Etiopathogenic models of psychosis spectrum illnesses must resolve four key features

Abstract

Etiopathogenic models for psychosis spectrum illnesses are converging on a number of key processes, such as the influence of specific genes on the synthesis of proteins important in synaptic functioning; alterations in how neurons respond to synaptic inputs and engage in synaptic pruning; and microcircuit asynchrony that leads to more global cortical information processing vulnerabilities. Disruptions in prefrontal operations then accumulate and propagate over time, interacting with environmental factors, developmental processes, and homeostatic mechanisms, eventually resulting in symptoms of psychosis and disability.

However, there are four key features of psychosis spectrum illnesses which are of primary clinical relevance but have been difficult to assimilate into a single model and have thus far received little direct attention: 1) The bidirectionality of the causal influences for the emergence of psychosis; 2) The catastrophic clinical threshold seen in first episodes of psychosis and why it is irreversible in some individuals; 3) Observed biotypes that are neurophysiologically distinct but both clinically convergent and divergent; and 4) A reconciliation of the role of striatal dopaminergic dysfunction with models of prefrontal cortical state instability.

In this selective review, we briefly describe these four hallmark features and we argue that theoretically-driven computational perspectives making use of both algorithmic and neurophysiologic models are needed to reduce this complexity and variability of psychosis spectrum illnesses in a principled manner.

The research highlighted in this special issue presents a compelling picture for the etiopathology of psychosis spectrum illnesses such as schizophrenia: multiple genes converge to influence the synthesis of proteins important in synaptic functioning, altering how neurons respond to synaptic inputs and engage in synaptic pruning (1,2); these synaptic changes create local microcircuit dysfunction leading to global cortical information processing vulnerabilities (3), such as disruptions in the time constant of sensory representation machinery and/or errors in higher-order prefrontal predictive coding operations (4)-(5). Prefrontal information processing vulnerabilities accumulate and propagate over time, interacting with environmental factors, developmental processes, and homeostatic mechanisms, eventually resulting in reduced prefrontal function, striatal hyperdopaminergia, symptoms of psychosis, and disability (6).

While this pathophysiologic sequence is compelling, it needs to account for four key features of psychosis spectrum illnesses which are of primary clinical relevance and thus far have been difficult to assimilate into a single model: 1) The bidirectionality of the causal influences for the emergence of psychosis; 2) The catastrophic clinical threshold seen in first episodes of psychosis and why it is irreversible in some individuals; 3) Observed biotypes that are neurophysiologically distinct but both clinically convergent and divergent; and 4) A reconciliation of the role of striatal dopaminergic dysfunction with models of prefrontal cortical state instability.

In this review, we briefly describe these four features and emphasize that emerging computational perspectives offer useful approaches to resolving and integrating them within more unified pathophysiological models. It has been argued that the non-linearity of the complex interactions across neurophysiologic levels within the brain, particularly the bidirectionality of interactions across levels, will require large data sets to capture and identify that variability (7,8). While we do not disagree with the importance of large data sets, they are unlikely on their own to resolve the complexity of psychosis spectrum illnesses. We argue that theoretically-driven computational perspectives are needed to reduce complexity and variability in a principled manner (9–11).

How do we account for the bidirectionality of the causal influences of psychosis?

It is tempting to view the pathway sketched above as causally unidirectional: e.g., genetic mutations lead to synaptic vulnerability and microcircuit asynchrony, which leads to changes in prefrontal macrocircuit functioning, which results in psychosis. However, all of these processes operate bidirectionally. Cellular firing patterns affect genetic transcription and consequential protein synthesis; experience-induced macrocircuit rewiring leads to changes in microcircuit physiology and cellular firing patterns (12,13). An individual organism’s environmental exposures and behavioral and emotional repertoire affect its sensory and cognitive processes and create experience-dependent plastic changes in representational systems that percolate back down the hierarchy to shape cellular and synaptic functions (14). Further, homeostatic plasticity must be considered, a process that allows neurons to modulate their intrinsic excitability and synaptic function in response to the intensity of their inputs; this can prevent the network from becoming (or drive a network to become) hypo- or hyperactive (15). Intrinsic excitability/plasticity mechanisms, synaptic responses, and the character and timing of inputs are linked in complex ways. These processes are tied to environmental sampling and sleep-wake cycles (16). These various causal influences differ in their weighting across individuals, such that different individuals will be best served by potentially different treatment levers.

