Immunoglobulin glycosylation profiling for early identification of patients with severe influenza pneumonia

Abstract

Background

Uncontrolled inflammation can result in severe status and even death in influenza patients, and there is a lack of early clinical evaluation models.

Methods

We recruited patients with influenza pneumonia and healthy controls from the emergency departments of three urban teaching hospitals in Beijing, China, during the winter of 2018–2019. Donated plasma samples were screened using protein and lectin microarrays to assess changes in the glycosylation patterns of immunoglobulins. These changes were used to develop and validate an Immunoglobulin Glycosylation Profile for Severe Status Identification Algorithm (IGPSSIA). A combined model of IGPSSIA score and clinical indicators was constructed to identify severe influenza pneumonia cases.

Results

We enrolled 114 patients, including 56 in the mild and 58 in the severe groups, and recruited 27 volunteers as healthy controls. We screened out the differentially expressed glycan moieties between the mild and the severe groups and included them in the LASSO regression analysis. In the training set (70% of patients, n = 80), IGPSSIA = 1.113 × [GNL‑IgG4 Man] − 2.499 × [LCA‑IgG1 Man] + 0.029 × [BanLec‑IgA2 Man] + 0.529 × [HHL, AL‑IgM Man] − 2.210 × [sWGA‑IgG GlcNAc] + 0.001 × [PSA‑IgG4 Man] + 0.027 × [Ricin B Chain‑IgG2 Gal & GalNAc]. Finally, we constructed a clinical diagnostic model using age, time interval from onset to admission, lymphocyte count, platelet count and IGPSSIA score, and achieved an AUC of 0.839 (95% CI 0.767–0.911).

Conclusions

Changes in immunoglobulin glycosylation profiles can be a promising tool for identifying severe status in patients with influenza pneumonia.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12879-025-11538-6.

Introduction

Seasonal influenza is an acute respiratory infection caused by influenza viruses, which is a serious cause of morbidity and mortality worldwide [1]. Several factors including the characteristics of the virus, vaccine strategies, and use of neuraminidase inhibitors (NAIs) affect the influenza severity. In virus-infected patients, appropriate antibody-dependent inflammation (ADI) plays an important physiological role in counteracting virus infections. However, undesired overactivation of immune responses and even cytokine storms, common with influenza and COVID-19, can lead to various disorders [2, 3].

Immunoglobulins or antibodies play a central role in various types of ADI. Immunoglobulin glycosylation is involved in cytokine storms and immune imbalances in patients with severe viral infections [4, 5]. The glycosylation of proteins is a dynamic post-translational modification that modulates the spatial conformations and biological functions of the majority of cell surface and secreted proteins [6]. Previous studies have shown a correlation between immunoglobulin glycosylation and immune diseases. For example, alterations in the galactosylation, sialylation, and fucosylation of immunoglobulin A (IgA)1 and IgA2 are associated with IgA nephropathy and glomerular function [7]. Anti-spike immunoglobulin G (IgG) from patients with severe COVID-19 intrinsically causes a greater proinflammatory response because of the different glycosylation pattern of the IgG [8].

Recent studies have highlighted the role of immunoglobulin glycosylation in influenza. One study found that while IgG galactosylation remained stable in influenza patients, sialylation, fucosylation, and bisecting GlcNAc levels varied at different stages of the disease [9]. Another investigation reported that following influenza vaccination, high responders exhibited a decrease in high mannose glycans, a change not observed in nonresponders, though the underlying reasons remain unclear [10]. These findings suggest that immunoglobulin glycosylation undergoes dynamic modifications during both influenza infection and vaccination. However, whether plasma immunoglobulin glycosylation profiles can serve as markers for disease severity in influenza pneumonia remains unclear.

Emerging proteomics technology has allowed high-throughput, high-sensitivity, and high-specificity lectin microarrays to be used in clinical research. By utilizing agglutinins’ sugar structure recognition properties, lectin microarrays can simultaneously screen and cross-validate a range of glycan moieties [11]. In this study, we analyzed plasma immunoglobulin glycosylation profiles in patients with influenza pneumonia and healthy controls using a lectin microarray assay, aiming to evaluate the potential of aberrant glycosylation patterns and clinical indicators in identifying severe disease status.

