Schizophrenic inpatients confined in Baiyun Jingkang Hospital, Guangzhou, China from 2020 to 2021 were recruited in this study. The inclusion criteria for patients were as follows: (1) patients who had been diagnosed with schizophrenia by experienced psychiatrists based on the International Classification of diseases-10 (ICD-10); (2) patients with stable psychotic symptoms for more than 2 weeks and able to finish the investigation and the cognitive function test; and (3) patients who were 18 years old or above. Meanwhile, the exclusion criteria for patients were as follows: (1) patients whose blood lipid measurements were missing; and (2) patients with history of cerebral trauma, stroke, myocardial infarction or dementia. A total of 489 inpatients were recruited in the current study. Based on the above exclusion criteria, 42 inpatients were excluded, and 447 inpatients were included in the analysis. The study abided by the Declaration of Helsinki principles. This current study was approved by the Medical Ethics Committee of Baiyun Jingkang Hospital. Written informed consent was obtained from each patient and family before performing any procedures related to this study.

The blood samples were collected after an overnight fast by medically trained nurses in the hospital. The full blood was extracted using an evacuated and promoting coagulating tube (Banbiao Medical Technology Co., Ltd., Guangzhou, Guangdong, China), and serum was obtained by centrifugation of 3000 rpm (DM0506 Centrifuge, Dalong Xingchuang Experimental Instruments Co., Ltd., Beijing, China). TC, LDL-C, HDL-C, TG, ApoA1, and ApoB were measured using an automatic Biochemical Analyzer (Mindray Medical International Co., Ltd., Shenzhen, Guangdong, China). NHDL-C was calculated by subtracting HDL-C from TC. Based on the 2016 Chinese guidelines for the management of dyslipidemia in adults, the high levels of TC, LDL-C, TG, NHDL-C, ApoA1, and ApoB are 5.2 mmol/L, 3.4 mmol/L, 1.7 mmol/L, 4.0 mmol/L, 1.9 mmol/L, 1.17 mmol/L, and above, respectively. Meanwhile, the low level of HDL-C is less than 1.0 mmol/L (21).

A total of 15 patients with cognitive impairment were randomly selected, and then another 15 patients without cognitive impairment were matched based on age, sex, education, and use of antipsychotics such as clozapine, olanzapine, risperidone and quetiapine. Then, metabolomics tests were conducted on these 30 patients. The pre-treated serum was stored at −80°C. The frozen serum was thawed at room temperature before the metabolomics test. Then, 50 μL of the sample was transferred to an EP tube (Banbiao Medical Technology Co., Ltd., Guangzhou, Guangdong, China) and mixed with 200 μL of extract containing an isotopically labeled internal standard mixture (acetonitrile: methanol = 1:1, Banbiao Medical Technology Co., Ltd., Guangzhou, Guangdong, China). The samples were vortexed for 30 s, and sonicated for 10 min in an ice-water bath and incubated for 1 h at −40°C to precipitate the proteins. Subsequently, the sample was centrifuged at 12,000 rpm for 15 min at 4°C and then the supernatant was transferred to a fresh tube for analysis. Four quality control (QC) samples were then prepared by pooling equal amounts of the supernatant from all samples. In addition, blank samples interspersed throughout the experiment were used to evaluate cross-contamination among the samples.

A liquid chromatography-mass spectrometry (LC-MS) metabolomics test was performed using an ultra-high-performance liquid chromatography (UPLC) system (Vanquish, Thermo Fisher Scientific, Shanghai, China) with a UPLC-bridged ethyl hybrid amide column (2.1 mm × 100 mm, 1.7 μm) coupled with Q Exactive HFX MS (Orbitrap MS, Thermo Fisher Scientific, Shanghai, China) equipped with an electrospray ionization (ESI) source to obtain a comprehensive serum metabolic feature. The A phase of the mobile phase included 25 mmol/L of ammonium acetate and 25 mmol/L of ammonia hydroxide in water (pH 9.75), and the B phase was acetonitrile (Banbiao Medical Technology Co., Ltd., Guangzhou, Guangdong, China). The auto-sampler temperature was set at 4°C and the injection volume was 3 μL.

Q Exactive HFX MS can acquire tandem MS data in the information-dependent acquisition mode of the acquisition software (Xcalibur, Thermo Fisher Scientific, Shanghai, China). The acquisition software continuously evaluates the full-scan MS spectrum. In this study, the detailed parameters of the ESI source were as follows: sheath gas flow rate, 30 Arb; aux gas flow rate, 25 Arb; capillary temperature, 350°C; full MS resolution, 60,000; tandem MS resolution, 7,500; collision energy (CE), 10/30/60 in normalized CE mode; and spray voltage, 3.6 kV (positive) or −3.2 kV (negative).

