Background
Schizophrenia is a severe psychiatric disorder that is characterized by positive, negative, and cognitive symptoms, which involve distributed regions in the brain [
1]. A long-standing neurodevelopmental hypothesis has greatly contributed to our understanding of the development of schizophrenia [
2]. The dopamine hypothesis suggests that a hyper-response occurs in schizophrenia; this is consistent with the mechanism of current antipsychotic drugs, which block dopamine D2 receptors [
3,
4]. Furthermore, although no histopathological evidence has yet met the definition of neurodegeneration in schizophrenia, a neurodegenerative hypothesis has been proposed to interpret the progressive course that is observed in this disease [
5]. To date, many genes have been recognized as important in different periods of the clinical course of schizophrenia [
6]. People tend to seek clinical help when their symptoms meet the clinical diagnosis threshold; they usually show predominantly positive symptoms at this stage and respond well to antipsychotic drugs. Unfortunately, however, antipsychotic drugs cannot effectively control negative and cognitive symptoms, which are the main symptom types as the disease progresses [
7]. Heterogeneous symptoms exist in different stages of schizophrenia, and the disease progression involves different structural and functional abnormalities.
Magnetic resonance imaging (MRI) studies have provided multimodal evidence to indicate abnormalities in distributed brain regions in schizophrenia and have highlighted the network properties of the disease [
8,
9]. Many studies have investigated the detailed anatomical features of schizophrenia and provide compelling evidence of striatum-dominated atrophy and associated morphological abnormalities [
10]. Furthermore, longitudinal studies of individuals with first-episode schizophrenia have demonstrated progressive gray matter loss in this disease [
11]. Similarly, one of our previous studies demonstrated a progressive reduction in gray matter in patients with schizophrenia [
12]. The duration of the disease and antipsychotic treatments are also associated with progressive morphological changes in the brain [
13]. In resting-state functional MRI, multiple functional indicators (FIs) have been proposed to illustrate multiple-view abnormalities of brain functional activity in schizophrenia [
14‐
16], thus providing evidence of the underlying pathology of this disease. Although abnormalities are not completely consistent across indicators, these different functional features can be referred to as representations of complex pathological mechanisms in different functional dimensions. In contrast, progressive functional changes are relatively less studied in schizophrenia. There are several possible reasons for this discrepancy: (1) the difficulties obtaining long-term longitudinal functional data from large samples, (2) the existence of large functional heterogeneity in patients with schizophrenia, and (3) the relatively poor stability of functional signals relative to structural data.
It would be interesting to explore the intrinsic characteristics of the disease itself through multiple FIs. A prior meta-analysis of multiple FIs integrated findings across publications in an attempt to identify duration-associated functional features [
17]. However, correlation analyses in a cross-sectional analysis are unable to reveal specific patterns of dysfunction over different disease courses. Additionally, although a longitudinal analysis is the best way to reveal the progression of a disease, decades of longitudinal data are extremely difficult to obtain. There is thus a need to develop an alternative research framework to address this issue. Moreover, to establish macro- and micro-scale understanding of diseases, studies have linked neuroimaging profiles with gene transcriptomic data across psychiatric disorders [
18,
19]. Gray and white matter microstructures are associated with polygenic risk for schizophrenia and have been further suggested to affect the psychiatric symptoms of patients [
20,
21]. Moreover, associations between genotypes and clinical phenotypes are complex over different periods of schizophrenia [
2,
22]. Therefore, integrating neuroimaging and gene transcriptomic data may provide further insights into the pathological mechanisms of the disease.
To address these issues, we used a longitudinal-substituted approach to investigate the progressive dysfunction that occurs in schizophrenia using cross-sectional datasets. This approach was conducted using a duration-sliding dynamic analysis framework, in which multiple FIs were integrated to characterize whole-brain voxel-wise dysfunction. Affinity propagation clustering and Liptak–Stouffer approaches were then used to acquire duration-labeled specific stages of dysfunction in the disease. We hypothesized that gene expression levels would be related to the patterns of dysfunction as well as disease progression. We therefore used a regression model to identify dysfunction-associated genetic factors in distinct stages using the Allen Human Brain Atlas (AHBA) database [
23]. Moreover, enriched networks were identified based on merged genes across stages to uncover schizophrenia-related pathways using Metascape [
24]. The overall aim of the present study was to identify disease-related multi-trajectories, from symptoms to neuroimaging to genes. The study flow chart is illustrated in Additional file
1: Fig. S1.
