Introduction

Recurrent respiratory tract infections (rRTIs) affect 10–15% of young children [1, 2] and increase the risk of lung damage and COPD later in life [3, 4]. Current clinical assessment of children with rRTIs focuses on identifying underlying conditions and risk factors, such as allergies, asthma, chronic lung diseases and immunological defects [1]. To identify immune deficiencies, clinicians measure serum antibody levels. However, systemic immune deficiencies are diagnosed in only a minority of cases [4, 5]. This diagnostic approach falls short in evaluating the mucosal immune system, even though the respiratory mucosa is the main entry point for pathogens causing RTIs and the primary site of immune system interaction. To the best of our knowledge, there are no available diagnostic tests that specifically assess the mucosal immune system.

The predominant mucosal antibody is secretory IgA (sIgA), which exhibits superior antigen binding, agglutination and neutralisation capabilities compared to serum IgA [6]. Mucosal antibodies may have broader binding capacity than serum antibodies, a phenomenon known as polyreactivity. While polyreactivity has been documented for intestinal sIgA [7–9], emerging evidence suggests the presence of polyreactive antibodies in saliva as well [10–12].

In children, specific respiratory microbiota profiles can precede or accompany RTIs [13, 14]. This indicates that microbial alterations may contribute to the onset and severity of RTIs as well as the spread of pathogens to the lower airways [15, 16]. Studies involving germ-free mice have demonstrated that the production of faecal sIgA is initiated and modulated by the resident microbiota in the intestinal tract [17]. Conversely, mucosal antibodies play a role in regulating the composition of the microbiota by binding to and eliminating bacteria from the mucosal surface and maintaining microbial homeostasis [7, 18]. While the interaction between the microbiome and the immune system has been extensively studied in the context of the gut, similar investigations within the respiratory mucosa are relatively scarce.

The main objective of this study was to examine salivary antibodies and the respiratory microbiota as potential indicators of both severity and burden of RTIs in young children with rRTIs. Additionally, our aim was to investigate the association of these mucosal immune factors with RTI severity and burden in comparison to the traditional measurement of serum antibody levels.

MethodsStudy population

As part of the DIMER study [19], we included two cohorts. The exploration cohort comprised children with rRTIs, their family members and healthcare workers without rRTIs or underlying diseases. This cohort was used to explore serum and salivary antibody characteristics. We included all subjects from the DIMER cohort who had sufficient saliva volume to perform all laboratory assays, with keyhole limpet haemocyanin (KLH) measurement performed on the subset of subjects who still had available saliva. The paediatric clinical cohort consisted of children <10 years of age with rRTIs who were referred to a participating hospital for immunological screening. These children were included between September and December from 2016 to 2019. rRTIs were defined according to the guideline of the Dutch Section of Paediatric Infectious Diseases and Immunology [20], based on definitions by Gruber et al. [21], as: ≥11 upper RTIs per year for children aged up to 2 years of age, ≥8 upper RTIs per year for children aged between 2 and 5 years, ≥6 upper RTIs per year for children aged between 5 and 8 years, or ≥2 pneumonia episodes diagnosed in 1 year or ≥3 pneumonia episodes diagnosed during the child's lifetime. Recurrent lower RTIs were defined as ≥2 physician-diagnosed pneumonia episodes in a lifetime. Antibody deficiencies were defined as 2sd below age-appropriate reference values (supplementary methods).

The PID study, of which the DIMER study is a substudy, received ethical approval from the Medical Ethical Committee of the Erasmus MC, Rotterdam, The Netherlands (METC:NL40331.078). All (legal guardians of) subjects signed informed consent and the study was conducted in accordance with the Declaration of Helsinki.

Sample collection

In the exploration cohort, a saliva and a serum sample were collected from every subject. Healthcare workers provided serum and saliva samples before and 3–4 weeks after influenza virus vaccination in the winter of 2018 and 2019 to investigate the effect of intramuscular vaccination on antibody levels and dynamics. In the clinical paediatric cohort, saliva was collected at baseline and three nasopharyngeal swabs were collected in the first month after inclusion. Afterwards, daily RTI symptoms were registered for four consecutive winter months using a mobile phone application (figure 1 and supplementary methods). Nasopharyngeal swabs were used for microbiota 16S rRNA sequencing and detection of 12 respiratory viruses with quantitative PCR (qPCR) and Haemophilus subtyping qPCR (supplementary table S2 and supplementary methods). For children with rRTIs, serum immunoglobulin levels were measured at immunological screening and additional serum samples were only collected when there was a clinical indication.

