Metabolomics and machine learning approaches for diagnostic biomarkers screening in systemic light chain amyloidosis

Abstract

Delayed diagnosis of systemic light chain (AL) amyloidosis is common and associated with worse survival and early mortality. Current diagnosis still relies on invasive tissue biopsies, highlighting the need for sensitive, noninvasive biomarkers for early diagnosis. This study aims to identify promising biomarkers for the early diagnosis of AL amyloidosis. Peripheral venous blood samples from 70 newly diagnosed systemic AL amyloidosis patients, 48 newly diagnosed multiple myeloma (MM) patients, and 29 healthy controls (HCs) were analyzed using high-performance liquid chromatography-mass spectrometry. Metabolomic profiling revealed distinct metabolic differences between the AL group and the controls (HCs and MM). Machine learning further identified that phytosphingosine and asymmetric dimethylarginine were significantly up-regulated in the AL group compared with HCs group, with area under curve (AUC) values of 0.990 and 0.904, sensitivity and specificity of (97%, 100%) and (88%, 93%), respectively. Compared with MM group, phytosphingosine was also significantly up-regulated in the AL group, with an AUC value of 0.779, sensitivity and specificity of (62%, 88%). Pathway analysis showed significant changes in starch and sucrose metabolism pathway, as well as pentose and glucuronate interconversions pathway between the AL and the controls. Metabolomics combined with machine learning identified phytosphingosine as a promising biomarker for early diagnosis of AL amyloidosis. Two metabolic pathways (starch and sucrose metabolism, pentose and glucuronate interconversions) may reflect the key pathological processes involved in the development and progression of AL amyloidosis. Further confirmation studies are warranted to validate its value in this field.

Supplementary information

The online version contains supplementary material available at 10.1007/s00277-025-06302-4.

Introduction

Systemic light chain (AL) amyloidosis is a clonal plasma cell disorder caused by the misfolding of light chains to aggregate into insoluble amyloid fibrils, causing severe and life-threatening organs dysfunction [1]. AL amyloidosis is a rare disease, with an estimated incidence of about 8–12 cases per million per year in Western countries [2]. Clinical manifestations of AL amyloidosis are complex and heterogeneous, depending on the extent and degree of organ involvement, making it challenging for physicians to diagnose this rare disease. Currently, there are no specific laboratory or imaging tests available for the diagnosis of AL amyloidosis, contributing to frequent delays or misdiagnosis, even among patients with preexisting plasma cell disorders [3–5]. Delay in diagnosis is associated with worse prognosis and early mortality [6], with a median overall survival (OS) of only about 12 months before effective treatment [7]. Over the past few decades, advances in treatment strategies, including refined selection criteria for autologous hematopoietic stem cell transplantation (ASCT) and the development of novel agents, have revolutionized the management of AL amyloidosis, significantly improving the prognosis of these patients, with a gradual but notable increase in the median OS from 1.4 to 4.6 years [8]. Despite a significant improvement in survival, a substantial proportion of patients still face a high risk of early mortality, with up to 25% of patients dying within 6 months of diagnosis [9, 10]. In an era of effective treatment and improved prognosis for AL amyloidosis, the importance of early diagnosis should be further emphasized. Therefore, there is an urgent need to identify sensitive and specific biomarkers for early diagnosis to improve prognosis.

Metabolomics is a novel omics platform that allows the simultaneous and relative quantification of thousands of different metabolites using sensitive and specific methods, such as high-performance liquid chromatography-mass spectrometry (HPLC–MS) [11, 12]. While metabolomics has gained attraction in cancer research for biomarkers discovery [13], there remains a dearth of studies investigating its potential in AL amyloidosis. Our study aims to fill this gap by leveraging metabolomics to explore diagnostic biomarkers and shed light on the pathogenesis of AL amyloidosis.

Methods and materials

Participants

70 patients with biopsy-proven newly diagnosed systemic AL amyloidosis were recruited at Peking University First Hospital (PKUFH) from June 2021 to June 2023. Additionally, 48 patients with newly diagnosed multiple myeloma (MM) and 29 healthy controls (HCs) matched for age and gender, were enrolled during the same period. Systemic AL amyloidosis and MM were diagnosed according to the criteria proposed by the International Myeloma Working Group [14]. All patients with MM routinely undergo bone marrow and abdominal fat biopsy with Congo red staining to exclude secondary AL amyloidosis. Systemic AL amyloidosis patients concomitant with MM, or other related disorders were excluded from the study. This study was approved by the institutional review board of PKUFH (2021yan323) and conducted in accordance with the declaration of Helsinki.

