A total of 31 AR patients who are allergic to autumnal pollen allergens were recruited from the outpatient allergy clinic of Peking Union Medical College Hospital. AR was diagnosed according to the diagnostic criteria proposed by the World Health Organization (WHO) workshop (Bousquet et al., 2008) and Chinese experts from Allergic Rhinitis Committee (Zhang, 2015). Inclusion criteria were as follows: (i) patients diagnosed with AR, (ii) adult men and women (≥ 18 years old), (iii) patients with a history of classic rhinoconjunctivitis symptoms during the autumn pollen season for at least 2 years and without symptoms in other seasons, (iv) positive intradermal skin test (wheal ≥ 10 mm) to grass and tree pollens spread in autumn (Mugwork, Japanese Hop, Firebush, Cocklebur, Daisy, Dandelion, Plantain, Goosefoot, and Common Goldenrop) (Allergen Manufacturing Laboratory of Peking Union Medical College Hospital, Beijing, China), and (v) corresponding serum allergen-specific IgE (sIgE) of ≥ 0.7 KUA/L (ImmunoCAP, ThermoFisher Scientific Inc. Waltham, MA, USA). Exclusion criteria were as follows: (i) patients with a history of asthma, immunodeficiency, chronic infection, or tumor; (ii) pregnancy or lactation, (iii) patients with a history of allergen immunotherapy, (iv) regular consumption of probiotic-supplemented foods, and (v) administration of antibiotics, immunosuppressive medications, monoclonal antibodies, or a large dose of corticosteroids in a month preceding this study. This study was approved by the Medical Ethics Committee of the Peking Union Medical College Hospital (JS-1649). Before enrollment, all participants were provided with written informed consent.

The probiotics used in this study were a multi-strain mixture that composes of 3 × 10⁹ colony-forming units (CFU) L. rhamnosus HN001, 2 × 10⁹ CFU L. acidophilus NCFM, 3 × 10⁹ CFU B. lactis Bi-07, 1 × 10⁹ CFU L. paracasei LPC-37, and 1 × 10⁹ CFU L. reuteri LE16. Moreover, the products are synthesized by a pharmaceutical company that is specialized in probiotic-fortified food products. Participants were given either the probiotic powder or the placebo. The active sample consisted of the probiotic mixture (10¹⁰ CFU/g) and galactooligosaccharides, while the placebo contained only galactooligosaccharides. Products were provided in 2 g small packs and were identical in flavor and appearance.

This was a randomized, placebo-controlled, double-blind study conducted from May 1, 2017 to October 15, 2017. In May and June, AR patients meeting the inclusion criteria were enrolled in this study. The sample size was chosen according to the criteria of phase 1 clinical trials. Participants were allocated into two groups (namely, the probiotic group and the placebo group) using block randomization. Participants were administered probiotics or a placebo of 2 g pack each day starting from July 1, 2017.

They were given a uniform medication regimen to relieve symptoms during the pollen season. The routine medication involves oral antihistamines, nasal corticosteroids, and antihistamine eye drops. Participants were asked to record the medication score, symptom score, and adverse effects in the electronic daily diary. Blood samples were collected at two time points: before the probiotic or placebo intervention in May or June (T1) and during the peak of the pollen season when patients suffer from the most severe AR symptoms (T2). A portion of the blood samples was randomly selected and processed for later protein profiling analyses. The whole landscape of the study is presented in Figure 1.

The visual analog scale (VAS score) provides details about the baseline severity of AR. The VAS was assessed by an allergist, and the score was based on rhinitis symptoms in the past pollen seasons. The score ranged from 0 (no discomfort) to 10 (maximal discomfort).

The symptom assessment scale (rhinoconjunctivitis symptom score) evaluates the frequency and severity of nasal and eye symptoms. The 4-point severity scale is presented in Table 1. The daily symptom score was the sum of all patients' symptom scores per day, ranging from 0 to 15.

Rescue medication score evaluates the alleviating effect generated by taking medicine. Considering the different effects of anti-inflammatory drugs, they are assigned different scores. For example, oral antihistamine, 6 points per tablet; nasal corticosteroid, 1 point per spray; antihistamine eye drop, and 0.5 points per drop. The daily medication score was the sum of all subjects' medication scores per day.

The pollen counts were evaluated through gravity sedimentation as previously reported (Candeias et al., 2021). The pollen granule-capture device was installed at Peking Union Medical Hospital, which was in the center of Beijing, China. Pollen counts were documented from August 1, 2017 to September 30, 2017 and were presented as the number of grains per 1,000 square millimeters.

PBMCs were isolated using Ficoll density gradient centrifugation, as previously described (Boyum, 1964). After washing twice with phosphate buffer saline (PBS), cells were lysed with 8M Urea in PBS, 1% Protease Inhibitor Cocktail tablet (Roche), and 1 mM phenylmethanesulfonyl fluoride (PMSF). Protein concentrations were measured by BCA assay (Thermo Fisher Scientific). Each sample was diluted to 1 mg/ml for subsequent tandem mass tag (TMT) labeling.

