Association of serum BMP2 with delayed memory impairment in first-episode, drug-naïve schizophrenia

Abstract

Background

Cognitive impairments, particularly in the domain of memory, are core and enduring features of schizophrenia. Bone morphogenetic proteins (BMPs), members of the transforming growth factor-β superfamily, are involved in neurodevelopment and synaptic plasticity, but few studies have investigated their relevance to cognitive deficits in schizophrenia.

Methods

In this study, we measured circulating levels of BMP2, BMP9, and BMP10, and evaluated correlations with cognitive performance and psychopathological symptoms among first-episode, drug-naïve (FEDN) patients with schizophrenia. Ninety FEDN patients were assessed for psychiatric symptoms using the Positive and Negative Syndrome Scale (PANSS). Patients and ninety demographically matched healthy controls (HCs) also completed the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) and provided serum samples for Luminex-based BMP assays.

Results

After covariate adjustment, patients exhibited significantly elevated serum BMP2 (p = 0.015) and BMP9 (p = 0.004) concentrations. Within the patient group, higher serum BMP2 was significantly negatively associated with RBANS delayed memory (RBANS-DM) scores after adjustment for demographic and clinical covariates (β = −0.236, p = 0.014), accounting for a modest 7% of variance (partial R² = 0.071). No robust associations of BMP9 or BMP10 with cognitive domains or symptoms were detected beyond exploratory trend-level findings.

Conclusions

These results suggest a potential association between elevated serum BMP2 and poorer delayed memory performance in early-stage schizophrenia. Given the cross-sectional design and modest effect size, the findings should be interpreted as hypothesis-generating rather than causal. BMP2 may represent a candidate peripheral correlate of delayed memory performance, warranting replication in larger, longitudinal, multicenter studies.

Clinical trial number

Not applicable.

Introduction

Schizophrenia is a severe, chronic psychiatric disorder that typically emerges in late adolescence or early adulthood and progresses in severity thereafter. Clinical manifestations are highly heterogeneous, but generally include positive symptoms such as hallucinations and delusions as well as negative symptoms such as affective flattening and avolition [1, 2]. Although current diagnostic criteria are primarily based on these psychiatric symptoms, cognitive impairments are widely recognized as core pathological features [3]. Indeed, deficits in multiple cognitive domains, including attention, working memory, and episodic memory, are already present in the earliest stages of the illness or even before clinical onset, and may persist during symptom remission [4]. Among these cognitive deficits, memory impairment is particularly prominent and is one of the key factors predicting poor long-term functional outcome and difficulty reintegrating into society [5]. Moreover, current antipsychotic medications primarily target dopamine D2 receptors to ameliorate positive symptoms, but show limited efficacy for cognitive impairments [6]. Therefore, elucidating the biological bases of cognitive deficits and identifying clinically accessible peripheral correlates are critical for long-term clinical management and for developing targeted interventions [7].

The bone morphogenetic proteins (BMPs), members of the transforming growth factor-β (TGF-β) superfamily, are well described regulators of skeletal development. In addition, BMP ligands and receptors are widely expressed in the adult central nervous system (CNS), where they modulate cell survival, differentiation, and pathological repair processes [8]. In particular, BMP ligands and receptors are expressed in the hippocampal dentate gyrus (DG), which retains adult neurogenesis capacity. Newborn DG neurons are integrated into local circuits and may contribute to episodic memory formation [9]. Moreover, BMP signaling also regulates blood-brain barrier (BBB) formation, influencing local metabolism and immunity [10].

Among BMP subtypes, BMP2 is involved in neural stem cell quiescence, and altered BMP2 signaling has been hypothesized to influence hippocampal neurogenesis, which could relate to memory processes, although this has not been directly demonstrated in humans [11, 12]. Importantly, BMP signaling in the adult hippocampal niche should not be viewed as uniformly detrimental. Rather, pathway activity appears to be tightly regulated and context-dependent: appropriate BMP signaling is required to maintain the quiescent neural stem cell pool and long-term neurogenic capacity [11], whereas BMP2/4 signaling may influence neuronal specification of adult hippocampal neural stem and progenitor cells through interaction with WNT signaling [13]. Therefore, BMP2 was selected as the primary BMP candidate in the present study because of its biological relevance to hippocampal neurogenesis and memory-related processes. In contrast, BMP9 and BMP10 primarily function as vascular quiescence factors and may contribute to cerebrovascular homeostasis, but their roles in adult cognitive function remain unclear [14, 15]. BMP9 and BMP10 were therefore included as exploratory BMPs to examine whether vascular-related BMP signaling might also be associated with cognitive or clinical features in early-stage schizophrenia.

