1. Introduction
Post-mortem studies have observed reduced expression of insulin receptors and decreased phosphorylation of the downstream signaling proteins AKT, GSK3β and mTOR in frontal cortices from patients with schizophrenia [Reference Chadha and Meador-Woodruff1–Reference Zhao, Ksiezak-Reding, Riggio, Haroutunian and Pasinetti3]. These findings were interpreted as indications of impaired cerebral insulin sensitivity. Wijtenburg et al. [Reference Wijtenburg, Kapogiannis, Korenic, Mullins, Tran and Gaston23] suggested that brain insulin resistance plays a role in learning and memory dysfunction in schizophrenia. The latter study analyzed differences in brain glucose using magnetic resonance spectroscopy (MRS) and biomarkers of neuronal insulin resistance in blood extracellular vesicle (EVs) enriched for neuronal origin (nEVs) from chronic schizophrenia patients and controls. EVs are nanoparticles comprised of a lipid bilayer, containing RNA and protein cargo. nEVs cargo also contains numerous intercellular signaling molecules and can cross the blood-brain-barrier in both directions [Reference Mustapic, Eitan, Werner, Berkowitz, Lazaropoulos and Tran5, Reference Saeedi, Israel, Nagy and Turecki6].
Notably, chronic disease and long-term medication use were potential confounders of the aforementioned studies. However, studies of peripheral insulin resistance using the homeostatic model assessment of insulin resistance (HOMA-IR) or oral glucose tolerance tests (OGTT) suggest abnormal glucose metabolism may also be present in drug-naive first-episode schizophrenia (DNFES) patients, independently from stress, smoking or obesity [Reference Perry, McIntosh, Weich, Singh and Rees7–Reference Thakore11]. Moreover, impaired OGTT has been observed in unaffected siblings [Reference Chouinard, Henderson, Dalla Man, Valeri, Gray and Ryan12]. Therefore, we hypothesized that impaired cerebral insulin signaling occurs in DNFES.
To gain insight into neuronal insulin-signaling in vivo, we analyzed peripheral blood nEVs which have been considered as a “message in a bottle” from neurons [Reference Dubal and Pleasure13], by a well-established protocol used in multiple previous studies [14–Reference Wijtenburg, Kapogiannis, Korenic, Mullins, Tran and Gaston23] and presented in detail in Mustapic et al. [Reference Mustapic, Eitan, Werner, Berkowitz, Lazaropoulos and Tran5].
Our primary outcome nEV biomarker was the upstream insulin signal transduction protein pS312-IRS-1 (Fig. 1a), an established marker of neuronal insulin signaling that has shown diagnostic potential for Alzheimer’s disease (characterized by insulin resistance) [Reference Kapogiannis, Boxer, Schwartz, Abner, Biragyn and Masharani18, Reference Kapogiannis, Mustapic, Shardell, Berkowitz, Diehl and Spangler19], association with gray matter volume [Reference Mullins, Mustapic, Goetzl and Kapogiannis20], cognition [Reference Kapogiannis, Mustapic, Shardell, Berkowitz, Diehl and Spangler19, Reference Wijtenburg, Kapogiannis, Korenic, Mullins, Tran and Gaston23], and dynamic response to insulin signaling-modifying interventions, such as diet, intranasal insulin and exenatide [Reference Athauda, Gulyani, Karnati, Li, Tweedie and Mustapic14, Reference Mustapic, Tran, Craft and Kapogiannis21, Reference Eitan, Tosti, Suire, Cava, Berkowitz and Bertozzi24]. Generally, serine phosphotypes stimulate uncoupling of IRS-1 leading to its degradation [Reference Pederson, Kramer and Rondinone25, Reference Sun, Goldberg, Qiao and Mitchell26]. In an explanatory fashion, we analyzed the functional counterpart of pSer312-IRS1, pY-IRS-1, which generally promotes insulin-stimulated responses [Reference Gual, Le Marchand-Brustel and Tanti27] and downstream serine-threonine kinases (AKT, GSK3β, mTOR, p70S6K; Fig. 1a) to examine the state of the entire insulin signaling cascade, as previously done [Reference Athauda, Gulyani, Karnati, Li, Tweedie and Mustapic14, Reference Mustapic, Tran, Craft and Kapogiannis21]. We hypothesized a lower ratio of phoshorylated to total levels for these proteins in patients vs. controls, indicating reduced signaling activity as previously suggested by postmortem studies (see above). Moreover, we aimed to test if blood insulin measures and the severity of clinical symptoms showed a statistical interaction with the phosphorylation ratios of the tested insulin signaling proteins in DNFES. Finally, we aimed to explore a possible genetic predisposition of the DNFES cohort for insulin signaling anomalies, e.g. whether genetic variants located in genes coding for the tested signal transduction proteins are present with varying probability in patients and controls.