Participants and methods

Study design and participants

We conducted a prospective and observational study during the winter of 2018–2019, and consecutively screened patients with influenza-like complicated pneumonia in the emergency departments (EDs) of Beijing Chao-Yang Hospital, Beijing Ditan Hospital, and Miyun Teaching Hospital, all affiliated to Capital Medical University, Beijing, China. All participating hospitals adhered to the same inclusion and exclusion criteria, and standardized protocols were followed for patient recruitment, sample collection, and laboratory testing to ensure methodological consistency and minimize potential biases arising from inter-hospital variations. The inclusion criteria and procedures were as follows: (i) We screened patients with acute onset, fever, and/or newly developed lower respiratory tract infection symptoms. (ii) We tested all enrolled patients for influenza A virus by reverse transcription polymerase chain reaction of nasopharyngeal swab, sputum or endotracheal aspiration specimens in accordance with guidelines [12], and selected patients with a positive result for one specimen. (iii) We performed chest imaging (CT or X-ray) on these enrolled patients, and patients with lung texture thickened with blurred margins, patchy infiltrating shadows, leaf/segment consolidation shadows, ground glass shadows, or interstitial changes were were included in the study cohort. Exclusion criteria included: (i) patients who died within 24 h of admission or were in an unresuscitable terminal state and (ii) patients whose standardized treatment was terminated due to various reasons or accidents. Healthy controls were recruited from volunteer staff members at Beijing Chao-Yang Hospital. Inclusion criteria for these controls were:​ (i) No history of acute or chronic diseases in the past 6 months.​ (ii) No influenza vaccination in the past year.

We used 72 h of admission as the study endpoint. The selected patients were divided into two groups: (i) the mild group was defined as having an oxygenation index (PaO₂/FiO₂) consistently greater than 200 mmHg within 72 h of admission, and (ii) the severe group was defined as having a PaO₂/FiO₂ ≤ 200mmHg upon admission or deteriorating within 72 h.

This study was reviewed and approved by the Institutional Review Board and Medical Ethics Committee of Beijing Chao-Yang Hospital (Ethical approval number: 2019-ke-301), and the other two hospitals mutually recognized this approval. Informed consent from all participants was acquired.

Plasma sample Preparation and data collection

At admission, peripheral blood (2 mL) was collected in an ethylenediaminetetraacetic acid-containing anticoagulant tube, and then blood samples were centrifuged at 1300 g at 4°C for 10 min. After centrifugation, the supernatant (plasma) was placed in a 1.5-mL centrifuge tube and stored at − 80°C.

Two physicians independently collected and reviewed data on medical identification, demographic characteristics, medical history, co-morbidities, clinical symptoms, physical examinations, laboratory results, and radiological findings. Any discrepancies in the data collected by the two doctors were corrected by a senior physician.

Plasma Immunoglobulin quantification with protein microarrays

Diluted plasma samples (10 to 500-fold) and serially diluted immunoglobulin standards were printed onto a 3D-modified slide surface (Capital Biochip Corp, Beijing, China) with two replicates using an Arrayjet microarrayer (Roslin, UK). Phosphate-buffered saline (PBS) and bovine serum albumin (BSA; 1 mg/mL) (Sigma-Aldrich, St. Louis, MO, USA) were used as negative controls. The slides were scanned using the Genepix 4000 A microarray scanner (Molecular Devices, San Jose, CA, USA). The signal intensity was extracted using GenePix Pro image analysis software (Molecular Devices). A four-parameter logistic standard curve was generated for each immunoglobulin subclass to describe the relationship between absolute immunoglobulin concentration and signal intensity. Diluted plasma samples were interpolated by the least-squares method. The absolute concentration of each immunoglobulin subclass in the plasma samples was imputed using the Newton-GC method according to the fitted standard curve.