A series of QC was performed in metabolite examination. First, instrument stability was demonstrated to be excellent via total ionization chromatography (Supplementary Figure 1). Data acquisition stability was also excellent based on the differences in the internal standard peaks amongst the QC samples (Supplementary Figure 2). There were no remarkable peaks observed for all internal standards in the blank samples, thus indicating that no cross-contamination occurred (Supplementary Figure 3).

The raw data were converted to mzXML format by using ProteoWizard and processed with an in-house program, which was developed using R 4.1.3 (R Foundation for Statistical Computing, Vienna, Austria) based on XCMS for peak detection, extraction, alignment and integration. Then, an in-house MS2 database (BiotreeDB) was applied for metabolite annotation with a cut-off of 0.3. The respective metabolic pathways of the different metabolites were annotated using the MetaboAnalyst.¹

The global cognitive function of each patient was tested using the Mini-Mental State Examination (MMSE) by a well-trained psychiatrist within the psychiatric hospital. As a cognitive screening tool, the MMSE test is extensively applied for the early detection of cognitive impairment and it has been validated in schizophrenia (22, 23). The total score of MMSE is 30 and the cut-off point used for cognitive impairment screening was 26, which was considered to be good at discriminating cognitive impairments using MMSE in a previous scoping review (24). Thus, patients with MMSE scores of 26 and below were considered to have cognitive impairments.

Covariates were collected from the medical records of each inpatient and the course of schizophrenia was measured in years. Education level was categorized as elementary school or below, and middle school or above. Meanwhile, body mass index (BMI) was calculated based on the standard definition (weight/height², kg/m²). The use of antipsychotics was from the prescription of the psychiatrist, and the actual usage of antipsychotics was corrected based on the test results of antipsychotics in the blood. Blood pressure was measured before the cognitive test, and hypertension was determined by recording a history of hypertension or systolic blood pressure ≥ 140 mmHg or diastolic pressure ≥ 90 mmHg or taking antihypertensive drugs. In addition, fasting blood glucose (FBG) and C-reactive protein (CRP) were measured using an automatic Biochemical Analyzer (Mindray Medical International Co., Ltd., Shenzhen, Guangdong, China). Moreover, diabetes was determined by recording the history of diabetes or FBG ≥ 7.0 mmol/L. Given the strict management of the hospital, the diet mode, physical activity level and sleeping habits of inpatients were similar, and they were not allowed to smoke or drink in the hospital. Hence, these factors were not selected as covariates.

Given that continuous variables are non-normally distributed in the study, continuous variables were presented as the median and interquartile range (IQR). Gender and education level were presented as categorical variables. The Kruskal–Wallis test was employed for continuous variables test between cognitively impaired and cognitively normal groups, and the Chi-square test was used for categorical variables. A binary logistic regression model was applied to identify the associations of various blood lipid parameters with cognitive impairment. Model 1 was adjusted based on age and sex. Model 2 was adjusted based on age, sex, the course of schizophrenia, education, BMI, hypertension, FBG, CRP, and the use of clozapine, olanzapine, risperidone, and quetiapine. In addition, the subgroup analysis was used to explore the associations of blood lipids with cognitive function stratified by age (45 years) and sex. In sensitivity analysis, patients with history of hypertension and diabetes were excluded followed by the assessment of associations of blood lipids with cognitive impairment. The results of logistic regression were shown as odds ratios (ORs) and 95% confidence intervals (CIs). All statistical tests were two-sided, and the significance level was P < 0.05.

Meanwhile, the metabolomics data were separately processed in positive and negative ionization modes. Principal component analysis (PCA) and the correlation and differences in the internal standard responses of the QC samples suggested excellent data quality (Supplementary Figures 4–6 and Supplementary Table 1). The resultant data matrices were log-transformed and centralized then subjected to automated modeling analysis using the SIMCA software (V16.0.2, Sartorius Stedim Data Analytics AB, Umea, Sweden). Multivariate PCA was applied to determine the stability of the entire analysis process. Afterward, orthogonal projection to latent structures-discriminant analysis (OPLS-DA) was used to distinguish intergroup differences in the metabolic profile. The OPLS-DA model was validated by sevenfold cross-validation with R² and Q² and further validated by a permutation test with R²Y and Q². The differences in peaks were tested by Wilcoxon rank-sum test and visualized in a volcano plot with P < 0.05 and variable importance in projection (VIP) > 1 in the first principal component in OPLS-DA. Moreover, a heat map was used to present the hierarchical clustering of the different metabolites that were successfully matched in the Human Metabolome Database. The data were processed by R version 4.1.3 software (R Foundation for Statistical Computing, Vienna, Austria).

