Clinical dimensions and categories of psychotic symptoms and disorders

The DSM-5-TR includes a section on “schizophrenia spectrum and other psychotic disorders.” This encompasses SCZ, schizoaffective disorder, schizophreniform disorder, other specified/unspecified psychotic disorders, and schizotypal personality disorder (SPD). With the exception of SPD, these diagnoses share the requirement for the presence of one or more positive symptoms, including delusions, hallucinations, or disorganized speech. Conversely, SPD diagnostic criteria include less severe positive symptoms; for example, unusual ideas (e.g., persistent belief in extrasensory perception) rather than delusions and unusual sensory experiences (e.g., repeatedly misperceiving noises as voices) rather than hallucinations. The subclinical positive symptoms present in SPD are not as distressing, severe, or persistent as those required to meet criteria for a clinical delusion or hallucination. These features are similar to the criteria for CHR-P status. Relatedly, youth at CHR-P often meet criteria for SPD, and the conversion rate of SPD to PSD in young adults is similar to that for CHR-P to PSD (Caballero et al., Reference Caballero, Machiraju, Diomino, Kennedy, Kadivar and Cadenhead2023; Debbané et al., Reference Debbané, Eliez, Badoud, Conus, Flückiger and Schultze-Lutter2015).

Before discussing research findings, we note the weaknesses of the categorical model put forth by the DSM-5-TR. The DSM-5-TR has clinical utility for diagnosis and communication among providers and researchers, but many researchers have advocated for a dimensional model of PSDs. There is no single symptom required for PSDs, thus increasing variation in symptoms among patients with the same diagnosis. Further, individuals with different diagnoses may nonetheless share most symptoms. These differences within diagnostic categories and similarities across categories are not compatible with the arbitrary thresholds inherent in a categorical framework. Finally, DSM categories lack treatment specificity in that the diagnosis does not necessarily facilitate treatment planning given the heterogeneity in symptomology and etiology within categories.

Psychotic-like experiences

As stated, population rates of PSDs (1%–2%) and CHR-P syndromes (4%) are low. However, especially during adolescence, the population base rates of psychotic-like experiences (PLEs) are much higher. For PLEs assessed with structured interviews, the prevalence ranges from 8% to 17% in the general population (Schultze-Lutter et al., Reference Schultze-Lutter, Renner, Paruch, Julkowski, Klosterkotter and Ruhrmann2014). Due to limitations in access to clinical interviews and concerns about sample representativeness, recent investigations have relied on self-reported PLEs. For example, the Adolescent Brain Cognitive Development (ABCD) Study, including over 11,000 preadolescents, demonstrated that 65% experienced at least one PLE (Karcher et al., Reference Karcher, Paul, Johnson, Hatoum, Baranger, Agrawal, Thompson, Barch and Bogdan2022). PLEs involve perceptions, thoughts, or beliefs that are considered unusual, odd, or unreal, but not clinically significant and not necessarily distressing or disorganizing (Hinterbuchinger & Mossaheb, Reference Hinterbuchinger and Mossaheb2021). Thus, PLEs are an index of the mildest end of the continuum of PSDs.

It is quite common for children to experience transient PLEs, but the occurrence of PLEs typically decreases with age through adolescence (van Os, Reference Van Os2013). Recent work with large samples has demonstrated that, compared to children without PLEs, those endorsing PLEs are at greater risk for developing PSDs and other psychiatric disorders in adulthood (Lindgren et al., Reference Lindgren, Numminen, Holm, Therman and Tuulio-Henriksson2022). This suggests that self-reported PLEs can be transdiagnostic predictors of risk. Further, PLEs among adolescents aged 12–14 years are associated with impaired social and cognitive functioning, as well as harmful drinking and later substance use (Davies et al., Reference Davies, Segre, Estradé, Radua, De Micheli, Provenzani, Oliver, Salazar de Pablo, Ramella-Cravaro, Besozzi, Dazzan, Miele, Caputo, Spallarossa, Crossland, Ilyas, Spada, Politi, Murray, McGuire and Fusar-Poli2020). Other studies have found that features of PLEs, such as persistence and distress, are associated with cognitive deficits and increased mental health service utilization (Karcher et al., Reference Karcher, Paul, Johnson, Hatoum, Baranger, Agrawal, Thompson, Barch and Bogdan2022). Considered together, positive symptoms of psychosis are best conceptualized as a continuum ranging from PLEs to CHR-P syndromes to PSDs (Insel et al., Reference Insel, Cuthbert, Garvey, Heinssen, Pine, Quinn, Sanislow and Wang2010).

Genetics and epigenetics

The search for a single or even small number of inherited genes with a major impact on psychosis risk was very informative, but did not prove successful, and this led to studies concerned with a variety of other genetic mechanisms. Genome wide association studies (GWAS), in particular, have revealed that numerous genes of small effect influence psychosis risk. The common genetic variants found to be statistically associated with psychosis in GWAS can be combined as an index of genetic liability. Polygenic risk scores (PRS) for various psychiatric disorders are weighted combinations of alleles that significantly differentiate patient samples from unaffected comparisons (Abdellaoui & Verweij, Reference Abdellaoui and Verweij2021). While an important advance, the proportion of the variance in ‘caseness’ accounted for by PRS has been modest, and much lower than estimates of heritability from twin studies. This small proportion holds for even the largest (e.g., > 69,000) GWAS to date, which shows less than 20% of genetic vulnerability to SCZ being attributable to common variants (Owen et al., Reference Owen, Legge, Rees, Walters and O’Donovan2023). Also, PRS for general psychopathology and SCZ account for a small but significant proportion of the variance in distressing PLEs (Karcher et al., Reference Karcher, Paul, Johnson, Hatoum, Baranger, Agrawal, Thompson, Barch and Bogdan2022) making them a powerful tool in research on gene-environment interactions.