In total, 200 μg of protein from each sample was used to conduct labeling. Protein extract was reduced with 5 mM dithiothreitol (DTT) at room temperature for 1 h and alkylated with 12.5 mM iodoacetamide (IAM) for 1 h in the dark. Samples were diluted with PBS, followed by trypsin digestion at a 100:1 protein/enzyme ratio overnight. The digested samples were desalted by Sep-Pak C18 columns (Waters, MA). Peptides from different samples were labeled with TMT reagents (Thermo Fisher Scientific), according to the manufacturer's instructions. Labeling reactions were quenched by using 5% hydroxylamine. All the TMT-labeled peptides from different samples were mixed and desalted by a Sep-Pak column.

Offline separation was conducted by a UPLC3000 system (Thermo Fisher Scientific) with an XbridgeTM BEH300 C18 column (5 μm, 150 Å, 250 mm × 4.6 mm i.d., Waters, MA). In brief, mobile phase buffer A is 2% acetonitrile and phase B is 98% acetonitrile. Both mobile phase buffers were adjusted by ammonium hydroxide to pH 10. The solvent gradient was set as follows: 8–18% phase B, 30 min; 18–32% phase B, 32 min; 32–95% phase B, 2 min; and 95% B, 6 min. Peptides were monitored at 214 nm and were collected. The collected fractions were dried by speed vac, reconstituted into 12 fractions, and re-dissolved in 0.1% (v/v) formic acid.

A 120-min gradient elution separated peptides from each fraction at a flow rate of 0.250 μl/min with EASY-nLCII™integrated nano-High Performance Liquid Chromatography System (Proxeon, Denmark). A Thermo Orbitrap Fusion Lumos mass spectrometer was used. The analytical column was a fused C-18 silica capillary column (75 μm ID, 150 mm length; Lexington, MA). Mobile phase A consisted of 0.1% formic acid, and mobile phase B consisted of 100% acetonitrile and 0.1% formic acid. The Orbitrap Fusion Lumos mass spectrometer was operated in the data-dependent acquisition mode. The scan range was from 350 to 1,550 m/z with 120,000 resolution with automatic gain control (AGC) target value of 2e5. A data-dependent acquisition method was performed to collect MS/MS spectra at 30,000 resolution and a maximum injection time (IT) of 60 ms for the top 20 ions observed in each mass spectrum. The exclusion duration was 8 s, and the collisional energy was 35%.

The MS/MS spectra were searched against the UniProt Human database using the SEQUEST searching engine in Proteome Discoverer software version 2.1. Parameters for database searching were set as full tryptic specificity, and one missed cleavage was allowed. Carbamidomethylation (C) and TMTsixplex (K- and N-terminal) were set as fixed modifications. The oxidation (M) was set as variable modification. Precursor ion mass tolerances were set at 10 ppm for all MS acquired, and the fragment ion mass tolerance was set at 20 mmu for MS2 spectra acquired. The peptide spectral matches (PSMs) were validated at a 1% false discovery rate (FDR) under the percolator of the Proteome Discoverer software. Quantification was performed only for proteins with two or more unique peptide matches. The original proteomic data were taken pre-processing, such as removing outliers, filling missing values, and monitoring batch effect through R software.

The results were expressed as “mean ± standard deviation (SD)” if the parameters complied with a normal distribution or as “median [P₂₅, P₇₅]” if they presented as an abnormal distribution. Statistical analyses include paired-sample t-test and independent-sample t-test. All the above statistical analyses were computed using SPSS statistical software and R software.

For bioinformatics analyses, the R package “clusterProfiler” was used to perform the Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG), while the R package “ggplot2” was applied to visualize the enriched pathways. PPI was constructed using the STRING online database (https://cn.string-db.org), and the network model was virtualized in Cytoscape (V3.9.1). CentiScaPe plugin was used to estimate the degree, betweenness, and closeness of nodes in the unweighted interaction network. The R package “glmnet” was used to conduct LASSO regression and variable selection was chosen using linear module strategy.

A total of 31 patients suffering from autumn seasonal AR were recruited, who were then randomly allocated into the probiotic group (n = 16) or placebo group (n = 15). As a participant missed multiple doses of probiotics, it was deemed as dropping out of the trial, and related recordings were not enrolled in later statistical analyses. There were 30 subjects, including 20 men and 10 women, who seriously followed the requirement and completed the clinical trial. Their age ranged from 23 to 62 years. Moreover, there was no statistically significant difference between the two groups in any aspects of the demographic and clinical features (Table 2). Notably, EOS% presented a statistical significance between the two time points (T1 and T2) in the placebo (P = 0.001) group while not presented in the probiotic group (P > 0.05). Although EOS% altered dramatically in comparison to the two groups, there was no statistical significance indicated by the independent-sample t-test. A detailed description is presented in Supplementary material 1.