FIGURE 1

Methods of DIMER studies separated for the exploration cohort and the clinical paediatric cohort of children with recurrent respiratory tract infections (RTIs). SARS-CoV-2: severe acute respiratory syndrome coronavirus 2; RSV: respiratory syncytial virus.

Measurement of serum and salivary antibodies

ELISA was performed on serum and saliva. Plates were coated overnight (supplementary table S3). After washing and blocking, samples, standard preparations and controls were added in a duplicate manner. Batches were corrected based on differences of the positive controls. Serum antibody concentrations were measured with a turbidimetric assay (supplementary methods). Antibody avidity was assessed using ELISA with sodium thiocyanate treatment (supplementary methods).

16S rRNA sequencing of nasopharyngeal samples

Bacterial DNA was isolated and quantified as previously described [22]. DNA isolated from nasopharyngeal swabs with a bacterial density of 0.2 pg·µL−1 above the negative controls was analysed. Amplicon libraries of the V4 region of the 16S rRNA gene were generated with primers 515F–806R. Paired-end reads were trimmed, merged, denoised, chimera filtered and binned into amplicon sequence variants (ASVs) with the DADA2 implementation (supplementary methods). All microbiota analyses were corrected for age and based on total sum scaled (relative) abundances, unless otherwise stated.

We removed non-bacterial ASVs and contaminants, only including >10 000 reads samples and ASVs with >0.1% RA in ≥2 samples. ASVs with 99% species-level nucleotide sequence similarity were clustered with the IdClusters() function and annotated using the Basic Local Alignment Search Tool (BLAST) (supplementary table S4). ASVs present in ≥10% of children with ≥1% RA were further investigated in relation to RTI severity.

Serum and saliva proteomics

In a subset of children with rRTIs, we used the Olink Inflammation panel (https://olink.com) to provide qualitative measurement of 92 analytes in serum (n=20) and saliva (n=71). We only included serum samples collected a maximum of 3 months away from the saliva sample collected. Saliva proteomics was performed on saliva stored in glycerol and analyses were corrected for weight of the sample.

Statistics

Data were analysed using R(-Studio) (www.rstudio.com). Spearman correlations were determined between antibody levels against different respiratory viruses and KLH, corrected for multiple comparisons (Benjamini–Hochberg method). Baseline characteristics and univariate analyses were conducted with the Chi-squared/Fisher's exact test and the t-test/Mann–Whitney U-test. Latent class analysis can identify patterns over time and thereby create clusters of subjects in longitudinal data. We used this analysis to analyse daily RTI symptoms measured over four winter months and bacterial RA in time (supplementary methods).

All microbiota and proteomics analyses that were performed ≥3 times were corrected for multiple comparisons (Benjamini–Hochberg method). We measured α diversity with the Shannon index and β diversity with the Bray–Curtis dissimilarity matrix. Microbial stability per subject was based on the mean Bray–Curtis dissimilarity between time-points 1 and 2 and 2 and 3. Spearman correlations were used for specific ASVs and antibody levels. For multivariable analyses, groups were compared using logistic or multinomial regression analyses with log-transformed antibody levels. For microbiota RAs and qPCR data, we used centred log ratio from the compositions package. We replaced all zeros in the microbiota dataset with a constant value smaller than the detection limit (RA of 0.00001) prior to transformation [23]. For analyses of longitudinal data, we performed univariate and multivariable negative binomial mixed models with daily RTI symptom count as the outcome variable (supplementary methods). Statistical significance was defined as p<0.05.

ResultsHigh proportion of polyreactive antibodies present in saliva compared to serum

We included 10 children with rRTIs, 18 parents, eight siblings and 25 healthcare workers in an exploration cohort (figure 1, supplementary figure S1 and study protocol) sampled between 2016 and 2019, predating the onset of the coronavirus disease 2019 (COVID-19) pandemic. In both saliva and serum samples, we quantified levels of IgA, IgG and IgM targeting influenza A, influenza B, respiratory syncytial virus (RSV) and severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). We investigated the correlation of antibody levels against one virus with those against other viruses. Salivary antibodies showed strong correlations across all viruses, whereas this was less pronounced for serum antibodies (figure 2a). The correlations were strongest for salivary IgA (ρ>0.90), followed by IgG (ρ>0.72) and IgM (ρ>0.51). While antibody levels were generally lower in children than in adults, when analysed separately, we observed consistently strong correlations of virus-reactive salivary IgA levels in both adults and children (supplementary figure S2a).