The modest predictive power of PRS is due to a number of factors, including copy number variations (CNVs), both deletions and duplications in the genome sequence which are usually spontaneous (i.e., de novo) mutations, thus not shared with the parental germline, yet can confer risk for PSDs. Interest in the role of mutations in PSD increased following the discovery that a particular recurrent CNV, 22q11 deletion syndrome (DS), confers a dramatically increased risk for psychosis, as well as a range of other psychiatric and physical disorders (Fiksinski et al., Reference Fiksinski, Hoftman, Vorstman and Bearden2023). This genetic syndrome is associated with elevated risk rates for intellectual deficits, psychosis, autism spectrum disorders (ASDs), and anxiety disorders to levels of 50%, 20%–25%, 10%–40%, and 36%, respectively (Fiksinski et al., Reference Fiksinski, Schneider, Zinkstok, Baribeau, Chawner and Vorstman2021). All of these rates are well above the population base rates for these disorders. Similarly, other genetic variants, including 3q29 DS, 7q11.2 DS, and 15q11.2–13 DS, have been identified and shown to be associated with increased rates of psychotic disorders (Mulle et al., Reference Mulle, Gambello, Russo, Murphy, Burrell, Klaiman and Li2021; Rutkowski et al., Reference Rutkowski, Schroeder, Gafford, Warren, Weinshenker, Caspary and Mulle2017). Although less common than 22q11 DS, all these CNVs are associated with a significant increase in risk for the syndromes associated with 22q11 DS, especially PSDs and ASDs. In addition to these well-studied CNVs, there are numerous other less common CNVs, and there is evidence that the majority of people (about 70%) have at least one rare CNV (Singh et al., Reference Singh, Olsen, Lavik, Vold, Drabløs and Sjursen2021). Recent studies of large samples of SCZ patients have revealed additional, less common CNVs occur at a higher rate than in healthy control samples (Kato et al., Reference Kato, Kimura, Kushima, Takahashi, Aleksic and Ozaki2023).

Recurring CNVs that confer risk for a range of psychiatric syndromes lend support to the notion of multifinality; they illustrate the pleiotropic nature of genetic risk factors and suggest many pathways to psychosis. Most notably, they illustrate the developmental comorbidity observed in psychiatric disorders and challenge us to explore the neurodevelopmental and epigenetic processes that modify the expression of genetic vulnerabilities. Further, they highlight the importance of identifying gene-environment interactions that trigger the manifestation of psychotic symptoms in vulnerable individuals. In sum, consistent with the assumption of etiological heterogeneity in SCZ and other PSDs, research has identified multiple CNVs that are associated with PSDs, illustrating that multiple genotypes can confer risk, and that differing developmental trajectories can precede clinical onset. Research on CNVs also has the potential to elucidate shared, final common pathways that subserve neurobiological mechanisms in the emergence of PSDs. For example, microglia activation, which is central to immune response in the brain, is elevated in PSDs, ASDs (Jutla et al., Reference Jutla, Foss-Feig and Veenstra-VanderWeele2022), and both 22q11 DS and 3q29 DS have the potential to affect microglia and contribute to brain dysfunction (Kato et al., Reference Kato, Kimura, Kushima, Takahashi, Aleksic and Ozaki2023). Nongenetic sources of psychosis-risk also modify microglial activation; this includes inflammatory processes and adverse prenatal exposures (Dietz et al., Reference Dietz, Goldman and Nedergaard2020) as well as trauma and/or stress exposure (Yue et al., Reference Yue, Huang and Duan2022). Thus, identifying convergent pathways in well-established genetic risk factors for psychosis has the potential to not only shed light on etiology, but also shared mechanisms and treatment (Besterman, Reference Besterman2023). Of course, genetic influences on psychosis-risk are not only a function of DNA, but also a consequence of gene expression (i.e., epigenetics). Epigenetics refers to the processes that determine whether genes are functionally expressed. Epigenetics entails changes in gene expression without changes in DNA, but modifications in the expression of DNA and its protein products. Evidence that epigenetic processes are relevant to PSD is illustrated by research on discordant monozygotic twin (MZ) pairs. Studies of discordant MZ twins reveal differences in epigenetic profiles, and some of the genes characterized by greater epigenetic changes in psychotic disorders are also candidate genes that are associated with risk for psychosis (van Dongen et al., Reference van Dongen, Odintsova and Boomsma2021). Among the factors that can alter the expression of genes are hormonal changes involved in prenatal and pubertal maturation and environmental factors that are related to risk for PSD, including cannabis and stress-related changes in adrenal hormones (Walker et al., Reference Walker, Trotman, Goulding, Holtzman, Ryan, McDonald, Shapiro and Brasfield2013).

Transdiagnostic neurodevelopmental trajectories

As described above, recurrent CNVs are known to be associated with PSDs that are preceded by a variety of other psychiatric syndromes. Since the advent of CHR-P research, it has become more apparent that multi-syndrome developmental trajectories are the norm, rather than the exception, in the prodrome to psychosis. Yet, in some cases, psychosis is preceded by a normal developmental course (Černis et al., Reference Černis, Vassos, Brébion, McKenna, Murray, David and MacCabe2015). As described above, CHR-P status is based on the administration of a structured clinical interview, such as the Structured Interview for Psychosis-Risk Syndromes (SIPS; McGlashan et al., Reference McGlashan, Walsh and Woods2010). Numerous studies of youth at CHR-P youth have revealed that they often experienced mood, anxiety, and obsessive-compulsive disorders (OCD) prior to meeting CHR-P criteria. For example, in the North American Prodrome Longitudinal Study (NAPLS), one of the largest multi-site longitudinal studies of CHR-P and healthy youth (ages 12–30), those at CHR-P showed a range of developmental trajectories (Allswede et al., Reference Allswede, Addington, Bearden, Cadenhead, Cornblatt, Mathalon, McGlashan, Perkins, Seidman, Tsuang, Walker, Woods and Cannon2020), with 60% meeting criteria for a current or past diagnosis of depression at baseline (Kline et al., Reference Kline, Seidman, Cornblatt, Woodberry, Bryant, Bearden, Cadenhead, Cannon, Mathalon, McGlashan, Perkins, Tsuang, Walker, Woods and Addington2018). While the presence of depression at baseline was not predictive of conversion to psychosis, it was associated with a lower likelihood of remission from CHR-P status. A similar baseline occurrence of various anxiety disorders was also observed in those at CHR-P (Perkins et al., Reference Perkins, Jeffries, Addington, Bearden, Cadenhead, Cannon, Cornblatt, Mathalon, McGlashan, Seidman, Tsuang, Walker, Woods and Heinssen2015). These findings echo those of a recent report on a 5-year cohort study addressing clinical profile, conversion to psychosis, and prognostic accuracy of DSM-5 attenuated psychosis syndrome, which revealed that 54% and 40% experienced major depression and generalized anxiety, respectively (Mensi et al., Reference Mensi, Molteni, Iorio, Filosi, Ballante, Balottin, Fusar-Poli and Borgatti2021).