Clinical data collection

Demographic and clinical data for both systemic AL amyloidosis and MM patients were collected from the medical records system of PKUFH at the time of diagnosis. Prognostic staging for systemic AL amyloidosis was assessed using the Mayo 2004 and Mayo 2012 criteria [15, 16]. The renal staging for systemic AL amyloidosis is based on 24 h urine protein and the estimated glomerular filtration rate (eGFR) proposed by Palladini et al. [17]. Organ involvement for systemic AL amyloidosis was established following the 2005 criteria from the International Society of Amyloidosis [18]. Prognostic staging for MM was assessed using the International Staging System (ISS) and Revised-International Staging System (R-ISS) [19, 20].

Plasma samples collection and preparation

Peripheral venous blood samples were collected from newly diagnosed AL amyloidosis patients, newly diagnosed MM patients, and HCs at the time of diagnosis. None of these patients received anti-neoplastic therapy prior to samples collection. The samples were then centrifuged at 1500 × rpm for 5 min at 4 ℃. The supernatants (e.g., plasma samples) were transferred to Eppendorf tubes and stored at –80 °C until further use. For sample preparation, a previously described method was employed [21]. Briefly, 50 μL of each sample was mixed with 150 μL of cold methanol/ethanol (v:v = 1:1) and vortexed for 30 s. The mixture was incubated at − 20 °C overnight, followed by centrifugation at 4 °C for 10 min at a speed of 14, 000 × g. The supernatants were then transferred to 0.2 mL high-performance liquid chromatography (HPLC) vials for analysis. Additionally, to ensure data quality during metabolic profiling, quality control (QC) samples were prepared by pooling equal quantities of the supernatants from all samples. QC samples were interspersed every 10 samples to evaluate the stability and reproducibility of the system throughout the HPLC–MS analysis.

HPLC–MS analysis and data processing

HPLC–MS analysis was performed using a Q Exactive Orbitrap mass spectrometer (Thermo Scientific, Waltham, MA) coupled with a Waters Acquity UPLC system (Waters, USA). Chromatographic separation was achieved on a BEH Amide analytical column (1.0 mm × 100 mm, 1.7 μm particle size; Waters, Milford, MA). Separation was performed under a gradient elution with a binary mobile phase system consisting of A (5 mM ammonium formate in water) and B (acetonitrile). The gradient elution program was as follows: 95% B for 0.5 min; 95‒90% B for 2 min; 90‒40% B for 7 min; 40% B for 1 min; 40‒95% B for 2 min; and 95% B for 1 min. The flow rate was set at 0.2 ml/min, and the column temperature was maintained at 35 °C. The mass spectrometer was capable of the mass spectra acquisition with an Ion Max HESI II. The instrument parameters were set as follows: sheath gas flow 10 Arb; auxiliary gas flow 5 Arb; ion spray voltage 3.5 kV; capillary temperature 300 °C; and auxiliary gas temperature 100 °C. In full MS scan mode, the instrument was set to acquire the m/z in the range of 100 − 1500 with a max inject time of 200 ms, 5 micro scans, an automatic gain control target of 5 × 10⁵, and full MS resolution at 70, 000.

The raw data obtained from the ESI positive ion mode were processed using Progenesis QI software (Waters, Milford, MA, USA). This software facilitated several key steps, including sample alignment, peak picking, peak grouping, deconvolution, and the export of final information. Metabolic features identified in blank samples that showed unsatisfactory linearity upon dilution in QC samples (r < 0.6) were excluded. To streamline the metabolome profile, only those metabolic features detected in more than two-thirds of the samples were retained for further analysis. Subsequently, metabolic features present in over 50% of the study samples were selected for additional analysis. Missing values were addressed by imputing them with half of the minimum value. The data were then log-transformed, and z-score normalization was applied to eliminate the effects of outliers and ensure consistent analysis.

Statistical analysis

Statistical analysis and data visualization in our study were performed using the R language (version 4.2.2) and MetaboAnalyst (version 6.0). Quantitative variables with normal distribution were expressed by means with standard deviation and compared using the Student’s t-test. Quantitative variables without normal distribution were expressed by median (minimum, maximum) and compared using the Mann–Whitney U test. Qualitative variables were expressed by percentage and compared using Chi-square or Fisher’s exact tests. All statistical tests were two-sided and P < 0.05 was considered statistically significant.