Fig. 1. a) Schematic illustration of insulin receptor (IR) signaling (adapted from: http://erc.endocrinology-journals.org/content/18/4/R125/F1.large.jpg): Ligand binding stimulates the intrinsic tyrosine kinase activity of the IR, which in turn phosphorylates several substrates including insulin receptor substrate / IRS-1. IRS proteins interact with the regulatory subunit of phosphatidylinositol-3-kinase / PI3K leading to the activation of AKT (=Protein kinase B / PKB). AKT propagates insulin’s metabolic effects (glycolysis, glycogen synthesis / lipogenesis, triglyceride secretion / gluconeogenesis, lipolysis, and glucose uptake) by targeting glucokinase, glycogen synthase kinase 3 beta / GSK3β, Forkhead-box-protein A2 / Foxa2, phosphodiesterase / PDE3B, and glucose transporter / GLUT-4. AKT regulates protein translation by activating ribosomal protein S6 kinase, 70 kD / p70S6K via mTOR. However, AKT also represents a crucial node in mediating cell proliferation, growth, synaptic plasticity and survival effects of insulin. Several mediators downstream of AKT are implicated in these effects: β-catenin (via GSK3β), the B-cell lymphoma 2 / Bcl2 antagonist of cell death (BAD), ForkHead transcription factor (FKHR), nuclear factor NFκB, Jun N-terminal kinase (JNK), JUN, mouse double minute 2 (Mdm2), and the tumor suppressor protein p53. b) Group differences: Overall, the ratio of phosphorylated to total insulin-related signal transduction proteins was decreased (p = 0.013) in nEVs from drug-naïve first-episode schizophrenia patients compared to controls. Post-hoc-tests indicated in particular a reduced phosphorylation ratio for mTOR (p = 0.027). Bars represent the mean ± SEM, *p < 0.05. ln (ECL-intensity) natural logarithm of the respective electrochemiluminescence intensity.
2. Methods
2.1 Samples
We studied plasma samples from our previous HOMA-IR-based study [Reference Steiner, Berger, Guest, Dobrowolny, Westphal and Schiltz9]. These were collected from February 2008 to March 2010 from all available sequentially admitted acutely ill DNFES inpatients (n = 24). Controls (n = 24; healthy blood donors and hospital staff and their relatives) came from the same collection period (Table 1). Procedures were IRB approved and written informed consent was obtained. nEV isolation and assays were performed by National Institute on Aging investigators blinded to group.
Table 1 Demographic data, clinical scores, insulin and glucose measures in schizophrenia patients and controls. All participants were Caucasian and groups were matched for gender, age, body mass index (BMI), and waist-hip ratio. Plasma samples were taken at 8 AM after overnight fasting (stored at −80 °C).
Annotations: Data are presented as mean ± standard deviation. T = Student’s t-test value. Significant p values <0.05 are displayed in bold letters. Statistical tests were two-tailed. HOMA-IR / Homeostasis model assessment (HOMA) of insulin resistance = (insulin [mU/L] * glucose [mmol/L]) / 22.5 (after overnight fasting); corrected PANSS / Positive and negative syndrome scale (P = positive scale / N = negative scale / G = general psychopathology scale) scores: subtraction of minimum scores representing “no symptoms” from the PANSS scores.
Psychopathology was assessed using the Positive and Negative Syndrome Scale (PANSS). Psychosis resulting from other medical conditions and substance-induced psychosis was excluded by a thorough medical history, physical examination, routine blood analysis, and screening for illegal drugs [Reference Steiner, Fernandes, Guest, Dobrowolny, Meyer-Lotz and Westphal28]. The same examinations were carried out for the controls. Controls were cleared for personal or family history of psychiatric and neurological disorders using the Mini-International Neuropsychiatric Interview [Reference Sheehan, Lecrubier, Sheehan, Amorim, Janavs and Weiller29]. Exclusion criteria consisted of the presence of immune diseases, immunomodulating treatment, cancer, chronic terminal disease, cardiovascular disorders, manifested diabetes mellitus or severe trauma [Reference Steiner, Fernandes, Guest, Dobrowolny, Meyer-Lotz and Westphal28].
Blood samples were obtained from fasting subjects around 8:00 a.m. and collected into BD Vacutainer™ tubes (Becton Dickinson; Heidelberg, Germany), as previously described. Plasma tubes were centrifuged immediately at 1000g for 10 min; supernatant aliquots were stored at −80 °C.