Several studies also implicate BMP signaling in neuropsychiatric disorders. The BMP signaling pathway has been identified as a common downstream target of various antidepressant drugs [16–18], and abnormalities in BMP signaling have been reported to suppress hippocampal neurogenesis and contribute to the evolution of pathological protein networks in Alzheimer’s disease [19, 20]. However, research on aberrant BMP signaling in schizophrenia remains surprisingly limited. One study reported altered BMP signaling by glial progenitor cells derived from patient induced pluripotent stem cells (iPSCs) [21] while others have found reduced circulating BMP6 in elderly schizophrenia patients [22] and reduced circulating BMP8b following antipsychotic treatment, particularly with clozapine [23]. Prior studies often included medicated or chronic patients, potentially confounding BMP levels. Furthermore, most studies focused on single BMPs or intracellular signaling, whereas multiple BMPs may act cooperatively to regulate neurovascular function. Evidence using comprehensive cognitive batteries, such as the RBANS, remains scarce.

To address these gaps, we examined associations between specific cognitive domains and circulating BMP2, BMP9, and BMP10 in a cohort of first-episode, drug-naïve (FEDN) schizophrenia patients. We focused a priori on delayed memory as the primary cognitive outcome because delayed memory is closely related to hippocampal structure and function, and memory impairment is a clinically important cognitive deficit in schizophrenia [5, 9]. Based on the hippocampal neurogenesis hypothesis, we hypothesized that serum BMP2 would be altered in FEDN patients and that higher BMP2 levels would be associated with poorer RBANS delayed memory performance. Exploratory analyses were conducted for BMP9 and BMP10 to examine potential links with cognitive domains and psychotic symptoms.

Methods

Participants

The study was conducted at Nantong Mental Health Center. Preparatory procedures were initiated in March 2022, and participant recruitment and data collection were conducted after ethics approval, from May 2022 to November 2024. A total of 90 FEDN patients with schizophrenia were recruited as the patient group, and 90 healthy volunteers as the control group. The diagnosis of schizophrenia was established by two senior psychiatrists according to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM‑5). Healthy controls (HCs) were evaluated by psychiatrists using clinical interviews to exclude any current or past major psychiatric disorders.

Inclusion criteria for patients were as follows: (1) age 18 to 65 years, (2) currently experiencing a first acute psychotic episode, and (3) no prior exposure to any antipsychotic medication. First-episode and drug-naive status were confirmed through clinical interviews with patients and their family members, review of available medical records, and independent confirmation by two senior psychiatrists. Duration of untreated psychosis was assessed as the interval between the onset of psychotic symptoms and study enrollment before the initiation of antipsychotic treatment.

Healthy controls were recruited from the local community and individually matched for age, sex, and years of education. Exclusion criteria for all participants included a history of major psychiatric disorders with the exception of schizophrenia in the patient group (e.g., bipolar disorder, major depressive disorder), neurological diseases, substance abuse or dependence, serious medical illnesses (e.g., diabetes, hypertension), and pregnancy or breastfeeding.

All enrolled participants completed the clinical assessment, cognitive evaluation, and serum BMP measurement; therefore, no participant was excluded after enrollment or omitted from the final analysis.

The study protocol was approved by the Ethics Committee of the Fourth People’s Hospital of Nantong (approval No. 2022-K015; approval date: May 6, 2022), and written informed consent was obtained from all participants or their legal guardians prior to testing. All study procedures were conducted in accordance with the principles of the Declaration of Helsinki.

Clinical and cognitive assessments

The severity of psychotic symptoms was assessed using the PANSS, which consisted of three subscales evaluating positive symptoms (PANSS-P), negative symptoms (PANSS-N), and general psychopathology (PANSS-G). Cognitive function was evaluated for all participants using the Chinese version of the RBANS, which has been validated in Chinese populations [24]. The RBANS consisted of 12 subtests and yields index scores for five cognitive domains, immediate memory (RBANS‑IM), visuospatial/constructional abilities (RBANS‑VS), language (RBANS‑LAN), attention (RBANS‑ATT), and delayed memory (RBANS‑DM), as well as a total score (RBANS‑T). All clinical and cognitive assessments were administered by two trained raters with standardized training on both scales, and were blinded to all clinical and demographic characteristics of the participants. Inter-rater correlation coefficients for repeated assessments were > 0.80, supporting the reliability and consistency of the ratings.