Univariate and multivariate statistical analyses were applied to these normalized datasets in order to identify the differential metabolites that can discriminate the AL amyloidosis group from the controls. Univariate statistical methods utilized a fold change (FC) and employed the paired Student’s t-test for normally distributed data. In cases where the data were not normally distributed, the Wilcoxon signed-rank test was used instead. After applying a false discovery rate correction with the Benjamini–Hochberg procedure, we selected differential metabolites that showed a significant combination of |log2 FC|> 1.5) and P < 0.05, and the findings were illustrated using a volcano plot.

Additionally, multivariate statistical methods such as principal component analysis (PCA) and orthogonal partial least squares discriminant analysis (OPLS-DA) were applied to examine the overall distribution of metabolites within the study groups and to assess their effectiveness in distinguishing the AL amyloidosis group from the controls. Unlike PCA, which is an unsupervised multivariate method, OPLS-DA is a supervised approach that requires cross-validation to prevent overfitting of the data. The validity of the OPLS-DA model was confirmed by generating 1000 random permutation models and comparing their performance (R2Y, which indicates goodness of fit, and Q2, which measures predictive ability) with that of the original model. Additionally, the variable importance in projection (VIP) score from the OPLS-DA model was used to rank the metabolites based on their contribution to the separation of the AL amyloidosis group from the controls in the score plot. Metabolites with VIP > 1 were deemed to significantly contribute to the model interpretation.

Finally, the criteria for differential metabolites in our study were defined as meeting the following three standards simultaneously: P < 0.05, |log2 FC|> 1.5, and VIP > 1.0. The differential metabolites were further analyzed to explore potential diagnostic biomarkers for AL amyloidosis by random forest (RF) algorithm. Receiver operating characteristic (ROC) curve analysis was carried out to identify the biomarkers with highest specificity, sensitivity and accuracy to segregate the AL amyloidosis group from the controls.

Furthermore, these differential metabolites were correlated with metabolomic pathways using the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis (http://www.kegg.jp/). Fisher’s exact test was used to determine the relevance of these metabolites to various metabolic and signal transduction pathways.

Results

Demographic and clinical characteristics

Our study enrolled 70 patients with newly diagnosed AL amyloidosis, 48 patients with newly diagnosed MM, and 29 HCs. The baseline characteristics of all patients at the time of diagnosis were summarized in Table 1and Table S1. The mean age of AL, HCs, and MM was 57.0, 55.7, and 60.0 years, respectively (P > 0.05). The proportions of males were 58.6%, 48.3%, and 58.3% for AL, HCs, and MM groups, respectively (P > 0.05). There were no significant differences in age and gender among these groups (P > 0.05). In the AL group, 87.1% were classified as the lambda subtype. The kidney and heacrt were the most involved organs, involved in 88.6% and 54.3% of patients, respectively. Patients in Mayo 2004 stage III and Mayo 2012 stage IV accounted for 36.4% and 10%, respectively. In the MM group, all patients met the criteria for symptomatic MM (e.g., CRAB or SLim). Among them, 60.9% and 35.6% of patients were classified as ISS stage III and R-ISS stage III, respectively.

Table 1.

Demographic and clinical characteristics of participants

Variables	AL amyloidosis N = 70	Healthy controls N = 29	Multiple myelomaN = 48
Age, years	57.0 ± 8.6	55.7 ± 10.5† §	60.0 ± 7.5‡ §
Male, n (%)	41 (58.6)	14 (48.3)† §	28 (58.3)‡ §
Involved light chain, lambda, n (%)	61 (87.1)
Troponin I, ng/ml	0.017 (0.001, 1.271)
Brain natriuretic peptide, pg/ml	147 (14, 3541)
24-h urine protein, g/d	6.30 (0.03, 20.90)
eGFR, ml/min/1.73m2	85.8 (8.7, 124.0)
dFLC, mg/L	103.8 (3.6, 4543.8)
Organ involvement, n (%)
Kidney	62 (88.6)
Heart	38 (54.3)
Liver	10 (14.3)
Peripheral nervous system	12 (17.1)
Gastrointestinal tract	7 (10.0)
Mayo 2004 stage, n (%)
I/II/III	24 (36.4)/18 (27.2)/24 (36.4)
Mayo 2012 stage, n (%)
I/II/III/IV	21 (42.0)/14 (28.0)/10 (20.0)/5 (10.0)
Renal stage, n (%)
I/II/III	25 (36.8)/39 (57.3)/4 (5.9)

dFLC: the absolute difference between involved and uninvolved free light chain; eGFR: estimated glomerular filtration rate; N: number.^† AL amyloidosis vs healthy controls; ^‡ AL amyloidosis vs multiple myeloma; § P > 0.05.