For example, while subtle synaptic dysfunction may play an important permissive role for the development of psychosis, experienced clinicians know that the identity, beliefs, and skills the individual forms around emerging symptoms is one of the most important factors that mitigates self-stigma and disability (17). Current successful treatments for early phases of psychosis rely heavily on teaching a person adaptive meta-cognitions that create new appraisals of environmental stimuli (18,19). Successful psychotherapy treatment creates compensatory and/or restorative cognitions and behaviors (new macrocircuit configurations) that then override the aberrant perceptual and cognitive representations induced by psychosis (20). Similarly, cognitive training that increases the fidelity and speed of perceptual processing and predictive coding and induces restorative and compensatory cortical circuit activation is associated not just with improved cognition but also with reduced psychosis symptoms in early-phase illness (21,22).

Furthermore, maladaptive mental/emotional/physiological states can drive perceptual and cognitive changes, including aberrant salience, that then contribute to psychosis, presumably through enduring macrocircuit dysfunction and epigenetic changes that alter neuronal function (23–27). Prolonged sleep deprivation, childhood maltreatment, and solitary confinement, for example, are all associated with the emergence of psychosis. Conversely, adoptees at high genetic risk for schizophrenia are protected by a healthy rearing-family environment but show psychiatric problems when reared in dysfunctional families (28). Cultural milieu is also related to long-term outcomes, with some developing countries faring better than industrialized nations (29). In sum, a wealth of clinical and epidemiologic work indicate the importance of environmental factors in the genesis of psychosis spectrum illnesses, while basic science has firmly established that plastic changes can be driven throughout the brain based on the nature and timing of environmental inputs (30).

Two good examples of the bidirectionality of causal factors are perceptual attention and efferent copy/ corollary discharge. In terms of attention, we live in cue-rich environments providing a plethora of high-dimensional perceptual features at every moment in time. Living adaptively within these environments and responding appropriately to environmental demands requires attending to specific and relevant aspects of these high-dimensional features. Errors in perceptual processing could degrade the allocation of attention to ecologically important cues, resulting in impaired interactions with the environment. Deficits in auditory tone discrimination, for instance, are associated with impaired auditory emotion processing and poorer social interaction abilities in individuals with schizophrenia (31); while poorer social interaction abilities in turn are causally related to lower motivation and worse social and occupational outcomes (32). Intensive “perceptual attention training” to improve speed and accuracy on basic social cognition exercises (face emotion recognition, auditory prosody discrimination, eye-gaze detection) can drive enduring improvements in anticipatory pleasure and social functioning (33).

Current computational explanations of cortical function suggest that information is encoded across cell assemblies– firing patterns across the neural set– and that both perceptual categorization and memory recall occur through recapitulation of self-consistent firing patterns (34,35). Hierarchical models suggest both bottom-up signals carrying perceptual components and top-down signals carrying abstract information to enforce that consistency (36–38). Moreover, it is possible that there is a bidirectional interaction between these two conditions, such that having prefrontal cortex in a part of representation space that has not been previously aligned to world-states leads to an impaired efferent copy and slower and less accurate perceptual prediction, or that timing problems in prefrontal cortex could impair that top-down enforcement of consistent representations.

As another example, corollary discharge is a key brain mechanism of sensorimotor coordination that allows animals to distinguish self-generated signals from external inputs. This mechanism is altered in individuals with psychosis, consistent with the clinical observations that such individuals often experience difficulties recognizing the origin of their own perceptions and actions and believe they are coming from outside of the self (39,40). These deficits could arise from synaptic timing problems, leading to unexpected perceptual observations that then need to be explained– giving rise to consequential reasoning dysfunctions. Alternatively, these deficits could arise from cortical firing patterns appearing in unusual parts of the representation space, that is, patterns of activity with subcomponents encoding inconsistent representations– thus leading to an inability to predict perceptions.

Computational models of psychosis spectrum illness will need to account for the bidirectional nature of interactions among all levels of the brain’s information processing hierarchy through well-conceptualized intrinsic, experience-dependent, and homeostatic plasticity mechanisms. For example, algorithmic computational models have explored how subtle changes in plasticity timing and ion channel function produce macro-level changes in the stability of representations in feedback networks (41–43). Neurophysiological computational modeling along with slice electrophysiology has been used to show how chronic activity deprivation of hippocampal neurons drives changes in the functioning and distribution of calcium channels in order to support homeostatic plasticity (44). Well-designed basic science experiments that study perturbations at various levels of scale, supported by algorithmic and neurophysiological computational modeling, will be required to shed light on these complex interacting processes.