Serum collection and measurement of BMPs

Venous blood samples were obtained from all participants between 07:00 and 09:00 AM after an overnight fast. For patients, blood samples were obtained before the initiation of any antipsychotic treatment. Approximately 5 mL of venous blood was drawn from the antecubital vein into tubes without anticoagulant and allowed to clot at room temperature for 30 min. Samples were then centrifuged at 3,000 rpm for 15 min to separate the serum, which was aliquoted and stored at − 80 °C until analysis.

Serum concentrations of BMP2, BMP9, and BMP10 were measured using a Luminex liquid suspension array assay on the Luminex 200 system (Luminex Corporation, Austin, TX, USA) with a Human Premixed Multi-Analyte Kit (catalog No. LXSAHM-18), according to the manufacturer’s instructions. Serum samples were centrifuged at 10,000 rpm for 10 min, and the supernatant was diluted two-fold before analysis. Each sample was assayed in duplicate, and intra-assay coefficients of variation were all less than 5%. Standard curve quality control met the manufacturer’s criteria, with observed-to-expected concentration ratios within the acceptable range of 70–130%. Values below the lower limit of detection, if present, were handled according to the manufacturer’s instructions. Blood samples were labeled with study identification codes before laboratory processing. Personnel responsible for blood collection, sample processing, and BMP assays were blinded to diagnostic group, clinical information, and cognitive test results. Patient and control samples were collected and processed using the same standardized protocol and identical assay procedures to minimize potential batch effects.

Statistical analysis

All statistical analyses were performed using SPSS version 26.0 (IBM Corp., Armonk, NY, USA), and figures were generated with GraphPad Prism version 10.0 (GraphPad Software, San Diego, CA, USA). Sample size was estimated using G*Power 3.1 [25]. Based on non-normal and positively skewed distributions, serum BMP2, BMP9, and BMP10 concentrations were natural log-transformed and are presented as ln[BMP2], ln[BMP9], and ln[BMP10]. Continuous variables were compared using independent-samples t-tests or Mann-Whitney U tests, and categorical variables using the χ² test.

Group differences in ln[BMP2], ln[BMP9], and ln[BMP10] were examined using MANCOVA with group (patient vs. control) as the fixed factor and age, sex, years of education, BMI, and smoking status as covariates; Pillai’s trace was used for the overall multivariate effect, followed by univariate ANCOVAs. The Benjamini-Hochberg procedure was used to control the false discovery rate (FDR) across the three ANCOVA group main effects, and FDR-adjusted p-values (pFDR) are reported.

Within the patient group, partial correlations between each ln[BMP] measure and PANSS scores (total and subscales) and RBANS indices (total and domain scores) were computed controlling for age, sex, years of education, duration of illness (months), BMI, and smoking status. The association between serum ln[BMP2] and the RBANS-DM index was defined a priori as the primary hypothesis-driven analysis, based on the relevance of BMP2 to hippocampal neurogenesis and memory-related processes; therefore, it was not included in the FDR correction. All other correlations were considered exploratory, and the Benjamini-Hochberg procedure was applied to these exploratory tests, with FDR-adjusted p values (q values) reported. Multivariable linear regression was used to further evaluate the association between ln[BMP2] and RBANS-DM with the same covariates. For this primary regression model, standardized residuals, Cook’s distance, and variance inflation factors (VIFs) were examined to assess potential outliers, influential observations, and multicollinearity. All tests were two-tailed, and p < 0.05 was considered statistically significant.

Results

Demographic and clinical characteristics

There was no significant difference in age, sex ratio, smoking status, BMI, or years of education between patients and HCs (all p > 0.05), indicating that the two groups were well matched for subsequent comparisons (Table 1).

Table 1.

