An Update on Promising Biomarkers in Schizophrenia

The role of the immune system in the pathogenesis of schizophrenia has been an area of active interest, with evidence from epidemiology; preclinical studies of maternal immune activation models; genetic studies; and studies of inflammation, oxidative stress, and the complement system converging to support a putative “immunophenotype” of schizophrenia (14). In terms of potential biomarkers related to the immune system, there have been a number of areas of investigation that warrant consideration, including acute-phase proteins (e.g., C-reactive protein; CRP), inflammatory cytokines, and markers of oxidative stress.

Since the initial Johnstone study demonstrating increased ventricular size on computerized tomography (CT) for individuals with schizophrenia when compared with controls (34), findings have generated significant interest in neuroimaging as a method to identify biomarkers and endophenotypes for schizophrenia. A variety of neuroimaging techniques, including CT, magnetic resonance imaging (MRI), functional MRI (fMRI), diffusion tensor imaging (DTI), positron emission tomography (PET), single-photon emission CT (SPECT), and magnetic resonance spectroscopy, have helped to contribute to our understanding of schizophrenia. However, common limitations in neuroimaging studies include small sample sizes, a clinically heterogeneous population, and challenges accounting for comorbidities (35). To address these issues, collaborative efforts with shared protocols across large groups are becoming increasingly common (36, 37).

Structural MRI is one of the most widely studied brain endophenotypes in psychiatry (38). For clinical high-risk individuals who ultimately develop psychosis, rates of gray matter (GM) loss over time have been of interest. The NAPLS group found that UHR individuals who converted to psychosis demonstrated a greater rate of GM loss in the right superior frontal, middle frontal, and medial orbitofrontral cortices, as well as a greater rate of expansion of the third ventricle, than those who did not convert (39). Increased rates of cortical GM loss in individuals who convert to psychosis have been replicated in other groups (40, 41). In individuals with chronic schizophrenia, Shenton et al. (42) conducted a review of 193 studies from 1988 to 2000 demonstrating ventricular enlargement, decreased volume in medial temporal lobe structures (hippocampus, parahippocampal gyrus, and amygdala), decreased volume in the superior temporal gyrus (STG), and subcortical involvement (including the cerebellum, basal ganglia, thalamus, and corpus callosum). Including data from 1998 to 2012, in a volumetric meta-analysis, Haijma et al. (43) found that individuals with schizophrenia who were taking medication had a small but significant reduction in overall intracranial (Cohen’s d=−0.17) and total brain (d=−0.30) volume, as well as decreased volume for total GM (d=−0.49), frontal lobe GM (d=−0.49), hippocampus (d=−0.52), STG GM (d=−0.58), and fusiform gyrus and an increased volume of the third and lateral ventricles. When comparing antipsychotic-naïve individuals with schizophrenia to control individuals, they found similar, though smaller, effect sizes for brain volumes, concluding that volume loss in schizophrenia was part of both a neurodevelopmental process and an illness progression (43). Then, using standardized protocols across 15 international centers, the ENIGMA consortium conducted a study of 2,028 individuals with schizophrenia and 2,540 healthy controls and measured differences in subcortical brain structures (44). For individuals with schizophrenia, they also found a slight reduction in intracranial volume (d=−0.12) and found reduced volumes in the hippocampus (d=−0.46), amydgala (d=−0.31), thalamus (d=−0.31), and accumbens (d=−0.25), while volumes were increased in the pallidum (d=0.21) and lateral ventricles (d=0.37) (44).

Schizophrenia has been characterized as a disorder of connectivity between brain regions (45), and the imaging modalities most commonly used to assess this are resting state fMRI and DTI (46). DTI, first described in the early 1990s (47, 48), is a type of diffusion-weighted MRI that enables researchers to study white matter (WM) tractography in vivo as well as neural connectivity (46). For individuals with schizophrenia, studies have shown a theme of decreased fractional anisotropy (FA) in the superior longitudinal fasiculus, cingulum bundle, unicate fasiculus, inferior longitudinal fasiculus, and arcuate fasiculus (49–51), although findings have been inconsistent across studies (46, 52, 53). The corpus callosum, the largest WM tract in the brain, is responsible for communication between brain hemispheres (54) and is also thought to be impaired in schizophrenia. Two recent meta-analyses have compared healthy controls to individuals with schizophrenia, with Zhuo et al. (55) finding a decrease in the FA in the both genu and splenium regions of the corpus callosum and Shahab et al. (56) finding a reduction in FA in only the genu.

