Unmasking schizophrenia: synaptic pruning in adolescence reveals a latent physiological vulnerability in prefrontal recurrent networks

Theories of the pathophysiology in schizophrenia are converging on the synapse. This special issue collects papers from leading researchers that suggest a hypothesis that multiple factors spanning many levels lead to synaptic dysfunction in recurrent circuits in prefrontal cortex. The hypothesis (Figure 1) suggests that the defect within prefrontal cortex is established during early development, but remains latent through childhood. Pruning of excitatory synapses during adolescence, resulting in lower synaptic density in persons with schizophrenia, triggers emergence of symptoms. The lower synaptic density results in deficits in cognitive processes, including working memory, because over-pruned neural circuits cannot support the relevant computations (1). Additionally, excessive prefrontal pruning leads to striatal hyperdopaminergia through increased activation of midbrain dopamine neurons by prefrontal input. The decreased synaptic density following adolescent pruning could be due to genetic or epigenetic effects that lead to an inability to build stable synapses or excessive pruning of established synapses. The excessive pruning could follow from over-active pruning mechanisms, or decreased excitatory synaptic activity due to synaptic changes local to prefrontal cortex or prolonged decreases in thalamic input across development. All of these factors may converge to alter spike timing dynamics in prefrontal recurrent circuits in adolescence, reducing synchronous neural activity, accelerating synaptic disconnection via activity-dependent mechanisms (2). Further, synaptic pruning is mediated by components of the immune system, and in-utero immune challenge may affect adolescent pruning. Decreased synaptic activity can also lead to increased pruning as pruning removes weak synapses. Thus, multiple effects converging on the synapse could lead to a prefrontal cortex with fewer excitatory synaptic connections, and this decreased recurrent connectivity could lead to cognitive deficits, and increased striatal dopamine and psychosis.

Figure 1.

Proposed sequence of pathogenic events leading to schizophrenia synthesizing some key points from this Special Issue. 1. In utero: A latent physiological and circuit vulnerability, that is influenced by genetic as well as epigenetic variation, and is exacerbated by maternal immune activation, is embedded in prefrontal local circuits during neural development. 2. Adolescence: Synaptic pruning drives vulnerable prefrontal local circuits past an E/I (excitatory/inhibitory) threshold, disrupting synchronous spiking as well as persistent activity mediated by recurrent connections dependent on NMDAR synaptic mechanisms in persons predisposed to schizophrenia, thereby unmasking the latent circuit vulnerability. Altered firing patterns in neurons and synchrony in local circuits engage activity-dependent plasticity mechanisms that could drive further local circuit disconnection. 3. Psychosis: Disruption of prefrontal local circuits leads to altered prefrontal output that either directly or indirectly increases activity in midbrain dopamine neurons (SNc, substantia nigra pars compacta). Resulting increase in dopamine release in the dorsal striatum alters reinforcement learning in striatal circuits. Altered feedback of basal ganglia through thalamus to PFC disrupts learned cortical attractor states in PFC, leading to emergence of positive symptoms.

Genetic and epigenetic factors increase risk.

In this Special Issue, Girdhar, Roussos and colleagues (3) summarize findings that suggest that epigenetic modifications of the genome play a causal role in schizophrenia. Epigenetic modifications include heritable patterns of DNA methylation and modification of histone proteins that put chromatin in an ‘open’ state associated with active transcription or a ‘closed’ state that blocks transcription. Analysis of open chromatin regions revealed enrichment of GWAS schizophrenia risk loci, and differential analysis by cell type revealed that open chromatin regions in glutamatergic neurons were most enriched in risk loci. An important question going forward will be to understand the factors, genetic or environmental, that lead to the differential epigenetic modifications.

In-utero developmental insults affect adolescent prefrontal circuits.