Demographic and clinical characteristics of patients and HCs

	Patients (n = 90)	HCs (n = 90)	χ2/Z/t	p
Age (years)	32.0 (24.8,42.3)	33.0 (22.8,42.0)	-0.376a	0.707
Sex (male/female)	28/62	29/61	0.026b	0.873
Smoking (yes/no)	5/85	6/84	0.097b	0.756
BMI (kg/m2)	21.94 ± 3.39	22.34 ± 2.50	-0.895c	0.372
Years of education	16 (12,16)	15 (14,16)	-0.036a	0.971
Duration of illness (months)	3.5 (0.5,12.0)	—	—	—
PANSS-T score	85.73 ± 5.86	—	—	—
PANSS-P score	21.63 ± 4.49	—	—	—
PANSS-N score	19.86 ± 4.90	—	—	—
PANSS-G score	44.24 ± 3.58	—	—	—

^aMann–Whitney U test; ^bχ² test; ^cindependent-samples t-tests; BMI: body mass index; PANSS: Positive and Negative Syndrome Scale

Cognitive performance and serum BMP levels in patients vs. controls

As expected, all RBANS cognitive indices and the total score were lower in patients with schizophrenia than HCs (all p < 0.05), with effect sizes ranging from moderate to very large. The largest between-group differences were observed in RBANS-ATT, RBANS-VS, RBANS-DM, and RBANS-T, with Cohen’s d values of approximately − 3.04, −2.28, −2.00, and − 2.36, respectively, indicating that these FEDN patients already exhibited widespread and severe cognitive impairments. Both serum ln[BMP2] and ln[BMP9] concentrations were significantly higher in patients than HCs (ln[BMP2]: p = 0.018; ln[BMP9]: p = 0.011), whereas serum ln[BMP10] concentration did not differ significantly between groups (p = 0.394) (Table 2). Raw BMP concentrations are also presented in Table 2 as median (IQR) for descriptive transparency, whereas log-transformed values were retained for statistical analyses because of their skewed distributions.

Table 2.

Comparisons of cognitive performance metrics and serum BMP concentrations between patients and HCs

	Patients (n = 90)	HCs (n = 90)	t	p	Cohen’s d (patients − controls)
RBANS-IM	82.09 ± 12.52	93.91 ± 13.96	-5.981	< 0.001	-0.89
RBANS-VS	74.63 ± 12.09	102.76 ± 12.60	-15.280	< 0.001	-2.28
RBANS-LAN	97.84 ± 6.83	101.14 ± 10.07	-2.573	0.011	-0.38
RBANS-ATT	76.08 ± 11.74	114.86 ± 13.71	-20.384	< 0.001	-3.04
RBANS-DM	71.10 ± 16.69	98.92 ± 10.49	-13.387	< 0.001	-2.00
RBANS-T	74.80 ± 12.70	102.77 ± 10.90	-15.853	< 0.001	-2.36
BMP2, pg/mL	28.23 (23.04, 30.20)	25.29 (21.05, 29.31)	—	—	—
BMP9, pg/mL	9.65 (6.04, 13.50)	7.61 (5.34, 9.27)	—	—	—
BMP10, pg/mL	40.22 (33.13, 47.48)	47.07 (36.46, 59.88)	—	—	—
ln[BMP2]	3.26 ± 0.23	3.12 ± 0.51	2.381	0.018	0.35
ln[BMP9]	2.24 ± 0.70	2.00 ± 0.54	2.578	0.011	0.38
ln[BMP10]	3.65 ± 0.44	3.73 ± 0.69	-0.854	0.394	-0.13

RBANS scores and ln[BMP] concentrations are presented as mean±standard deviation, whereas raw BMP concentrations are presented as median (interquartile range) because of skewed distributions. The p-values and Cohen’s d values were derived from independent-samples t-tests for RBANS scores and ln[BMP] concentrations. Raw BMP concentrations are shown for descriptive interpretability and were not used for group-comparison testing

Furthermore, the effects of group on ln[BMP2] and ln[BMP9] concentrations remained significant after controlling for potential confounders by multivariate analysis of covariance (MANCOVA, with ln[BMPs] as dependent variables, group as the fixed factor, and age, sex, years of education, BMI, and smoking status as covariates). The MANCOVA revealed a significant overall multivariate effect of group on the three BMP measures [Pillai’s trace = 0.074, F(3,171) = 4.548, p = 0.004, partial η² = 0.074], while follow-up univariate analyses of covariance (ANCOVAs) showed that serum ln[BMP2] and ln[BMP9] concentrations remained significantly higher in patients than HCs after adjustment for the above covariates (Table 3): ln[BMP2]: F(1,173) = 6.017, p = 0.015, p_FDR = 0.023, partial η² = 0.034; ln[BMP9]: F(1,173) = 8.656, p = 0.004, p_FDR = 0.012, partial η² = 0.048. Estimated marginal means indicated that ln[BMP2] and ln[BMP9] concentrations were approximately 0.15 and 0.27 log units higher in patients than HCs, respectively, after covariate adjustment. In contrast, ln[BMP10] did not differ between groups either before or after covariate adjustment (F(1,173) = 0.456, p = 0.501, p_FDR = 0.501, partial η² = 0.003).