FIGURE 2

Exploration of polyreactive salivary IgA. a) Polyreactive potential of saliva and serum antibodies across different respiratory viruses from 10 children with recurrent respiratory tract infections, 18 parents, eight siblings and 25 healthcare workers samples prior to influenza virus vaccination. Dots represent the Spearman correlation coefficient corrected for multiple comparisons. Only correlations with a p-value <0.05 are depicted. Influenza strains used: H1N1 A/California/04/2009 and H3N2 A/Singapore/INFIMH-16-0019/2016. b) Saliva IgA levels from 12 healthcare workers before and after vaccination against the three strains in the influenza virus vaccine and three unfamiliar viruses. Influenza strains from left to right: H1N1 A/California/04/2009, H3N2 A/Singapore/INFIMH-16-0019/2016, B/Maryland/15/2016, H3N8 A/duck/Ukraine/1/1963 and H5N1 A/Vietnam/1204/04. p-values from paired Wilcoxon signed-rank test. c) Antibody avidity is measured as percentage of antibody bound to antigen after treatment with sodium thiocyanate (NaSCN) compared to untreated wells. In these samples collected from 12 healthcare workers prior to the coronavirus disease 2019 outbreak and 3–4 weeks after influenza virus vaccination, specific antibody levels were defined as antibodies targeting vaccine-type influenza A H1N1 (H1N1 A/California/04/2009) and polyreactive antibodies as targeting severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). p-value from Wilcoxon signed-rank test. RSV: respiratory syncytial virus; XLA: X-linked agammaglobulinaemia.

To differentiate between cross-reactive antibodies (that bind to different but structurally related antigens) and polyreactive antibodies (that bind to multiple unrelated antigens), we determined the reactivity of salivary IgA and IgG targeting KLH, an unrelated antigen derived from a seasnail residing off the Californian coast, in children and adults with sufficient saliva samples available (n=40: five children with rRTIs, four siblings, 11 parents and 20 healthcare workers prior to influenza virus vaccination). Binding of salivary IgA and IgG to KLH correlated strongly with binding to the four respiratory viruses (ρ>0.70) (supplementary figure S2b and c), suggesting that polyreactive antibodies, with the capacity to bind to unrelated and previously unencountered antigens, are indeed present in saliva. However, the concentrations of salivary IgA targeting KLH were lower compared to concentrations targeting respiratory viruses. Specifically, the median (interquartile range (IQR)) proportion of salivary IgA binding to KLH, expressed as a percentage of the concentration found for the respiratory viruses, ranged from 47% (23–97%) for influenza A, 51% (27–94%) for influenza B, 65% (29–131%) for RSV to 81% (36–146%) for SARS-CoV-2.

Since all samples were collected before the onset of the COVID-19 pandemic in 2019, no specific antibody response against SARS-CoV-2 was expected. Therefore, we classified the detected antibody levels targeting SARS-CoV-2 as polyreactive antibodies.

The proportion of polyreactive IgA (percentage of total IgA) was significantly higher in saliva than in serum, with median (IQR) values of 28% (18–76%) and 2% (0–4%), respectively (p<0.001). Polyreactive antibody levels were higher in adults compared to children (median (IQR) 308.5 (146.2–566.3) versus 46.0 (16.3–137.9), respectively), while the proportion of polyreactive IgA (percentage of total IgA) was lower in adults compared to children (median (IQR) 10.5% (5.5–14.1%) versus 33.0% (11.9–90.9%), respectively; p<0.001). In addition, in siblings without rRTIs, polyreactive IgA levels positively correlated with age (ρ=0.37, p=0.02), whereas this correlation was not observed in children with rRTIs (ρ=0.10, p=0.33). This indicates that the dynamics of polyreactive antibodies are influenced by age, but this effect may be disrupted in children with a high burden of RTIs due to altered production and/or consumption of polyreactive antibodies.

