Niche-specific immune and microbial signatures across the healthy upper respiratory tract mucosa

Summary

Host-microbe interactions in the upper respiratory tract (URT) niches that form the first line of defense against respiratory infections are incompletely understood. We profiled immune and microbial features using minimally invasive samples from 44 healthy adults (20–65 years), including nasopharyngeal (NPS), oropharyngeal (OPS), and mid-turbinate nasal swabs (MTSs), mucosal lining fluid (MLF), and saliva. Multiplex cytokine and antibody assays, 16S rRNA sequencing, and pathogen detection by PCR were performed. Immune and microbial profiles differed by niche: nasal samples showed consistently higher antiviral cytokine levels and Corynebacterium dominance. Oral samples had greater microbial diversity with distinct cytokine and antibody patterns. Saliva and MLF had the highest antibody concentrations, predominantly IgA. Integrated analyses identified site-specific microbe-immune associations. These findings show the feasibility of non-invasive, integrated profiling to uncover compartment-specific host-microbe relationships, providing a scalable framework for respiratory mucosal immunology and microbiome research with applications in infection surveillance and vaccine evaluation.

Highlights

• •Non-invasive upper airway samples enable integrated immune and microbial profiling
• •Nasosorption enables antigen-specific IgG profiling with high participant acceptance
• •Microbe-immune associations are niche-specific across nasal and oral sites

Immunology; Medical Microbiology

Introduction

Respiratory tract infections (RTIs) remain a global health burden, with lower RTIs (LRTIs) accounting for considerable morbidity and mortality across all ages. Despite overall declines, LRTIs caused approximately 2.18 million deaths in 2021, with the highest burden observed in low- and middle-income countries.^1,2 While young children and older adults are especially vulnerable, RTIs impact individuals across the lifespan, influencing long-term health outcomes and contributing to broader public health concerns.¹ Although vaccines exist for some respiratory pathogens, most causative agents lack immunization options.^3,4 Even for pathogens with available vaccines, coverage and effectiveness vary by age and region, and suboptimal mucosal immunity contributes to a substantial residual burden of RTIs.^4,5,6,7,8

The upper respiratory tract (URT) mucosa represents the primary site of pathogen entry, acting as a critical first line of defense.⁹ Epithelial cells lining the mucosa not only form a physical barrier but also contribute actively to immune defense by producing immunoregulatory proteins and by transporting secretory IgA (sIgA), facilitating pathogen neutralization and clearance via mucociliary action.^10,11 Innate immune cells associated with the URT mucosa (e.g., macrophages and dendritic cells) detect antigens and, in response, secrete cytokines and chemokines, which, together with signals from the epithelial cells, can recruit and activate other immune cells, including T and B cells.¹¹ URT compartments—comprising nasal passages and the nasopharynx/oropharynx—harbor distinct microbiota and immune cells that collectively influence infection susceptibility and mucosal homeostasis.^10,12,13 The nasopharynx, rich in lymphoid tissue, such as the adenoids, is central for orchestrating immune responses against invading pathogens and is typically dominated by Corynebacterium or Dolosigranulum species in a healthy state.^10,12 Meanwhile, the oropharynx, exposed to both inhaled and ingested materials, contains immune structures like the palatine tonsils that are richer in differentiated B cells and harbors a more diverse microbial community, including Fusobacterium, Neisseria, Prevotella, and Veillonella species.^12,13 However, URT samples in humans are most often used for pathogen detection rather than for exploring host-microbe interactions.¹² This limits understanding of how local immune and microbial networks contribute to mucosal defense and how they could inform improvements in vaccines, diagnostics, and treatments.

Although several studies have characterized either immune or microbial features of the respiratory mucosa, few have integrated both within the same URT sample or compared profiles across different niches of the same individual.^{14,15,16,17,18,19} Consequently, key questions remain regarding the coordination of mucosal immunity and the microbiome across URT sites. Non-invasive and self-collection sampling approaches could help address these gaps by enabling repeated, large-scale, and cost-effective studies, which are especially valuable in low-resource settings and longitudinal investigations.²⁰ However, conventional URT sampling methods, such as nasopharyngeal (NPS) or oropharyngeal swabs (OPS), are poorly tolerated due to discomfort and participant burden and are further limited by restricted sample volumes, posing challenges for conducting such studies.^21,22

To our knowledge, few studies have enabled integrated within-subject comparison of microbial and immune features across multiple URT niches. Here, we systematically characterized the healthy URT ecosystem through simultaneous microbial and immunological profiling of distinct compartments using non-invasive and self-collection methods. Our aim was to establish a scalable, participant-friendly framework for integrative microbial and immunological profiling that captures site-specific immune-microbial signatures and interactions. This approach enables future translational studies into respiratory infection susceptibility, mucosal vaccine responses, and population-level respiratory health surveillance.

Results

Non-invasive samples were better tolerated

To inform sampling strategy selection for future studies, we first assessed participant perception of comfort and willingness to participate in repeated sampling across the five collection methods whose demographic characteristics are summarized in STAR Methods Table 1 (Figure 1A). Participants experienced saliva and mucosal lining fluid (MLF) sample collections to be most comfortable, with no significant differences between the two methods. In contrast, NPS collection was rated as the most uncomfortable. Mid-turbinate nasal swab (MTS) and OPS sample collections showed no significant differences and were generally perceived as neutral to comfortable (Figure 1B). This pattern was confirmed in the results of the second question, where all participants indicated that they would (strongly) agree to participate in a study requiring regular saliva or MLF sampling. (Figure 1B).

Table 1.

Summary of the age distribution and biological sex of participants stratified by sampling strategy (professional vs. self-collection)

	Professional (N = 44)	Self (N = 16)
Age group (years)
20-25	4 (9.1%)	1 (6.3%)
25-30	19 (43.2%)	8 (50.0%)
30-35	3 (6.8%)	1 (6.3%)
35-40	2 (4.5%)	1 (6.3%)
40-45	7 (15.9%)	3 (18.8%)
45-50	6 (13.6%)	2 (12.5%)
50-55	0 (0%)	0 (0%)
55-60	1 (2.3%)	0 (0%)
60-65	2 (4.5%)	0 (0%)
Biological sex
Female	28 (63.6%)	11 (68.8%)
Male	16 (36.4%)	5 (31.3%)

Data are depicted as counts with percentages (%), calculated relative to the total number of participants in each group (shown in parentheses). Participants who self-collected samples were a subset of those who were sampled professionally.

Figure 1.

Sampling protocol and participant-reported comfort across the different upper respiratory tract sampling methods

(A) Order of sample collection. MLF and MTS samples were first collected from the participant’s left and right nostrils, respectively, followed by saliva and OPS samples. Next, MLF and MTS samples were collected from the right nostril and left nostril, respectively, with the final sample being the NPS. The saliva swab illustrated here is a schematic representation only and does not depict the exact collection device or protocol used in this study. 60″ = nasosorption strip held in nostril for 60 s, 2′ = saliva swabs held in mouth for 2 min. The figure was created using BioRender (https://biorender.com).

(B) Questionnaire results comparing comfort perception of the participants (N = 44) across different sampling techniques. Statistical analysis was performed using the Friedman rank-sum test and Wilcoxon pairwise signed-rank test (post hoc). Only non-significant comparisons are shown (ns); all other pairwise comparisons were statistically significant. MTS, mid-turbinate nasal swab; MLF, nasosorption strip; NPS, nasopharyngeal swab; OPS, oropharyngeal swab.