Preparation of lectin microarrays

A panel of 47 lectins (Vector Laboratories/EY Laboratories, USA) was prepared at a concentration of 1 mg/mL in PBS (pH 7.4) containing 0.2 mg/mL BSA. All lectins were printed on a 3D-modified slide (Capital Biochip Corp, Beijing, China) in two replicates by an Arrayjet microarrayer to form an array. PBS and BSA (1 mg/mL) (Sigma-Aldrich, MO, USA) were printed as negative controls. Pooled human immunoglobulins (IgA, IgA1, IgA2, IgG, IgG1, IgG2, IgG3, IgG4, and IgM) were used as positive controls. The prepared lectin microarrays were stored at − 20°C until use. The recognition relationship between lectins and glycogroups was shown in Supplementary Table 1.

Detection of plasma Immunoglobulin glycosylation with lectin microarrays

All plasma samples were diluted 1:100 with 1% BSA in PBS containing 0.05% (v/v) Tween 20 (PBST). The lectin microarrays were blocked with 500 µL of PBST containing 1% BSA for 1 h and then incubated with the diluted plasma for 2 h. After washing with PBST, the microarrays were then incubated for 1 h with the corresponding fluorescent-labeled antibody (0.5 µg/mL) [Donkey anti-human IgG Fc-Alex 555, Rabbit anti-human IgA Fc-Alex 647, AffiniPure Goat anti-human IgM FC5µ-Alexa fluor^® 647 (Jackson ImmunoResearch, USA); Mouse anti-human IgG1 Hinge-Alexa Fluor^® 488, mouse anti-human IgG2 Fc-Alexa Fluor^® 488, mouse anti-human IgG3 Hinge-Alexa Fluor^® 647, mouse anti-human IgG4 Fc-Alexa Fluor^® 647, mouse anti-Human IgA1-Alexa Fluor^® 647, and mouse anti-Human IgA2-Alexa Fluor^® 488 (SouthernBiotech, USA)]. After washing with PBST and deionized water, the array was dried and scanned using a Genepix 4300 A microarray scanner (Molecular Devices). The signal intensity was extracted using GenePix Pro image analysis software (Molecular Devices).

Statistical analysis

A total of 423 (47 × 9) lectin fluorescence images were read and quantified as signal intensity values. To eliminate the effect of immunoglobulin levels, each lectin signal value of the microarray was corrected by dividing the corresponding quantitative value of the immunoglobulin concentration to obtain a lectin signal value per unit immunoglobulin concentration.

Student’s t-test was used to compare two groups with normally distributed variables, while Mann-Whitney U test was employed to compare two groups with non-normally distributed variables. For multiple comparisons, one-way analysis of variance (ANOVA) and Kruskal-Wallis test were respectively used for parametric and non-parametric data. Chi-square test or Fisher’s exact test was used for analyzing categorical variables. Spearman correlation analysis was conducted to determine the correlation coefficient between two variables. Statistical significance was defined as a two-tailed p value < 0.05. All statistical analyses were performed using R software (version 4.1.0; Rstudio, Boston, Massachusetts).

Variable selection and model construction

Firstly, differentially expressed glycoproteins between the mild and severe groups were initially identified using uncorrected p-values (p < 0.05) in exploratory analyses. Next, enrolled influenza patients were randomly assigned in a 7:3 ratio to a training set (70%) for score derivation and a validation set (30%) for performance assessment. Glycoproteins with p < 0.05 were selected for the least absolute shrinkage and selection operator (LASSO) regression analysis using the R package “glmnet”. We performed 1000 iterations of LASSO regression analysis in the training set and selected glycoproteins that appeared more than 300 times. Finally, a multivariable logistic regression model was used to construct the Immunoglobulin Glycosylation Profile Severe Status Identification Algorithm (IGPSSIA), and the estimated regression coefficients were obtained for each sample. The IGPSSIA scoring formula was established as follows:

n, coef, and exp corresponded to the number of glycoproteins, regression coefficients, and expression levels of the associated glycoproteins, respectively.

Receiver-operating characteristic (ROC) curves were generated to evaluate the model, and the areas under the curves (AUCs) were calculated in the training and the validation sets.