2.2 nEV isolation and enrichment
As described [Reference Mustapic, Eitan, Werner, Berkowitz, Lazaropoulos and Tran5], nEVs were isolated by a two-step methodology including particle precipitation using Exoquick™ (System Biosciences) to acquire a pellet of total EVs, followed by enrichment for L1 neural cell adhesion molecule (L1CAM) expression using immunoprecipitation with biotinylated antibodies (CD171, clone 5G3) and Pierce™ Streptavidin Plus UltraLink™ Resin (ThermoFisher). nEV concentrations were determined by Nanosight NS500 (Malvern).
2.3 Quantification of insulin signal transduction proteins and SNP data
Upstream insulin signal transduction proteins pS312-IRS-1, pY-IRS-1 and downstream phosphorylated (pS473-AKT, pS9-GSK3β, pS2448-mTOR, pT389-p70S6K) and total insulin signal transduction proteins AKT, GSK3β, mTOR, p70S6K were quantified in lysed nEVs by electrochemiluminescence in duplicate (MesoScale Diagnostics). Participant genotypes for IRS1, AKT1, GSK3B, MTOR and RPS6KB1 (encoding p70S6K) were available from a previous study [Reference Chan, Cooper, Heilmann-Heimbach, Frank, Witt and Nothen30].
2.4 Statistics
Statistical analyses were performed using SPSS 25.0 (IBM, Armonk, NY, USA) and G*Power (www.gpower.hhu.de). Protein measures were natural logarithm (ln)-transformed to correct skewness. Regression analysis showed significant effects of age, sex and ln-nEV concentration. Group differences were assessed using a linear mixed effects model (LMM), with ln-(ECL intensity) as dependent variable, diagnosis and sex as fixed factors, age, waist-hip-ratio, cigarette smoking, and ln-nEV concentration (to normalize for differential nEV yield) as covariates, with participant ID as random effect. The primary outcome pS312-IRS-1 was analyzed first, alongside pY-IRS-1. Exploratory omnibus LMM analysis was carried out for phosphorylated and total signal transduction protein ratios (AKT, GSK3B, p70S6K) to better understand downstream cascade effects, followed by one-sided post-hoc LMM analyses for respective individual protein ratios. Next, LMM was performed with blood insulin as dependent variable and ln-nEV of the respective phosphorylated to total signal transduction protein ratios as additional covariate. Pearson correlation coefficients were calculated to assess associations of protein measures with the severity of clinical symptoms (Positive and negative syndrome scale / PANSS). Due to the exploratory nature of this study we applied a significance threshold of p < 0.05 without correction for multiple comparisons. P < 0.10 was considered as trending towards significance.
To compare the genotype frequencies between DNFES and controls, SNP*diagnosis interactions were determined by F-tests. The respective p-values were corrected for false discovery rate (FDR).
3. Results
3.1 Diagnosis-related differences regarding insulin signal transduction proteins in isolated nEVs
pS312-IRS-1 was lower at trend level in DNFES vs. controls (p = 0.071), but there was no difference for pY-IRS. The omnibus analysis of downstream insulin signaling proteins revealed decreased phosphorylated/total signaling protein ratios in nEVs from patients vs. controls (p = 0.013). Post-hoc-tests indicated in particular a reduced phosphorylation ratio of mTOR (p = 0.027; Fig. 1b). Regression analyses results are presented in detail with effect estimates, confidence intervals, and power analyses in Supplementary Table 2.
3.2 Protein phosphorylation ratios* insulin and SNP*diagnosis interactions
The phosphorylation ratios of p70S6K (p = 0.029), GSK3β (p = 0.039), and at trend level AKT (p = 0.061), showed diagnosis-dependent statistical interactions with insulin blood levels. No significant SNP*diagnosis interactions were identified after FDR significance adjustment (Supplementary Table 1).
3.3 Association of insulin signal transduction protein measures with the severity of clinical symptoms
Within the DNFES group, the ratio pS473-AKT/AKT showed a significant negative correlation with PANSS-G (r=-0.50, p = 0.016) and PANSS-total scores (r=-0.50, p = 0.015) and other downstream phosphorylation ratios showed similar statistical trends (correlation of PANSS-G ratios with GSK3β: r=-0.038, p = 0.085; mTOR: r=-0.42, p = 0.055; p70S6K: r=-0.41, p = 0.054; and of PANSS-total ratios with GSK3β: r=-0.037, p = 0.087; mTOR: r=-0.24, p = 0.290; p70S6K: r=-0.40, p = 0.058).