Discussion
In schizophrenia, longitudinal studies of the prodromal phase to the first episode are important for revealing the pathological mechanisms of the disease, as well as a basis for its clinical treatment [
43]. After being diagnosed, the vast majority of people with schizophrenia experience a lifelong regimen of alternating positive and negative symptoms or negative-dominated symptoms [
1]. Subtypes of schizophrenia can be classified according to the brain functional characteristics of patients, which may lead to the application of appropriate neuromodulation [
44,
45]. Studying the progression of brain function in patients receiving regular antipsychotics after clinical diagnosis is therefore necessary to develop antipsychotics and neuromodulation technologies.
Using an innovative dynamic analysis framework for multiple FIs in the present study, we identified five progressive stages of schizophrenia in patients after diagnosis. Consistent with prior studies, a trajectory of clinical symptoms was identified from predominantly positive to predominantly negative symptoms, with five distinct stages that were classified as positive-dominated, negative ascendant, negative-dominated, positive ascendant, and negative surpassed. A hypofunction trajectory also emerged from the primary sensorimotor and visual cortices to the salience system and then to the default mode network, which contributes to abnormal external sensory gating in schizophrenia. Furthermore, a hyperfunction trajectory was identified from subcortical regions to the hippocampus and then to the dorsal frontoparietal network, which is associated with a disrupted internal excitation–inhibition equilibrium. We also identified that the regions located in primary and salience systems were strongly associated with behavior in patients at relatively early stages, whereas regions located in higher-order and subcortical systems were significantly associated with behavior in the later stages of disease progression. The trajectories of progressive dysfunction were nonlinearly associated with cerebral gradients, which were specifically concordant with the trajectories of positive symptoms. These findings further indicate that the dysfunction identified in the present study is essentially the same as that of previously reported cerebral gradients, and reflects integrated brain functional profiles. Furthermore, dysfunction-associated genetic analysis revealed that neurodevelopmental and neurodegenerative factors were associated with specific patterns of dysfunction over the disease progression and specifically highlighted the dopamine-centered synaptic system in the early disease stages. Together, these findings indicate the existence of progressive neuroimaging dysfunction with associated genetic factors in schizophrenia and will be useful for establishing an integrated framework to better understand the pathomechanisms of this disease.
A previously proposed clinical course of schizophrenia defined three representative phases (treatment, relapse, and chronic phases) after first-episode psychosis [
7]. In the treatment phase, positive symptoms are well-controlled, with a sharp slope, and negative symptoms present a gradual slope. Next, both the positive and negative symptoms ascend as the disease progresses (with a greater slope observed in positive symptoms). This is followed by a descent into the chronic phase, characterized by predominantly negative symptoms. Similarly, our work identified five stages—from predominantly positive to predominantly negative symptoms—after first-episode psychosis; this is largely consistent with the recognized clinical course of schizophrenia and supports the rationality of our analysis framework. Furthermore, in line with the conventionally defined stages of relapse, we characterized dynamic changes in the dominance of positive and negative symptoms. Symptoms from the positive-dominated, negative-ascendant, and negative-dominated stages correspond to the phase that is well-controlled with antipsychotics [
46]. The positive-ascendant stage corresponds to the relapse phase, and the negative-surpassed stage indicates the beginning of the chronic phase [
47,
48].