Results

Participant characteristics

We screened 188 patients with suspected influenza-like onset combined with lower respiratory tract infection, and excluded 70 of them without laboratory evidence of influenza A or radiographic evidence of pneumonia. Of the 118 selected patients, four were excluded, including one died within 24 h, one was in an unresuscitable terminal state, and two were lost to follow-up due to transfers. Finally, we enrolled 114 patients in the study cohort, of which 56 were in the mild group and 58 were in the severe group (Fig. 1). The distribution of patients across the three participating hospitals was as follows: Beijing Chao-Yang Hospital contributed 92 patients (45 mild, 47 severe), Beijing Ditan Hospital enrolled 11 patients (5 mild, 6 severe), and Miyun Teaching Hospital included 11 patients (6 mild, 5 severe). At admission, demographic and clinical characteristics of the two groups were compared in Table 1. None of the participants had influenza vaccinations documented in their medical records. During the same period, 27 volunteers were recruited as healthy controls: the average age was 29.2 ± 4.2 years; 15/27 were male; the average body mass index (BMI) was 22.8 ± 2.8; and the volunteers had no underlying medical conditions.

Fig. 1.

Study flowchart. Mild group: PaO₂/FiO₂ >200 mmHg; Severe group: PaO₂/FiO₂ ≤ 200 mmHg at admission or deterioration

Table 1.

Characteristics of patients in mild and severe groups

Characteristics	Mild group n = 56	Severe group n = 58	p value
Age (years), (mean ± SD)	60.9 ± 17.9	64.2 ± 16.7	0.307
Sex (male), n (%)	37 (66.1)	35 (60.3)	0.526
BMIa, (mean ± SD)	24.55 ± 4.05	23.65 ± 3.67	0.216
Time interval from onset to admission (days), (mean ± SD)	4.2 ± 2.8	5.7 ± 2.8	0.005
NAIsb administration within 48 h, n (%)	7 (12.5)	3 (5.2)	0.293
Co-morbidities, n (%)
Immunodeficiencyc	6 (10.7)	5 (8.6)	0.705
Chronic respiratory diseased	7 (12.5)	8 (13.8)	0.838
Diabetes mellitus	12 (21.4)	18 (31.0)	0.244
Clinical indicators, (mean ± SD)
Lymphocyte count (×109/L)	15.38 ± 8.82	10.92 ± 7.58	0.004
Platelet count (×109/L)	187 ± 86	167 ± 79	0.189
D-dimer concentration (mg/L FEU)	2.93 ± 7.40	5.21 ± 11.41	0.244

^aBMI Body mass index

^bNAIs Neuraminidase inhibitors

^cImmunodeficiency active cancer, long-term use of glucocorticoids, anti-rejection therapy of solid organ transplantation

^dChronic respiratory disease Asthma, chronic obstructive pulmonary disease, pulmonary interstitial fibrosis, bronchiectasis

Glycosylation profiling using lectin microarrays

Using protein microarrays, we quantified the plasma IgA, IgA1 and 2, IgG, IgG1-4, and IgM levels of each subject’s sample as described in a previous publication [13]. We obtained 423 (47 × 9) types of lectin signal intensity values of each subject’s sample using lectin microarrays. We calculated the lectin signal values for unit immunoglobulin concentrations in each subject’s sample, reflecting the corresponding number of glycans on the lectins.

Firstly, we compared the glycosylation profiles between healthy individuals and influenza patients and performed False Discovery Rate (FDR) correction. The results showed a total of 78 differentially expressed glycan moieties. We visualized them using a volcano plot (Fig. 2A), where red indicated upregulated glycan moieties in influenza patients, and blue represented downregulated glycan moieties. The heatmap of differentially expressed glycan moieties revealed significant expression differences between the two groups (Fig. 2B).

Fig. 2.