DTI has also been used for illness classification purposes. With a combination of FA and mean diffusivity (MD), individuals with schizophrenia could be differentiated from controls with 96% sensitivity and 96% specificity (57), though the sample size was small and included a mixture of individuals with chronic and first-episode schizophrenia. Considering heterogeneity of the illness as a possible reason for inconsistent results, researchers have begun to divide individuals with different WM patterns into different subtypes of schizophrenia. This approach has been used to differentiate “non-deficit” and “deficit” subtypes (58), to differentiate between groups of individuals with schizophrenia not taking medication (59), and to identify biotypes of schizophrenia that correspond to clinical symptoms (60). Arnedo et al. (60) used a generalized factorization model to identify four groups of schizophrenia in which areas of low FA in certain regions corresponded to different clinical phenotypes (e.g. low FA in the genu corresponded to bizarre behavior).

PET and SPECT imaging allow investigators to better understand schizophrenia on a molecular/neurotransmitter system level. For example, PET/SPECT studies have helped to further elucidate the dopamine hypothesis in schizophrenia, one of the most enduring ideas in psychiatry (61). In a meta-analysis conducted by Howes et al. (62), individuals with schizophrenia, in comparison with controls, had increased presynaptic dopaminergic function (d=0.79) and a small elevation in dopamine subtype 2/3 (D_2/3) receptor availability (d=0.26) but no change in dopamine transporter activity. In a meta-analysis of DOPA PET studies using [¹⁸F] and [¹¹C] radiotracers, individuals with schizophrenia had a 14% increase in striatal dopamine synthesis capacity when compared with controls (63). Taken together, these findings suggest that drug development in schizophrenia should target presynaptic dopamine targets. PET studies could also be used to guide dose ranges for new treatment options (64) or help to detect diagnostic biomarkers for schizophrenia. For example, increased reactivity to amphetamine challenge is present in clinical high-risk individuals, prodromal individuals, and individuals with chronic schizophrenia (65–68). Findings from PET could be used to help determine which at-risk individuals ultimately convert to psychosis, but currently, these studies are underpowered for clinical use (69).

As mentioned earlier, various neuroimaging techniques have been used to differentiate individuals with schizophrenia from those without it. Conventional pattern classification approaches have typically used voxel-based morphometry (VBM) with general linear models to identify discriminating factors in localized regions of the brain (70). However, there is often considerable overlap between cases and noncases at the group level (71). Additionally, combining the intertwined nature of structural and functional abnormalities in schizophrenia has been a challenge, and to overcome methodological issues with traditional univariate data analyses, multivariate analyses have been used (72). In a meta-analysis using multivariate pattern analysis (MVPA) involving 1,602 first-episode and individuals with chronic schizophrenia, patients could be differentiated from normal controls (N=1,637) with a sensitivity and specificity of 80% (73). MVPA has also been used to differentiate women with schizophrenia from women without schizophrenia in a DTI study with an accuracy of 72%–88% (74). Through machine learning models, studies have also differentiated between bipolar disorder and schizophrenia using structural MRI (75, 76) and fMRI (77), with rates of accuracy around 80% or greater. Machine learning models have also been used to predict individuals at UHR of converting to psychosis, as well, with reasonable accuracy (78). Although these results are intriguing, classification experiments should be externally and independently validated (79) and include sample sizes of over 130 individuals (80). Currently, the performance for these classification systems is too low to be used in clinical practice (81). Various modalities of neuroimaging, as well as multimodal approaches, remain an active area of research to identify biomarkers and endophenotypes in schizophrenia.