4. Discussion
These findings support the hypothesis of insulin signaling abnormalities in neuronal cells in DNFES, small sample sizes notwithstanding. The observed trend towards reduced pS312-IRS-1 in DNFES differs from observations in type-2 diabetes where chronic overstimulation by insulin induces an increased phosphorylation of pS312-IRS-1 via an inhibitory bottom-up feedback by elevated pT389-p70S6K [Reference Copps and White31, Reference Zhang, Gao, Yin, Quon and Ye32]. Reduced pS312-IRS-1 in DNFES is at first glance counterintuitive and surprising, since this condition amplifies IRS-1 signaling [Reference Copps and White31]. However, this finding may result from adaptive feedback mechanisms to primary downstream insulin signaling disturbances. In line with previous postmortem studies that point to downstream insulin signaling as the primary abnormality [Reference Chadha and Meador-Woodruff1–Reference Zhao, Ksiezak-Reding, Riggio, Haroutunian and Pasinetti3], the exploratory analysis of downstream signaling serine-threonine kinases (AKT, GSK3β, mTOR, p70S6K) revealed lower phoshorylated to total protein ratios in patients vs. controls, indicating diminished signaling pathway activity; such a state could result in decreased pS312-IRS-1 via the feedback loops of pGSK3β and p70S6K. In previous nEV studies examining change over time in pS312-IRS-1, AKT, GSK3β, mTOR, and p70S6K in response to interventions, we observed similar direction of change [Reference Athauda, Gulyani, Karnati, Li, Tweedie and Mustapic14, Reference Mustapic, Tran, Craft and Kapogiannis21].
Phosphorylation of serine-thereonine kinases is ATP- and pH-dependent. Thus, considering the known mitochondrial dysfunction and increased lactate production in schizophrenia [Reference Steiner, Bernstein, Schiltz, Müller, Westphal and Drexhage10], we hypothesize that more fundamental neuronal metabolic abnormalities underlie the observed hypophosphorylations. Phosphorylated signal transduction proteins, which are subject to dynamic regulation, were found to be reduced in DNFES, as opposed to the respective total proteins, the levels of which depend on translation (see Supplementary Figure). This discrepancy between phosphorylated and total protein levels suggests potential reversibility of neuronal insulin resistance by factors and interventions that promote insulin signaling. Accordingly, in the light of non-significant SNP*diagnosis interactions, genetic effects do not sufficiently explain the observed differences in nEV signaling proteins, pointing to either dominance of lifestyle/environmental factors in producing insulin signaling abnormalities in schizophrenia or the implication of unexamined and undetected variants of other genes.
Besides its brain energy metabolism role, insulin is linked to leading hypotheses of schizophrenia since 1) it governs brain development / maturation as well as the complexity of dendritic branching and synaptic plasticity, particularly via mTOR [Reference Steiner, Bernstein, Schiltz, Müller, Westphal and Drexhage10, Reference Ryskalin, Limanaqi, Frati, Busceti and Fornai33], which showed the strongest differences among downstream signaling mediators, and 2) insulin’s transduction network is shared by neurotransmitters, BNDF, and proinflammatory cytokines [Reference Zheng, Wang, Zeng, Lin, Little and Srivastava34]. Thus, insulin resistance in schizophrenia may partially reflect reduced signal transduction downstream of GABA- or NMDA-receptors, dopamine-D2-recepter hyperactivity, reduced BDNF or low-grade inflammation [Reference Steiner, Bernstein, Schiltz, Müller, Westphal and Drexhage10, Reference Zheng, Wang, Zeng, Lin, Little and Srivastava34]. Cerebral insulin resistance may cause partial cerebral glucose deprivation via GLUT-4 downregulation and the aforementioned shared pathways may contribute to a worsening of disease-related neurotransmitter changes [Reference Steiner, Bernstein, Schiltz, Müller, Westphal and Drexhage10]. Therefore, we hypothesized that the severity of psychosis symptoms might correlate inversely with the phosphorylation ratio (i.e. activity) of the tested signal transduction in our study. Indeed, changes in nEV insulin signaling biomarkers appeared to be clinically relevant, as we found associations of higher PANSS-G and PANSS-total scores with reduced phosphorylation ratios for AKT, and similar trends were observed for GSK3β, mTOR and p70S6K. In addition, phosphorylation ratios of p70S6K, GSK3β and at trend level AKT showed diagnosis-dependent statistical interactions with insulin blood levels, which might contribute to our earlier observation of peripheral insulin resistance in these DNFES patients vs. controls [Reference Steiner, Berger, Guest, Dobrowolny, Westphal and Schiltz9].