In addition to mood and anxiety disorders, it is also now recognized that early childhood onset disorders, especially ASD and ADHD, often precede the onset of psychosis. A recent review of the literature concluded that ASD youth are 3 to 6 times more likely to develop SCZ than age-matched neurotypical youth (Jutla et al., Reference Jutla, Foss-Feig and Veenstra-VanderWeele2022). Consistent with other longitudinal studies, the association was not specific to SCZ, but rather associated with a range of PSDs. ASD is also elevated in youth at CHR-P, although the presence of a previous ASD diagnosis does not alter the rate of conversion to psychosis (Foss-Feig et al., Reference Foss-Feig, Velthorst, Smith, Reichenberg, Addington, Cadenhead, Cornblatt, Mathalon, McGlashan, Perkins, Seidman, Stone, Keshavan, Tsuang, Walker, Woods, Cannon and Bearden2019). Similarly, children diagnosed with ADHD are more likely to later manifest a psychotic disorder (Nourredine et al., Reference Nourredine, Gering, Fourneret, Rolland, Falissard, Cucherat, Geoffray and Jurek2021), and it is estimated that 50%–60% of youth at CHR-P have a current or lifetime diagnosis of ADHD (Olde Loohuis et al., Reference Olde Loohuis, Mennigen, Ori, Perkins, Robinson, Addington, Cadenhead, Cornblatt, Mathalon, McGlashan, Seidman, Keshavan, Stone, Tsuang, Walker, Woods, Cannon, Gur, Gur, Bearden and Ophoff2021).

In sum, longitudinal studies yield strong support for transdiagnostic pathways to many cases of psychosis. We now know that this is, in part, a consequence of shared and nonshared genetic risk factors. In addition, there is increasing evidence that other risk factors, including environmental exposures, can trigger the onset of developmental trajectories that are transdiagnostic and lead to PSDs.

Environmental risk factors

Environmental factors affect risk for psychopathology beginning in the prenatal period. Research on PSDs has long implicated prenatal, perinatal, and early life complications, aligning with the neurodevelopmental hypotheses of SCZ. A recent review and meta-analysis identified 98 potential risk factors associated with psychosis during the prenatal and perinatal periods (Davies et al., Reference Davies, Segre, Estradé, Radua, De Micheli, Provenzani, Oliver, Salazar de Pablo, Ramella-Cravaro, Besozzi, Dazzan, Miele, Caputo, Spallarossa, Crossland, Ilyas, Spada, Politi, Murray, McGuire and Fusar-Poli2020). These included maternal infections (e.g., toxoplasma and influenza) and nutritional deficits, season of birth, obstetric complications, hypoxic birth, rupture or premature rupture of membranes, and multiparity exceeding three.

Beyond the proximal prenatal environment, recent research on the distal postnatal environment has shed light on neighborhood-level or area-level risk factors. Neighborhood-level characteristics can be derived from publicly available sources; some are derived from the measurement of air pollutants, and others from aggregated surveys from thousands who live in a particular catchment area. Therefore, neighborhood characteristics provide an objective measure of the environment not likely to be subject to recall bias.

Urban upbringing (i.e., having grown up in cities as compared to rural areas) is one of the main risk factors for SCZ in high-income countries (March et al., Reference March, Hatch, Morgan, Kirkbride, Bresnahan, Fearon and Susser2008). Growing evidence suggests that childhood area-level characteristics may explain this association and contribute to the development of psychosis (Anglin et al., Reference Anglin, Ereshefsky, Klaunig, Bridgwater, Niendam, Ellman, Devylder, Thayer, Bolden, Musket, Grattan, Lincoln, Schiffman, Lipner, Bachman, Corcoran, Mota and Van Der Ven2021; Ku et al., Reference Ku, Compton, Walker and Druss2021a). Since 1939, studies have shown that SCZ rates are higher among racial minorities residing in neighborhoods with less ethnic density (i.e., neighborhood-level proportions of others belonging to the same ethnic group) and youth residing in neighborhoods with greater indices of social fragmentation (Faris & Dunham, Reference Faris and Dunham1939). Neighborhood social fragmentation has been measured with a combination of area-level indices that empirically index the disruption of social ties and relationships among residents and families in a community (Ku et al., Reference Ku, Compton, Walker and Druss2021a). A recent systematic review demonstrated a 4- and 12-fold increase in SCZ prevalence and admission rates, respectively, in neighborhoods with the highest as compared to the lowest indices of social fragmentation (Allardyce & Boydell, Reference Allardyce and Boydell2006; Ku et al., Reference Ku, Compton, Walker and Druss2021a). Greater area-level residential instability, or the proportion of people who moved in a catchment area, was also associated with earlier onset of psychosis among individuals with FEP as well as conversion to psychosis among CHR-P youth (Ku et al., Reference Ku, Walker, Druss, Murray and Compton2020, Reference Ku, Compton, Walker and Druss2021b). In addition, the timing of social adversity and/or disconnection may be relevant. For example, the association between moving (i.e., greater number as well as the distance of moves) and psychosis-risk has been shown to be more relevant during childhood and adolescence compared to adulthood (Ku et al., Reference Ku, Addington, Bearden, Cadenhead, Cannon, Compton, Cornblatt, Druss, Guloksuz, Mathalon, Perkins, Tsuang, Walker, Woods and Carrion2023; Paksarian et al., Reference Paksarian, Trabjerg, Merikangas, Mors, Børglum, Hougaard, Nordentoft, Werge, Pedersen, Mortensen, Agerbo and Horsdal2020; Price et al., Reference Price, Dalman, Zammit and Kirkbride2018a).