BDNF is the most widely expressed neurotrophin in the human brain and is involved in a number of crucial neurodevelopmental mechanisms, including neurogenesis, neuronal differentiation, and neuronal survival, as well as synapse formation and maturation (82–84). As schizophrenia is recognized as a neurodevelopmental disorder whose pathogenesis involves alterations in neuroplasticity and synaptogenesis (85, 86), it is of no surprise that BDNF would be considered a putative biomarker. Indeed, BDNF has been found to be decreased in patients with schizophrenia (87), though this finding has either been inconsistent (88) or been found to be related to other factors, such as substance use (89). There is also some evidence to suggest that BDNF levels may increase with antipsychotic treatment in some (90), but not all (91), studies. Furthermore, postmortem studies have found decreased BDNF mRNA expression in the hippocampus (92) and prefrontal cortex (93). However, BDNF cannot be considered a disease-specific biomarker, as it has been found to be reduced in both MDD and bipolar disorder as well (94, 95).

A recent meta-analysis has provided some clarity and direction to the heterogeneity of BDNF findings in the literature (96): Fernandes et al. included 41 studies, with over 7,000 participants, that measured peripheral levels of BDNF. Importantly for its consideration as a putative biomarker, the authors note that BDNF crosses the blood-brain barrier (97) and that both serum and plasma levels correlate highly with BDNF concentrations in the cerebrospinal fluid (98–100). The results of the meta-analysis showed an overall moderate decrease in serum and plasma BDNF levels in patients with schizophrenia compared with healthy controls. This finding was confirmed in both first-episode and chronic patients with schizophrenia. The decrease in BDNF levels was found to be more pronounced with length of illness (though not associated with age), suggesting the possibility that the relative decrease in BDNF is involved in neuroprogression, as is seen in both MDD and bipolar disorder (101, 102). Despite this, there was no relationship between BDNF levels and either positive or negative symptoms, though other studies have found relationships between BDNF levels and cognition (103, 104). It therefore remains to be seen whether BDNF could be considered a useful biomarker for disease severity.

The meta-analysis also demonstrated that, in longitudinal treatment studies, BDNF concentrations show a small increase after antipsychotic treatment in both patients who responded to medications (defined as a 40% reduction in Positive and Negative Syndrome Scale [PANSS] scores) and in those who did not respond. This suggests that BDNF may not be a useful biomarker to assess treatment response. (Of note, this treatment effect was only found in plasma, not in serum). The meta-analysis found no dose-related relationships, though individual studies have demonstrated such an effect, such as with clozapine dose, but not with typical antipsychotics (105). As there was a high degree of heterogeneity in the studies included in the meta-analysis, these results should be interpreted with some caution, though they suggest that BDNF may predict the presence of disease. It remains unclear whether BDNF may be predictive of illness phase or treatment response. Its relationship with severity of illness remains questionable, though further work will be necessary to understand whether it may be predictive of improvement in cognition.

Developments in statistical methodologies and computational technologies, robust epidemiological studies, and genome projects such as the Schizophrenia Psychiatric Genome-Wide Association Study Consortium (PGC-SCZ), have allowed for significant advances in the ability to detect genetic and epigenetic markers of schizophrenia. These advances have also dramatically increased our knowledge and understanding of the complex epidemiology of schizophrenia and the factors that contribute to its neurophysiological, cognitive, and behavioral phenotypes and have allowed for the formulation of multiple, developmentally driven conceptual models of schizophrenia as well as suggested new possibilities for development of pharmacological interventions. Schizophrenia is a highly heritable disorder and has been investigated through numerous twin and familial studies (106–108). Characteristic alterations of gene expressions in schizophrenia lead to abnormal phenotypic markers (109).

Vulnerability to neurodevelopmental abnormalities associated with schizophrenia are linked to an array of genetic markers found on as many as 108 chromosomal sites; thus, schizophrenia is currently thought to be a polygenic psychiatric condition (107). Presence of the 3q29 microdeletion, for example, is considered the largest genetic risk factor for schizophrenia (110). RELN and GAD1 genes related to GABAergic neuronal function; glutamate receptor and transporter genes; serotonergic receptor gene HTR2A; COMT enzyme gene; and BDNF, important to cognitive function, are among those genes more extensively discussed (108). Other genes linked to schizophrenia include ARC, NMDAR, and VGCC, all critical neurobiological pathway genes that influence neuronal excitability, long-term potentiation, and cognitive processing (110). Extensive lists of top genetic markers identified from genomewide studies are included in Ayalew et al. (111), Flint and Munafó (112), and Rodrigues-Amorim et al. (109), and they illustrate the challenges and complexities involved in understanding genetic risk factors. Genetic markers of schizophrenia are numerous, interrelated, likely interdependent, and, as a result, form a complex genomic profile.