Polyreactive antibody response can be distinguished from antigen-specific response

Next, we investigated to what extent the induction of polyreactive antibodies differed from antigen-specific antibodies. In saliva collected before and after intramuscular influenza vaccination of 12 healthcare workers in the exploration cohort, we measured IgA targeting the three vaccine subtypes (specific salivary IgA response), a non-human virus (duck influenza A) and two human viruses to which the healthcare workers had not been exposed (1963 influenza A and SARS-CoV-2). While reactivity against all viruses was observed (supplementary figure S2d), salivary IgA levels were higher against the vaccine strains than against the unfamiliar strains (p<0.001) (figure 2b). Systemic vaccination induced salivary IgA levels targeting all viruses, although the induction of vaccine strain-specific IgA was highest (supplementary figure S2e). To determine whether polyreactivity was directly caused by IgA, we included a saliva sample from a patient with X-linked agammaglobulinaemia. No polyreactive IgA was detected in this sample, confirming that the polyreactivity was not due to a saliva matrix effect (figure 2b). Finally, we compared the avidity (binding strength) of polyreactive and specific antibodies in serum and saliva. Serum antibodies displayed higher avidity compared to salivary antibodies (median (IQR) 58% (51–73%) versus 15% (8–21%), respectively; p<0.001) (figure 2c). In saliva, polyreactive IgA targeting SARS-CoV-2 showed lower median (IQR) avidity (10% (6–17%)) than specific IgA (targeting vaccine-type influenza A H1N1 28% (20–38%); p<0.001).

Serum antibody levels are not associated with increased RTI severity, RTI symptoms or viral burden

Following our findings in the exploration cohort, subsequent experiments were conducted in a clinical patient cohort comprising 100 children with rRTIs, 37 of their siblings and 45 parents (figure 1, supplementary figure S1 and study protocol). Children with rRTIs had a median (IQR) age of 3.9 (2.6–5.7) years, 46% were assigned female sex at birth, 24% had recurrent lower RTIs and 50% had a serum antibody deficiency (supplementary table S1). We observed no significant associations between serum IgA or IgG antibody deficiencies or levels and RTI severity as defined as the prevalence of recurrent lower RTIs, while serum IgM levels were higher in children with recurrent lower RTIs (table 1 and supplementary figure S3a). To assess RTI symptom burden, caregivers of 85 out of the 100 children with recurrent RTIs recorded daily symptoms through a dedicated mobile phone application over four consecutive winter months. Adherence to RTI symptom registration was high with a median (IQR) of 122 (119–124) winter days filled in by parents. A latent class analysis that can identify different trajectory groups over time was used to categorise children into severity levels (mild (n=10), moderate (n=60) or severe (n=15)) based on the number of symptoms that were present on a daily basis over the 4-month period (supplementary figure S3b). Serum antibody levels did not differ between RTI symptom severity groups (supplementary figure S3c) and neither serum antibody deficiencies nor serum antibody levels were associated with viral burden in winter (supplementary figure S3d).

TABLE 1

Recurrent lower respiratory tract infections (RTIs) and viruses detected in children with recurrent RTIs (rRTIs)

Significant correlations between the mucosal polyreactive antibody response and the microbiota

In the nasopharyngeal microbiota, 4604 ASVs were detected. After removal of 571 possible contaminants, 431 non-bacterial ASVs and the application of a presence-abundance filter (only including ASVs with >0.1% RA in ≥2 samples), 113 ASVs remained in the dataset for downstream analyses.

Corrected for age, we observed a significant association between both polyreactive salivary IgA and IgG levels and α diversity (supplementary figure S4a), while β diversity was not significantly associated with polyreactive antibody levels (supplementary figure S4b).

Among the most abundant ASVs, only H. influenzae/haemolyticus was negatively correlated with polyreactive salivary IgA levels (ρ= −0.24, p=0.03) (figure 3a). This association was not observed for polyreactive IgG or IgM, nor for any of the serum antibodies (supplementary figure S4c).

FIGURE 3

Nasopharyngeal microbiota in relation to polyreactive antibody levels. a) Polyreactive IgA and IgG associated with Haemophilus influenzae/haemolyticus relative abundance (RA). Statistics from Spearman correlations. b) Hierarchical clustering of microbial profiles of children with recurrent respiratory tract infections based on Bray–Curtis dissimilarities. Hierarchial clusters are shown in the colour of the most abundant bacteria for that cluster. High and low groups are based on polyreactive IgA levels relative to the median value for the paediatric cohort. c) H. influenzae abundance in time trajectories was identified using latent class analyses. p-values from multinomial regression analysis corrected for age and recent antibiotic courses. ASV: amplicon sequence variant.