Respiratory pathogens were detected with low concordance across sample types

Having established that non-invasive methods were overall better tolerated, we next examined whether these methods captured comparable respiratory pathogen carriage to conventional sampling using multiplex PCR across all sample types (RespiFinder). The respiratory sample types analyzed using each assay are summarized in STAR Methods Table 2. Carriage of one or more respiratory pathogens was detected in 56.8% of participants (25/44). In total, two bacterial species (i.e., Bordetella pertussis and Chlamydophila pneumoniae) and four viruses (i.e., bocavirus, influenza A, rhino/enterovirus, and severe acute respiratory syndrome coronavirus 2 [SARS-CoV-2]) were detected. The majority of the positive results originated from MTS, MLF, and NPS samples. Detections were also observed in the self-collected samples (Table 3). However, the concordance in results among sample types was low, with no participant being positive for a specific respiratory pathogen in the entirety of the collected sample types. Four participants were positive for a specific respiratory pathogen in two or more URT sample types (Figure S1). Given the potential influence of pathogen carriage on immune and microbial profiles, samples positive for respiratory pathogens were excluded from subsequent microbe-immune association analyses. Sensitivity analyses confirmed the robustness of findings when pathogen-positive samples were included as a stratification variable (Figure S2).

Table 2.

Study design overview indicating the assays performed per sample type, sampling strategy and use case

Sample type	Assays
	Molecular detection by RespiFinder PCR	Total Ig	Immune markers by LegendPlex	Immune markers by Olink	16S rRNA sequencing
Self-sampling (home)
MTS	X	X	X	X	X
Saliva	X	X	X	–	X
MLF	X	X	X	–	X
Professional sampling
MTS	X	X	X	X	X
Saliva	X	X	X	–	X
OPS	X	X	X	–	X
MLF	X	X	X	–	X
NPS	X	X	X	X∗	X

The symbol “X” indicates that the corresponding sample type was analyzed with the respective assay, whereas the empty cells indicate that these sample types were not analyzed with the corresponding assays. Asterisk (∗) indicates that only a subset of these samples were analyzed with the corresponding assay. MTS, mid-turbinate nasal swabs; MLF, nasosorption strips; OPS, oropharyngeal swabs; NPS, nasopharyngeal swabs.

Table 3.

Positive results per pathogen, sample type, and sampling strategy presented as counts and the respective percentages in parentheses

	Total	B. pertussis	Bocavirus	Influenza A	Rhino/enterovirus	SARS-CoV2	C. pneumoniae
MLF
Professional	87	1/87 (1.15%)	1/87 (1.15%)	2/87 (2.29%)	3/87 (3.45%)	1/87 (1.15%)	0/0 (0%)
Self	16	2/16 (12.50%)	0/0 (0%)	0/0 (0%)	1/16 (6.25%)	0/0 (0%)	0/0 (0%)
MTS
Professional	88	7/88 (7.95%)	0/0 (0%)	0/0 (0%)	3/88 (3.41%)	5/88 (5.68%)	1/88 (1.14%)
Self	16	1/16 (6.25%)	0/0 (0%)	0/0 (0%)	1/16 (6.25%)	0/0 (0%)	0/0 (0%)
NPS
Professional	44	1/44 (2.27%)	1/44 (2.27%)	0/0 (0%)	1/44 (2.27%)	4/44 (9.09%)	0/0 (0%)
OPS
Professional	44	1/44 (2.27%)	0/0 (0%)	0/0 (0%)	1/44 (2.27%)	0/0 (0%)	0/0 (0%)
Saliva
Professional	42	0/0 (0%)	0/0 (0%)	0/0 (0%)	1/42 (2.38%)	0/0 (0%)	0/0 (0%)
Self	16	2/16 (12.50%)	0/0 (0%)	0/0 (0%)	0/0 (0%)	0/0 (0%)	0/0 (0%)

Data are depicted as counts with percentages (%) in parentheses. Percentages were estimated using the total number of samples available for this assay (total) per sample type and sampling strategy (professional or self). MTS, mid-turbinate nasal swab; MLF, nasosorption strip; NPS, nasopharyngeal swab; OPS, oropharyngeal swab. MLF and MTS samples had two samples per participant (from the two nostrils). One MLF sample did not have sufficient volume for this assay.

Cytokine patterns were broadly similar with compartment-specific differences

Following characterization of pathogen carriage across sample types, we profiled cytokines and effector molecules across all URT samples to identify compartment-specific signatures and evaluate whether the different sampling methods captured similar immune profiles. APRIL and CXCL8 were detectable in all samples, while IFN-γ, IFN-α2, and IL-33 were among the least detectable immune markers for the majority of the samples, with the exception of IFN-γ in saliva (Table S1). Overall, the cytokines involved in cytotoxic antiviral, neutrophil-related, or broader inflammatory responses displayed similar concentration patterns in all URT niches (Figures 2A and 4A). Cytokines including IL-6, IL-33, and type I/III interferons demonstrated more variable or limited cross-site concordance. As expected, no differences between nostrils were observed (Figure S3). Professionally and self-collected samples showed similar concentrations, particularly for MLF and MTS, whereas salivary CCL2 and CXCL1 levels were significantly higher in the self-collected samples (Figure S4A). Despite overall concordance, within analytes, differences in concentration profiles were observed between sample types (Figure 2B; Table S2). OPS consistently showed the lowest concentrations across all analytes as compared to the other sample types (Figures 2A and 2C). NPS had the highest concentrations for the majority of the analytes (8/13), particularly those involved in antiviral responses. Results from the log-linear mixed-effects model indicated variations in specific analyte concentrations per sample type (Figure 2C; Table S3). Notably, while MTS samples had significantly lower concentrations compared to NPS across analytes, they showed similar immune marker patterns (Figures 2B and 2C).

Figure 2.

Immune marker (LegendPlex) concentrations in the different samples

(A) Boxplot of the concentrations of each immune marker per sampling method. The horizontal line shows the median, the box represents the interquartile range (IQR), and the whiskers extend to 1.5× the IQR.

(B) Principal component analysis (PCA) plot showing clustering of immune marker profiles across different sample types. Ellipses represent 95% confidence regions around group centroids. Black arrows indicate immune markers with the strongest loadings. Statistical significance of group differences was assessed by Permutational Multivariate Analysis of Variance (PERMANOVA), with the F statistic, R², and p value shown in the plot.

(C) Geometric mean ratios (GMRs) by sample type with 95% confidence intervals (CIs) per immune marker, derived from a log-linear mixed-effects model that included fixed effects for the interaction between sample type and immune marker and a random intercept for participant ID to account for repeated measures (red line placed at GMR = 1, indicating the reference, i.e., NPS). Vertical lines represent the 95% CI. Sample size for all sample types was 44, with the exception of saliva (N = 40) and MTS and MLF, which had 88 samples each—44 from the left and 44 from the right nostril. The corresponding mean concentration of each immune marker measured in the left and the right nostril samples was used for the depicted analyses. MTS, mid-turbinate nasal swab; MLF, nasosorption strip; NPS, nasopharyngeal swab; OPS, oropharyngeal swab.

Figure 4.

Pairwise Spearman’s rank correlation coefficients of analyte concentrations across and within respiratory sample types

(A) Heatmap showing pairwise Spearman correlation coefficients (rho) for each immune marker and total antibody measured across sample type comparisons. Correlation coefficients are color-coded, indicating positive (brown) or negative (teal) correlations.

