and early childhood neurodevelopment, which has been reviewed at length elsewhere (Batalle, Edwards, & O'Muircheartaigh, Reference Batalle, Edwards and O'Muircheartaigh2017; Gilmore, Knickmeyer, & Gao, Reference Gilmore, Knickmeyer and Gao2018).

Structural imaging of the fetal brain

As discussed previously, the CNS anomaly that currently has the most robust association with SSD is enlargement of the ventricular system, a finding that predates disease onset. It has not yet been determined at what gestational age ventriculomegaly can first be discerned, but it can be readily observed in utero with MRI (Kyriakopoulou et al., Reference Kyriakopoulou, Vatansever, Elkommos, Dawson, McGuinness, Allsop and Rutherford2014). The ventricles emerge at 4 weeks gestation, and its shape is influenced by the overall brain morphology and adjacent deep gray matter structures (O'Rahilly & Muller, Reference O'Rahilly and Muller1990). Volumetrically, the ventricles remain static from midgestation to term (Grossman, Hoffman, Mardor, & Biegon, Reference Grossman, Hoffman, Mardor and Biegon2006), and so enlargement as determined by a two-dimensional measurement at the level of the atrium can be detected at any gestational age (Nadel & Benacerraf, Reference Nadel and Benacerraf1995).

It has been shown that fetal ventriculomegaly persists into childhood (Gilmore et al., Reference Gilmore, Smith, Wolfe, Hertzberg, Smith, Chescheir and Gerig2008; Lyall et al., Reference Lyall, Woolson, Wolfe, Goldman, Reznick, Hamer and Gilmore2012). An early study using retrospective case notes and ultrasound films review showed ventricular volume to be significantly different in both monozygotic (n = 41) and dizygotic (n = 103) twin pairs detectable from the second to third trimester (Gilmore et al., Reference Gilmore and Jarskog1997). This indicates an environmental component to ventricular development that the authors suggested may be related to the in utero position of each twin and blood flow of the placenta, which are not shared even in monochorionic pregnancies. There were limitations to this study including technical limitations of ultrasound acquisition, difficulty in assigning zygosity in a large number of cases, which had to be excluded, and not being able to fully account for gender differences in dizygotic pairs. A more recent prospective MRI study in neonates in same-sex twins (monozygotic n = 41; dizygotic n = 50 twin pairs) indicated heritability for lateral ventricles in the order of 71% and total cerebrospinal fluid of 63% (Gilmore et al., Reference Gilmore, Schmitt, Knickmeyer, Smith, Lin, Styner and Neale2010). This indicates a substantial, although not exclusive, genetic component to ventricular in the neonatal period.

Other structural targets for biomarkers include total gray matter volume, white matter volume, and total brain volume. Neonates at risk of SSD have increased total brain parenchyma, and gray matter in particular; conversely, first-episode psychosis subjects show reductions in these indices.

It is important to note from previous fetal MRI studies that the brain has a microstructure that evolves over gestation. The lamination pattern varies in thickness across the anlage, and not all layers are identifiable (Rados, Judas, & Kostović, Reference Rados, Judas and Kostović2006). Brain volumes segmented from MRI data show an approximate ten-fold increase in the latter half of gestation in a fairly linear manner (Kyriakopoulou et al., Reference Kyriakopoulou, Vatansever, Davidson, Patkee, Elkommos, Chew and Rutherford2017; Link et al., Reference Link, Braginsky, Joskowicz, Ben Sira, Harel, Many and Ben Bashat2017), and the rate of cortical volume growth appears to accelerate toward term (Kyriakopoulou et al., Reference Kyriakopoulou, Vatansever, Davidson, Patkee, Elkommos, Chew and Rutherford2017).

Additional targets requiring further investigation include cortical thickness, cortical surface area, and gyrification. Cortical markers have been shown to be abnormal in high-risk infants, whereas abnormalities in the gyrification connectome are demonstrated in first-episode patients and are indicative of prognosis (Palaniyappan et al., Reference Palaniyappan, Park, Balain, Dangi and Liddle2015). Fetal MRI has shown cortical asymmetries that emerge in the second trimester, which is more pronounced in males (Habas et al., Reference Habas, Scott, Roosta, Rajagopalan, Kim, Rousseau and Studholme2012; Kyriakopoulou et al., Reference Kyriakopoulou, Vatansever, Davidson, Patkee, Elkommos, Chew and Rutherford2017). Cortical surface area shows an almost four-fold increase from the second trimester onward, with a corresponding increase in the gyrification index from 0.8 to 1.5 (Clouchoux et al., Reference Clouchoux, Kudelski, Gholipour, Warfield, Viseur, Bouyssi-Kobar and Limperopoulos2012; Garel et al., Reference Garel, Chantrel, Brisse, Elmaleh, Luton, Oury and Hassan2001; Zilles, Palomero-Gallagher, & Amunts, Reference Zilles, Palomero-Gallagher and Amunts2013). Just prior to term age, all of the primary, and the majority of secondary, sulci are visible on MRI (Wright et al., Reference Wright, Kyriakopoulou, Ledig, Rutherford, Hajnal, Rueckert and Aljabar2014, Reference Wright, Makropoulos, Kyriakopoulou, Patkee, Koch, Rutherford and Aljabar2015).