For differential metabolites screening, the AL and HCs samples were randomly divided into training (n = 50) and validation cohorts (n = 49) at a 1:1 ratio. Similarly, the AL and MM samples were randomly divided into training (n = 59) and validation cohorts (n = 59) at a 1:1 ratio. The baseline characteristics of patients in the training and validation cohorts were presented in Table S2-S3, showing that the baseline characteristics of the two cohorts were well-matched (P > 0.05).

Comparative analysis of metabolic profiles of AL, HCs, and MM

In the metabolic analysis, a total of 850 metabolites were detected. Through database searches, we identified 322 metabolites that were present in at least two-thirds of the samples. The PCA score plot indicated that the metabolic profiles of AL group generally differed from those of HCs group (Fig. 1A), while did not show a clearer separation of the AL and MM groups (Fig. 1B). Next, OPLS-DA was performed to maximize variations between different groups and screen for the metabolites that contributed discriminant to this variation. The OPLS-DA score plot showed better separation of metabolic profiles between the AL and controls groups (HCs and MM) (Fig. 1C-D), though there was only minimal overlap between the AL and MM groups (Fig. 1D). A permutation test was performed to evaluate the validity of the OPLS-DA model. The model parameters were as follows: AL vs HCs: R2Y = 0.851, Q2 = 0.742 (Fig. 1E), AL vs MM: R2Y = 0.627, Q2 = 0.514 (Fig. 1F), illustrating the model’s high predictive efficacy and absence of overfitting.

Fig. 1.

The metabolites difference between AL amyloidosis, healthy controls (HCs), and multiple myeloma (MM) in the training cohort. A, B: The Principal Component Analysis (PCA) score plot between the AL (red circle) and controls groups (HCs and MM) (blue circle). C, D: The Orthogonal Partial-Least Squares Discrimination Analysis (OPLS-DA) score plot between the AL (red circle) and controls groups (HCs and MM) (blue circle). E, F: Permutation 1000 test of the OPLS‐DA model (E: AL amyloidosis vs healthy controls, F: AL amyloidosis vs multiple myeloma). The values of R2Y and Q2 represent the goodness of fit and predictability of the model, respectively

Screening differential metabolites between AL, HCs, and MM

We further screened differential metabolites using the VIP values from the OPLS-DA model (Fig. 2A, C). Metabolites with larger VIP values are considered more significant, those with VIP > 1 were selected in our study. Volcano plots were employed to visualize differential metabolites with P < 0.05 and |log2 FC|> 1.5 (Fig. 2B, D). Ultimately, significantly differential metabolites were identified based on a combination of multivariate and univariate statistical criteria: P < 0.05, |log2 FC|> 1.5, and VIP > 1.0. Applying these criteria, we identified 12 differential metabolites between the AL and HCs groups (Table S4), including 6 up-regulated and 6 down-regulated metabolites. Additionally, 13 differential metabolites were identified between the AL and MM groups (Table S5), with 12 up-regulated and 1 down-regulated metabolites.

Fig. 2.

Differential metabolites between AL amyloidosis, healthy controls, and multiple myeloma in the training cohort. A, C: Top 20 metabolites based on the variable importance in the projection (VIP) value of the OPLS-DA model; B, D: Volcano plots of the differential metabolites. The Y coordinate was -log10 (P value) and the X coordinate was log2 FC. Red dots were the upregulated metabolites of significant expression levels. Blue dots were downregulated metabolites of significant expression levels. Gray dots were metabolites of non-significant difference. The cutoff standard was set to P < 0.05 and |log2 FC|> 1.5