Why is crossing the threshold into psychosis so often clinically catastrophic?

An important feature of psychosis spectrum illnesses is that circuit-level and cognitive vulnerabilities remain clinically masked until they cascade into the physiologically and psychologically damaging state of psychosis, where an individual experiences a disconnection from reality. Epigenetic changes are aligned to the time of first psychosis (45), with evidence of both oxidative and inflammatory stress responses in many patients (46). The fact that the experience of an episode of psychosis is itself extremely stressful speaks to the bidirectionality of causality across levels-- stress-induced epigenetic and inflammatory changes will affect neural function, which changes the information processing accomplished by that neural network, which can then make it even more difficult for an individual to respond adaptively to the confusing flood of sensory and cognitive changes occurring during the early phases of illness.

Current theories suggest that the individual’s neural circuitry is able to compensate for pre-existing vulnerabilities until sufficient stressors (biological and/or socio-emotional) induce a phase-change (47). These stressors include the normal prefrontal synaptic pruning, neuroendocrine changes, and social challenges that occur during adolescence. Once the impairment in prefrontal functioning exceeds the brain’s global compensatory capacities, the result is a catastrophic disruption in perceptions and cognitions. But neurophysiological and computational models must explain why crossing this threshold can occur suddenly in some individuals, as an acute state-change, but also gradually over months in other individuals, as a slowly progressive dysfunction in brain representational systems. Models must also account for the role of concurrent oxidative stress and inflammation (46) as they seek to explain why the threshold into psychosis at times represents a neurocognitive and psychological Rubicon from which it can be difficult to fully recover. Finally, models must address the dynamics of some of the neurophysiologic differences that have been observed in early-stage psychosis compared to later stages– such as increased rather than decreased resting-state prefrontal connectivity and gamma-band power, consistent with increased glutamatergic activity/increased cortical excitation in the earliest phases of illness (48,49).

Algorithmic attractor network models in which cortical representations suddenly shift into new basins of attraction, which are then learned into place, may provide important starting points for addressing these issues (41,50,51). If a physiological “hyperexcitation” threshold is crossed in the at-risk individual, in which aberrant perceptions, cognitions, and salience are now being generated, then this state of aberrant network configurations becomes a state of aberrant network and spiking synchrony– in other words, a period of aberrant Hebbian associative learning for the brain. Theories which suggest that the delusions (and possibly hallucinations) within the first episode of psychosis arise from sudden transitions to a different part of the neural representation space, but that subsequent plastic changes lock that new firing pattern in as an attractor state, may provide the start to an explanation for both the arbitrary nature of the first psychosis symptoms and also why relapses tend to “fold in” new events into the original set of aberrant beliefs (50,52,53). Models of impaired neural ensemble tuning that indicate the role of homeostatic processes in response to increased cortical excitation can help to explain the progressive impairment in cortical structure and function observed as illness continues (6). Models of the emergent effects of E/I imbalance that take into account pre-existing cortical functional hierarchy demonstrate how widespread microcircuit dysfunction can induce network-preferential disruptions across cortical sectors (54). However, much work remains to be done in order to understand why the hypothesized abnormal attractor states and the putative homeostatic/allostatic processes (synaptic downscaling, programmed synaptic elimination, subsequent macrocircuit dysfunction) are so irreversible in many individuals.

A detailed understanding of the dynamics that underlie these “dysplasticity” processes will be critical in order to develop meaningful and well-timed preemptive interventions and perhaps some day cures for these illnesses. It is clear that– if the timing of neuronal ensemble firing is misaligned– the brain will experience difficulties in adaptively representing the state of the world in a manner which supports efficient predictive coding (55). What is less clear is why, how, and when–during the physiological progression of these activity-timing and localized representational impairments–the brain transitions into such an altered global state. Answers will require a combination of large longitudinal human behavioral and neurophysiologic data sets, along with basic science perturbations of microcircuit and mesocircuit functioning, and algorithmic models of the dynamics of phase changes or state changes across levels of scales in neural systems.