It is also likely that childhood social adversity (in concert with other vulnerabilities) leads to more observable phenotypic expression of psychosis later in development during the peak age period for onset. For example, childhood area-level social fragmentation was associated with social functioning later in adulthood, but not during childhood (Ku et al., Reference Ku, Addington, Bearden, Cadenhead, Cannon, Compton, Cornblatt, Druss, Guloksuz, Mathalon, Perkins, Tsuang, Walker, Woods and Carrion2023b). In studies conducted in Spain and the US, the association between neighborhood-level social characteristics and worse social functioning has been shown to be stronger among those with early psychosis versus healthy comparisons (Izquierdo et al., Reference Izquierdo, Cabello, Leal, Torio, Madrigal, MacDowell, Rodriguez-Jimenez, Rentero, Ibáñez, Ayora, Díaz-Caneja, Abregú-Crespo, Mellor-Marsá, Díaz-Marsá, Malpica, Bravo-Ortiz, Baca-García, Arango and Ayuso-Mateos2023; Ku et al., Reference Ku, Addington, Bearden, Cadenhead, Cannon, Compton, Cornblatt, Druss, Guloksuz, Mathalon, Perkins, Tsuang, Walker, Woods and Carrion2023b). Further, these effects may be mediated by neurobiological factors. Childhood area-level residential instability was shown to be associated with the reduction of brain gray matter volume in the same brain regions (i.e., caudal middle frontal gyrus and rostral anterior cingulate cortex of the right hemisphere) as those associated with urban upbringing (Haddad et al., Reference Haddad, Schäfer, Streit, Lederbogen, Grimm, Wüst, Deuschle, Kirsch, Tost and Meyer-Lindenberg2015; Ku et al., Reference Ku, Aberizk, Addington, Bearden, Cadenhead, Cannon, Carrión, Compton, Cornblatt, Druss, Mathalon, Perkins, Tsuang, Woods and Walker2022). These findings demonstrate various and complex social factors that may contribute to the onset of psychosis.

Stress and inflammation

As described above, numerous environmental risk factors for psychosis have been identified and they may act on the same pathways that have been hypothesized to account for the adverse effects of psychosocial stressors. Stress exposure, especially chronic exposure, has been found to be associated with elevated risk for a range of psychiatric disorders (Cohen et al., Reference Cohen, Murphy and Prather2019). A large body of population-based studies suggests that early childhood trauma (i.e., sexual, physical, and emotional abuse, and neglect) is associated with psychosis risk (Schäfer & Fisher, Reference Schäfer and Fisher2011). In studies of individuals at CHR-P, measures of life event stress (LES) exposure have been shown to be related with an increased likelihood that individuals will subsequently develop PSD (Trotman et al., Reference Trotman, Holtzman, Walker, Addington, Bearden, Cadenhead, Cannon, Cornblatt, Heinssen, Mathalon, Tsuang, Perkins, Seidman, Woods and McGlashan2014: Cullen et al., Reference Cullen, Addington, Bearden, Stone, Seidman, Cadenhead, Cannon, Cornblatt, Mathalon, McGlashan, Perkins, Tsuang, Woods and Walker2020). Similarly, daytime basal cortisol levels are elevated in nonmedicated individuals with psychosis (Misiak et al., Reference Misiak, Pruessner, Samochowiec, Wiśniewski, Reginia and Stańczykiewicz2021) and are associated with risk for conversion in youth at CHR-P (Ristanovic et al., Reference Ristanovic, Vargas, Damme and Mittal2023; Walker et al., Reference Walker, Trotman, Goulding, Holtzman, Ryan, McDonald, Shapiro and Brasfield2013).

The mechanisms for these relationships are not known, yet potential vulnerabilities include genetic and biological vulnerabilities (Pastore et al., Reference Pastore, De Girolamo, Tafuri, Tomasicchio and Margari2022) and social environmental factors (Allardyce & Boydell, Reference Allardyce and Boydell2006; van Os & McGuffin, Reference van Os and McGuffin2003). The neural mediators have been explicated in neural diathesis-stress models linking the hypothalamic-pituitary-adrenal axis and glucocorticoid release with downstream neurotransmitter changes that can alter brain structure and function (Walker & Diforio, Reference Walker and Diforio1997; Walker et al., Reference Walker, Mittal and Tessner2008). All of the neurotransmitter systems (dopamine, glutamate, and GABA:gamma-aminobutyric acid) that have been hypothesized to be involved in psychosis can be altered by the increased release of glucocorticoids (e.g., cortisol in humans).

Importantly, recent work suggests that elevated LES and basal cortisol are associated with neuroanatomical changes in healthy individuals and those at CHR-P. A report from the 2nd cohort of NAPLS revealed significant inverse relations between cumulative LES and thickness of bilateral middle and superior temporal cortex, as well as aspects of right occipital, orbitofrontal, temporal, and parietal cortex in both healthy comparisons and those at CHR-P. A select set of brain regions were also “double hit,” or demonstrated significant main effects for both LES and diagnostic group status (i.e., right lateral temporal and occipital, and left anterior temporal lobes), suggesting that LES may contribute to cortical thinning in brain areas associated with diagnostic outcomes (Aberizk et al., Reference Aberizk, Collins, Addington, Bearden, Cadenhead, Cornblatt, Mathalon, McGlashan, Perkins, Tsuang, Woods, Cannon and Walker2022). Further, in a recent application of structural equation modeling in the combined NAPLS cohorts significant inverse relations between basal cortisol and bilateral HV were observed (Aberizk et al., Reference Aberizk, Addington, Bearden, Cadenhead, Cannon, Cornblatt, Keshavan, Mathalon, Perkins, Stone, Tsuang, Woods, Walker and Ku2024), lending support to the notion that elevated cortisol is a nonspecific risk factor for reduced HV (Merz et al., Reference Merz, Desai, Maskus, Melvin, Rehman, Torres, Meyer, He and Noble2019). Significant reductions in HV remain the most well-replicated brain morphological feature of psychotic illness, a conclusion supported by meta-analysis (Gutman et al., Reference Gutman, van Erp, Alpert, Ching, Isaev, Ragothaman, Jahanshad, Saremi, Zavaliangos-Petropulu, Glahn, Shen, Cong, Alnaes, Andreassen, Doan, Westlye, Kochunov, Satterthwaite, Wolf, Huang and Wang2022).