Potential diagnostic biomarkers selection by machine learning

Differential metabolites selected based on P < 0.05, |log2 FC|> 1.5 and VIP > 1.0 were further analyzed using the random forest (RF) algorithm to identify potential diagnostic biomarkers for AL amyloidosis. For the AL vs HCs comparison, the top three differential metabolites identified by RF were phytosphingosine, N-Alpha-acetyllysine, and asymmetric dimethylarginine (ADMA). Phytosphingosine and ADMA were significantly up-regulated in the AL group compared with HCs group (P < 0.001; Fig. 3A, C). ROC curve analysis determined the effectiveness of these biomarkers. In the training cohort, the AUC values of phytosphingosine and ADMA were 0.990 and 0.904, with sensitivity and specificity of (97%, 100%) and (88%, 93%), respectively (Fig. 3D, F). In the validation cohort, the AUC values were 0.994 and 0.834, with sensitivity and specificity of (100%, 93%) and (82%, 100%), respectively (Fig. S1D, F).

Fig. 3.

Expression difference (A-C) and ROC curves (D-F) of the top three differential metabolites by random forest between AL amyloidosis and healthy controls in the training cohort. The red dots were the optimal cut-off (sensitivity, specificity). *** P < 0.001

For the AL vs MM comparison, the top three differential metabolites identified by RF were ophthalmic acid, glycochenodeoxycholic acid 3-glucuronide, and phytosphingosine. Compared with MM group, phytosphingosine was significantly up-regulated in the AL group (P < 0.001, Fig. 4C). In the training cohort, the AUC of phytosphingosine was 0.779, with sensitivity and specificity of (62%, 88%) (Fig. 4F). In the validation cohort, the AUC of phytosphingosine was 0.799, with sensitivity and specificity of (74%, 79%) (Fig. S2F).

Fig. 4.

Expression difference (A-C) and ROC curves (D-F) of the top three differential metabolites by random forest between AL amyloidosis and multiple myeloma in the training cohort. The red dots were the optimal cut-off (sensitivity, specificity). *** P < 0.001

Pathway analysis of differential metabolites

To explore the metabolic pathways that potentially contribute to AL amyloidosis, differential metabolites (Table S4-S5) from AL vs HCs, and AL vs MM comparisons were analyzed using MetaboAnalyst 6.0. The metabolic pathway analysis results were shown in Fig. 5and Table S6-S7. Pathways with P < 0.05 or impact values > 0.1 were considered highly influential. Compared with HCs group, AL group exhibited disturbances in four metabolic pathways, including starch and sucrose metabolism (P < 0.001), ascorbate and aldarate metabolism (P = 0.023), pentose and glucuronate interconversions (P = 0.047, impact = 0.108) and fructose and mannose metabolism (P = 0.049) (Fig. 5A). Compared with MM group, AL group showed disturbances in three metabolic pathways: starch and sucrose metabolism (P = 0.005), taurine and hypotaurine metabolism (P = 0.049), and pentose and glucuronate interconversions (impact = 0.108) (Fig. 5B). Notably, starch and sucrose metabolism, and pentose and glucuronate interconversions were significantly altered in both comparisons (AL vs HCs and AL vs MM).

Fig. 5.

Pathways analysis of the differential metabolites. A: AL amyloidosis and healthy controls; B: AL amyloidosis and multiple myeloma. Node color and radius are based on the P value and pathway impact value, respectively. Labeled nodes denote significantly changed pathways with P < 0.05 or pathway impact > 0.1

Discussion

With recent technological breakthroughs, multi-omics approaches have become increasingly prominent in research. The metabolome represents the most downstream level of the genome, transcriptome, proteome, and environmental changes, providing the most complete and dynamic phenotype [22]. There are two main approaches in metabolomics: targeted metabolomics and non-targeted metabolomics. Targeted metabolomics focuses on quantifying specific metabolites of interest, often based on prior knowledge or hypotheses. In contrast, non-targeted metabolomics aims to characterize as many metabolites as possible within a sample without prior knowledge, making it particularly useful for discovering new biomarkers and exploring unknown metabolic changes [23].

Our study, using HPLC–MS technology, identified significant metabolomic differences between the AL amyloidosis and the controls (MM and HCs). Applying strict filtration criteria, 12 differential metabolites between the AL and HCs groups and 13 differential metabolites between the AL and MM groups were identified. To optimize the screening of diagnostic biomarkers, machine learning was employed to refine and reduce the dimensionality of the data, enhancing the accuracy of potential biomarkers. Two metabolites, including phytosphingosine and ADMA, emerged as promising biomarkers for distinguishing AL amyloidosis from HCs (AUC 0.904 and 0.990), while phytosphingosine also showed potential in differentiating AL from MM (AUC 0.779). These biomarkers also had good diagnostic performance in the validation cohort, suggesting their clinical application for screening AL amyloidosis from MM and HCs.