How can different biotypes produce both overlapping and divergent clinical presentations?

Clinicians distinguish among individuals with a psychosis phenotype using diagnostic criteria that highlight different symptom features–schizophrenia vs. schizoaffective disorder vs. bipolar disorder with psychosis. Cluster analyses on EEG, MRI and behavioral data on a large sample of individuals with the psychosis phenotype reveal three biologically-defined biotypes, but each distinct biotype consisted of a mixture of individuals carrying the three different clinical diagnoses (56). While it is not surprising that different ensembles of neurophysiological signatures are all associated with psychosis spectrum illness, it is puzzling that each biotype gave rise to a wide range of behavioral presentations. Additional analyses on this same transdiagnostic group of individuals have derived dimensionality-reduced symptom axes that map onto distinct, reproducible brain maps with potential individualized clinical applications (57)—highlighting yet again the complex heterogeneity of the clinical syndromes.

As a more specific example, the B-SNIP Biotype 1 individuals showed lower intrinsic EEG activity and lower ERP responses to sensory stimuli, along with significant and widespread reductions in gray matter volume (58), consistent with widespread deficits in synaptic plasticity mechanisms and accelerated aging (see (59) for a relevant review). But Biotype 2 individuals showed high levels of intrinsic EEG activity and higher reactivity to sensory stimuli, with less extensive reductions in gray matter volume. This suggests a potential role for psychological stress and trauma, which has been associated with intrinsic sensory hyperactivity and bottom-up inhibition deficits (60) as well as neuroinflammation (61) (B-SNIP individuals with neuroinflammatory markers show less gray matter loss/ possible gray matter thickening (62)). Biotype 2 also recalls the sensory hyperexcitability and prefrontal hyperconnectivity reported in early phases of psychosis (49,63), along with potential hyperplasticity processes (64). Yet both Biotypes 1 and 2 showed reduced cognitive control performance, indicating that the processes that support the use of context by modulating perceptual systems may be impaired in different ways between subgroups. In other words, either abnormally high or abnormally low intrinsic cortical neural activity may compromise prefrontal functions, leading to an impaired ability to represent task-relevant stimuli and engage in efficient predictive coding. However, Biotype 1 may represent a more relentlessly neurologically progressive form of illness, while Biotype 2 may be more static in nature, though perhaps no less functionally impaired. Finally, Biotype 3 showed normal cognitive control, only modestly impaired cortical neuronal activity, better-than-normal visual orienting, and reductions in anterior limbic structures – suggesting less involvement of mechanisms related to cortical signal-to-noise ratio, less primary widespread synaptic dysfunction or neuroinflammation, and possibly a greater role for subcortical circuit vulnerabilities.

A primary challenge, then, is to understand how significantly different underlying neurophysiologic processes drive both similar and divergent clinical phenomena. Perturbations in neural macrocircuit function can often show an “inverted-U” shape in terms of behavioral outputs, whereby both too little and too much of an underlying physiological process produce maladaptive behavior, though for different reasons and depending on the pre-existing baseline physiology (51,65). Many models have suggested that such behavioral impairments could arise from imbalances in the excitatory and inhibitory circuits of attractor networks (41,43,66), but there are many pathways to create that imbalance, some of which would reduce the overall firing rate of the excitatory neurons (perhaps below the level of optimal neural stability) while others would increase that overall activity (perhaps above the level of optimal neural stability) (42,43,67). The key question, however, is how this range of plausible physiological and computational perturbations– which converge on a handful of maladaptive behavioral and cognitive outputs observed in psychosis– then diverge to produce overlapping clusters of widely varying clinical phenotypes in terms of symptom profile, illness trajectory, and treatment responsiveness. So far there has been no one-to-one mapping of macrocircuit physiological disturbances to symptom expression. This may be due to as-yet-unidentified latent variables that underlie these macrocircuit disturbances and that interact in complex ways (10,68,69). Such latent variables likely include the prior history of the organism, including both pre-existing neural system configurations as well as compensatory and adaptive representational processes, intrinsic resilience factors, and environmental factors (70).