(B) Spearman correlations between total antibodies (IgA, IgG, and IgM) and cytokine concentrations within each sample type/niche. Analytes are ordered by predefined functional groups (antiviral/cytotoxic: red, chemokine: green, humoral/B cell: blue, inflammatory: yellow). MTS, mid-turbinate nasal swab; MLF, nasosorption strip; NPS, nasopharyngeal swab; OPS, oropharyngeal swab; FDR, false discovery rate. ∗FDR < 0.05; ∗∗FDR < 0.01; ∗∗∗FDR < 0.001. Cells that are empty (no asterisk) indicate no significant correlation after Benjamini-Hochberg correction. For MTS and MLF samples, the mean concentration of the left and right nostrils per analyte was used. Sample size (N = 36).

Total and antigen-specific antibody levels varied across URT compartments

To complement cytokine profiling, we proceeded to quantify total and antigen-specific antibody concentrations across the five sample types to investigate humoral immune signatures and assess URT sample type comparability. The majority of measurements were within the limits of quantitation for total antibodies (Table S4). Across nasal cavity samples, the patterns of total IgA, IgG, and IgM concentrations were similar, with total IgA concentrations being the highest, followed by IgG and subsequently IgM. While total IgA remained the most abundant isotype in oral cavity samples (i.e., OPS and saliva), total IgG and IgM concentrations did not differ significantly between OPS and saliva (p value > 0.05) (Figure 3A). Spearman correlation analysis showed positive correlations of total antibody concentrations across sample types, which were particularly strong among MLF and MTS, NPS and saliva, and to a lesser extent between OPS and saliva (Figure 4A). OPS samples had the lowest concentrations for all antibodies. The log-linear mixed-effects model showed that, when compared to NPS, MLF and saliva samples had significantly higher concentrations across all total antibodies (Figure 3B; Table S5). No significant differences were observed between professionally and self-collected samples for total antibody concentrations (Figure S4B).

Figure 3.

Total and antigen-specific antibody concentrations

(A) Boxplots of the concentrations of each total antibody measurement per sample type. The horizontal line shows the median, the box represents the IQR, and the whiskers extend to 1.5× the IQR. Statistical testing was performed using the Wilcoxon signed-rank test for paired data.

(B) Bar plot of GMRs and corresponding 95% CIs (red line placed at GMR = 1, indicating the reference, vertical lines depict 95% CIs) derived from a log-linear mixed-effects model that included fixed effects for the interaction between sample type and total antibody and a random intercept for participant ID to account for repeated measures.

(C) Boxplot with log-10 concentrations for RSV-PreF, SARS-CoV-2 S1, and B. pertussis PT IgG per sample type. Statistical testing was performed using the Wilcoxon signed-rank test for paired data, except for PT due to most data being below the lower limit of detection. The horizontal line shows the median, the box represents the IQR, and the whiskers extend to 1.5× the IQR. Sample size for all sample types was 44, with the exception of saliva (N = 40) and MTS and MLF, which had 88 samples each—44 from the left and 44 from the right nostril. The corresponding mean concentration of each immune marker measured in the left and the right nostril samples was used for the depicted analyses. MTS, mid-turbinate nasal swab; MLF, nasosorption strip; NPS, nasopharyngeal swab; OPS, oropharyngeal swab. ns, p > 0.05; ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001; ∗∗∗∗p < 0.0001. AU, arbitrary units; IU, international units.

Next, we explored pre-existing antigen-specific humoral immunity reflecting prior pathogen exposure or vaccination by measuring IgG concentrations against three epidemiologically relevant respiratory pathogens: respiratory syncytial virus (RSV), SARS-CoV-2, and B. pertussis. RSV PreF and SARS-CoV-2 S1 IgG were reliably quantified across sample types, while B. pertussis PT detection was more limited. Regarding PreF and S1, MLF samples had significantly higher IgG concentrations than NPS samples. NPS samples had no significant differences in IgG concentrations compared with MTS and saliva samples but had significantly higher concentrations compared to OPS samples and significantly lower concentrations for IgG against PreF and S1 compared to MLF samples (Figure 3C).

Immune marker correlations differ between oral and nasal compartments

Given that individual immune markers differed across niches, we next inquired whether these differences reflected broader organizational patterns in immune mediator co-expression, both across and within URT compartments. Spearman correlation analysis revealed that chemokines such as CXCL8, CXCL10, CXCL1, CCL2, and CCL4 showed strong positive correlations across swab types, particularly among MLF, MTS, and NPS samples (Figure 4A). In contrast, saliva exhibited weaker cross-site correlations overall, though CXCL8 and IFN-λ1 remained significantly associated with other compartments. Similarly, total antibody concentrations showed positive correlations across sample types, which were particularly strong among MLF and MTS, NPS and saliva, and to a lesser extent between OPS and saliva (Figure 4A). To assess within-sample relationships among immune mediators, we performed pairwise Spearman correlations and visualized analytes grouped by functional category (Figure 4B). Within individual sample types, total antibodies (IgA, IgG, and IgM) correlated strongly with one another. Chemokines (CCL2, CCL4, CXCL1, and CXCL8) and APRIL also associated with total antibodies, especially in MLF, OPS, and MTS, while NPS and saliva showed similar but not significant patterns (Figure 4B). More broadly, chemokines consistently correlated within their group, and the same was observed for cytotoxic/antiviral mediators (Figure 4B). Inflammatory cytokines (IL-6 and IL-33) correlated with chemokines and antiviral/cytotoxic mediators, especially in saliva, MLF, MTS, and OPS. In oral cavity samples, antiviral/cytotoxic cytokines correlated strongly with chemokines, but this relationship was not observed in nasal cavity samples. Interestingly, antibodies did not correlate with cytokines or chemokines in NPS and saliva, but antibodies did correlate with chemokines in the nasal MLF and MTS samples and in OPS also with antiviral cytokines.

Nasal and oral niches display distinct microbial communities

Beyond immune profiling, microbial community composition represents a key determinant of mucosal physiology. We, therefore, sought to characterize and compare microbial communities across the five URT samples. In the nasal cavity, NPS and MTS showed no differences in alpha diversity, while MLF samples were significantly more diverse than MTS and NPS. OPS and saliva showed higher alpha diversity than nasal samples (Figure 5A). The nasal and oral cavity niches showed distinct community composition profiles, with 51.1% (R²) of the total variance explained by the URT region (p value < 0.01) (Figure S5A). Oral samples (i.e., OPS and saliva) were also distinct from each other in microbial composition (beta diversity) (OPS vs. saliva, R² = 25.7 %, p value < 0.05) (Figures 5C and S5B). Only for saliva, professionally collected samples were significantly different in both alpha- and beta-diversity measures when compared to self-collected counterparts (Figure S6C). No significant differences were observed between nostrils (Figure S7). In addition, the most abundant genera differed between the URT niches (Figure S8), where nasal samples were more comparable among each other in the top abundant representative genera than oral samples (Figure 5B).

Figure 5.