Glycosylation profile analysis. A-B Volcano plot and Heatmap of differential immunoglobulin glycosylation profiles between influenza patients (n = 114) and healthy controls (n = 27). C-D Volcano plot and Heatmap of differentially expressed glycosylation profiles between severe (n = 58) and mild (n = 56) influenza patients

Development and validation of an immunoglobulin glycosylation profile severe status identification algorithm (IGPSSIA)

In order to explore the relationship between immunoglobulin glycosylation profiles and severe influenza, we initially compared the levels of glycan moieties in the entire patient cohort. There were 20 differentially expressed glycan moieties between the mild and the severe groups, as shown in Fig. 2C and D. Next, we performed 1000 iterations of LASSO regression analysis on these 20 variables in the training set (70% of patients, n = 80) and selected 7 variables with more than 300 appearances, which were shown in Fig. 3A. These 7 lectins along with their recognized glycan types and structures was presented in Table 2.

Fig. 3.

IGPSSIA model development. A LASSO-selected features: Red: >300 iterations, Blue: <100 iterations. B-C ROC curves: Training (AUC = 0.786) and validation (AUC = 0.850) sets of IGPSSIA score

Table 2.

Lectin variables associated with severe influenza selected by LASSO regression and their corresponding Immunoglobulin glycan moieties

Variables	Full name of lectins	Glycan Binding Specificity
GNL‑IgG4	Galanthus Nivalis Lectin	α−1,3-linked mannose oligosaccharides
LCA‑IgG1	Lens Culinaris Agglutinin	non-terminal, α-linked mannose residues
BanLec‑IgA2	Musa Paradisiaca (Banana) Lectin	α−1,2-, α−1,3-, α−1,6-linked mannose; β−1,3- and β−1,6-linked glucosyl structures
HHL, AL‑IgM	Hippeastrum Hybrid (Amaryllis) Lectin	N-glycan structures with Man5–Man8; α−1,3- and α−1,6-linked mannose structures
sWGA‑IgG	succinylated Wheatgerm Agglutinin	N-acetylglucosamine
PSA‑IgG4	Pisum Sativum Agglutinin	α-linked terminal mannose residues
Ricin B Chain‑IgG2	Ricinus Communis Agglutinin B Chain	terminal β-linked galactose and N-acetylgalactosamine residues

Finally, we constructed a multivariable logistic regression model based on these 7 glycoproteins and calculated their corresponding coefficients. The IGPSSIA score for each patient was then determined as follows:

• IGPSSIA =

• 1.113 × [GNL‑IgG4 Man]

• − 2.499 × [LCA‑IgG1 Man]

• + 0.029 × [BanLec‑IgA2 Man]

• + 0.529 × [HHL, AL‑IgM Man]

• − 2.210 × [sWGA‑IgG GlcNAc]

• + 0.001 × [PSA‑IgG4 Man]

• + 0.027 × [Ricin B Chain‑IgG2 Gal & GalNAc]

where each bracketed term is the normalized assay value for the specified lectin‑ immunoglobulin glycosylation.

To evaluate the efficiency of the IGPSSIA score, we calculated the AUC of the training set was 0.786 (95% CI 0.686–0.885) (Fig. 3B), and the AUC of the validation set (30% of patients, n = 34) was 0.850 (95% CI 0.72–0.979) (Fig. 3C).

Diagnostic value of IGPSSIA score combined with clinical indicators

We chose several potential clinical indicators, including age, BMI, time interval from onset to admission, NAIs administration within 48 h, D-dimer concentration, lymphocyte count, platelet count in conjunction with IGPSSIA score to recognize severe status in patients with influenza pneumonia. These clinical variables were first screened via univariate logistic regression (p < 0.10 threshold), which excluded BMI (p = 0.111), D-dimer (p = 0.869), and NAIs administration within 48 h (p = 0.556). The remaining variables (age, time interval from onset to admission, lymphocyte count, and platelet count) were further refined using ​multivariable logistic regression analysis, and their ROC performance in predicting influenza severity was individually delineated (Fig. 4A-D). The selected clinical parameters were combined with the IGPSSIA score to establish a clinical diagnostic model with an AUC of 0.839 (95% CI 0.767–0.911) (Fig. 5).

Fig. 4.

Clinical variable ROC curves. A Age (AUC = 0.549). B Time interval from onset to admission (AUC = 0.672). C Lymphocyte count (AUC = 0.671). D Platelet count (AUC = 0.577)

Fig. 5.