Even the most elegantly and reliably defined computational and neurophysiological biotypes will be of limited use if clinicians cannot translate them into the experience and needs of a given individual. Future algorithmic models, built on longitudinal human data sets, will need to be developed that fold in the prior characteristic of the information processing system, as well as the development of compensatory mechanisms and agent-environment interactions, in order to fully address what is observed in the clinic. The distinction between the propensity to the state of psychosis vs. a specific clinical diagnosis or syndrome–such as schizophrenia or schizoaffective disorder–will be an important challenge for models that seek to explain both heterogeneity and equifinality.

What is the relationship of striatal dopamine to cortical dysfunction?

Finally, many years of evidence point to an important role for striatal hyperdopaminergia in the pathophysiology of many (though not all) psychosis spectrum illnesses (71). The transition to a first episode of psychosis is associated with elevated striatal dopamine synthesis and clinicians primarily treat psychosis with medications that reduce striatal dopamine via D2-receptor blockade (72). In contrast, many neurophysiological models–such as those focused on dysfunction in glutamatergic signaling, GABA-ergic interneurons, or microglial-mediated synaptic pruning–emphasize the role of cortex in the genesis and perpetuation of psychosis (73–75). Similarly, most computational models of psychosis begin from the hypothesis of changes in cortical function, particularly in the ability to form and maintain stable attractor states representing reliable information about the world (41–43,50,51). In contrast, computational models of dopaminergic function in striatum are based on reward-prediction error (RPE) signals and reinforcement-learning and complexifying variations therein. Recent data have shown that dopamine modulates the precision weighting of unsigned prediction errors in prefrontal cortex, and that individuals with early psychosis demonstrate abnormal precision-weighting that is associated with symptom severity (76). However, the role for striatal hyperdopaminergia in relation to cortical dopaminergic and glutamatergic function in psychosis is an open question, as highlighted by findings that in first-episode individuals, striatal dopamine synthesis capacity shows an inverse relationship with anterior cingulate glutamate concentrations (77).

At the cellular and circuit levels, striatal dopamine levels change the excitability and plasticity of corticostriatal interactions (78). These cellular mechanisms are demonstrated to affect the behavioral expression of previously learned values or associations and the learning of new values. While computational formalizations of dopamine have emphasized its role in learning by signaling RPEs that drive plastic changes in the circuit, there is a stronger recognition that dopamine provides dual contributions to both ongoing and future computations of decisions (79–81). Recent studies indicate that dopamine signals vary as a function of the striatal target (82–85) and report its role in salience-attribution (84) and enhancing learning of more generalized actions rather than just rewarding ones (85,86). But an important gap in the literature is how regional changes in dopamine shape the expression of ongoing adaptive processes, such as risk preference (87), inferring agency (83), or perceptual confidence (88). These region-specific ‘online’ effects also interact with persistent behavioral effects of dopamine that reinforce executed actions or thoughts by shaping corticostriatal plasticity. If these circuit-interactions are perturbed, they may promote the learning of potentially maladaptive behaviors that include incorrectly assessing risk, aberrant perceptual confidence, and dysfunctional agentic control.

Striatal dopaminergic contributions to psychosis may especially relate to cognitive frontostriatal loops that connect prefrontal cortical regions to the dorsal striatum (89,90). In rodents, the dorsomedial striatum receives input from the prefrontal cortex (91,92), whereas the analogous caudate nucleus in primates is strongly connected with rostral premotor cortical areas and lateral aspects of prefrontal cortex (93). These sub-circuits are critical for planning and learning how our actions affect the world (i.e. action-outcome or agency learning), recognizing changes in world-state, and extinction processes (94–97), all of which include representing and monitoring state-representation changes (98). Our understanding of regional dopaminergic influences on these processes and how they become dysregulated in psychosis remains incomplete. One possibility is that striatal dopamine’s role in shaping the arbitration process during performance may be spatially and temporally segregated from error signals that provide policy evaluation (83). In particular, dopamine bursts during the feedback-epoch encode RPEs, but persistent ramps during anticipation and performance epochs co-vary with motivational vigor and risk preference (99,100). Moreover, the striatal formation, and the dorsomedial striatum in particular, receive regionally tailored dopamine inputs across both epochs: feedback-epoch error signals arrive as dopamine waves and performance-epoch ramps are expressed in form of evolving regional bumps of dopamine (83). Such temporal and regional restriction of dopamine decision-signals are hypothesized to be facilitated by striatal cholinergic interneurons that pause at reward to provide a ‘plasticity window’ (78,101,102), and themselves display regionally complex patterns of activation across the striatum (103). While yet to be fully elucidated, these functional interactions may dictate how striatal dopamine levels (hypo- or hyperdopaminergia) and regional coordination (spatiotemporal waves, ramps, and transients) could refine cortical belief representations that support behavioral learning and performance. Finally, in addition to striatal dopamine levels affecting cortical dynamics, descending prefrontal projections into the striatum can directly modulate the magnitude and coordination of striatal dopamine signals (104)); these twin mechanisms may both break down in psychosis spectrum illnesses (89,90,104–106).