Table 3.

ANCOVA results comparing serum ln[BMP] concentration between patients and HCs with adjustment for covariates

Dependent variable	Adjusted mean ± SE (Patients)	Adjusted mean ± SE (HCs)	Group mean difference 95% CI	F	p	p FDR	partial η²
ln[BMP2]	3.26 ± 0.04	3.12 ± 0.04	0.15 (0.03, 0.27)	6.017	0.015	0.023	0.034
ln[BMP9]	2.25 ± 0.06	1.99 ± 0.06	0.27 (0.09, 0.44)	8.656	0.004	0.012	0.048
ln[BMP10]	3.66 ± 0.06	3.72 ± 0.06	-0.06 (-0.23, 0.11)	0.456	0.501	0.501	0.003

Values represent estimated marginal mean ± standard error after adjustment for covariates. Covariates included age, sex, BMI, years of education, and smoking status (0 = no, 1 = yes). Group mean differences in serum BMPs were calculated as patients minus controls, with positive values indicating higher concentrations in patients. p values were derived from one‑way analysis of covariance (ANCOVA). p_FDR values represent false discovery rate-adjusted p values for the three ANCOVA group effects calculated using the Benjamini-Hochberg procedure

Correlations of serum BMP concentrations with symptom and cognitive performance metrics in the patient group

As specified a priori, the association between serum BMP2 concentration and RBANS-DM was defined as the primary hypothesis of this study, whereas associations between BMPs and other cognitive domains or clinical symptoms were considered exploratory. The partial correlations of the three BMP measures with PANSS subscale scores and most RBANS indices were generally weak, and the vast majority of partial correlation coefficients did not reach statistical significance after controlling for age, sex, years of education, duration of illness (months), BMI, and smoking status (Fig. 1). After Benjamini-Hochberg FDR correction across the 29 exploratory partial correlations as a single family, no exploratory association reached statistical significance, with the smallest q_FDR value being 0.502. Among clinical symptoms, ln[BMP10] concentration showed a trend‑level negative association with the PANSS-G score (partial r = − 0.200, p = 0.068) (Table 4A) but did not survive FDR correction (q_FDR > 0.05).

Fig. 1.

Partial correlations of serum ln[BMP] concentrations with clinical and cognitive measures [adjusted for age, sex, years of education, duration of illness (months), BMI, and smoking status] displayed as a heat map. Red indicates a positive correlation and blue indicates a negative correlation, with color intensity reflecting correlation strength. ^*denotes the pre-specified primary partial correlation (ln[BMP2]–RBANS-DM), evaluated at two-tailed p < 0.05 and not included in the FDR correction

Table 4.

A. Partial correlations between serum BMP concentrations and PANSS scores. B. Partial correlations between serum ln[BMP] concentrations and RBANS scores

A
		PANSS-P score	PANSS-N score	PANSS-G score	PANSS-T score
ln[BMP2]	r	0.036	-0.119	0.020	-0.060
	p	0.745	0.280	0.859	0.588
	q	0.783	0.551	0.856	0.718
ln[BMP9]	r	0.069	-0.194	0.136	-0.023
	p	0.532	0.077	0.217	0.834
	q	0.673	0.502	0.502	0.856
ln[BMP10]	r	-0.081	0.112	-0.200	-0.096
	p	0.462	0.312	0.068	0.384
	q	0.627	0.554	0.502	0.585
Partial correlation analyses were performed with adjustment for age, sex, years of education, duration of illness (months), BMI, and smoking status. q values denote Benjamini-Hochberg FDR-adjusted p values calculated across all exploratory partial correlations involving PANSS and RBANS outcomes as a single family, excluding the pre-specified primary ln[BMP2]–RBANS-DM association.