The endoplasmic reticulum (ER) is essential for protein maturation, quality control, and trafficking [24]. As a key organelle of the cell, the ER cannot be synthesized de novo and is generated only from pre-existing ER [25]. To ensure that daughter cells inherit a functional ER, cells employ an endoplasmic reticulum stress surveillance (ERSU) mechanism [26, 27]. When functional demands on the ER exceed its capacity, unfolded and misfolded proteins accumulate in the ER, resulting in ER stress that activates the ERSU pathway to prevent transmission of a potentially damaged ER into the daughter cells [28]. Pina et al. study has identified phytosphingosine as a key inducer of the ERSU pathway. Under ER stress, phytosphingosine levels increase, which then activates all the marker events of the ERSU pathway. These include ER inheritance block, septin ring mislocalization, and Slt2 phosphorylation, all of which contribute to halting the transmission of functionally impaired ER to daughter cells and stopping cell division [29]. Previous studies have found that amyloidogenic plasma cells secrete amyloidogenic light chains, leading to ER stress [30]. Therefore, we speculate that the significant increase of phytosphingosine in AL amyloidosis may be a response to ER stress. This response likely involves the activation of the ERSU pathway to ensure the survival and proper functioning of daughter cells under these stressful conditions.

ADMA is synthesized when arginine residues in proteins are methylated by protein arginine methyltransferases. Protein arginine methylation is a post-translational modification [31]. The metabolism of ADMA is primarily carried out by dimethylarginine dimethylaminohydrolase (DDAH) [32]. DDAH is highly sensitive to oxidative stress, therefore, pathological stimuli that induce oxidative stress in endothelial cells, such as inflammatory cytokines, hyperglycemia, infectious factors, and oxidized low-density lipoprotein, may reduce DDAH activity and consequently increase ADMA levels [33]. ADMA acts as an endogenous inhibitor of nitric oxide (NO) synthase. Elevated ADMA levels lead to decreased NO production, which adversely affects endothelial function and raises the risk of cardiovascular events such as atherosclerosis and hypertension [34]. ADMA has been identified as an effective and independent prognostic marker for several cardiovascular diseases and chronic kidney disease [35]. A recent study has investigated the potential role of ADMA in exacerbating cognitive dysfunction associated with cerebrovascular β-amyloidosis [36]. It has been reported that the amyloidogenic light chain can induce oxidative stress, which may contribute to microvascular dysfunction [37]. We speculate that this oxidative stress, driven by the amyloidogenic light chain, could lead to reduced DDAH activity and elevated ADMA levels, further impairing endothelial function.

To better elucidate the mechanism of AL amyloidosis pathogenesis, a metabolic pathway analysis of differential metabolites was conducted. Two pathways (starch and sucrose metabolism, pentose and glucuronate interconversions) showed significant changes in both comparisons (AL vs HCs and AL vs MM). These alterations may reflect the key pathological processes involved in the development and progression of AL amyloidosis. Further confirmation studies are warranted to validate its value in this field.

However, this study also has several limitations: (1) AL amyloidosis is a rare disease, and therefore, our sample size is limited. Future research will require the inclusion of a larger cohort. (2) The potential biomarkers identified in this study need to be validated in independent sample sets, along with functional experiments to investigate their role in the onset and progression of AL amyloidosis.

Supplementary information

Below is the link to the electronic supplementary material.

Author contributions

Conceptualization, Yujun Dong, Zeyin Liang and Huihui Liu; methodology, Weiwei Xie and Zhizhen Lai; validation, Qian Wang; formal analysis, Zhizhen Lai and Zeyin Liang; data curation, Wenqiong Wang and Jin Wang; writing—original draft preparation, Weiwei Xie; writing—review and editing, Yunjun Dong; supervision, Huihui Liu. All authors have read and agreed to the published version of the manuscript.

Funding

This study was supported by Beijing Natural Science Foundation No. 7232175 and No. J230024.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval

This study was approved by the institutional review board of PKUFH (2021yan323) and conducted in accordance with the declaration of Helsinki.

Competing interests

The authors declare no competing interests.

Consent to participate

Written informed consent was obtained from the patients.