Microbial community composition and diversity across URT sampling sites

(A) Alpha diversity measured by Shannon (left) and Simpson (right) indices across the five sample types (N = 22). Violin plots overlay boxplots, with the horizontal line indicating the median, the box representing the IQR, and the whiskers extending to 1.5× the IQR. Horizontal bars with “ns, FDR > 0.05; ∗FDR < 0.05; ∗∗FDR < 0.01; ∗∗∗FDR < 0.001; ∗∗∗∗FDR < 0.0001” denote Wilcoxon signed-rank tests for paired data or paired t test significance after Benjamini-Hochberg correction, depending on whether the data followed a normal distribution or not.

(B) Boxplots of relative abundances (%) for the top six genera in nasal (MLF, MTS, NPS, N = 27, top row) and oral (OPS, saliva, N = 36, bottom row) cavity samples. Each iamge shows one genus, with the box indicating the IQR, the horizontal line representing the median, the whiskers extending to 1.5× the IQR, and points indicating outliers. Statistical testing was performed with the Wilcoxon signed-rank test for paired data (ns, p > 0.05; ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001; ∗∗∗∗p < 0.0001).

(C) Principal coordinates analysis (PCoA) of Bray-Curtis dissimilarities, colored by sample type (sample sizes: 71 MLF, 80 MTS, 36 NPS, 44 OPS, and 36 saliva). Ellipses show 95% confidence intervals for each sample type. Percentages on axes denote explained variance. Samples clustering distinctly on the right represent OPS and saliva, reflecting their divergent microbial composition from nasal sampling sites. Pairwise PERMANOVA p values adjusted for multiple testing (Benjamini-Hochberg) are shown (∗FDR < 0.05). MTS, mid-turbinate nasal swab; MLF, nasosorption strip; NPS, nasopharyngeal swab; OPS, oropharyngeal swab.

Integrated analysis identified niche-specific microbe-immune profiles

Having separately characterized immune and microbial profiles across URT niches, we next sought to integrate these datasets to identify major drivers of variation between niches. Multi-omics factor analysis (MOFA) identified three latent factors capturing variation across domains (i.e., microbiome, cytokine, and total antibodies) (Figure 6A). Factor 1 explained the largest proportion of variance across cytokines and total antibodies, whereas factor 2 accounted for microbiome variance and showed a more balanced contribution of all three domains (microbiome, cytokines, and total antibodies). Factor 3 explained a smaller proportion of variance and was driven mainly by microbiome and antibody features, with limited contribution from cytokines (Figure 6A). Associations with metadata showed that sample type was strongly associated with factor 2 (Figure 6B). Notably, pathogen detection status contributed minimally to the explained variance across data domains, suggesting that asymptomatic pathogen carriage did not substantially alter the niche-specific immune and microbial signatures observed in this cohort (Figure 6B). Samples clustered distinctly by collection site (Figure 6C), with oral cavity samples separating from nasal cavity samples along factor 2, as well as OPS clustering distinctly from saliva. Inspection of the feature weights highlighted the microbial genera Corynebacterium and Staphylococcus, alongside immune markers such as CXCL8, granzyme A, and total IgG, as top contributors to factor 2, which separated nasal cavity samples from oral cavity samples. In contrast, oral cavity samples (OPS and saliva) were characterized by genera such as Rothia, Actinomyces, and Veillonella, together with cytokines IFN-γ and CCL2 (Figure 6D). Constrained ordination with redundancy analyses (RDAs) provided confirmatory evidence that anatomical compartment was the major driver of microbial-immune variation (Figure S9).

Figure 6.

MOFA of the upper respiratory tract dataset

(A) Variance explained by each factor across the three data domains (microbiome, cytokines, and total antibodies).

(B) Associations of MOFA factors with sample metadata, shown as −log10 adjusted p values.

(C) Latent factor space (factor 1 vs. factor 2) showing clustering of samples by sample type.

(D) Top cross-modal features driving factor 2, highlighting microbial genera, cytokines, and total antibodies. Negative loadings correspond to features associated with nasal cavity samples (NPS, MTS, and MLF), whereas positive loadings correspond to features associated with oral cavity samples (OPS and saliva). For this analysis, 40 MLF, 40 MTS, 36 NPS, 44 OPS, and 35 saliva samples were used. MTS, mid-turbinate nasal swab; MLF, nasosorption strip; NPS, nasopharyngeal swab; OPS, oropharyngeal swab.

Microbe-immune marker associations varied per sampled site

To further characterize the niche-specific host-microbe interactions suggested by the MOFA results, we examined specific microbe-immune marker associations within each URT site using a non-parametric permutation-based approach. In MTS samples, Corynebacterium was positively associated with several pro-inflammatory cytokines (e.g., IL-8 and TNF), chemokines (e.g., CCL4), as well as total IgG. In MLF samples, Anaerococcus and Finegoldia were positively associated with total IgM. Within NPS, Capnocytophaga showed negative associations with CCL2, and Haemophilus with granzyme A and CXCL8. In the oropharynx, Streptococcus was negatively associated with IL-6, whereas positive associations with total IgA and IgG were observed with certain oral genera. Similarly, in saliva, Streptococcus was negatively associated with CXCL1 and CCL2, while Leptotrichia showed negative links to CCL2 and total IgM (Figure 7). To confirm robustness of these associations, analyses were repeated with pathogen-positive samples included as a stratification variable, yielding largely consistent network architecture (Figure S2).

Figure 7.

Site-specific microbial-immune association network

Each image shows the associations for the specific sample type/niche as a force-directed network of amplicon sequence variant (ASV)-immune marker associations filtered for p value < 0.05 and FDR < 0.2. Nodes represent genera (black circles) or cytokines and total antibodies (i.e., total Ig) (gray circles). Edges encode significant positive (orange) or negative (blue) associations, with line thickness proportional to |effect size| (strength). Layout is Fruchterman-Reingold, weighted by |effect size|. For this analysis, 39 MLF, 32 MTS, 29 NPS, 41 OPS, and 34 saliva samples were used. MTS, mid-turbinate nasal swab; MLF, nasosorption strip; NPS, nasopharyngeal swab; OPS, oropharyngeal swab.

Discussion

We applied a systematic integrative approach to explore immune and microbial ecosystems across five URT niches in healthy adults. Here, we demonstrated that it is possible to perform multiple assays in a single URT sample regardless of whether it was collected more abrasively or non-invasively. This allowed us to identify distinct immune and microbial patterns across URT niches and host-microbe interactions, revealing both shared and site-specific features of mucosal immunity. Our results additionally indicate that self-collected samples (primarily MTS, MLF, and to a lesser extent saliva) provide reasonably comparable results to professionally collected samples, suggesting that self-sampling could serve as an acceptable alternative. Non-invasive approaches, such as MTS and MLF, may, therefore, be particularly well suited for repeated immune profiling and longitudinal surveillance in populations where tolerability is critical, including young children and older adults. Saliva may provide a practical alternative for antibody-based or population-level surveillance, although its greater biological heterogeneity should be considered during study design. Together, these findings demonstrate the potential of non-invasive, integrative URT profiling to investigate niche-specific host-microbe interactions as well as pathogen carriage and diagnostics.

Interestingly, 32% of the asymptomatic participants tested positive for B. pertussis. This may be explained by our sample collection coinciding with a rise in whooping cough cases across the Netherlands.²³ These detections, mainly in the most anterior nasal samples, likely reflect asymptomatic exposures in individuals with pre-existing immunity, which may similarly explain our other positive PCR findings. Latent, past or imminent infection coupled with the high sensitivity of PCR techniques (87.3% sensitivity for the RespiFinder) may, indeed, explain positive PCR outcomes in asymptomatic individuals.^24,25,26 We propose that the inconsistent positive PCR results across sample types are caused by low viral or bacterial loads.²⁶ These technical and biological factors likely account for the low concordance of pathogen detection across sampling sites and should be considered when interpreting site-specific PCR results in this study.