Consistent with previous meta-analyses of FIs [
49], the results of the current study suggest that hypofunction occurs in the primary sensorimotor and visual cortices in the early stage of schizophrenia. Previous studies have documented sensory and perceptual deficits in early-stage processing and cognitive behavior [
50] that are associated with specific clinical symptoms, such as delusions, hallucinations, and decreased voluntary motion. In the present study, decreased FIs were observed in lower-order regions (involving the primary sensorimotor and visual cortices) in this positive-dominated stage, which are associated with a potential mechanism for self-disorder in schizophrenia [
51] and suggest a disconnect with real external stimuli in the world [
52]. Significant associations between drug equivalents and FIs were also observed in visual cortices in the present study; this finding may be related to the response of positive symptoms to antipsychotics. There is substantial evidence of a hyperresponsive dopamine system in schizophrenia, and striatal dopamine pathways are strongly involved in disease progression [
7]. It is therefore unsurprising that we identified striatal hyperfunction in all disease stages in the present study; this finding is in line with the recognized hyperresponsive dopamine system in schizophrenia. Furthermore, abnormal striatal dopamine synthesis is a specific feature of the prodromal stage and worsens as the disease progresses [
53,
54]. The patterns of dysfunction were significantly spatially correlated with the distribution of striatal dopamine synthesis, release capacity, and transporters in the current study; these results further support the fundamental role of striatal dopamine pathways in the progression of schizophrenia. Striatal hyperfunction and hypofunction in lower-order regions might therefore be representative phenotypes of functional disturbance in the early positive-dominated phase of disease progression after clinical diagnosis.
The blockade of dopamine receptors can improve clinical symptoms in patients with schizophrenia, suggesting a common disruption of dopaminergic pathways [
55]. A hyperresponsive dopamine system in schizophrenia has been suggested to result from a tendency of the brain to overrespond to external salient stimuli, independent of their importance, thus allocating disturbance salience when processing signals [
56]. Similarly, the sensory gating hypothesis suggests that the pathological basis of schizophrenia is the absence of sensory gating, which leads to a large amount of irrelevant information entering the brain and disrupting its function [
57]. The salience network is involved in the process of differentiating relevant from irrelevant stimuli and assigning salience to stimuli-focused information [
58]. In schizophrenia, however, the salience process is disrupted by aberrant dopamine signaling related to irrelevant stimuli [
59].
In the present study, as the disease progressed, we identified hypofunction in the insula. This finding supports the theory of a disrupted salience-monitoring system in schizophrenia. Furthermore, insular hypofunction was accompanied by hypofunction of the default mode network, which is associated with abnormalities of ongoing information processing (including of inner reference, memory, and emotions) in schizophrenia [
60]. An abnormal functional interaction among the so-called triple network (the salience, default mode, and central executive networks) has been proposed to explain the pathophysiological dysfunction underlying psychiatric disorders [
61], and as a marker to understand the vulnerability of external and internal perceptions in patients with schizophrenia [
62]. The interaction between the salience and default mode networks has been suggested to be strongly linked to positive symptoms, but accumulating evidence also indicates that a complex association exists between triple network profiles and positive, negative, and cognitive symptoms in schizophrenia [
63]. Consistent with those of sensory gating, these findings might partially explain the cascade of impairments from lower- to higher-level functions in patients with schizophrenia.
In the present study, the orbitofrontal cortices worsened as the disease progressed. The orbitofrontal cortices play important roles in value processing and positive affect; their abnormal function might therefore induce apathy and lack of effect, which are typical clinical symptoms of schizophrenia. Thus, our findings are in line with the recognized aberrant motivation and reward-based learning that occur in schizophrenia [
64].
As the disease progresses, negative symptoms ascend to be dominant after a phase of well-controlled symptoms. In the current study, marked hyperfunction was observed in the thalamus, cerebellum, and hippocampus as the disease progressed. The striatal–thalamocortical network is important for integrating complex information and has been associated with cognitive and emotional deficits in patients with schizophrenia [
65]. Notably, the dopaminergic system is not the only aberrant system in schizophrenia; its modulatory γ-aminobutyric acid (GABA)ergic and glutamatergic synaptic systems also contribute to the disease. The hippocampus plays a crucial role in driving the dopaminergic system and contributes to the underlying pathophysiology of schizophrenia [
2,
66,
67]. Increased hippocampal glutamate function is associated with increased striatal dopamine function in schizophrenia, as well as with dysfunction across symptom domains [
67]. Hippocampal hyperfunction might therefore be a neuroimaging phenotype that implies long-term memory impairment in schizophrenia [
66]. From a molecular perspective, a magnetic resonance spectroscopy study reported that higher GABA concentrations in the cerebellum are associated with greater behavioral impairments in patients with schizophrenia. Increased cerebellar function gradually emerges as the disease progresses, implying that cerebellar GABAergic modulation plays a role in schizophrenia. An excitatory–inhibitory imbalance in the cerebello-thalamo-cortical and striato-thalamo-cortical loops has been proposed to explain the pathology and development of schizophrenia, in which the dopamine and GABA systems make predominant contributions [
7,
68,
69]. The findings of the present study also suggest that the striatum and cerebellum may predominantly contribute to the progressive excitatory–inhibitory disruption, which is in line with the “cognitive dysmetria” theory of schizophrenia [
70]. In general, the trajectory of hyperfunction might reflect the progressive dysfunction of the dopaminergic system and its upstream GABA and glutamate modulator systems in patients with schizophrenia, which may correspond to their complex clinical symptoms.