In this Special Issue, Dienel, Schoonover, and Lewis (4), and Hansen, Bauman and colleagues (5) review evidence that circuit vulnerabilities that lead to schizophrenia may be established early. However, they do not disrupt function of prefrontal networks until synaptic density falls below a critical threshold as synapses are pruned in adolescence. There is evidence that cognitive deficits in schizophrenia are present long before emergence of symptoms (4). Changes in somal size, dendritic arborization, and dendritic spines of cortical pyramidal neurons may occur in utero (4), rather than later in adolescence, although the balance between early and late changes remains to be determined. Hansen, Bauman and colleagues (5) build on this by reviewing evidence that maternal immune activation (MIA) disrupts development of prefrontal cortex increasing schizophrenia risk, perhaps by interacting with genetic risk. In a nonhuman primate model of MIA they found that offspring later exhibit disrupted social interactions and differences in performance on cognitive tasks as well as reduction in the volume of prefrontal cortex. Collectively, these data implicate early neurodevelopmental events in setting the stage for schizophrenia later in life.

Synaptic pruning in late adolescence is overly aggressive.

There is accumulating evidence that synaptic pruning of cortical networks may be altered in schizophrenia. For example, one of the strongest GWAS findings links schizophrenia risk to C4, within the major histocompatibility complex (6). Risk variants lead to increased expression of C4A, an immune system signaling molecule that tags dendritic spines for engulfment by microglia during synaptic pruning. In this Special Issue, Dienel, Schoonover and Lewis (4) review post mortem evidence that dendritic spines are reduced in schizophrenia, indicative either of excessive synaptic pruning, insufficient synaptogenesis, or both. They also find that spine loss is most prominent in layer III in prefrontal cortex, and absent in the deeper layers (V/VI) or other cortical areas such as primary visual cortex. Synapses within layer III likely underlie recurrent computations that support cognitive processes. In addition, in this Special Issue, Sheridan, Horng, and Perlis (7) review evidence that microglia-mediated synaptic pruning is more aggressive in schizophrenia using in vitro models. Synaptic engulfment is more aggressive when either the microglia or neurons in cell culture are derived from people with schizophrenia in comparison to control subjects. Furthermore, the degree of elevated engulfment by microglia correlates with the level of C4 expression in cultures derived from people with schizophrenia. Although these lines of evidence support excessive synaptic pruning in schizophrenia, important questions remain. In a commentary in this Special Issue, Johnson and Hyman (8) critically review the synaptic pruning hypothesis. Decreased dendritic spines in schizophrenia may result from insufficient synaptogenesis rather than excessive pruning. Genetic studies implicate elevated complement C4 expression in schizophrenia, but it is not known whether this accounts for the pattern of spine loss, or what governs C4 deposition to target specific synapses for elimination. It is often presumed that loss of synapses drives the reduction of gray matter in schizophrenia, but spines are only 1.5% of gray matter, making this unlikely. These remain important questions for future research.

Recurrent circuits in layer III of prefrontal cortex have unique properties.

In this Special Issue, Arnsten and colleagues (9) review work that shows that recurrent excitation in layer III depends preferentially on NMDAR and acetylcholine nicotinic α-7 receptors, both of which allow influx of Ca²⁺ into dendritic spines. The influx is amplified by intracellular mechanisms including the opening of voltage-gated Ca²⁺ channels (Cav_1.2) on dendritic spines, and the release of Ca²⁺ from internal stores. Both NMDAR and Cav_1.2 are implicated as causal in schizophrenia by GWAS (9), and both channels are voltage sensitive, linking the rise in intracellular Ca²⁺ to near synchronous pre- and postsynaptic depolarization (spike synchrony). Thus disruptions of spike timing (2) could accelerate disconnection of prefrontal recurrent networks by altering calcium dynamics in spines. Dienel, Schoonover and Lewis (4) further suggest that deficient synaptogenesis or excessive pruning lowers activity in recurrent DLPFC networks. GABA neuron activity levels downregulate to compensate and restore recurrent activation in prefrontal local circuits (4). However, progressive loss of recurrent excitation may exceed the compensatory capacity of these GABAergic mechanisms.