Epigenetics describes modifications to gene function and resulting phenotypic changes not explained by DNA sequencing (112). DNA methylation, translation of mRNA, and histone modification are among the most studied epigenetic mechanisms, and several key events in genomic development have been associated with disruptions in these mechanisms (106, 108, 109, 113–115). Methylation, the addition of a methyl (CH3) group to DNA, results in the modification of DNA function and appears to be a critical epigenetic mechanism for controlling gene expression (106, 109). Several studies following schizophrenia phenotypic development have investigated the role of the methylation process and have suggested that disruptions in the methylation of genes linked to schizophrenia may lead to neurobiological abnormalities such as dysregulation of the dopamine, NMDA, and GABA signaling pathways (108, 115, 116). Increased methylation of reelin (i.e., RELN), for example, has been implicated in prefrontal cortex dysfunction (109, 117). It has also been suggested that abnormalities in DNA methylation may also be heritable, as well as a dynamic process that can continue throughout the lifespan, making it vulnerable to environmental influence (108).

Current developmental models of schizophrenia supported through human and animal studies, as well as models such as the dopamine hypothesis and the water-shed hypothesis, describe multifactorial relationships that consider the contributions of and complexities among several genetic and environmental factors (107, 108, 113, 118). The conclusions of recent investigations of the relationships among genetic, epigenetic, and phenotypical markers of schizophrenia appear to agree that multiple genetic risk factors—combined with adverse environmental stimuli, including a range of gestational conditions, caregiver and familial experience, exposure to and abuse of substances, and traumatic events—can lead to a cascade of epigenetic abnormalities that may then result in phenotypic abnormalities associated with schizophrenia, as well as potentially other comorbid conditions (107, 111, 115, 119). The implications of these collective findings for risk detection, early monitoring, and therapeutic interventions for schizophrenia are that the most effective course may involve participation of multiple specialties and a longitudinal, systemic approach that is customized to the individual.

Much of the recent research has focused on investigating deficits in the cognitive processing of speech sounds and language production in those diagnosed as having schizophrenia; primarily, studies have centered on the abilities to comprehend and produce specific types of verbal utterances and language mechanics. Differences have been found, for example, in the ability to integrate audio-visual information (120). Other research has suggested that auditory processing is altered in a manner that impairs the further processing of speech stimuli, including preattentive processing of speech sounds (121) and vowel phonemes (122), as well as the assignment of meaning to sounds (123). Imaging technologies such as fMRI and electroencephalography (EEG) have supported findings indicating neural correlates of lower sensory-cognitive performance as being measured through traditional cognitive assessment and experimental devices, such as the “cocktail party” condition (124). In general, much of this research suggests that neural mechanisms related to schizophrenia are also related to detectable limitations in sensory-cognitive processing ability and that initial auditory processing of sounds, including verbal stimuli, may be implicated.

Language mechanics has been shown to be abnormal in schizophrenia, suggesting impairment in semantic processing (125, 126). The content of autobiographical narratives in those diagnosed as having schizophrenia also appears to be associated with lower expressivity and complexity but greater self-referencing and repetition of words (127). Differences in frequencies of words categorized as representing emotional content have been found; for example, in a study by Minor et al. (128), use of words lexically associated with anger predicted greater positive and negative symptoms. Innovative technologies, such as social media and electronic messaging platforms, as well as voice recognition software capabilities, present a unique opportunity to further explore and define language-based detectors of psychiatric symptoms using existing knowledge of associated verbal markers.