Neuroinflammatory processes are another mediating factor in the relation of stress with psychosis. Both experimental animal studies and clinical research with humans have demonstrated a relation of psychosocial stress with inflammatory processes, and both factors heighten vulnerability to physical and mental disorders. A recent meta-analysis of research with healthy participants revealed that the adverse effects of stress exposure on inflammation become increasingly apparent with age from childhood into adolescence and young adulthood, suggesting there are sensitive periods for stress-induced inflammation (Chiang et al., Reference Chiang, Lam, Chen and Miller2022). Numerous studies have also shown that signs of neuroinflammation (e.g., cytokines) are elevated in individuals with psychosis (de Bartolomeis et al., Reference De Bartolomeis, Barone, Vellucci, Mazza, Austin, Iasevoli and Ciccarelli2022). Potential mechanisms for the relation of stress with inflammation include the activation of microglia, phagocytes and immune cells in the brain that can affect neuronal circuitry through synapse modification, including the addition, elimination, maintenance, and plasticity of synapses (Comer et al., Reference Comer, Carrier, Tremblay and Cruz-Martín2020). Thus, the stress-induced release of glucocorticoids can compromise immune function, alter microglia function, and compromise brain structure and function.

The association of neuroinflammatory indicators and disorders with psychosis has been demonstrated in research spanning the past 20 years and is now gaining increasing attention (Bechter, Reference Bechter2023). The global COVID pandemic is providing new evidence of the potential for neuroinflammation to trigger psychosis. A recent study, based on an international database from over 80 million patients from the US and eight other countries, including over 1 million who had contracted COVID-19, demonstrated that a past COVID-19 diagnosis was associated with a significant increase in psychiatric disorders. There was a significant increase in risk of psychotic and cognitive disorders that persisted throughout the 2-year follow-up period and were most evident in children (age < 18 years) and older adults (age > 64 years). This pattern of findings led the authors to conclude that underlying neuroinflammatory processes can increase risk years after the acute infection. They also suggest that vulnerability to psychosis is not necessarily congenital (Taquet et al., Reference Taquet, Sillett, Zhu, Mendel, Camplisson, Dercon and Harrison2022).

The findings from prospective longitudinal studies of post-viral infection patients are consistent with other evidence indicating that some psychosis patients may not have a congenital vulnerability, but rather acquire vulnerability via exposure to environmental factors. For example, one study using cluster analysis with a large sample of FEP patients identified one subgroup that manifested normal childhood/adolescent premorbid functioning and also had the lowest PRS scores for SCZ as well as other psychiatric disorders (Ferraro et al., Reference Ferraro, Quattrone, La Barbera, La Cascia, Morgan, Kirkbride, Cardno, Sham, Tripoli, Sideli, Seminerio, Sartorio, Szoke, Tarricone, Bernardo, Rodriguez, Stilo, Gayer-Anderson, De Haan and Velthorst2023). Similarly, another study found that high-IQ (> = 120) SCZ patients showed less severe negative and disorganized symptoms, better global functioning, and greater insight than other SCZ patients, leading the authors to invoke the notion of a psychosis spectrum or continuum that cuts across standard diagnostic categories (Černis et al., Reference Černis, Vassos, Brébion, McKenna, Murray, David and MacCabe2015).

Brain structure and networks

Advances in neuroimaging resulted in techniques with greater sensitivity to and increased efficiency in delineating cortical morphology. It has been shown that two indices that comprise cortical volume thickness and surface area are uncorrelated and differ in cytoarchitecture, genetic influences and development. Cortical thickness (CT) indexes neuronal cell bodies in addition to the dense, extracellular space between neurons, including dendrites and glia. Surface area (SA) represents the number of vertical cortical columns, which act as information-processing units. Due to differences in developmental course, CT and SA should be considered separately when investigating brain morphology.

To elucidate relations of global brain structure with psychosis progression, normative trajectories of CT and SA have been compared with those at CHR-P who develop a psychotic illness. One of the earliest studies investigating progressive changes in the cortex among youth at CHR-P found that those who developed a psychotic disorder exhibited a steeper rate of volume loss among prefrontal regions when compared to healthy comparisons (Sun et al., Reference Sun, Phillips, Velakoulis, Yung, McGorry, Wood, Van Erp, Thompson, Toga, Cannon and Pantelis2009). This observed volumetric loss was driven by changes in CT. Both NAPLS cohorts demonstrated that psychosis progression is associated with a steeper decline in CT (Cannon et al., Reference Cannon, Chung, He, Sun, Jacobson, Van Erp, McEwen, Addington, Bearden, Cadenhead, Cornblatt, Mathalon, McGlashan, Perkins, Jeffries, Seidman, Tsuang, Walker, Woods and Heinssen2015; Collins et al., Reference Collins, Ji, Chung, Lympus, Afriyie-Agyemang, Addington, Goodyear, Bearden, Cadenhead, Mirzakhanian, Tsuang, Cornblatt, Carrión, Keshavan, Stone, Mathalon, Perkins, Walker, Woods, Powers and Cannon2023). Contrasts of mean annualized rates of thinning between youth who converted versus not highlighted right prefrontal regions (i.e., right superior frontal, middle frontal, and medial orbitofrontal regions) as associated with progression to a psychotic disorder (Cannon et al., Reference Cannon, Chung, He, Sun, Jacobson, Van Erp, McEwen, Addington, Bearden, Cadenhead, Cornblatt, Mathalon, McGlashan, Perkins, Jeffries, Seidman, Tsuang, Walker, Woods and Heinssen2015), partially overlapping with regions implicated in significant associations between LES and reduced CT (Aberizk et al., Reference Aberizk, Collins, Addington, Bearden, Cadenhead, Cornblatt, Mathalon, McGlashan, Perkins, Tsuang, Woods, Cannon and Walker2022). Furthermore, in the last NAPLS cohort, regions within the medial orbitofrontal and superior frontal cortices, left caudal anterior cingulate, and left lateral temporal cortex demonstrated excessive thinning among people at CHR-P who converted relative to healthy participants and people who did not convert (Collins et al., Reference Collins, Ji, Chung, Lympus, Afriyie-Agyemang, Addington, Goodyear, Bearden, Cadenhead, Mirzakhanian, Tsuang, Cornblatt, Carrión, Keshavan, Stone, Mathalon, Perkins, Walker, Woods, Powers and Cannon2023).