ROC curve of the clinical diagnosis model (AUC = 0.839)

Discussion

In this study, we profiled plasma immunoglobulin glycosylation using proteomics and lectin microarray technologies. Significant differences in binding signals were observed between influenza patients and healthy controls, as well as between mild and severe cases. We developed and validated the IGPSSIA score to identify severe influenza. Finally, we constructed a model combining the IGPSSIA score with age, time from onset to admission, lymphocyte count, and platelet count. This model shows potential as a screening tool for identifying severe or progressive influenza pneumonia.

Glycosylation affects immune homeostasis by modulating the pro- and anti-inflammatory functions of antibodies. In this study, we observed that patients with severe influenza exhibited elevated levels of glycoforms bearing mannose N-glycans identified by HHL, AL, GNL, PSA, and BanLec on IgM, IgG4, and IgA2 subclasses. These lectins preferentially bind to outer-arm high-mannose structures [14–17]. Conversely, a negative correlation was found between disease severity and the presence of core α-linked mannose structures (Man₃GlcNAc₂ core) on IgG1, as identified by LCA [18]. These findings suggest that severe influenza is characterized by subclass-specific alterations in immunoglobulin glycosylation patterns, notably an increase in outer-arm high-mannose structures on IgM, IgG4, and IgA2, and a decrease in core mannose residues on IgG1. Interestingly, a similar pattern has been observed in COVID-19, where patients with severe disease exhibit increased high-mannose IgM despite a decline in total mannose levels [19]. Influenza infection has been reported to upregulate high-mannose glycans on the surface of host cells. These structures are recognized by mannose-binding lectin 2 (MBL2), triggering innate immune activation and potentially amplifying inflammation, leading to lung damage and worse clinical outcomes [20]. Such glycosylation changes may influence immune responses and disease progression. Notably, a study has observed that high responders to influenza vaccination exhibit a decrease in high-mannose glycans postvaccination [10]. This finding suggests that lower levels of high-mannose glycans may be associated with a more effective immune response, indirectly supporting the notion that elevated high-mannose levels could be detrimental. Besides, our results show that severe influenza is negatively correlated with IgG GlcNAcylation recognized by sWGA. This finding is similar to a previous study that IgG GlcNAc decreased in the early stages of influenza [9], although the specific mechanism has not yet been clarified. A study on glycosylation profiles of IgG4-related disease holds that while the mannose level of serum IgG4 reflects the degree of inflammation, the reduced level of GlcNAc was associated with damage to multiple organs [21]. In addition, our results suggest that IgG2 Gal and GalNAc recognized by Ricin B Chain are associated with influenza severity. The IgG galactosylation is heterogeneous, which may result in distinct pro- or anti-inflammatory effects, indicating that there may be more complex biological pathways behind glycosylation. Notably, the observed proinflammatory effect of IgG galactosylation in certain contexts may depend on its contribution to antibody-dependent cellular cytotoxicity (ADCC) and complement-dependent cytotoxicity (CDC) [22, 23].

This study investigated how alterations in the complex glycosylation network of immunoglobulins may serve as prognostic biomarkers during the acute phase of influenza pneumonia. Acute viral infections are known to trigger an initial pro-inflammatory response, often followed by a compensatory anti-inflammatory phase. The predominance of specific pro- or anti-inflammatory glycan structures on immunoglobulins may reflect the stage of the inflammatory response. Notably, immunoglobulin glycosylation profiles hold promise not only as early prognostic markers but also as potential targets for therapeutic intervention.