Some limitations of our study should be noted. First, L1CAM was originally selected as target for enrichment due to its high and relatively specific expression in neural tissue and early research demonstrating high expression on EVs derived from cultured neurons [Reference Mustapic, Eitan, Werner, Berkowitz, Lazaropoulos and Tran5, Reference Faure, Lachenal, Court, Hirrlinger, Chatellard-Causse and Blot35]. Multifaceted evidence for neuronal and brain enrichment (i.e. increased concentration compared to control EV subpopulations for L1CAM, NCAM, synaptophysin, neurofilament-light, neuronal enolase, Tuj-1 and many other neuronal and some brain specific proteins) has been provided in four previous publications [Reference Mustapic, Eitan, Werner, Berkowitz, Lazaropoulos and Tran5, Reference Athauda, Gulyani, Karnati, Li, Tweedie and Mustapic14, Reference Kapogiannis, Mustapic, Shardell, Berkowitz, Diehl and Spangler19, Reference Pulliam, Sun, Mustapic, Chawla and Kapogiannis22]. Nevertheless, as we previously recognized [Reference Pulliam, Sun, Mustapic, Chawla and Kapogiannis22], L1CAM is not entirely brain-specific, since it is also highly expressed in kidney tubular epithelium [Reference Debiec, Christensen and Ronco36]. Moderate L1CAM expression has been observed in peripheral nerves, intestinal crypt cells, and glandular cells of the seminal vesicle and fallopian tube (www.proteinatlas.org/ENSG00000198910-L1CAM/tissue) and low L1CAM expression has been detected in other cell types such as lymphoid and myelomonocytic cells [Reference Pancook, Reisfeld, Varki, Vitiello, Fox and Montgomery37]. Therefore, the observed transduction protein differences may not be solely attributed to neurons but could also reflect disturbances of insulin signal transduction in other tissues in DNFES. Further technical optimization for specific neuronal EV enrichment is warranted to clarify this issue. Second, limited by the available sample size, the presented study was underpowered for analysis of diagnosis-related differences in individual signal transduction proteins (Supplementary Table 2) and thus we may have been unable to identify some diagnosis-related differences (false negative results). Due to the exploratory nature of our study and the relatively small sample size, correction for multiple testing for each set of analyses was not applicable, limiting the generalizability of the results. Therefore, future replication using larger cohorts is essential. The inclusion of unaffected relatives and other disorders may elucidate the etiopathological underpinnings of the present findings. However, an important strength of the study is the inclusion of well-characterized samples, with a focus on DNFES to exclude potential confounding effects of medication.
In conclusion, to our knowledge, this is the first study providing in vivo evidence for impaired insulin signaling by a comprehensive analysis of insulin signal transduction proteins in DNFES. The observed phosphorylation pattern implies that the signaling pathway activity in DNFES is compromised further downstream compared to type-2 diabetes.
Author contributions
Drs Steiner and Kapogiannis and Dipl.-Ing. Dobrowolny had full access to all study data and take responsibility for the integrity of the data and the accuracy of the data analysis (Dr Kapogiannis was unblinded after nEV analysis was completed). Concept and design: Drs Steiner, Kapogiannis, Bernstein, Frodl, Schiltz. Acquisition, analysis, and interpretation of data: All authors. First manuscript draft: Dr Steiner. Critical revision of the manuscript for important intellectual content: All authors. Statistical analysis: Drs Kapogiannis, Steiner, Schiltz and Dipl.-Ing. Dobrowolny. Administrative, technical, or material support: Drs Steiner, Kapogiannis, Mrs Tran, Mrs Mustapic. Sample collection and characterization: Mrs Meyer-Lotz and Dr. Steiner.
Additional contributions
Paul C. Guest (Cambridge, UK), PhD provided language editing of our manuscript as a native speaker. Dipl.-Psych. Josef Frank (Department of Genetic Epidemiology in Psychiatry, Central Institute of Mental Health, Mannheim, Germany) supported the analyses of participant SNP*diagnosis interactions and the search in scientific GWAS databases.
Declaration of Competing Interest
The authors declare no conflict of interest.
Acknowledgments
This research was supported in part by the Intramural Research Program of the National Institute on Aging, National Institutes of Health (Dimitrios Kapogiannis, Joyce Tran, Maja Mustapic). and by the German Federal Ministry of Education and Research (Marcella Rietschel 01EW1810).
Appendix A Supplementary data
Supplementary material related to this article can be found, in the online version, at doi:https://doi.org/10.1016/j.eurpsy.2019.08.012.
:){kind=link}
Comments
No Comments have been published for this article.