Overall, we observed that the nasal cavity compartments displayed concordant results in all assays, with especially strong agreement between NPS and MTS samples in terms of relative immune and microbial patterns. Indeed, immune marker detection patterns, especially those involved in cytotoxic antiviral responses (i.e., granzyme A, perforin, IFN-λ1, and CXCL10), were shared among NPS, MTS, and MLF samples, albeit with lower concentrations than the former two. Total antibody patterns were nearly identical among these samples in terms of isotype distribution, with total IgA present at the highest concentrations, followed by total IgG and total IgM at the lowest, mirroring the results of a previous study.²⁷ Interestingly, MLF exhibited the highest antigen-specific IgG and total antibody concentrations except for IgA, which was most abundant in saliva, as observed in previous studies.^28,29 Acknowledging that the smaller MLF elution volume may partly contribute to higher absolute concentrations, the finding that MLF captured detectable antigen-specific IgG against epidemiologically relevant respiratory pathogens, combined with its high participant acceptability demonstrated in this study, suggests that nasosorption represents a promising non-invasive approach for large-scale seroprevalence surveillance and mucosal vaccine immunogenicity studies where participant compliance is critical. Saliva appeared to be distinct from the oropharynx in cytokine, microbiome, and antibody assays, with the oropharynx exhibiting consistently lower cytokine and antibody concentrations. Notably, while IFN-γ was detectable in saliva, this cytokine was largely undetectable in NPS, OPS, MLF, and MTS samples, a pattern previously reported in healthy cohorts.^30,31 This may reflect the composite nature of saliva, which includes contributions from nasal drainage, the oral mucosa, gingival crevicular fluid, salivary glands, and tonsillar tissue.³² Consistent with this, the relatively elevated CCL2 levels observed in saliva may reflect the broader biological heterogeneity of this sample type, potentially including differential stimulation by local oral microbial communities.³³

Correlation analyses indicated that within niches, total antibodies were strongly interrelated and—particularly in the nasal cavity—tended to group with chemokines and APRIL, whereas cytotoxic antiviral cytokines and effector molecules tended to correlate within a separate group. This organization was broadly conserved across compartments but varied in strength, suggesting both shared immune features and site-specific influences. Although similar immune correlation patterns were also observed in the oropharynx, antiviral cytokines additionally correlated with total antibodies in this niche. Despite this, the oropharyngeal samples consistently displayed the lowest antibody and cytokine concentrations. This may indicate that immune mediators in the oropharynx remain functionally coordinated through underlying lymphoid tissues, while their overall abundance is reduced. Possibly, this reflects its tolerogenic role due to continuous exposure to food and microbiota or its distinct non-keratinized squamous epithelium compared with the ciliated, goblet cell-rich columnar epithelium of the nasopharynx.³⁴ By contrast, the higher immune marker levels in the nasal cavity immune profiles support a surveillance-oriented immune state against inhaled pathogens. Collectively, these results support a model in which URT niches occupy distinct positions along a surveillance-tolerance axis, with nasal cavity sites reflecting immune surveillance against inhaled pathogens and the oropharynx adopting a more tolerogenic profile, shaped by epithelial architecture, local microbial communities, and anatomical location.

One explanation for the observed differences across sites may lie in the collection procedures. MLF samples were eluted in only 300 μL and saliva was undiluted, while swab samples were placed in 1 mL universal transport medium (UTM). This difference in dilution could partly account for the higher concentrations of immune mediators and microbial abundances observed in MLF and saliva, although not just levels but also profiles varied between sample types. Yet, despite OPS, MTS, and NPS sharing the same collection medium, clear differences remained between OPS and the nasal swabs, indicating that biological variation between niches contributed to the observed patterns. Dilution correction was deliberately not applied as the aim was to reflect the biological information obtainable from each sample type under standard collection conditions, providing a practically relevant comparison for future study design.

The microbiome has been recognized to shape mucosal immunity and influence resistance to invading pathogens.³⁵ We observed a clear separation in the microbial composition of the nasal and oral cavity, similar to a previous study.³⁶ Oral sites exhibited higher diversity compared to nasal cavity samples, in line with the current knowledge on URT bacterial diversity.^12,35 Among the nasal cavity samples, MLF showed greater microbial diversity than NPS and MTS, possibly reflecting the richer mucus content in MLF samples, which may harbor both resident and transient bacterial species.³⁷ Microbe-cytokine associations were predominantly negative across niches, with the exception of Corynebacterium in the mid-turbinate nasal cavity, which showed positive associations with pro-inflammatory cytokines. This finding may, in part, reflect the additional resolution provided by the Olink dataset available for MTS samples, although this would not affect the directionality of the associations. In addition, positive associations between commensal genera and total IgG or IgM were also observed in OPS and MLF samples. To investigate whether positive PCR results influenced the observed microbe-immune associations, we repeated the analyses with pathogen-positive samples stratified, which yielded largely consistent associations indicating that these are robust.

To investigate the driving factors explaining URT niche differences, we integrated microbiome, cytokine, and total antibody datasets with MOFA. This confirmed that sample type (i.e., niche) was the primary driver of variation, with nasal sites clustering together and separating from oral sites. MOFA identified Corynebacterium, Staphylococcus, CXCL8, granzyme A, and total IgG as key features distinguishing nasal from oral niches, while oral niches were characterized by Rothia, Actinomyces, Veillonella, IFN-γ, and CCL2. In contrast, total IgA and IgM had little discriminatory properties. To further confirm our observations, we employed distance-based RDA and constrained RDA, which both corroborated our findings. These reinforce our observations in the separate cytokine, total antibody, and microbiome analyses. In addition, the observation that defined host niches associated with microbiome composition suggests that host responses may regulate microbial presence, although regulation of immune profiles by microbes cannot be excluded.

A strength of this study is that five different URT samples were collected from each participant, including more abrasive and non-invasive methods. This allowed for direct comparisons between sample types and niches. In addition, we employed four independent integrative approaches to robustly characterize microbe-immune signatures in the healthy URT. We show how multiple immunological and microbial analyses in different niches could provide a foundational reference for studying RTIs and their potential determinants.