In China, autumnal pollen spores and grains generally spread from August to October (Yao and Zhang, 2009; Li et al., 2015). Therefore, we monitored the pollen grain concentrations to define “pollen season” in 2017. According to the EAN definition of pollen season (Bastl et al., 2015, 2018), the autumn allergy season lasted for 43 days in Beijing, China, starting on 13 August 2017. The median is 116 (36, 228) grains per 1,000 square millimeters. Moreover, the peak of pollen concentration was from the last 7 days of August to the first 14 days of September (Figure 2A). Correspondingly, the clinical manifestations were more prominent at the end period of August and the onset period of September as indicated by the daily recorded symptom assessment scales (Figure 2A). The average rhinoconjunctivitis symptom score of the probiotic group (3.11 ± 1.03) was significantly lower than that of the placebo group (3.54 ± 0.85, P = 0.037) during the pollen season (Figure 2B). The correlation analysis between pollen counts and clinical manifestations was then performed, and the correlation coefficient showed that these two factors were highly related (0.742 in the placebo group and 0.864 in the probiotic group).

Considering that the alleviated symptoms in the probiotic group may be caused by pharmaceutical intervention or baseline severity of the AR, we, then, analyzed medication usage and the initial AR severity by calculating medicine score and VAS score. The medicine scores and VAS scores were not statistically significantly different between the two groups (P > 0.05; Figures 2C, D). The above results suggested that the ameliorating effect from probiotics was not caused by pharmaceutical intervention or baseline symptom severity. A more detailed description is presented in Supplementary material 2.

During the probiotic mixture or placebo administration period, no adverse event was reported. Just one participants in the probiotic group (1/15, 6.7%) experienced diarrhea after eating probiotic. However, the symptom was mild and relieved quickly without interrupting the continuous use of probiotics. Moreover, one subject in the placebo group also experienced diarrhea (1/15, 6.7%). No other adverse drug reactions (e.g., nausea, vomiting, fever, abdominal pain, and hematochezia) were observed, and none of the patients dropped out because of side effects.

A total of six participants were randomly sampled from the placebo group, and the whole blood was collected to isolate PBMCs for further proteomics. DEPs were, then, screened following the criteria: (i) ≥ 95% confidence level, (ii) the number of matching peptides ≥ 2, (iii) fold change > 1.2 or < 0.83, and (iv) false discovery rate (FDR) < 0.05. A total of 12,448 proteins were identified with 967 proteins statistically expressed in all samples (Supplementary material 3). There were 60 DEPs, 35 downregulated and 25 upregulated, screened in patients with placebo administration when comparing T1 and T2 (Supplementary material 4), reflecting the pathogenic condition of AR. In total, 60 DEPs, then, underwent GO and KEGG enrichment analyses. GO analysis informed that these DEGs mainly are enriched in the coagulation process, transcription and translation, cell phagocytosis, and nutrient metabolic pathways (Figure 3A). The KEGG enrichment result showed that the regulation of the actin cytoskeleton, ribosome activity, platelet activation, inflammatory signal pathways, lipid metabolism, and amino acid metabolism was significantly enriched (Figure 3B). PPI co-expression network highlighted functionally interacted proteins. The network, including four separate clusters, contains 60 nodes and 82 edges. To obtain key proteins that play dominant roles, the network was then analyzed by the CentiScaPe plugin to calculate the degree, betweenness, and closeness between nodes. Key proteins were, then, filtered in various clusters and marked red (Figure 3C). To identify AR signature proteins, LASSO regression was applied to extract the main elements correlated with AR symptoms. In this study, the response type was set as Gaussian, and the alpha was set to 1. Figure 3D describes the coefficient profile of six indicators, which are generated using five-fold cross-validation. Figure 3E presents 10-fold cross-validation for tuning parameter selection in the model via minimum criteria. As a result, glutathione S-transferase omega-1 (GSTO1) (coeff. = −0.0998), sarcoplasmic/endoplasmic reticulum calcium ATPase 2 (ATP2A2) (coeff. = −0.107), minichromosome maintenance complex component 7 (MCM7) (coeff. = 0.0409), vitamin K-dependent protein S (PROS1) (coeff. = −0.0248), and E3 ubiquitin-protein ligase TRIM58 (TRIM58) (coeff. = −0.0522) were screened (Figures 3D, E).

To explore the potential targets of probiotics, six samples were randomly assigned from the probiotic group for further proteomic analyses. In a comparison of the time points of T1 and T2, 17 DEPs, composed of 6 downregulated and 11 upregulated proteins, were obtained in the probiotic group. The schematic landscapes of DEPs were notable differences in the two groups as indicated by principal component analysis and complex heap map (Figures 4A, B). There were only three common DEPs (MYH9, HMGN2, and Histone H1.2) in both groups (Figure 4C). Different from MYH9, whose expression levels were identical in two groups, HMGN2 and Histone H1.2 exist a significant reverse trend in groups with different intervention methods (Supplementary material 4).