Table 1 describes the demographic characteristics of the patients. A total of 82.3% (368/447) of patients with schizophrenia were diagnosed to have cognitive impairment. The median score of cognition among overall patients was 18 points (IQR: 10–24). Meanwhile, the score of the cognitive impairment group and the cognitively normal group were 15 points (IQR: 10–21) and 29 points (IQR: 28–30), respectively. The age of the patients ranged from 19 to 73 years, with a median age of 47 years (IQR: 39–56). The median age in the cognitive impairment group was 48 years (IQR: 39–57), and slightly higher than that in the cognitively normal group (46 years, IQR: 38–49, P < 0.001). Furthermore, the cognitively normal group had significantly better education, a lower rate of olanzapine use, and a higher rate of risperidone use than the cognitive impairment group. There were no differences observed in sex, course of schizophrenia, BMI, FBG, CRP, the prevalence of hypertension and diabetes, as well as the use of clozapine and quetiapine between the two groups.

Patients with cognitive impairment had lower concentrations of LDL-C, TG, and ApoB, and higher concentrations of HDL-C than patients with normal cognition. There were no significant differences found in concentrations of TC and NHDL-C between the two groups (Table 1).

Significantly higher concentrations of TC, HDL-C, NHDL-C, and ApoA1 were observed in female versus male inpatients in the subgroup. As for age subgroups divided by the cut-off point of 45 years, TC, LDL-C, NHDL-C, and ApoB were significantly higher in patients aged 45 years or older (Supplementary Table 2).

A higher level of HDL-C was associated with higher odds of cognitive impairment after age and sex were adjusted, whereas, higher levels of NHDL-C and ApoB were associated with lower odds of cognitive impairment as shown in Figure 1. After further adjustment for education level, BMI, hypertension, FBG, and CRP, as well as the use of clozapine, olanzapine, risperidone and quetiapine, higher levels of NHDL-C, and ApoB were still associated with lower odds of cognitive impairment.

As shown in Figures 2, 3, a negative association of NHDL-C with cognitive impairment was observed in inpatients who were over 45 years or female in age and sex subgroups; ApoB was negatively associated with cognitive impairment in all subgroups; higher LDL-C was associated with lower odds of cognitive impairment in females, and higher TG was associated with lower odds of cognitive impairment in inpatients who were below 45 years old. HDL-C was positively associated with cognitive impairment only in patients below 45 years old.

The associations between ApoB, NHDL-C, and cognitive impairment in the sensitivity analysis were still significant in the inpatients without hypertension or diabetes, as shown in Figure 4.

The demographic characteristics among 15 patients with cognitive impairment and 15 cognitively normal were similar after they were matched (Supplementary Table 3).

The positive ionization mode was adopted because the OPLS-DA model in the negative ionization mode displayed low reliability (R²Y = 0.948, Q² = –0.034). A total of 5,567 peaks were extracted in the positive ionization mode, and 3,933 peaks were used for actual analyses after filtration for single chromatographic peaks and data normalization (Supplementary Dataset 1).

The OPLS-DA model showed a remarkable separation of metabolism between schizophrenia with and without cognitive impairment (R²Y = 0.929, Q² = 0.053, Figure 5A). The permutation test revealed that the OPLS-DA model did not produce overfitting (R²Y = 0.92, Q² = –0.27, Supplementary Figure 7). There were 349 down-regulated and 16 up-regulated peaks found in cognitive impairment patients, with VIP > 1 in OPLS-DA and P < 0.05 (Figure 5B). However, only 37 metabolites (34 down-regulated and 3 up-regulated) were previously identified (see Figure 5E and Supplementary Table 4).

The bubble plot shows the pathways that have evident enrichment. In lipid metabolism, sphingolipid metabolism was mainly involved by differential metabolites (P < 0.05, Figure 5C). The metabolites, including sphinganine and 3-dehydrosphinganine, decreased in the cognitive impairment group (Figure 5D and Supplementary Table 5), and positively correlated with TC, NHDL-C, LDL-C, and ApoB (Figure 6). In addition, D-glutamine and D-glutamate metabolism, glycine, serine and threonine metabolism were also involved, with significantly decreased metabolites of D-glutamine, pyrrolidonecarboxylic acid, choline and creatine in the cognitive impairment group.