An alternative hypothesis arises from computational models in which memory-recall is viewed as an action in itself, enabling agents to either act within the world or to pause and act within memory (107–109). In fact, some of these models have suggested that which memory gets recalled can arise from processes similar to which action gets taken in an action-selection process. These models would suggest that basal ganglia circuits could guide changes in neural representations (a cognitive path) just as such circuits have been well-modeled to guide changes in one’s position in the world (an action path). However, these models have not been applied to psychosis nor to a possible role for striatum. Current action-based dorsolateral striatal reinforcement learning models–in which the D1-dependent direct-path circuit enhances an action, while simultaneously the D2-dependent indirect path circuit focuses attention by reducing selection of alternative actions–could be expanded into a memory-model in which the striatal circuit focuses reasoning pathways (110). In such a model, D1-dependent direct-path circuits would enhance cortical representational sequences while D2-dependent indirect-path circuits would focus the potential representational paths, making some paths more likely while cutting off others. Dysfunction in these representational sequences would appear behaviorally as disorganized or dysfunctional reasoning, much like what is seen in psychosis. This model aligns with work suggesting that mental recall and mental search are decision processes in and of themselves (108,109,111–114). An early model of the basal ganglia as a filter on cortical representations may provide a starting point for these hypotheses (89,90).

Importantly, the first-line treatments for psychosis are striatal D2 dopaminergic antagonists. There are two intriguing possibilities through which dysfunctions in striatal D2 pathways could cause representational dysfunctions given the memory-recall-as-action model detailed in the previous paragraph. First, hyperdopaminergia could cause an over-focus of a specific cortical proposal or memory-recall representational journey by emphasizing D1 over D2 processes (i.e. dopamine excitability effects), and not allowing the consideration of alternative actions, alternative memories, or alternative representational justifications and explanations, which would manifest as reasoning biases or delusions that were hard to dislodge. Such a dopaminergic modification of the gating operation is computationally analogous to inflating the decision-confidence by reducing network entropy (level of disagreement between the actions driven by the D1 and D2 pathways) during the arbitration epoch. Implementationally, because this decision process evolves in topographically organized, parallel frontostriatal circuits, the hyperdopaminergia-induced behavioral over-focusing, or high-confidence belief state (which D2 antagonists alleviate) may be supported by distinct spatiotemporal signatures of striatal dopamine patterns and accompanying D1/D2 medium spiny neuron recruitment. A clearer understanding of these neuromodulator and circuit states may be critical for refining therapies. Second, the transition into the first episode of psychosis is characterized neurochemically by elevated striatal dopamine synthesis capacity, and clinically by an initial state of confusion and fleeting/attenutated symptoms that then consolidate into a specific mental explanation and/or aberrant sensory experiences (72). These observations hint at a sequence of changes in gating/arbitration operations due to persistent hyperdopaminergic states: the presynaptic D2-related striatal dysfunction could disrupt normal cortical attractor dynamics, leading to unfocused representational changes that then take novel trajectories and interact further with hyperdopaminergia to become over-focused and hard to dislodge, as seen clinically.

One alternative is that striatum exerts effects on prefrontal cortex through feedback projections mediated by cortical dopamine, in which striatal projections to midbrain dopamine structures modulates the dopaminergic inputs to prefrontal cortex, leading to prefrontal processing changes (7,71). A second alternative is that hyperdopaminergic effects on D2-receptors and the indirect path create instances of behavioral extinction that then engender new states (98). Thirdly, there is a potential role for the indirect path in the corollary discharge mechanism that has been described earlier (90,115). All of these hypotheses remain plausible within our current functional knowledge of the system. Moreover, these hypotheses are not inconsistent with each other; some or even all of them may be producing variations of psychosis spectrum illnesses, singly or in concert. They may be producing similar behavioral outcomes across groups of individuals, or they may occur serially within a single individual, accounting for the changing symptom patterns that are often seen over time. They may very well be bidirectional. Striatal function depends on the cortical states being passed into it; cortical states likely depend on the striatal decisions being made, particularly if we consider cognitive narrative as a form of semantic action–a mental path taken through the set of plausible explanations for one’s experiences of the world. Future studies that bring these cortical and striatal models together may be particularly relevant and may provide novel insights into the bidirectionality of the causes of psychosis spectrum illnesses noted at the start of this discussion.