Some studies have investigated aging, genetics, chronic inflammation, and the environment as contributing factors to abnormal glycosylation in individuals [24, 25]. In our study, however, the IGPSSIA score demonstrated an independent association with disease severity in patients with influenza pneumonia. This suggests that, during acute progressive inflammatory states, the impact of baseline physiological variations or chronic comorbidities on immunoglobulin glycosylation may be minimal. Nevertheless, age was included as a covariate in the final clinical diagnostic model to account for its potential confounding effect, ensuring that age-related factors are appropriately adjusted in the model’s predictions of severity. Some younger people enrolled in the study without co-morbidities unexpectedly underwent the worst outcome, which could be partially attributed to excessive uncontrolled inflammatory responses associated with abnormal glycosylation profiles. According to our findings and a recent report, determining the interaction mechanisms between severe influenza infections, human susceptibility, and the production of immunoglobulins with abnormal glycosylation is a complex challenge, but understanding these relationships is desirable for developing clinical management strategies for influenza infections [26].

In addition, we found that age and time interval from onset to admission were positively correlated with severe status, while lymphocyte count and platelet count were negatively correlated with severe status. Similarly, a study amongst adults hospitalized by influenza A H1N1 virus pneumonia showed that non-survivors were associated with lower lymphocytes [27]. In influenza and COVID-19 patients, platelet hyperreactivity and thrombocytopenia were often accompanied by a worse hyper-inflammatory response, and then were associated with a severe state and an adverse outcome [28]. Therefore, the IGPSSIA score combined with lymphocytopenia and thrombocytopenia can improve the predictive efficacy for severe influenza.

Several limitations of this approach need to be addressed. First, while lectin-based methods are valuable for initial glycan profiling due to their simplicity and cost-effectiveness, they have an inherent imprecise nature. High-resolution techniques, such as mass spectrometry, are essential for detailed structural characterization and accurate quantification of immunoglobulin glycosylation patterns. Second, the relatively small sample size may limit the generalizability of our findings. The variability in the “time interval from onset to admission” was not rigorously controlled, which may have resulted in patients being at different stages of inflammation. Larger cohort studies are needed to enhance the model’s robustness, particularly by stratifying patients based on this time interval. Third, all variables in this study reflect dynamic changes following influenza virus infection; however, data were collected at a single time point upon admission, which may not fully capture the disease progression. Longitudinal data collection at multiple time points would be valuable for elucidating the relationship between glycosylation dynamics and disease evolution.

Conclusions

Our findings demonstrate the development of a clinical diagnostic model that integrates patient age, time interval from onset to admission, lymphocyte and platelet counts, and the IGPSSIA score, yielding strong discrimination of disease severity, and it could be a promising clinical tool. However, because our glycosylation profiling remains exploratory, further refinement and validation are essential. Specifically, it is suggested to (1) characterize the precise structural alterations of immunoglobulin glycans and (2) rigorously validate the model’s performance in independent patient cohorts. These steps will ensure the model’s reliability in clinical practice.

Abbreviations

NAIs

Neuraminidase inhibitors

ADI

Antibody-dependent inflammation

IGPSSIA

Immunoglobulin glycosylation profile severe status identification algorithm

ROC

Receiver operating characteristic

AUC

Area under receiver operating characteristic curve

BMI

Body mass index

Man

Mannose

Glc

Glucose

GlcNAc

N-Acetylglucosamine

Gal

Galactose

GalNAc

N-acetylgalactosamine

Authors’ contributions

FT, MZ and S-BG had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. FT, W-XL and XL oversaw clinical data and plasma sample collection. D-GL, G-XL, HW and MZ contributed to data analysis. FT, H-HL, MZ and S-BG contributed to manuscript preparation and review. S-BG conceived and designed the research, guided the experiment and data analysis, revised the manuscript critically for important intellectual content and were responsible for the overall content as guarantor. All authors gave final approval of the version to be published and agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Funding

This work was supported by the Beijing Municipal Administration of Hospitals Incubating Program (PX2023012), the Capital Clinical Characteristic Applied Research Project (Z171100001017057), and the Open project of State Key Laboratory of Proteomics (SKLP-O201904).

Data availability

The datasets used and analyzed during the current study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

The Institutional Review Board and Medical Ethics Committee of Beijing Chao-Yang Hospital approved the study (Ethical approval number: 2019-ke-301), and Beijing Ditan Hospital and Miyun Teaching Hospital, both affiliated to Capital Medical University, mutually recognized this approval. Informed consent from all participants was acquired.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.