Attention should be paid to the development of deep gray matter structures implicated in psychosis, namely, the hippocampus, thalamus, and basal ganglia. Hippocampal volume changes are present at early stages of disease pathogenesis and are implicated in the neurodevelopmental model for psychosis (Adriano et al., Reference Adriano, Caltagirone and Spalletta2012). A complete hippocampal profile (volume, shape, seed based functional, and structural connectivity) should be interrogated to determine if hippocampus anomalies are a relevant fetal biomarker of psychosis.

MRI studies in healthy pregnancies show that the deep gray matter structures grow at a faster rate relative to other cerebral territories and can be used to index normative growth. Volumetric analysis of segmented deep gray structures show a doubling during the third trimester (Rajagopalan et al., Reference Rajagopalan, Scott, Habas, Kim, Corbett-Detig, Rousseau and Studholme2011). The thalamus grows slightly faster than the basal ganglia over the second and third trimesters (Kyriakopoulou et al., Reference Kyriakopoulou, Vatansever, Elkommos, Dawson, McGuinness, Allsop and Rutherford2014). The hippocampus, which is a challenging structure to segment on MRI during gestation due to its small size and partial volume effects, appears to increase three-fold by the start of the third trimester (Jacob et al., Reference Jacob, Habas, Kim, Corbett-Detig, Xu, Studholme and Glenn2011). Whether subcortical structures deviate from these growth trajectories remains to be established in those at risk of SSD.

Metabolic imaging

MRS is limited by the large size of the voxel, which limits characterization of local changes in the fetal brain. Several spectral peaks have been identified in the normal fetal brain, including choline, creatine, N-acetylaspartate, myo-inositol, and lactate (Story et al., Reference Story, Damodaram, Allsop, McGuinness, Wylezinska, Kumar and Rutherford2011). The choline peak appears to peak toward the end of the second trimester and subsequently declines perhaps related to the physiological large-scale neuronal pruning that occurs (Girard et al., Reference Girard, Fogliarini, Viola, Confort-Gouny, Le Fur, Viout and Cozzone2006). A recent cross-sectional study aimed to characterize the development of metabolite spectra during gestation and observed choline and myo-inositol peaks emerging first at 18 weeks gestation, followed by N-acetylaspartate and creatine peaks at 23 weeks; adjacent to the N-acetylaspartate, a glutamate signal was detected at GA 29. All metabolites increase in concentration with gestation (Urbanik, Cichocka, Kozub, Karcz, & Herman-Sucharska, Reference Urbanik, Cichocka, Kozub, Karcz and Herman-Sucharska2019).

MRS does allow for chemical characterization of the developing brain. Using MRS glutamate/glutamine spectra may contribute to supporting the glutamatergic theory of psychosis (Marsman et al., Reference Marsman, Van Den Heuvel, Klomp, Kahn, Luijten and Hulshoff Pol2013; Merritt et al., Reference Merritt, Egerton, Kempton, Taylor and McGuire2016; Poels et al., Reference Poels, Kegeles, Kantrowitz, Slifstein, Javitt, Lieberman and Girgis2014). MRS GABA signal is of interest given the phenomenon of the postnatal excitatory–inhibitory (Ben-Ari, Khalilov, Kahle, & Cherubini, Reference Ben-Ari, Khalilov, Kahle and Cherubini2012); however, the spectral peak is challenging to resolve even at 3T in adults (Mullins et al., Reference Mullins, McGonigle, O'Gorman, Puts, Vidyasagar, Evans and Wilson2014). MRS lactate, a transient marker of fetal distress and hypoxia, may indicate abnormal neurodevelopment and SSD risk.