B
		RBANS-IM score	RBANS-VS score	RBANS-LAN score	RBANS-ATT score	RBANS-DM score	RBANS-T score
ln[BMP2]	r	-0.139	-0.102	0.047	-0.109	-0.266	-0.154
	p	0.206	0.358	0.671	0.324	0.014	0.161
	q	0.502	0.585	0.774	0.554	—	0.502
ln[BMP9]	r	0.118	0.134	0.192	0.185	0.041	0.141
	p	0.285	0.224	0.080	0.092	0.709	0.199
	q	0.551	0.502	0.502	0.502	0.783	0.502
ln[BMP10]	r	-0.140	-0.172	-0.079	-0.205	-0.082	-0.157
	p	0.203	0.118	0.473	0.061	0.459	0.154
	q	0.502	0.502	0.627	0.502	0.627	0.502

Partial correlation analyses were performed with adjustment for age, sex, years of education, duration of illness (months), BMI, and smoking status. q values denote Benjamini-Hochberg FDR-adjusted p values calculated across all exploratory partial correlations involving PANSS and RBANS outcomes as a single family. The pre-specified primary correlation between ln[BMP2] and RBANS-DM was not included in the FDR correction and is indicated by “—”

However, the pre‑specified primary hypothesis of a significant negative correlation between serum ln[BMP2] and delayed memory performance (RBANS-DM index) was supported by partial correlation analysis (partial r = −0.266, p = 0.014, 95% CI: −0.454, −0.055). This association was domain-selective and modest in magnitude. In contrast, no significant partial correlation was detected between the three BMP measures and the other RBANS domain scores or RBANS-T score (Table 4B).

Multiple regression analysis of the ln[BMP2]–RBANS-DM association

To further examine the robustness of the association between ln[BMP2] and RBANS-DM among patients, we constructed a multivariable linear regression model with RBANS-DM index as the dependent variable, serum ln[BMP2] concentration as the main independent variable, and age, sex, years of education, BMI, smoking status, and duration of illness as covariates. The overall model fit was good [R = 0.599, R² = 0.359, adjusted R² = 0.304, F(7, 82) = 6.566, p < 0.001], and high ln[BMP2] remained a significant negative predictor of RBANS-DM score after controlling for the above covariates (unstandardized coefficient B = −16.976, standardized coefficient β = −0.236, p = 0.014). In this sample, a one-unit increase in ln[BMP2] was associated with an approximately 17-point decrease in RBANS-DM index. Based on a simple linear fit of the partial regression residuals, ln[BMP2] independently accounted for about 7% of the variance in RBANS-DM (partial R² = 0.071, F(1, 82) = 6.255, p = 0.014). Model diagnostics further supported the stability of this primary regression model. No participant had an absolute standardized residual greater than 3, and no Cook’s distance exceeded 1.0, with a maximum Cook’s distance of 0.075. No substantial multicollinearity was observed, as VIF values ranged from 1.043 to 1.357. In contrast to ln[BMP2], age and years of education were positively associated with RBANS-DM scores, whereas sex, BMI, smoking status, and duration of illness were not significantly related to RBANS-DM in the model (Table 5). The scatter plot in Fig. 2 visually depicts this adjusted association.

Table 5.

Multiple linear regression analysis of the associations between RBANS-DM index (dependent variable) and the indicated clinicodemographic factors

Predictor	B	SE	β	t	p	95% CI (lower)	95% CI (upper)
Constant	63.014	25.260	—	2.495	0.015	12.764	113.265
ln[BMP2]	-16.976	6.790	-0.236	-2.500	0.014	-30.483	-3.469
Age	0.361	0.131	0.258	2.762	0.007	0.101	0.621
Sex	0.052	3.692	0.001	0.014	0.989	-7.292	7.397
Years of education	3.577	0.664	0.517	5.388	< 0.001	2.256	4.898
BMI	-0.062	0.499	-0.013	-0.124	0.902	-1.054	0.930
Smoking status	-2.104	6.952	-0.029	-0.303	0.763	-15.934	11.725
Duration of illness	0.171	0.152	0.102	1.128	0.263	-0.131	0.473

Sex was coded as 1 = male and 0 = female. Smoking status was coded as 1 = smoker and 0 = non‑smoker. Multicollinearity diagnostics indicated no significant multicollinearity, with variance inflation factors (VIFs) for all predictors < 1.4

Fig. 2.