Given the prominence of H. influenzae/haemolyticus, we performed two qPCRs to identify H. influenzae and H. haemolyticus in the nasopharyngeal swabs from the clinical paediatric cohort. H. influenzae was detected in 41 samples and H. haemolyticus was co-detected in 17 of the 85 samples. H. haemolyticus had a significantly higher median (IQR) quantification cycle (Cq) value (i.e. lower abundance) compared to H. influenzae (33.8 (33.2–34.3) and 24.2 (22.6–27.6), respectively; p<0.001). H. influenzae abundance was quantified by qPCR and correlated strongly with absolute abundance of the H. influenzae/haemolyticus ASV detected by 16S rRNA sequencing (ρ=0.95, p<0.001) (supplementary figure S5a), which was not the case for H. haemolyticus (ρ= −0.14, p=0.09).

Through hierarchical clustering, we identified Moraxella catarrhalis (n=40), H. influenzae (n=19), Corynebacterium propinquum/pseudodiphtheriticum (n=13), Streptococcus pneumoniae/pseudopneumoniae (n=4) and Staphylococcus aureus/epidermidis dominant clusters (n=4) (figure 3b). Notably, polyreactive salivary IgA and IgG levels were lower in the H. influenzae dominated cluster compared to the other clusters combined (p=0.006 and p=0.035, respectively) (supplementary figure S5b).

Microbiome stability exhibited considerable variation across the three nasopharyngeal swabs collected within the first winter month (supplementary figure S5c), which was significantly associated with recent antibiotic use (p=0.02). After adjusting for age and recent antibiotic courses, children with consistently low H. influenzae abundance over time (n=50) had higher polyreactive salivary IgA compared to children with moderate H. influenzae abundance (n=17, p=0.08) and high H. influenzae abundance (n=18, p=0.01) (figure 3c and supplementary figure S5d).

Decreased levels of polyreactive IgA and IgG are associated with RTI disease severity

We compared the presence of recurrent lower RTIs with the presence of recurrent upper RTIs in children with rRTIs at the time of first presentation to determine severe versus mild RTI disease severity. In univariate analysis, polyreactive salivary IgA and IgG levels were negatively associated with recurrent lower RTIs (both p<0.001) (figure 4a). In contrast, total salivary IgA levels were higher in children with recurrent lower RTIs (p=0.03) compared to children with only upper RTIs. Polyreactive IgM showed no association with recurrent lower RTIs (supplementary figure S6a). In multivariable logistic regression analysis adjusting for confounders that could influence mucosal measurements and RTI severity, such as age, RTI symptoms at time of sampling and recent antibiotic use, low levels of polyreactive salivary IgA (adjusted OR (aOR) 0.80, 95% CI 0.67–0.94, per log-fold increase; p=0.010) and decreased H. influenzae abundance (aOR 0.79, 95% CI 0.68–0.91, per log-fold increase; p=0.002) (table 2) were associated with recurrent lower RTIs. H. influenzae concentration as measured with qPCR was also associated with recurrent lower RTIs (aOR 0.78, 95% CI 0.60–0.97, per log-fold increase; p=0.041) (supplementary table S5a).

FIGURE 4

Polyreactive salivary antibodies and nasopharyngeal microbiota in relation to respiratory tract infection (RTI) severity and symptom burden a) Recurrent lower RTIs were defined as ≥2 physician-diagnosed pneumonia episodes throughout life. p-values from Mann–Whitney non-parametric U-tests. b) RTI symptoms were measured with a daily diary mobile phone application with trajectories of RTI symptoms identified using latent class analyses. ASV: amplicon sequence variant.

TABLE 2

Multivariable analysis for respiratory tract infection (RTI) severity: logistic regression for children with recurrent lower RTIs compared to those with only upper RTIs

H. influenzae abundance is associated with RTI symptom burden

Salivary polyreactive antibody levels were not associated with RTI symptom burden (supplementary figure S6b). Upon investigation of the most abundant ASVs in relation to number of RTI symptoms over four winter months, H. influenzae was almost absent in the mild group and highly abundant in the moderate (p<0.001) and severe (p<0.001) RTI symptom severity groups (figure 4b). Next, we assessed the effect of polyreactive IgA and H. influenzae abundance (explanatory variables) on RTI symptoms (response variable) in a multivariable negative binomial mixed model with number of RTI symptoms per day over time as a continuous outcome (response variable), thus employing the in-depth data obtained from the daily diary RTI symptom registration over four winter months. We identified H. influenzae abundance as the strongest indicator of RTI symptom severity in young children with rRTIs (0.05, 95% CI 0.02–0.08) (table 3). Per natural log increase in the RA of H. influenzae, the average within-subject change in the log of the expected number of RTI symptoms per day increased by 0.05, with other factors in the model held constant. For H. influenzae concentration as measured with qPCR the average within-subject change increased by 0.04 (95% CI 0.01–0.07) (p=0.007) (supplementary table S5b).