Connectivity imaging—Structural

The potential to identify network abnormalities may be of interest. Use of diffusion tensor imaging approaches allows characterization of white matter microstructure and tractography, whereas resting-state functional connectivity may reveal differences in topographical structure using graph theory approaches. Of the major white matter tracts present in the mature brain, the limbic circuitry and anterior segment of the corpus callosum have emerged by the second trimester. In addition, the uncinate fasciculus—particularly implicated in neuropsychiatric disease—internal capsule and cerebral peduncle develop relatively early in gestation (Huang et al., Reference Huang, Zhang, Wakana, Zhang, Ren, Richards and Mori2006). The gradual maturation of white matter tracts and rudimentary myelination during gestation are reflected by a decline in apparent diffusion coefficient and increasing fractional anisotropy (Bui et al., Reference Bui, Daire, Chalard, Zaccaria, Alberti, Elmaleh and Sebag2006; Marami et al., Reference Marami, Mohseni Salehi, Afacan, Scherrer, Rollins, Yang and Gholipour2017). Myelination itself has been found to start during the second half of gestation, initiated in somatosensory pathways (Dubois et al., Reference Dubois, Dehaene-Lambertz, Kulikova, Poupon, Hüppi and Hertz-Pannier2014). Motor pathways begin myelinating at 33 weeks (Zanin et al., Reference Zanin, Ranjeva, Confort-Gouny, Guye, Denis, Cozzone and Girard2011).

Connectivity imaging—Functional

Fetal resting-state functional MRI demonstrates cross-hemispheric functional connectivity, notably in regions of the occipital, parietal, and cingulate cortex. The strength of bilateral networks appeared to increase with gestational age (Thomason et al., Reference Thomason, Dassanayake, Shen, Katkuri, Alexis, Anderson and Romero2013), which may underlie the development of a modular architecture that prioritizes a hub (regions of high connectivity) configuration earlier in gestation (van den Heuvel & Thomason, Reference van den Heuvel and Thomason2016). A network analysis approach has found dominant hubs located in primary motor and sensory regions, and a quarter of which are in the cerebellum. This finding is thought to reflect the relatively earlier myelination in these areas. Furthermore. of the dominant hubs, there appears to be age-related variation in some frontal and temporal hubs (van den Heuvel et al., Reference van den Heuvel, Turk, Manning, Hect, Hernandez-Andrade, Hassan and Thomason2018).

Extra-axial imaging

Finally, markers of SSD may not only be apparent in the brain. MPAs are a robust marker of SSD risk (Radua et al., Reference Radua, Ramella-Cravaro, Ioannidis, Reichenberg, Phiphopthatsanee and Amir2018; Xu et al., Reference Xu, Chan and Compton2011) and can be demonstrated on fetal MRI (Hayat, Martinez-Biarge, Kyriakopoulou, Hajnal, & Rutherford, Reference Hayat, Martinez-Biarge, Kyriakopoulou, Hajnal and Rutherford2018).

Spontaneous CNS activity underlies resting-state networks as described earlier, but may also result in motor behavior before the onset of voluntary activity several months postterm age (Hayat et al., Reference Hayat, Nihat, Martinez-Biarge, McGuinness, Allsop, Hajnal and Rutherford2011). It is currently unclear where spontaneous motor activity originates despite mounting evidence for a sufficiently mature sensorimotor network with peripheral connectivity (Allievi et al., Reference Allievi, Arichi, Tusor, Kimpton, Arulkumaran, Counsell and Burdet2016; Eyre, Miller, Clowry, Conway, & Watts, Reference Eyre, Miller, Clowry, Conway and Watts2000; Sarnat, Reference Sarnat1989). The environment of the fetus changes substantially during gestation, which will lead to an altered sensorimotor activation pattern. Preliminary work has suggested that fetuses with structural abnormalities show altered patterns of spontaneous motor activity (Hayat et al., Reference Hayat, Martinez-Biarge, Kyriakopoulou, Hajnal and Rutherford2018). As movements within the first 2 years of life have been noted to be associated with development of SSD (Walker et al., Reference Walker, Savole and Davis1994), visualizing movement in utero may also be a likely candidate biomarker for SSD risk.

Funding

Birth cohorts are expensive projects, and funding particularly through public bodies has ever been contentious since the Collaborative Perinatal Project (Hardy, Reference Hardy2003), one of the first neuroscience-based cohort studies. Longitudinal cohort designed studies have been criticized for the potential to detract from other scientific pursuits on the basis of resource allocation.

Our proposal is aided by falling costs of MRI. In the United Kingdom falling costs of MRI scanning has facilitated increased expansion of clinical scans from 2006 to 2016. The annual cost of the 2 million clinical scans undertaken in the National Health Service has remained stable at around £300 million GBP. In real terms, as cost has halved, utilization has doubled.