Partial regression plot showing the association between serum ln[BMP2] concentration and RBANS-DM index score after adjustment for covariates age, sex, years of education, BMI, smoking status, and duration of illness. The solid line indicates the fitted regression line, and the shaded area represents the 95% confidence interval. Partial R² = 0.071, p = 0.014

Discussion

In this study, we systematically assessed the associations of peripheral circulating BMP2, BMP9, and BMP10 concentrations with clinical symptoms and cognitive functions in FEDN patients with schizophrenia. Compared to demographically matched healthy controls, patients exhibited significantly elevated serum BMP2 and BMP9 concentrations, and these differences remained robust after adjustment for relevant demographic and lifestyle covariates. Adjusted analyses revealed a relatively specific association between higher BMP2 and lower RBANS-DM scores, whereas no consistent associations were observed for other cognitive domains or PANSS symptom scores. Importantly, the observed association between BMP2 and delayed memory was modest and should be interpreted in the context of the cross-sectional design.

While prospective studies with longitudinal follow-up are required to provide evidence for causal associations between BMP elevations and disease onset, prior evidence suggests that pathophysiological changes in signaling pathways within the CNS are often reflected by changes in peripheral circulating factors [26]. However, the present serum-based findings cannot determine whether peripheral BMP alterations directly reflect central BMP signaling. Given prior evidence linking BMP signaling to nervous system development and adult neurogenesis [27, 28], elevated BMP2 and BMP9 may be viewed as peripheral alterations associated with early-stage schizophrenia. Whether these changes reflect compensatory responses, systemic inflammation, or central pathway alterations remains unknown. Potential involvement of downstream pathways such as Smad signaling, hippocampal neurogenesis, and synaptic plasticity is biologically plausible but was not directly examined in this study and should therefore be considered hypothesis-generating.

Early-stage upregulation of neurotrophic factors such as BDNF may reflect protective mechanisms, but sustained signaling could alter neural circuit function, and these interpretations remain speculative [29, 30]. Alternatively, elevated BMP signaling may reflect systemic inflammation. Large-scale meta-analyses have demonstrated widespread abnormalities in peripheral cytokine profiles among first-episode patients with schizophrenia [31, 32], and pro-inflammatory cytokines such as interleukin-1 (IL-1) and tumor necrosis factor-α (TNF-α) can induce BMP2 expression via nuclear factor-κB (NF-κB) [33]. Therefore, the elevated serum BMPs observed in this study may partly reflect an activated systemic inflammatory state. In contrast, no significant between-group difference was observed for BMP10, which may reflect tissue-specific biological functions. Unlike BMP2 and BMP9, which are broadly involved in neural and metabolic regulation, BMP10 plays a highly specialized role in cardiovascular development and angiogenesis [34]. Future large-scale studies are needed to examine whether specific BMP patterns provide clues to pathogenesis or symptom profiles.

Elevated serum BMP2 concentration was an independent correlate of poor RBANS-DM performance in multivariable regression analysis, suggesting that peripheral BMP2 may be relevant to memory performance in schizophrenia. Delayed memory depends on hippocampal structure and function [9], and BMP signaling is a key negative regulator of hippocampal neurogenesis [11]. Specifically, the BMP-Smad pathway acts as a molecular brake within the neural stem cell niche. Upon BMP binding, Smad1/5/8 proteins are phosphorylated, translocate to the nucleus, and regulate gene transcription to maintain quiescence and suppress proliferation and differentiation [11, 12]. Excessive BMP signaling has been reported to reduce newly generated neurons [35], which are implicated in memory encoding and retrieval [36]. However, the present serum-based study cannot determine whether peripheral BMP2 directly reflects hippocampal BMP signaling. It also remains uncertain whether serum BMP2 crosses the BBB or whether BBB alterations in schizophrenia meaningfully influence peripheral BMP measurements [37]. Therefore, the observed association should not be interpreted as direct evidence of central BMP pathway activation. Given the modest effect size, this finding should be considered preliminary and hypothesis-generating. Further validation through multimodal imaging and experimental studies is warranted.

The study identified a significant negative association between serum BMP2 and RBANS-DM, but not with other PANSS or RBANS metrics. Serum BMP9 was elevated but not associated with cognitive or clinical measures, whereas BMP10 neither differed from controls nor correlated with phenotype, although a trend-level association with PANSS-G was observed. This pattern provides indirect comparative support for the specificity of the BMP2-delayed memory association, suggesting that the finding is unlikely to simply reflect a nonspecific change across all measured BMP isoforms. Additional associations may require larger cohorts, given the heterogeneity of schizophrenia. Overall, these findings suggest possible functional differences among BMP isoforms. The selective association between BMP2 and delayed memory, together with the lack of significant associations with PANSS scores, is consistent with previous evidence that cognitive impairment may be partly independent of psychotic symptom severity [38, 39]. Positive and negative symptoms have been linked to subcortical dopamine dysfunction [40], whereas memory deficits may involve cortical and hippocampal glutamatergic and synaptoplastic mechanisms [41]. Although BMP signaling can involve canonical Smad-dependent pathways related to gene transcription and structural remodeling [42], the present serum-based findings did not directly assess Smad activity or central neural mechanisms. Therefore, the BMP2-RBANS-DM association should be interpreted as a peripheral correlate of delayed memory performance rather than direct evidence of a specific pathophysiological mechanism.