Some of the observed structural differences between healthy individuals and those at CHR-P could reflect neural substrates of PSDs and likely vary in their origins, resulting from both environmental and genetic influences. SNVs that associate with SCZ also differentially associate with CT and/or SA (Cheng et al., Reference Cheng, Frei, Van Der Meer, Wang, O’Connell, Chu, Bahrami, Shadrin, Alnæs, Hindley, Lin, Karadag, Fan, Westlye, Kaufmann, Molden, Dale, Djurovic, Smeland and Andreassen2021). As described, environmental influences are known to impact neurodevelopment, particularly during the fetal and childhood periods, and cumulative stress over the lifespan is linked with reduced CT among individuals at CHR-P and healthy comparisons (Aberizk et al., Reference Aberizk, Collins, Addington, Bearden, Cadenhead, Cornblatt, Mathalon, McGlashan, Perkins, Tsuang, Woods, Cannon and Walker2022).

Despite group-level differences among these global cortical measures, SCZ patients demonstrate greater variability in average CT and overall SA compared to controls (Alnaes et al., Reference Alnæs, Kaufmann, van der Meer, Córdova-Palomera, Rokicki, Moberget, Bettella, Agartz, Barch, Bertolino, Brandt, Cervenka, Djurovic, Doan, Eisenacher, Fatouros-Bergman, Flyckt, Di Giorgio, Haatveit and Jönsson2019). Further, even with consistent levels of CT and SA, significant deviations from the norm are diverse in terms of cortical location. At most, 18% of people with SCZ demonstrate “infra-normal” deviations (<5th percentile within normative distributions of morphology) in CT in any single region (Lv et al., Reference Lv, Di Biase, Cash, Cocchi, Cropley, Klauser, Tian, Bayer, Schmaal, Cetin-Karayumak, Rathi, Pasternak, Bousman, Pantelis, Calamante and Zalesky2021). Similarly, among CHR-P youth, “infra-normal” CT deviations were found for 5% of participants in any single region (Allen et al., Reference Allen, Baldwin, Bartholomeusz, Chee, Chen, Cooper, De Haan, Hamilton, He, Hegelstad, Horton, Hubl, Klaunig, Koppel, Kwak, León-Ortiz, Loewy, McGorry and Frangou2024). Again, differences in brain morphology could result from differing environmental and/or genetic influences. Overall, these findings are consistent with other evidence of heterogeneity in psychotic disorders.

The hippocampus is a brain region that is most consistently observed to be characterized by reduced volume in PSD and mood disorders, and it also is among the most sensitive regions to adverse environmental effects (Khan et al., Reference Khan, Geiger, Wiborg and Czéh2020; Knight et al., Reference Knight, McCutcheon, Dwir, Grace, O’Daly, McGuire and Modinos2022; Modinos et al., Reference Modinos, Richter, Egerton, Bonoldi, Azis, Antoniades, Bossong, Crossley, Perez, Stone, Veronese, Zelaya, Grace, Howes, Allen and McGuire2021). Several reports on PSD describe reductions in GABAergic interneurons (Benes & Berreta, Reference Benes and Berreta2001; Konradi et al., Reference Konradi, Yang, Zimmerman, Lohmann, Gresch, Pantazopoulos, Berretta and Heckers2011), which comprise approximately 20% of all cortical neurons and 10% of hippocampal neurons in the mature human brain (Egerton et al., Reference Egerton, Modinos, Ferrera and McGuire2017; Vazquez et al., Reference Vazquez, Fukuda and Kim2018). This has been observed in both affective and nonaffective PSDs, which led to speculation that interneuron cell loss may be shared factor, like chronic stress, with nonspecific influence on psychiatric disorders (Benes & Berreta, Reference Benes and Berreta2001). We still do not know whether impaired GABAergic neurotransmission in PSDs is better characterized as losses of inhibition (e.g., cell loss) or disorganization of inhibition (e.g., reduced expression of relevant proteins) (Glausier & Lewis, Reference Glausier and Lewis2017; Konradi et al., Reference Konradi, Yang, Zimmerman, Lohmann, Gresch, Pantazopoulos, Berretta and Heckers2011). But evidence suggests that the principal enzyme responsible for GABA synthesis is downregulated in the hippocampus of those with SCZ (Egerton et al., Reference Egerton, Modinos, Ferrera and McGuire2017; Konradi et al., Reference Konradi, Yang, Zimmerman, Lohmann, Gresch, Pantazopoulos, Berretta and Heckers2011) and depression (Prévot & Sibille, Reference Prévot and Sibille2021).

In addition, individuals at CHR-P who later developed a PSD demonstrated reduced efficiency across several functional brain networks as indexed by an increase in path length between brain regions and increase in node (predefined region) diversity, which suggests declines in functional segregation (Cao et al., Reference Cao, Chung, McEwen, Bearden, Addington, Goodyear, Cadenhead, Mirzakhanian, Cornblatt, Carrión, Mathalon, McGlashan, Perkins, Belger, Seidman, Thermenos, Tsuang, van Erp, Walker, Hamann and Cannon2020). Moreover, those reductions in efficiency were driven by increases in path length within the default mode network (DMN), a bilateral group of brain regions showing heightened FC in the absence of externally focused cognitive demands (Buckner & DiNicola, Reference Buckner and DiNicola2019), which suggests that functional aberrations in the DMN drive global inefficiencies in neurotransmission as psychosis develops. These findings suggest that the prodrome to psychosis may be characterized by a substantial reorganization of functional brain networks (Cao et al., Reference Cao, Chung, McEwen, Bearden, Addington, Goodyear, Cadenhead, Mirzakhanian, Cornblatt, Carrión, Mathalon, McGlashan, Perkins, Belger, Seidman, Thermenos, Tsuang, van Erp, Walker, Hamann and Cannon2020) consistent with aberrations in a context-independent state (Cao et al., Reference Cao, McEwen, Forsyth, Gee, Bearden, Addington, Goodyear, Cadenhead, Mirzakhanian, Cornblatt, Carrión, Mathalon, McGlashan, Perkins, Belger, Seidman, Thermenos, Tsuang, van Erp, Walker and Cannon2019). Critically, the DMN has also been described as an apex transmodal network responsible for coordinating integrative cognitive processes (Buckner & DiNicola, Reference Buckner and DiNicola2019). This hierarchical organization putatively supports efficient neurotransmission because too many intermediaries may interfere with efficient information transfer (Bullmore & Sporns, Reference Bullmore and Sporns2009). Thus, aberrations to the function or structure of hub regions may dramatically affect neurotransmission within and between networks involving the implicated region (Gong et al., Reference Gong, He, Concha, Lebel, Gross, Evans and Beaulieu2009).