TABLE 3

Multivariable analysis for respiratory tract infection (RTI) symptom burden: negative binomial mixed model for amount of daily RTI symptoms over time

Polyreactive salivary IgA is negatively associated with viral burden

In univariate analysis, polyreactive salivary IgA and IgG levels were negatively associated with detection of multiple viruses compared to no viruses (p=0.004 and p=0.032, respectively) (figure 5a and b). Total salivary IgA levels and polyreactive IgM did not differ in relation to number of viruses detected (supplementary figure S6c). In multivariable analysis, polyreactive salivary IgA levels were indeed negatively associated with the detection of multiple viruses compared to no viruses during the first winter month (aOR 0.76, 95% CI 0.61–0.96, per log-fold increase; p=0.020) (table 4). H. influenzae was found to be more abundant in children with a virus detected compared to those with no viral detection in univariate analysis (p=0.049) (figure 5c), but this difference disappeared in multivariable analysis (table 4). H. influenzae concentration as measured with qPCR was associated with one virus detected compared to no viral detection (aOR 1.18, 95% CI 1.02–1.36, per log-fold increase; p=0.023) (supplementary table S5c).

FIGURE 5

Polyreactive salivary antibodies and nasopharyngeal microbiota in relation to viral burden. a) Number of viruses detected in three consecutive nasopharyngeal swabs collected at the start of winter. p-values from Kruskal–Wallis with post-hoc Dunn's tests. b) Number of viruses detected in children with high and low polyreactive salivary antibodies. High and low groups are based on polyreactive levels relative to the median value. p-values from Fisher's exact test. c) Microbiota composition in relation to number of viruses detected in three consecutive nasopharyngeal swabs collected at the start of winter. RSV: respiratory syncytial virus; ASV: amplicon sequence variant.

TABLE 4

Multivariable analysis for viral burden: multinomial regression comparing children without a virus detected with children with 1 and ≥2 viruses detected

RA of H. influenzae in the upper respiratory tract is positively associated with salivary inflammatory markers and negatively with serum inflammatory markers

Finally, to gain insight into the relationship between mucosal measurements and immune activation, we applied proteomic analysis (Olink Inflammation panel) to measure the expression of 92 inflammatory proteins in serum (n=20) and saliva (n=71) from children with rRTIs. Serum proteomics revealed that polyreactive salivary IgA was positively associated with CDCP1, CCL4, CCL25 and 4-1BB (or CD137L), while displaying a negative correlation with CXCL5 (p=0.01–0.02, adjusted for multiple testing) (supplementary figure S8a) and that polyreactive salivary IgG was associated with CCL28 (p=0.02, adjusted for multiple testing) (supplementary figure S8b).

The RA of H. influenzae was negatively associated with multiple proteins in serum, including several T-cell markers (CD8A, CD5, CD6, CD244, programmed death-ligand 1), natural killer (NK) cell markers (CD244) and inflammatory markers (tumour necrosis factor (TNF), TRAIL, interleukin (IL)-15RA, IL-18) (supplementary figure S8c). In contrast, in saliva, H. influenzae RA was positively associated with the T-cell marker CD8A and IL-33 (p=0.047 and p<0.01, respectively, adjusted for multiple testing) (supplementary figure S9). In addition, carriage of H. influenzae in time was associated with higher levels of the inflammatory marker interferon (IFN)-γ. Notably, serum 4-1BB was associated with both increased polyreactive salivary IgA (β=109.1, p=0.02) and decreased H. influenzae RA (β= −0.70, p<0.01), and there was a trend towards decreased 4-1BB levels with high H. influenzae abundance over time (compared to low abundance; p=0.06) (supplementary figure S10).