The different association patterns across BMP isoforms may reflect biologically distinct peripheral correlates, but this interpretation should remain cautious. Experimental evidence indicates that BMP2 can regulate hippocampal DG neurogenesis by maintaining neural stem cells in quiescence [43], BMP9 supports basal forebrain cholinergic neuron differentiation [44], and BMP10 is mainly implicated in cardiovascular biology [34]. In the present study, only BMP2 showed a robust association with delayed memory, which is consistent with, but does not prove, a memory-related BMP2 pathway. Systemic influences, unmeasured confounding, and limited statistical power may also have contributed to the observed pattern, particularly for BMP10.

Several limitations should be acknowledged. First, although including only FEDN patients minimized confounding by antipsychotics, the cross-sectional design precludes causal conclusions. Longitudinal studies are needed to examine dynamic BMP changes and cognitive trajectories. Second, only peripheral biomarkers were assessed, and it is uncertain how accurately these reflect hippocampal BMP signaling. Third, recruitment from a single center may limit generalizability. Fourth, because the number of smokers was small in both groups, smoking status could only be modeled as a binary covariate, which may not fully account for smoking-related confounding or potential model instability. Future studies should include more detailed measures of smoking exposure. Future multicenter studies incorporating fMRI or PET may help clarify mechanisms linking peripheral BMPs to hippocampal function.

Conclusions

In conclusion, FEDN patients with schizophrenia exhibited significantly elevated peripheral BMP2 and BMP9 concentrations. Serum BMP2 was independently associated with delayed memory performance, whereas BMP9 and BMP10 did not show robust relationships with cognitive function. These findings support serum BMP2 as a candidate peripheral correlate of delayed memory impairment in early-stage schizophrenia, rather than a confirmed mechanistic biomarker. They do not establish causal involvement of BMP2, central BMP signaling, BBB permeability, or hippocampal neurogenesis. Larger multicenter longitudinal studies, ideally combined with neuroimaging or experimental approaches, are needed to clarify the temporal stability, neural relevance, and clinical utility of this association.

Author contributions

Q.Y., J.H., Q.T., and X.Z. conceptualized the study. Q.Y. and J.H. wrote the first draft of the manuscript. J.H., F.J., and P.W. contributed to methodology. Q.Y. and X.Z. curated the data. S.J., F.J., Y.D., X.Y., S.L., and P.W. contributed to investigation and data collection. Q.T. performed formal analysis. X.Z. reviewed and edited the manuscript. S.J. supervised the study. All authors read and approved the manuscript.

Funding

This work was supported by Suzhou Clinical Medical Center for Mood Disorders (Grant No. Szlcyxzx202109); Suzhou Key Laboratory (Grant No. SZS2024016); Multicenter Clinical Research on Major Diseases in Suzhou (Grant No. DZXYJ202413, MR-32-25-054378); Suzhou Municipal Key Project for Applied Basic Research in Medical and Health Sciences (Grant No. SYW2025022); Nantong Municipal Science and Technology Program Guiding Projects (Grant Nos. MSZ2025192; MSZ2025021); Nantong Municipal Health Commission Research Project (Grant Nos. MS2024076; MSZ2025065; MS2023085).

Data availability

The datasets generated and/or analysed during the current study are not publicly available due to the presence of potentially identifiable patient information but are available from the corresponding author on reasonable request.

Declarations

Human ethics and consent to participate

We declare that all experiments on human subjects were conducted in accordance with the Declaration of Helsinki and that all procedures were carried out with the adequate understanding and written consent of the subjects. We also certify that formal approval to conduct the experiments described has been obtained from the human subject review board of our institution. All experimental protocols were approved by the Ethics Committee of the Fourth People’s Hospital of Nantong (approval No. 2022-K015; approval date: May 6, 2022). Informed consent was obtained from all subjects or their guardian. All methods were carried out in accordance with relevant guidelines and regulations.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.