Increased cerebral blood flow (CBF) to the hippocampus is one example of such a functional aberration and has been consistently reported in studies of individuals at CHR-P (Modinos et al., Reference Modinos, Richter, Egerton, Bonoldi, Azis, Antoniades, Bossong, Crossley, Perez, Stone, Veronese, Zelaya, Grace, Howes, Allen and McGuire2021; Schobel et al., Reference Schobel, Chaudhury, Khan, Paniagua, Styner, Asllani, Inbar, Corcoran, Lieberman, Moore and Small2013). Indeed, the hippocampus is one of several brain regions known to function as a hub of large-scale brain networks (Bassett & Sporns, Reference Bassett and Sporns2017; Knight et al., Reference Knight, McCutcheon, Dwir, Grace, O’Daly, McGuire and Modinos2022; Niu & Palomero-Gallagher, Reference Niu and Palomero-Gallagher2023). To further explore aberrant hippocampal function across the psychosis spectrum, some have examined the amplitude of blood-oxygen-level-dependent (BOLD) signal time series localized to the hippocampus during rest to determine whether hippocampal hemodynamics are altered in the early stages of psychotic illness. Findings corroborate earlier reports concerning increased CBF as evidenced by observations of increased amplitude of rsfMRI BOLD signal in the hippocampus. However, those changes appear to normalize throughout the first two years of illness, which suggests that hippocampal hyperactivity is a risk factor for psychotic disorder (McGrath et al., Reference McGrath, Baskerville, Rogero and Castell2022). Similarly studies of depressed patients show increased hippocampal rCBF compared to healthy controls (Chithiramohan et al., Reference Chithiramohan, Parekh, Kronenberg, Haunton, Minhas, Panerai, Robinson, Divall, Subramaniam, Mukaetova-Ladinska and Beishon2022). So this pattern is consistent with the high levels of depression observed in CHR-P youth.

As noted, there is greater variability among SCZ and CHR-P in brain region volumes than among healthy comparison groups (Allen et al., Reference Allen, Baldwin, Bartholomeusz, Chee, Chen, Cooper, De Haan, Hamilton, He, Hegelstad, Horton, Hubl, Klaunig, Koppel, Kwak, León-Ortiz, Loewy, McGorry and Frangou2024). Greater heterogeneity has also been observed in brain networks. In a recent study, investigators obtain structural and functional MRI data on a large transdiagnostic (ADHD, ASD, bipolar, depression, OCD and SCZ) sample and found regional volumes were highly heterogeneous, involving the same region in fewer than 7% within the same diagnostic group (Segal et al., Reference Segal, Parkes, Aquino, Kia, Wolfers, Franke, Hoogman, Beckmann, Westlye, Andreassen, Zalesky, Harrison, Davey, Soriano-Mas, Cardoner, Tiego, Yücel, Braganza, Suo, Berk and Fornito2023). But notably, volumetric deviations from the norm were in regions that are part common functional circuits/networks in more than half of the patient participants. Thus varied deviations in brain structure (i e., nodes) can be involved in the same circuits/networks, and the resulting dysfunction of common circuits/networks may underlie clinical similarities among people with the same DSM diagnosis. In sum, the complex circuitry of the human brain means that a variety of brain regional abnormalities can perturb a single circuit. The authors conclude that “These findings challenge classical views that distinct psychiatric diagnoses are associated with dysfunction in specific circuits and suggest that each disorder is associated with complex changes that affect diverse neural systems, often transdiagnostically.”

Machine learning in research on psychosis

At its inception, the CHR-P research paradigm was viewed as optimal for developing predictive algorithms with the ultimate aim of preventing psychosis. As methodologies for data collection have advanced, and consortia have become more common, there has been a growing need for sophisticated tools to analyze large amounts of data. Machine learning (ML) has provided a tool for dealing with this challenge and has been successfully applied to the prediction of conversion to psychosis.

Instead of a priori programing, ML algorithms use a data-driven approach to detect multivariate patterns in a dataset. Often, this involves the use of supervised ML, which “learns” on a training dataset to identify patterns. Following this, the algorithm, or set of patterns generated from the training dataset, is tested on a new dataset. This process allows researchers to know whether the ML algorithm is valid and generalizable via cross validation, external validation, or prospective validation. Unsupervised ML can also be useful to detect patterns in a bottom-up fashion and form/identify clusters of individuals with similar profiles.

With data obtained from the NAPLS project and another large multisite European CHR-P study, investigators attempted to test a ML prediction algorithm that had been developed on the initial NAPLS cohort to predict psychosis in CHR-P samples (Koutsouleris et al., Reference Koutsouleris, Dwyer, Degenhardt, Maj, Urquijo-Castro, Sanfelici, Popovic, Oeztuerk, Haas, Weiske, Ruef, Kambeitz-Ilankovic, Antonucci, Neufang, Schmidt-Kraepelin, Ruhrmann, Penzel, Kambeitz, Haidl and Rosen2021). The variables entered into the prediction model were included based on preliminary analyses for predictive power, and included age, SIPS severity ratings for unusual thought content and suspiciousness, scores from a symbol coding test and a measure of verbal learning, ratings of past year decline in social functioning, and family history of psychosis in a first-degree relative. Using the combined datasets of NAPLS and PRONIA, the prediction algorithm yielded indices of sensitivity and specificity of 68% and 66%, respectively (Koutsouleris et. al., Reference Koutsouleris, Dwyer, Degenhardt, Maj, Urquijo-Castro, Sanfelici, Popovic, Oeztuerk, Haas, Weiske, Ruef, Kambeitz-Ilankovic, Antonucci, Neufang, Schmidt-Kraepelin, Ruhrmann, Penzel, Kambeitz, Haidl and Rosen2021). It is anticipated that these indices will improve in future studies with larger samples and additional and/or improved predictor variables, especially developmental trajectories.

Another line of ML research has focused on language samples, which have the benefit of not requiring in-person clinical assessment. One notable area in CHR-P research employing ML has been language abnormalities. Some abnormalities in language reflect thought disorder (TD). Using ML, it has been possible to identify subtle forms of TD present before psychosis onset in CHR-P. A landmark study was able to predict conversion to psychosis using baseline CHR-P language samples with near perfect accuracy based on syntactic complexity and semantic coherence (Bedi et al., Reference Bedi, Carrillo, Cecchi, Slezak, Sigman, Mota, Ribeiro, Javitt, Copelli and Corcoran2015). Another study examined semantic density within language and predicted conversion with 90% accuracy (Rezaii et al., Reference Rezaii, Walker and Wolff2019).