Synaptic over-pruning in DLPFC is worsened by weaker inputs from the thalamus.

Synaptic pruning is activity dependent, and therefore weakening of thalamic inputs to the DLPFC that evoke correlated activity would lead to excessive pruning during adolescence. In this special issue, Benoit, Canetta and Kellendonk (10) review evidence that resting state functional connectivity between the thalamus and prefrontal cortex is reduced in schizophrenia and adolescents at high clinical risk. Further, the magnitude of the deficit in connectivity predicts conversion to psychosis. Interestingly, transient suppression of thalamic neurons projecting to the prefrontal cortex in adolescent, but not adult, mice led to lasting changes in PFC circuits and cognitive function (10). This suggests there may be a critical period for PFC maturation during which thalamic inputs engage plastic changes at the cortical level.

E/I imbalance in PFC causes excessive dopamine release in the dorsal striatum, triggering psychosis.

In this Special Issue, Howes and Shatalina (11) put forward a theory of schizophrenia pathogenesis that suggests that changes in E/I balance in DLPFC lead to elevated dopamine release. PET imaging studies have shown that DA release is elevated in the caudate, which receives excitatory input from DLPFC (11). Howes and Shatalina suggest that elevated striatal dopamine follows from increased glutamatergic output from the DLPFC to dopamine neurons in the midbrain that project to the striatum. DLPFC inputs to the dorsal striatum may also increase dopamine release through cholinergic interneuron mediated dopamine release mechanisms. Preclinical data has shown that loss of dendritic spines in the cortex can produce an increase in the activity of pyramidal neurons projecting to midbrain dopamine neurons (12). This provides support for the hypothesis that PFC local circuit malfunction can lead to excessive dopamine release in the dorsal striatum, triggering psychosis (11).

Constraints on biological theories of schizophrenia

In this Special Issue, Vinogradov, Hamid and Redish (13) identify four phenomena that theories of schizophrenia should explain. First, causal interactions between variables leading to schizophrenia are typically bidirectional, cautioning against one-way models leading from causes to consequences. For example, models proposing disruptions progressing from genes → development → synapses → firing patterns → computations (symptoms) in schizophrenia are too simple (as many of these arrows are likely bidirectional). Second, theories of schizophrenia must explain the threshold in brain function that individuals cross as they enter into psychosis, and why for many individuals that transition is abrupt and irreversible. Once stressors exceed compensatory capacity in prefrontal circuits, the functional state of the brain shifts and psychosis emerges. Third, theories of schizophrenia must account for convergent (and divergent) symptoms among individuals of different biotypes within the diagnosis. Fourth, theories of schizophrenia must explain the role of striatal hyperdopaminergia in the disease given the evidence that increase in striatal dopamine heralds the conversion to psychosis.

Conclusion

Although much remains to be explained, consensus is building around a model of schizophrenia pathogenesis in which synaptic pruning in adolescence reveals a latent physiological vulnerability in recurrent prefrontal circuits. In this framework, it is hypothesized that vulnerable individuals exhibit cognitive deficits, but not psychosis, before adolescence because the superabundance of synapses in prefrontal circuits partially masks the latent physiological vulnerability established in early development. However, as synaptic density starts to fall late in adolescence, synaptic communication in prefrontal local circuits falls below a critical physiological threshold in vulnerable individuals. Past that threshold, pathophysiological phenomena emerge in prefrontal local circuits, that (1) worsen cognitive deficits, (2) accelerate circuit disconnection by an activity-dependent mechanism, and (3) trigger psychosis by altering striatal dopamine and prefrontal-striatal interactions. From this perspective, understanding the relationship between synaptic communication in prefrontal local circuits and the cognitive functions these circuits support, as well as how those mechanisms are disrupted by schizophrenia risk factors, become questions of central importance.