Limitations of the study

Our study has certain limitations. Firstly, this study was conducted in a single cohort of healthy adults (20–65 years) and did not include comparative groups such as children, older adults, vaccinated individuals, or participants with active respiratory infection, which limits generalizability and constrains the clinical interpretation of the observed niche-specific immune and microbial patterns. This was, however, intentional, as the study was designed to establish a baseline reference framework for integrated URT profiling under steady-state conditions. Secondly, the study population consisted of individuals affiliated with biomedical research, which may have introduced a bias in the questionnaire responses favoring mucosal sampling. However, similar results were obtained by Salzano et al. regarding discomfort levels with different URT samples from a more diverse study population.³⁸ Additionally, the quality of self-collected samples may have been higher because of the professional background of the participants. However, previous studies point to high self-collection sample accuracy, especially regarding nasal samples.³⁹ The paired self-collection subset was limited and not powered to establish formal equivalence between collection strategies, but was sufficient to assess directional concordance and feasibility. Importantly, a key challenge in conducting multiple analyses on URT samples is the limited material available, which can lead to missing data for certain assays. Although this was also occasionally observed in our study, we were able to successfully perform parallel immunological, microbial, and molecular analyses from small sample volumes, demonstrating the feasibility of multiplexed profiling in minimally invasive samples. In addition, microbiome analyses were conservatively interpreted at the genus level, as the V4 region of the 16S rRNA gene does not provide sufficiently reliable species-level resolution, thereby limiting species-level functional interpretation.⁴⁰ Despite this, robust cross-niche comparisons of dominant microbial communities were still possible. Although nasal lavage is a common URT sample, we did not include it in this study. This sampling approach may provide higher sample volume and yield greater microbial diversity in some settings, but it also introduces variable dilution and may reduce sensitivity for soluble immune mediator detection compared with absorptive sampling (e.g., nasosorption strips).⁴¹ Finally, our study had a relatively small sample size, which should be considered when interpreting the generalizability of our observations. It also limited the statistical power needed for some subgroup and association analyses, particularly after exclusion of pathogen-positive participants. These analyses should, therefore, be interpreted as exploratory and hypothesis generating. However, many of our observations are consistent with previous reports.^{27,28,29,30,31,32,33,34,35,36} This, together with the inclusion of multiple URT niches and the use of complementary integrative analyses, supports the robustness of our findings and provides a reference framework for future, larger studies in more diverse populations.

In conclusion, this study demonstrates that immune and microbial signatures, and their interactions, across URT niches can be obtained from a single minimally invasive sample. These results provide a foundation for future work to validate and extend non-invasive, integrated profiling across ages and geographical settings, including through longer-read amplicon sequencing or metagenomics, with potential applications in vaccine, respiratory infection, and chronic airway disease research and surveillance.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Gerco den Hartog (gerco.den.hartog@rivm.nl).

Materials availability

This study did not generate new, unique reagents. All materials that were used are commercially available and previously published.

Data and code availability

• •Raw 16S rRNA gene sequencing reads have been deposited at the European Nucleotide Archive (ENA) and are publicly available (ENA: PRJEB121324).
• •Silva v138.1 taxonomic training data: https://zenodo.org/records/4587955.⁴²
• •Processed datasets generated during this study have been deposited and are available in the Mendeley Data repository (Mendeley Data: https://doi.org/10.17632/96nnvxsr32.1).
• •All code used for data processing and statistical analyses has been deposited on GitHub and is available at GitHub: https://github.com/NafsikaK/urt-multiomics-profiling.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

This study was funded by the Strategic Program RIVM (SPR) and the Dutch Ministry of Health, Welfare and Sport. The funders had no role in the study design, data collection, analysis, or manuscript preparation. We would like to thank all study participants and acknowledge the valuable support of colleagues and clinical personnel involved in sample collection and logistical coordination.

Author contributions

N.K., writing – original draft, writing – review and editing, visualization, investigation, formal analysis, methodology, data curation, conceptualization; J.K., writing – review and editing, formal analysis, methodology, validation, data curation, resources; M.A.N., writing – review and editing, methodology, validation; H.v.D., investigation, resources; E.R., investigation, resources. P.K., writing – review and editing, methodology, validation; D.E., methodology, resources; R.J., investigation, resources; L.v.E., investigation, resources; E.v.W., writing – review and editing, methodology; J.A.F., writing – review and editing, software, methodology, validation; S.F., writing – review and editing, methodology, resources, validation, conceptualization, supervision, funding acquisition; M.I.d.J., writing – review and editing, validation, resources, conceptualization, supervision; J.d.J., writing – review and editing, validation, resources, conceptualization, supervision, funding acquisition; L.M.V., writing – review and editing, validation, resources, conceptualization, supervision, funding acquisition; G.d.H., writing – review and editing, methodology, validation, resources, conceptualization, supervision, funding acquisition, project administration.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Mouse anti-human IgA capture antibody	BioLegend, San Diego, CA, USA	Cat# 411502; RRID: AB_2650697
Mouse anti-human IgG capture antibody	BD Biosciences, Franklin Lakes, NJ, USA	Cat# 555784; RRID: AB_396119
Mouse anti-human IgM capture antibody	Southern Biotech, Birmingham, AL, USA	Cat# 9020-01; RRID: AB_2796572
Goat anti-human IgA-PE detection antibody	Southern Biotech, Birmingham, AL, USA	Cat# 2052-09; RRID: AB_2795707
Goat anti-human IgG-PE detection antibody	Jackson Immunoresearch, Cambridge, UK	Cat# 109-116-170; RRID: AB_2337675
Goat anti-human IgM-PE detection antibody	Southern Biotech, Birmingham, AL, USA	Cat# 2022-09; RRID: AB_2795606
Biological samples
Human nasopharyngeal swab samples (healthy adults)	National Institute for Public Health and The Environment (RIVM), Radboudumc	Copan Minitip Size Flocked Swab, Cat# 3C057N
Human mid-turbinate swab samples (healthy adults)	National Institute for Public Health and The Environment (RIVM), Radboudumc, at their home	Copan Regular Size Flocked Swab, Cat# 3C064N
Human nasosorption strip (i.e., mucosal lining fluid) samples (healthy adults)	National Institute for Public Health and The Environment (RIVM), Radboudumc, at their home	Mucosal Diagnostics FX·I device, 7.00 mm Cat# NSFL-FXI-11
Human oropharyngeal swab samples (healthy adults)	National Institute for Public Health and The Environment (RIVM), Radboudumc	Copan Regular Size Flocked Swab, Cat# 3C064N
Human saliva samples (healthy adults)	National Institute for Public Health and The Environment (RIVM), Radboudumc, at their home	ORACOL®, Malvern Medical Developments Ltd, Cat# S10
Chemicals, peptides, and recombinant proteins
Recombinant RSV prefusion F (PreF) protein	Sino Biological, Beijing, China	Cat# 11049-VNAS
SARS-CoV-2 monomeric spike S1	Sino Biological, Beijing, China	Cat# 40591-V08H
Pertussis toxin (PT)	The Native Antigen Company, Oxford, UK	Cat# PT-TNF-100
Critical commercial assays
LEGENDplex Custom Human Panel 1717	BioLegend, San Diego, CA	Cat# 900007151
Olink Target 96 Inflammation	Olink Proteomics, Uppsala, Sweden	v.3024
RespiFinder 2Smart Kit	PathoFinder B.V., Maastricht, the Netherlands	Cat# PF2600-2S
Magplex Microsphere beads	Luminex, Austin, TX, USA	Cat# MC12060-01, Cat# MC10027-01, Cat# MC10009-01
Fluorescent carboxylated microspheres	Bio-Rad Laboratories, Hercules, CA, USA	Bio-Plex® COOH Microspheres
Fastprep-24	MP Biomedicals, Irvine, USA	Cat# 116004500
DNeasy PowerSoil Pro Kit	QIAGEN, Venlo, the Netherlands	Cat# 47014
Maxima SYBR Green/ROX qPCR Master Mix	Mix (Thermo Fisher Scientific, Waltham, MA, USA)	Cat# K0222
Phusion Hot Start II High-Fidelity DNA Polymerase	Thermo Fisher Scientific, MA, USA	Cat# F-549L
QIAxcel DNA High-Resolution Kit	Qiagen, Hilden, Germany	Cat# 929002
Agencourt AMPure XP magnetic beads	Beckman Coulter, the Netherlands	Cat# A63880
KAPA library quantification kit	Roche, USA	Cat# KK4824
MiSeq Reagent kit V3 (600 cycles)	Illumina, CA, USA	Cat# MS-102-3003
Deposited data
Silva (version: 138.1)	Quast et al.42	Zenodo: https://zenodo.org/records/4587955
Code used for data processing and statistical analyses	This study	GitHub: https://github.com/NafsikaK/urt-multiomics-profiling
Processed datasets (phyloseq objects, cytokine/antibody/multiplex PCR data)	This study	Mendeley Data: https://doi.org/10.17632/96nnvxsr32.1
Raw 16S rRNA sequencing reads	This study	ENA: PRJEB121324
Oligonucleotides
Forward primer Eub341F	Muyzer et al.43	5′- CCTACGGGAGGCAGCAG-3′
Reverse primer Eub534R	Muyzer et al.43	5′- ATTACCGCGGCTGCTGGC-3′
515F 16S V4 forward primer	Caporaso et al.44 Thompson et al.45 Kozich et al.46	5′-GTG CCA GCM GCC GCG GTA A-3′ (including Illumina adapters and barcodes)
806R 16S V4 reverse primer	Caporaso et al.44 Thompson et al.45 Kozich et al.46	5′-GGA CTA CHV GGG TWT CTA AT-3′ (including Illumina adapters and barcodes)
Software and algorithms
R Statistical Software (version: 4.4.1)	N/A	https://www.r-project.org
RStudio Server (version: 2024.09.0.375)	RStudio	https://posit.co
Biorender (2024)	N/A	https://www.biorender.com/
FlowJo (version: 10.10.1)	BD Biosciences, Franklin Lakes, NJ, USA	RRID: SCR_008520
Bio-Plex Manager (version 6.2.0.175)	Bio-Rad Laboratories, Hercules, CA, USA	https://www.bio-rad.com/en-nl/product/bio-plex-manager-software-standard-edition?ID=5846e84e-03a7-4599-a8ae-7ba5dd2c7684
DADA2 (version: 1.26)	N/A	https://benjjneb.github.io/dada2/
microbiome (version: 1.19.1)	N/A	https://github.com/microbiome/microbiome
phyloseq (version: 1.46.0)	McMurdie and Holmes47	https://joey711.github.io/phyloseq/
vegan (version: 2.7.5)	N/A	https://cran.r-project.org/web/packages/vegan/index.html
pairwiseAdonis (version: 0.4.1)	N/A	https://github.com/pmartinezarbizu/pairwiseAdonis
microbiomeutilities (version: 1.0.17)	N/A	https://github.com/microsud/microbiomeutilities/
decontam (version: 1.26.0)	N/A	https://github.com/benjjneb/decontam
ggplot2 (version: 4.0.0)	N/A	https://ggplot2.tidyverse.org/
nlme (version: 3.1.168)	N/A	https://cran.r-project.org/web/packages/nlme/index.html
factoextra (version: 2.0.0)	N/A	https://cran.r-project.org/web/packages/factoextra/index.html
coin (version: 1.4.3)	Hothorn et al.48	https://cran.r-project.org/web/packages/coin/index.html
MOFA2 (version: 1.18.0)	Argelaguet et al.49	https://biofam.github.io/MOFA2/
Bacterial and virus strains
ZymoBIOMICS Microbial Community DNA Standard	Zymo Research, CA, USA	Cat# D6305
ZymoBIOMICS Microbial Community Standard	Zymo Research, CA, USA	Cat# D6300
ZymoBIOMICS Gut Microbiome Standard	Zymo Research, CA, USA	Cat# D6331