Genome-wide association studies have identified more than 50 genes associated with the pathology of schizophrenia, which are mainly implicated in neurodevelopment and the dopamine-centered synaptic system [
71]. In the current study, specific genetic features were identified in duration-labeled progressive stages. Multiple factors relating to neurodevelopment were revealed as the disease progressed, including
AH1 and the phosphoinositide 3-kinase–protein kinase B signaling pathways. Dopaminergic, glutamatergic, and GABAergic synapses were markedly enriched as early risk factors in the positive-dominated stage, thus supporting a substrate role of the dopaminergic synapse, which has been widely recognized as a cause of schizophrenia [
72]. Our work thus provides explicit evidence to support the early involvement of dopaminergic synapses and highlights the role of the excito-inhibitory synaptic system as the disease progresses. The glutamatergic and GABA pathways are crucial in the upstream regulation of dopaminergic function [
73], and their interaction contributes to the overall dysfunction that occurs in schizophrenia [
74]. It has been proposed that imbalanced inhibitory and excitatory signaling complexes are a pathomechanism of schizophrenia [
75]. These interconnected synaptic systems regulate brain plasticity and are a potential intrinsic cause of schizophrenia, as well as being associated with the complex symptoms of patients [
76,
77].
Another notable finding of the present study was the emergence of neurodegenerative factors in relatively late stages of disease progression. A neurodegenerative model has been proposed to understand the mental decline that occurs in schizophrenia, along with its progressive clinical course [
78]. Although postmortem examinations have not found neurodegenerative pathological features (such as gliosis) in schizophrenia [
79], the neurodegenerative model suggests that pathological neuronal apoptosis (which does not cause gliosis) occurs in schizophrenia, whereas necrosis does not. In support of this model, apoptosis-related genetic factors were identified in the present study. Furthermore, a progressive neurodevelopmental hypothesis has redefined the boundaries of neurodevelopment and neurodegeneration in schizophrenia, and views schizophrenia as having components of both development and degeneration [
5,
80]. Although the present findings do not directly indicate that schizophrenia is a neurodegenerative disease, they do suggest that neurodegeneration-related genes might be involved later in the disease. Thus, the neurodegenerative hypothesis of schizophrenia requires further investigation.
Limitations
There are several limitations of the present study. First, our approach has high requirements for data. The sample size needs to be relatively large and the disease course distribution of patients needs to be relatively uniform, which may limit its application. Although we validated the associated genes in the non-progressive analysis using a replication cohort from the UCLA, we were unable to identify a suitable dataset to validate the progressive analysis. Moreover, although the drug equivalent represents the stable drug intake of patients within the last year of data collection, it only provides a partial possible drug effect. The drug effects in this study should therefore be analyzed using more powerful models. We sought to explore the association between neuroimaging and the transcriptome in patients with schizophrenia in the present study. The best way to do this would be to directly link each patient’s neuroimaging data to their transcriptome data. Unfortunately, however, transcriptome data were not available from the patients with neuroimaging data. A compromise was therefore made; the case–control differences in neuroimaging were linked to healthy transcriptome data from the AHBA database, to indirectly identify genes that are associated with the disease. This association analysis of the brain imaging data of disease samples and the gene expression levels of healthy samples was likely affected by sample heterogeneity, which should be improved in future studies (e.g., by using genetic data from disease samples). Moreover, the association among neuroimaging, genetic features, and behavior was characterized by the PLSR model. This model can be overfitted and is not stable across datasets, especially when the number of samples is small; these limitations severely hinder the interpretability and generalizability of the results [
81]. Therefore, a further validation analysis with larger data samples is needed.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.