Recent research has suggested that machine learning models are capable of analyzing text content from Twitter feeds to detect mood-related content associated with schizophrenia with a high degree of accuracy (129). The use of machine learning algorithms to analyze large data sets—including social media activity and neuroimaging data for the detection of mental health markers, including suicide risk and depression, in users (130–132)—is also a topic of emerging research. Studies such as these are currently being conducted through funding from such organizations as the U.S. Department of Defense (DoD) and have been recently publicized in online news articles (133–136). However, these studies represent the comparably sparse body of research available throughout the technology-centric open literature applying emerging technologies to the diagnosis and treatment of psychiatric conditions, and schizophrenia in particular. As a result, this continues to be a potential growth area for clinical investigation and interdisciplinary collaboration.

As reviewed herein, blood-based, neuroimaging, genetic/epigenetic, and speech patterns make up some of the more attractive biomarkers to aid in the prediction of disease state and response to treatment. Currently, none of these putative biomarkers appear ready to assist the clinician in identifying cases of schizophrenia, subtypes of the disorder, treatment choice, or treatment response. Biomarkers, such as CRP, may be most useful at this point in identifying those individuals who may be more highly inflamed, which could drive treatment choice, as may be the case in depression (137, 138). Similarly, detecting differences in speech patterns may be a future clinically useful, noninvasive, tool.

Instead of focusing on individual biomarkers, one area of increasing interest has been on the prognostic utility of combining biomarkers to determine which UHR individuals may convert to psychosis. In a meta-analysis of 2,502 subjects in 27 studies, Fusar-Poli et al. (139) found 29% of individuals meeting UHR criteria converted to psychosis at 2 years, but these studies do not take into account that these cases vary greatly in risk (140). As such, it is of great importance to develop tools that help predict conversion. For example, NAPLS used a multiplex blood assay using plasma analytes of inflammation, oxidative stress, hormones, and metabolism to predict conversion to psychosis (24). This approach of using multiple predictive biomarkers alongside clinical, demographic, and cognitive variables, as NAPLS has done using an externally validated individualized risk prediction tool (assigning an individual a probability of conversion in two years), may yield even stronger predictive value (141, 142). Prognostic biomarkers could play an important role in aiding individuals at higher risk for the disease in initiating interventions that may help to reduce conversion to psychosis (143, 144). If an individual does convert, connection with the mental health services and knowledge of risk prediction could help to reduce the duration of untreated psychosis, a variable that may help to improve long-term outcomes in schizophrenia (145, 146). Another important issue regarding the use of biomarkers in the prediction of conversion to psychosis is that in most CHR research, the diagnostic outcome is dichotomous—conversion or no conversion. As such, both schizophrenia spectrum disorders and affective psychosis are grouped as meeting criteria for conversion, and therefore studies are measuring a nonspecific psychosis outcome. Just as other studies have attempted to use biomarkers to differentiate schizophrenia from affective psychoses, diagnostic specificity may be an important consideration in identifying biomarkers for psychosis conversion.

Besides having prognostic value, the approach of combining various biomarkers may aid in the subtyping of patients with schizophrenia, which could have important future treatment implications. For example, the Bipolar-Schizophrenia Network on Intermediate Phenotypes (B-SNIP) consortium is a group working to further understand biological, rather than clinical, phenomenological differences between schizophrenia, schizoaffective disorder, and bipolar disorder with psychotic features (147). B-SNIP used several endophenotypes, including neuropsychological testing, stop signal, saccadic control, and auditory stimulation paradigms to identify three distinct psychosis biotypes that superseded clinical diagnoses (148). Importantly, identifying these “biotypes” may allow for further elucidation of biomarkers that are specific to each subgroup. This may help explain the heterogeneity in results presented earlier, as certain biomarkers may only be relevant for a specific subgroup of the disorder.

There is much promise in the identification of biomarkers in psychiatry, especially for a disorder as complex and heterogeneous as schizophrenia. Just as biomarkers help predict treatment choices and disease progression in other areas of medicine such as cancer and cardiology, the hope is that we can be more neurobiologically specific in our understanding of disease state and in our prediction of risk and treatment response, all supporting precision medicine. Given the enormous burden of illness of schizophrenia, the search for relevant and useful biomarkers is of great importance for improving the lives of patients with the disorder.

Over the past 12 months, Dr. Goldsmith and Cotes received research funding from Otsuka Pharmaceutical and Alkermes.

Dr. Cotes is on a speakers’ bureau for Otsuka Pharmaceutical and participated in an advisory panel for Alkermes.