Experimental model and study participant details

We conducted a cross-sectional study in the Netherlands (October 2023–February 2024) with 44 healthy adults (20–65 years) to investigate biological signatures across five URT niches and compare sampling STAR Methods (Table 1). Biological females (28/44) outnumbered biological males (16/44) in this study. We did not collect data on gender. Participants were required to be free of respiratory symptoms at the time of sample collection and provided written informed consent prior to sampling. Samples from healthy controls at the Radboud university medical center were collected under the PID study, which received ethical approval from the Medical Ethics Committee of Erasmus MC, University Medical Center Rotterdam (CCMO registration number: NL40331.078.12). Sample size was determined a priori based on the primary within-subject paired comparisons across URT sample types, assuming a conservative effect size of 0.4 for paired immune mediator differences, 80% power, and a two-sided α of 0.05, 52 participants with paired samples were estimated to be required. With 44 participants, the study remained adequately powered to detect moderate-to-large effects, in line with the upper range observed in prior data.

Two sampling strategies were employed during sample collection; professional and self-collection. All participants (N = 44) underwent professional collection, which was performed by professionally trained personnel. Professionally collected samples included nasopharyngeal swabs (NPS) (Copan, Murrieta, CA, USA) and oropharyngeal swabs (OPS) (Copan), placed in 1 mL Universal Transport Medium (UTM). Non-invasive samples included mid-turbinate nasal swabs (MTS, in 1 mL UTM) (Copan), nasosorption strips (collecting nasal mucosal lining fluid; MLF, eluted in 300 μL PBS/BSA 1%) (Mucosal Diagnostics FX·I device, 7.00 mm, Midhurst, UK), and unstimulated saliva (Oracol system, Malvern Medical Developments Ltd, Worcester, UK) (Table 2). To prevent cross-contamination, samples were collected in a specific order from the participants (Figure 1A). Personnel were equipped with FFP2 medical masks, gloves, and hair caps to prevent contamination. For MTS and MLF, samples were obtained from both the left and right nostrils. All samples were transported on ice, aliquoted, and stored at −80°C. Participants completed a short questionnaire at the end of the sample collection to assess their perception of each sample type, with questions rated on a five-point Likert scale (Data S1). This took place only at the professional sample collection. A subset of participants (N = 16) additionally self-collected MTS, MLF, and saliva at home later the same day. Participants were provided with detailed instructions and the necessary materials to perform the self-collection. The participants were instructed to store all samples in their home freezer and deliver them to the laboratory within one week of the sample collection. Upon arrival at the laboratory, the samples were stored at -80°C for further use.

Professional collection

Sample collection was performed by trained personnel under aseptic conditions. Before each session, staff wore protective equipment (FFP2 mask, head cap, lab coat, and gloves), washed and disinfected their hands, and disinfected the workstation. Gloves were disinfected between each sample collection. Participants were asked to refrain from eating and drinking for 30 minutes prior to the sample collection. For nasosorption strip (MLF) sampling, the flat end of the strip was inserted into the participant’s nostril until the plastic thickening of the strip and the participant was asked to pressed lightly against the mucosal lining for 60 seconds. Then, the strip was returned to its tube. For mid-turbinate nasal swabs (MTS), the swab was inserted ∼2 cm into the nostril, rotated 4–5 times in the nostril and placed into the UTM tube. The swab was rotated for 30 seconds in the UTM tube before being broken at the indicated breakpoint, and sealed. For saliva collection, two Oracol swabs were placed against the participant’s cheeks for 2 minutes before being returned to the collection tube. For the oropharyngeal swab (OPS), the participant’s head was tilted back, the tongue gently depressed with a tongue depressor, and the swab was rubbed against the posterior pharyngeal wall (avoiding the tongue and cheeks). Subsequently, the swab was transferred to the UTM tube, rotated for 30 seconds, broken at the breakpoint, and the tube sealed. For the nasopharyngeal swab (NPS), the swab was advanced to the nasopharynx, rotated while counting to five, and processed into the UTM tube as above. All samples were placed in a biohazard bag, sealed, and stored in a cooler with ice packs. A short questionnaire was completed by each participant.