A comprehensive review of the cost and funding of birth cohorts suggests that these studies should be viewed as long-term investments, necessitating an appropriate infrastructure for implementation (Doyle & Golding, Reference Doyle and Golding2009). The authors suggest that a priori setting of early milestones can form the basis of further funding requests and that the initial commitment should span 5 years as a minimum. Furthermore, the project can be secured by spreading the funding over several agencies, but it will require a carefully designed administrative setup to manage this.

Recruitment

One strategy might be a focused study recruiting 1,000 mothers with SSD to form the birth cohort, given a parental risk of approximately 10%. This would have challenges in case ascertainment and would be required to be a multicenter study. It would require follow-up of 20–25 years for final outcome to test those who developed SSD against those who did not, although surrogate outcomes such as neuromotor development, cognitive and school attainment, and subthreshold psychotic-like experiences would allow for valuable interim analysis as demonstrated in previous cohorts (Isohanni, Murray, Jokelainen, Croudace, & Jones, Reference Isohanni, Murray, Jokelainen, Croudace and Jones2004; Jones, Rodgers, Murray, & Marmot, Reference Jones, Murray, Jones, Rodgers and Marmot1994; Niarchou, Zammit, & Lewis, Reference Niarchou, Zammit and Lewis2015; Wallace & Linscott, Reference Wallace and Linscott2018). An alternative approach would be to recruit healthy controls matched for parental socioeconomic status, site, and fetal gender in a case-control design. An important caveat is that the vast majority of patients with SSD do not have parents with illness, and such a study may further not be generalizable to yield insights into the other neurodevelopmental disorders.

Conversely, a population-based cohort would be optimal to interrogate broader neurodevelopmental outcomes. However, this would require a sample of around 20,000 and would allow for 100–200 cases of SSD, consistent with the Prenatal Determinants of Schizophrenia and Northern Finland Birth cohort studies (Bresnahan et al., Reference Bresnahan, Schaefer, Brown and Susser2005; Jääskeläinen et al., Reference Jääskeläinen, Haapea, Rautio, Juola, Penttilä, Nordström and Miettunen2015). In keeping with current estimates, such a sample would detect 1,200 cases of all disability, 500 cases of moderate to profound learning disability, and 600–800 cases of neurodevelopmental disorders, including attention-deficit/hyperactivity disorder (200–400), autism (200), and epilepsy (60; Department of Health, 2013) although these figures do not factor for attrition.

Acceptability

A great deal of work will need to be done in ensuring MRI acceptability to ensure widespread enrollment. Some qualitative work has already been done. Mothers are interested in taking part in fetal MRI and clinical research but voice concerns about potential harm to fetus, and comfort in the scanner (Lie, Graham, Robson, & Griffiths, Reference Lie, Graham, Robson and Griffiths2018; Reed, Kochetkova, & Whitby, Reference Reed, Kochetkova and Whitby2016). Perhaps understandably those with more severe malformations experience the most anxiety about the scan (Leithner et al., Reference Leithner, Pörnbacher, Assem-Hilger, Krampl, Ponocny-Seliger and Prayer2008). Some concerns can be allayed with information (Leithner, Pörnbacher, Assem-Hilger, Krampl-Bettelheim, & Prayer, Reference Leithner, Pörnbacher, Assem-Hilger, Krampl-Bettelheim and Prayer2009), and ensuring attention to mothers’ concerns will be an important component of study recruitment.

Infrastructure

The United Kingdom has a long-term experience of fetal MRI; in 2005, 91% of survey sites responding to a national survey reported use of MRI in the second and third trimesters (De Wilde, Rivers, & Price, Reference De Wilde, Rivers and Price2005). The nature of ensuring access to MRI will require multiple centers. This will require sufficient capacity of dedicated MRI scanning facilities available at each site for the cohort. Although a hurdle, this is not unsurmountable, and a recent fetal MRI study collected data from 16 sites (Griffiths et al., Reference Griffiths, Bradburn, Campbell, Cooper, Graham and Jarvis2017). Nonetheless, identifying sites and ensuring adequate capacity will be a major feature of the planning phase.

Technical challenges

Regarding our proposal, the specific imaging techniques require robust validation and piloting prior to commencement. Fortunately, there is a large body of studies accumulating on the development and optimization of fetal imaging sequences and postprocessing techniques as described earlier including the Developing Connectome and Baby Connectome projects.