Discussion

Current diagnostic analysis for children with rRTIs primarily relies on serum antibody measurements to detect deficiencies. Our study in 100 young children with rRTIs emphasises our previous findings that neither serum antibody deficiencies nor serum antibody levels are associated with RTI disease burden [19]. Instead, our findings indicate a clear distinction between salivary and serum antibodies, with salivary polyreactive antibodies emerging as clinically relevant indicators, potentially influencing both RTI severity and viral burden. Moreover, this study underscores the role of H. influenzae in the respiratory microbiota, demonstrating its association with reduced levels of polyreactive IgA and increased RTI severity and burden. These results emphasise the importance of analysing the mucosal immune system to enhance our understanding of rRTIs in young children.

Previous studies have described the presence of mucosal polyreactive antibodies [7–9, 11], but their protective potential has been overlooked because they were considered a technical hurdle to overcome in order to measure specific antibody responses [11, 24]. We identified salivary antibodies capable of binding to multiple unrelated antigens, including novel targets such as SARS-CoV-2 antigens and KLH. It is important to differentiate this polyreactivity from cross-reactivity as a result of previously encountered similar antigens. While cross-reactivity based on previously encountered strains of viruses cannot be entirely ruled out for the respiratory viruses we measured, this is unlikely for KLH. The correlation coefficient between KLH and the respiratory viruses was lower than those between the viruses themselves. In addition, the level of salivary IgA binding to KLH was lower than levels for the respiratory viruses, with a proportion of KLH to viruses of 50–81%. This suggests that the presence of specific antibodies and/or cross-reactive antibodies targeting the respiratory virus antigens in our assays cannot be ruled out. However, given that a significant majority of salivary IgA bound to KLH and SARS-CoV-2 in our pre-pandemic samples, it is likely that we are predominantly measuring polyreactive antibodies with this target.

Among children with rRTIs, lower levels of salivary polyreactive IgA were clinically significant in relation to RTI severity and viral burden. In contrast, higher levels of total salivary IgA were associated with RTI severity. This indicates that higher polyreactive antibody levels might confer protective effects against viral and lower RTIs, which is not the case for total salivary IgA. The increased levels of total salivary IgA could result from an increase in specific antibody levels due to recurrent lower RTIs, which may expand the total IgA pool in the respiratory tract. The capacity of polyreactive antibodies to bind to unrelated and previously unencountered antigens suggests that they serve as a first line of innate defence against RTIs. Pathophysiologically, increased reactivity and lower avidity, as observed for polyreactive salivary IgA, could result from T-cell-independent IgA production [8, 25, 26].

While (polyreactive) antibody levels were measured in saliva, the microbiota was determined in nasopharyngeal swabs. Previous studies have shown that the microbiota composition differs between the oral and nasopharyngeal cavities [27, 28], as a result of different physiological factors such as pH and humidity [29]. A large prospective birth cohort study in the Netherlands that compared the salivary and nasopharyngeal microbiota in young children found that nasopharyngeal microbiota composition was most strongly associated with RTI disease severity [27]. Therefore, we focused on the nasopharyngeal microbiota in our cohort. Nasopharyngeal swabs were stored in RNAprotect reagent in order to effectively detect the 16S rRNA gene for microbiota sequencing. However, this medium can denature proteins [30], making antibody detection using ELISA unreliable. Since both saliva and nasopharyngeal swabs reflect the upper respiratory tract, we used saliva samples for immunoglobulin measurements. This approach has the additional advantage that saliva is more easily obtained from young children, making it a viable, child-friendly alternative to serum samples for clinical use.

The presence of H. influenzae in the respiratory tract was strongly associated with RTI symptom burden during the winter season. Early-life colonisation with Haemophilus has been linked to the development of (lower) RTIs [13, 15, 16, 31, 32] and increased risk of hospitalisation [33]. Extensive antibiotic use in early life, as in our cohort, is associated with a high abundance of Haemophilus [15], potentially shifting the microbiota towards an RTI-prone profile and creating a vicious cycle of increased susceptibility to subsequent RTIs. Interestingly, the presence of recurrent lower RTIs was associated with lower abundance of H. influenzae. While RTI symptom burden was prospectively measured during winter, RTI severity was defined retrospectively as ≥2 physician-diagnosed pneumonia episodes throughout life. Antibiotic courses for RTIs, particularly pneumonia, are high in the Netherlands, with an increasing prevalence of antibiotics that (partially) cover H. influenzae [34, 35]. While a recent antibiotic course was not associated with RTI symptoms and viral burden, it was positively associated with recurrent lower RTIs. This may contribute to the lower abundance of H. influenzae in the context of RTI severity, but not for RTI symptom burden in children with rRTIs. In addition, for RTI symptom burden, we selected a subgroup of children without recent prophylactic antibiotics in whom both upper and lower RTI symptoms were registered, potentially mitigating the effect of antibiotic prescription on this outcome.