Beyond language, ML algorithms are now employed for other modalities and biomarkers. One study using ML on structural MRI data predicted with 70% accuracy (after cross validation) whether individuals experienced continuous or episodic phases of illness over a 6-year period (Mourao-Miranda et al., Reference Mourao-Miranda, Reinders, Rocha-Rego, Lappin, Rondina, Morgan, Morgan, Fearon, Jones, Doody, Murray, Kapur and Dazzan2012). Subsequent research has used ML to identify subgroups with greater homogeneity in brain morphology (Koen et al., Reference Koen, Lewis, Rugg, Clementz, Keshavan, Pearlson, Sweeney, Tamminga and Ivleva2023). Two recent studies of large samples of SCZ patients used ML to identify inflammatory subgroups that differed in inflammatory marker levels (Enrico et al., Reference Enrico, Delvecchio, Turtulici, Aronica, Pigoni, Squarcina, Villa, Perlini, Rossetti, Bellani, Lasalvia, Bonetto, Scocco, D’Agostino, Torresani, Imbesi, Bellini, Veronese, Bocchio-Chiavetto and Gennarelli2023; Lalousis et al., Reference Lalousis, Schmaal, Wood, Reniers, Cropley, Watson, Pantelis, Suckling, Barnes, Pariante, Jones, Joyce, Barnes, Lawrie, Husain, Dazzan, Deakin, Weickert and Upthegrove2023). Because ML can integrate multiple streams of data from different domains at multiple time points, it may prove to be an indispensable tool in identifying etiologic subtypes in psychosis. This would involve using datasets representing multiple domains that are presumed to reflect etiology subtype determinants (e.g., genetics, inflammatory processes, prenatal exposures, trauma and/or stress and neighborhood characteristics) as well as phenotypic measures (brain morphology, symptoms, developmental course).

Summary and directions for future research

This article was not intended as a comprehensive review of research on PSDs, but rather an effort to highlight recent findings, including many from our research group, which have implications for our conceptual and investigative approaches to the origins of PSDs. The findings we described lend credence to the concepts of equifinality and multifinality. Three conclusions which should continue to modify scientific conceptualizations are apparent; (1) the developmental pathways to psychosis are varied and often transdiagnostic, (2) a range of environmental (social and biological) and genetic/epigenetic factors are relevant to vulnerability and psychiatric outcomes, and (3) the effects of the environment begin prenatally and cut across levels of analysis, such that even distal environmental factors (neighborhood) can influence the expression of the neurobiological substrates underlying vulnerability.

It is likely that the varied transdiagnostic trajectories to psychosis hold important information about etiology. Some PSDs are preceded by early onset disorders, such as ASD and developmental delays, others are preceded by ADHD and/or mood disorders that arise in middle childhood, some trajectories do not deviate from the norm until adolescence when mood disorders symptoms arise, and some arise with minimal prodromal signs. As described, we now know that developmental trajectories vary as a function of genetic factors, such as inherited risk genes, de novo CNVs, and epigenetics. Thus, considering transdiagnostic developmental trajectories as a variable that has implications for etiology will accelerate the discovery of their determinants.

Second, multiple levels of analyses are needed to understand relationships across domains. It is now clear that socioenvironmental factors influence brain development and can confer neurobiological vulnerability. While experimental animal research established this many decades ago, it is only recently that we have compelling evidence of this from research on humans. Thus we will need to incorporate both environmental and biological factors, including genetic factors, in our research and conceptual models of etiology. Finally, our conceptualization of the environmental influences must be expanded longitudinally to include the prenatal environment. Moreover, the effects of the environment cut across levels of analysis, such that even distal environmental factors (neighborhood) can influence the neurobiological substrates underlying vulnerability.

The field is fortunate that there are now many large longitudinal datasets available to researchers. In Europe, these datasets include measures of the distal and proximal environments, as well as life-span information from health databases. In the US, the Adolescent Brain Cognitive Development Study (ABCDStudy.org) is a prime example. Funded by the National Institutes of Health, it is the largest US longitudinal study of brain and behavioral development. Accessible to all qualified researchers, the ABCD dataset is providing an outstanding opportunity for examining developmental trajectories leading to PSDs. After these trajectory subtypes are identified and replicated, the next step will be to link them with risk factors.

Large data sets that include information on a host of potential risk factors, ranging from genetics to geocoding, will make it possible to examine the relations among the factors that confer vulnerability and the trajectories they precipitate. For example, in the case of PSDs, it is highly plausible that there are trajectories that are distinguishable based on genetics, prenatal factors, adolescent stress-sensitivity, and/or exposures such as cannabis and inflammatory triggers. It is also likely that these trajectories differentiate individuals with respect to their response to treatment and preventive intervention. Identifying these relationships will be a critical step.

To date, few studies of psychopathology have included transdiagnostic developmental trajectory as a dependent or independent variable in research. But it is now feasible with the availability of advanced statistical methods and large datasets. There are a variety of sophisticated analytic tools that will move this process forward, including latent class analysis, hierarchical clustering, and ML. With large datasets, ML is a very promising approach that has already revealed its ability to aid in the dissection of DSM diagnostic categories into more homogenous subtypes. As described above, this has been done at the level of behavior, brain morphology, and inflammatory indices. With advances in ML and other techniques for analyzing large datasets, it will be possible to incorporate more domains and variables in our analyses and have greater opportunities for identifying etiologic subtypes of PSDs that reflect direct and interactive determinants. Working towards this goal will require a transdiagnostic approach to psychopathology research that does not confine us to a search for a single domain of etiologic factor(s) that give rise a single DSM-defined disorder preceded by a uniform developmental trajectory. Instead, developmental trajectory should be viewed as a critical variable.

We believe a transdiagnostic developmental approach offers a very promising strategy for gaining traction into the determinants of PSDs. In sum, developmental trajectories, rather than the final DSM category, should be the focus of greater attention. We thank Dr Cicchetti for calling this to our attention decades ago and we hope that research on serious mental disorders will be increasingly fruitful as more transdiagnostic perspectives and methodologies are applied.