Conclusion

Tremendous progress has been made in our understanding of the neurophysiological underpinnings of psychosis spectrum illnesses and in developing models that provide insights into their features. However, key open questions remain that are of primary clinical relevance. 1. How do higher-level inputs— cognitive, emotional, socio-cultural– interact via plasticity mechanisms with lower-level processes such as synaptic asynchrony, homeostatic plasticity, and microcircuit vulnerabilities? 2. What creates the initially catastrophic and then often irreversible nature of many psychosis spectrum illnesses? 3. How do very different neurophysiologic signatures produce similar information processing failures but then divergent clinical presentations? And, finally, 4. How do we reconcile the role of striatal hyperdopaminergia with models of psychosis as state representation changes in cortex?

No single computational model is likely to integrate answers to all of these questions. Rather, the application of models to each of these questions will point in a disciplined way to parallels across levels of neural system scale that can be the focus of future investigations. Answering these questions through additional fundamental, human, and computational studies will help to identify new treatment handles that come from a deeper understanding of the neurophysiologic contributors to illness.

Table 1:

Summary of key unresolved questions and proposed research approaches to answer them

Key Questions for Etiopathogenic Models of Psychosis	Types of Study Required	Computational Approaches
The bidirectional nature of plasticity mechanisms and how perturbations reverberate across levels of scale The interrelationship of changes in intrinsic, homeostatic, and experience-dependent plasticity processes under pathophysiologic conditions	Basic science experiments to investigate the bidirectionality of plastic changes in synaptic and network functioning as plasticity processes interact with environmental inputs. Experiments to investigate how psychosis-relevant perturbations change those processes–e.g., ketamine administration in nonhuman primates, schizophrenia- relevant genetic variants in mice, neonatal ventral hippocampal lesions in rats; see (116,117) Clinical experiments to probe learning-induced neuroplastic changes in individuals with psychosis as a function of baseline neurophysiologic signatures.	Neurophysiological models of the effect of changes in synaptic function and intrinsic excitability properties to macrocircuit representations during experience-dependent plasticity. Models of the effect of changes in macrocircuit representations and neuronal inputs on synaptic properties and cellular functioning. Models of how specific psychosis-relevant “failure modes” change (and are changed by) plasticity responses.
The catastrophic and often irreversible threshold into psychosis	Basic science experiments of the timing of micro- and macro-circuit changes in relationship to psychosis-relevant perturbations and the reversibility or irreversibilty of those changes Clinical science investigations of large longitudinal human data sets in earliest phases of psychosis illness (behavioral, electrophysiological, imaging) to capture the psychosis threshold and its aftermath.	Models of phase-change criticality within interactive models, including both attractor-network (cortical) and neurophysiological interaction (striato-cortical) models. Models to address (ir)reversibility of changes.
How different biotypes result in various information processing abnormalities that have both convergent and divergent features	Basic science experiments to determine which cellular, micro-, and macro-circuit perturbations can produce differential biotype observations. Clinical science investigations of longitudinal human data sets to identify causal latent variables relating neurophysiologic signatures to cognitive and clinical phenotypes	Models to address how different cellular and micro-circuit observations can produce the same macro-circuit and behavioral abnormalities. Models to identify subtle behavioral consequences of those biotypes.
The relationship between cortical dysfunction and striatal dopaminergia in the genesis and trajectory of psychosis	Basic science experiments to determine how changes in striatal dopamine synthesis affect cortical representational processing, and the inverse. Clinical examinations of the early time course of striatal and cortical changes in relationship to key cognitive capacities and symptom expression (pharmacological fMRI and PET studies(76))	Models that include both striatal and cortical components, their interactions, and how the circuit functionality changes with dopaminergic consequences.