In this study, we integrated respiratory microbiota, salivary antibody and serum and saliva proteomic results to identify potential mucosal mechanisms associated with RTI outcomes in early childhood. We found a high abundance of H. influenzae in our clinical paediatric cohort. This high abundance could either contribute to or result from rRTIs and associated morbidities or underlying disease, such as recurrent wheezing or allergic disease. The frequent use of antibiotics in the paediatric cohort could also play a role, as antibiotic exposure in early infancy leads to Haemophilus dominant maturation of the nasal microbiome during the first 2 years of life [36]. Additionally, the high viral burden in our cohort, with predominantly rhinovirus detected in the viral qPCR, could increase the Haemophilus abundance in the respiratory tract [37, 38].

H. influenzae abundance in the respiratory tract was negatively associated with polyreactive salivary antibodies. Serum and salivary proteomics identified several T-cell associated targets in serum, but not in saliva, that were positively associated with salivary polyreactive IgA and IgG. All serum protein markers that were associated with H. influenzae, including multiple T-cell, NK cell and inflammatory markers, showed negative correlations. Conversely, H. influenzae was positively associated with saliva proteomic markers, including cytotoxic T-cell marker CD8A, and inflammation markers such as IFN-γ and IL-33, indicating that local activation and inflammation on the respiratory mucosa is associated with increased abundance of H. influenzae. We identified soluble 4-1BB in serum as a factor positively associated with polyreactive salivary IgA and negatively associated with H. influenzae abundance. 4-1BB, as a member of the TNF receptor superfamily, can be induced as a co-stimulatory molecule and may be involved in the production of polyreactive salivary IgA production, by stimulation of B-cells or plasma cells that produce these antibodies. However, soluble 4-1BB can also act as an antagonist of co-stimulation by competing with membrane-bound 4-1BB for binding to 4-1BB ligand expressed on antigen-presenting cells. Future studies are required to elucidate the role of 4-1BB in polyreactive antibody production and the implication of H. influenzae abundance.

Our findings may be relevant for long-term respiratory outcomes, as a H. influenzae dominated microbiota profile in infants has been associated with the development of asthma [39, 40]. H. influenzae in the respiratory tract has been associated with viral wheeze, which is prevalent in young children [39]. Several studies have also found an association between H. influenzae and viruses, particularly rhinoviruses, in relation to asthma symptoms [38, 40]. In our clinical paediatric cohort, both H. influenzae and rhinovirus were highly prevalent. As we have previously published, asthma or recurrent wheezing is suspected in over half of children with rRTIs in this cohort [19]. H. influenzae colonisation can be associated with low-grade systemic inflammation as identified by elevated levels of serum C-reactive protein [41] and TNF-α [39]. We found that elevated inflammatory markers in saliva, particularly IFN-γ, were associated with H. influenzae abundance. This inflammation could influence both short-term RTI outcomes and long-term risks such as asthma, as indicated by the association between H. influenzae and IL-33, a cytokine important in asthma pathogenesis [42]. The recurrent nature of RTIs in childhood could also lead to epigenetic changes. Several studies have shown that (recurrent) RTIs can alter microRNA regulation [43] and DNA methylation [44] via IFN signalling pathways, possibly increasing the risk of asthma [43, 44] and the severity of RTIs [44].

This study's major strength lies in its prospective measurement of RTI symptoms through a daily mobile phone application, creating a high-depth dataset that avoids recall bias. While the total sample size of our clinical cohort was relatively small, the detailed daily registration of RTI symptoms allowed for a comprehensive multivariable negative binomial mixed model analysis accounting for longitudinal data per subject. We corrected for potential confounding variables such as age, RTI symptoms at time of sampling and recent antibiotic courses. However, the high lifetime antibiotic use in our cohort may still influence our results, even after correction for recent antibiotic courses.

A limitation in our study is the use of the V4 region, which hampers species-level annotation. However, in an attempt to provide species-level annotation, we employed multiple annotation methods. Using BLASTn, we identified multiple species annotations based on multiple database entries for each individual ASV, demonstrating the uncertainty in our annotations. To enable annotation to approximate species level, we employed the DECIPHER package for clust