Self-collection at home

Participants were instructed to disinfect their work surface, wash or disinfect their hands, and arrange the collection materials before beginning. Additionally, participants were instructed to refrain from eating and drinking for 30 minutes prior to the sample collection. For the mid-turbinate nasal swab (MTS), the swab was inserted ∼2 cm into one nostril, rotated 4–5 times, then placed in the UTM tube, rotated for 30 seconds, broken at the breakpoint, and sealed. For saliva collection, two Oracol swabs were placed against the inner cheeks for 2 minutes before being returned to their tube. For the nasosorption strip (MLF), the strip was inserted into one nostril up to the stopper, the nostril was gently closed with light pressure for 60 seconds, and the strip was then returned to its tube. All self-collected samples were placed in a biohazard bag, sealed, and stored in the home freezer until transport to the study site with ice packs.

Method details

Laboratory assays

Immune marker measurements

Immune marker profiles were determined in all available samples using a Custom LegendPlex Kit (BioLegend, San Diego, CA, USA), featuring 13 markers involved in neutrophil-related responses, cytotoxic antiviral responses or broader inflammatory pathways as previously described.³¹ MTS (N = 44) and NPS samples (N = 26) were additionally analyzed with Olink’s 96 Inflammation Panel (Olink Proteomics, Sweden, Uppsala). Olink data were incorporated into the microbe-immune association analyses for MTS. Concordance between overlapping analytes across platforms was assessed by Spearman correlation (Figure S10).

Antibody measurements

Total IgA, IgG, and IgM concentrations were measured using monoplex Luminex immunoassays. For the total IgA, IgG and IgM assay, Magplex Microsphere beads (Luminex, Austin, TX, USA) are coupled to mouse anti-human IgA (BioLegend, San Diego, CA, USA), mouse anti-human IgG (BD Biosciences, Franklin Lakes, NJ, USA) and mouse anti-human IgM (Southern Biotech, Birmingham, AL, USA). The binding of antibodies from the samples was detected using goat anti-human IgA/IgG/IgM (Southern Biotech Birmingham, AL, USA/ Jackson Immunoresearch, Cambridge, UK/ Southern Biotech Birmingham, AL, USA) conjugated to R-phycoerythrin (PE), respectively. The samples were acquired using the FlexMap3D (Luminex). Four samples did not have sufficient volume to measure total IgM (2 NPS, 1 Saliva and 1 OPS). IgG specific to prefusion F (PreF) of RSV, Spike S1 subunit of SARS-CoV-2 and Pertussis toxin (PT) of Bordetella pertussis.^50,51,52 were also measured using Luminex multiplex immunoassays.

Nucleic acid extraction and analysis

Nucleic acids were extracted from all samples using QIAGEN’s PowerSoil Pro Kit (Venlo, the Netherlands), with a maximum of 250 μL of MLF and saliva. For NPS, OPS and MTS samples, a total of 500 μL of sample was used to increase the nucleic acid yield. One aliquot of the nucleic acid extract was used to perform a real-time PCR (PCR)–based multiplex ligation-dependent probe amplification assay on all samples available with the RespiFinder 2Smart Kit (PathoFinder B.V., Maastricht, the Netherlands). This assay simultaneously detects 24 respiratory pathogens, including 20 viral targets (influenza A, influenza B, influenza A(H1N1)pdm09, RSV-A, RSV-B, SARS-CoV-2, rhinovirus/enterovirus, bocavirus, adenovirus, human metapneumovirus, parainfluenza 1–4, coronavirus 229E/NL63/OC43/HKU1, MERS CoV) and four bacterial targets (Bordetella pertussis, Chlamydophila pneumoniae, Mycoplasma pneumoniae, Legionella pneumophila). The remaining aliquot was used for 16S rRNA gene amplicon sequencing of the V4 hypervariable region for analysis of the microbiome.

Quantification of bacterial DNA by quantitative PCR

DNA concentrations were measured using a Quantus Fluorometer (Promega, Madison, USA), and samples were stored at −20°C until further processing. Bacterial load in the purified samples was quantified by real-time PCR on a StepOnePlus Real-Time PCR System (Thermo Fisher Scientific, the Netherlands) using a universal 16S rRNA gene primer pair (Eub341F: CCTACGGGAGGCAGCAG and Eub534R: ATTACCGCGGCTGCTGGC).⁴³ Reactions were performed with SYBR Green chemistry in a total volume of 25 μL, comprising 12.5 μL of Maxima SYBR Green/ROX qPCR Master Mix (Thermo Fisher Scientific, Waltham, MA, USA), 0.5 μM of each primer, 2 μL of DNA (diluted 1:500 in HPLC-grade water), and 8 μL of HPLC-grade water. The PCR cycling conditions included an initial denaturation at 95°C for 10 minutes, followed by 40 cycles of denaturation at 95°C for 15 seconds, annealing at 60°C for 15 seconds, extension at 72°C for 15 seconds, and a final holding stage at 95°C for 1 minute and 60°C for 1 minute.

Library preparation and sequencing of the V4 region of the 16S rRNA gene

Quantitative PCR results were used to normalize bacterial load across samples, enabling input standardization to 500 pg of DNA for amplification of the 16S rRNA gene’s hypervariable V4 region. Amplification was performed using primers 515F (5′-GTG CCA GCM GCC GCG GTA A-3′) and 806R (5′-GGA CTA CHV GGG TWT CTA AT-3′), each containing Illumina flow cell adapters and unique 8-nt index sequences.^44,45,46 PCR reactions were prepared with 0.5 μL (1 U) Phusion Hot Start II High-Fidelity DNA Polymerase, 5 μL of 5× Phusion HF Buffer, 7 μL of HPLC-grade water, 2.5 μL of 2 mM dNTP mix (Thermo Fisher Scientific, Waltham, MA, USA), 0.5 μM of each primer, and 5 μL of template DNA, in a total volume of 20 μL. Thermocycling conditions included initial denaturation at 98°C for 30 seconds, followed by 30 cycles of 98°C for 10 seconds, 55°C for 30 seconds, and 72°C for 30 seconds, with a final extension at 72°C for 30 seconds. Amplicon size and concentration were checked using the QIAxcel DNA High-Resolution Kit on the QIAxcel Advanced System (Qiagen, Hilden, Germany), and equimolar pooling of samples was performed. The pooled library was purified twice with 0.9× AMPure XP magnetic beads (Beckman Coulter, the Netherlands) and quantified using the KAPA Library Quantification Kit (Roche, USA). Paired-end sequencing (2 × 300 bp) was conducted on an Illumina MiSeq platform using the V3 MiSeq reagent kit (600 cycles) (Illumina Inc., San Diego, CA, USA).

Quantification and statistical analysis

All data and statistical analyses were performed using R (version 4.4.1) within the RStudio Server v2024.09.0.375.

Questionnaire analysis

Responses were analyzed using Friedman’s test and Wilcoxon signed-rank post-hoc tests on five-point Likert scales from all 44 participants.

Immune marker and antibody measurement data analysis

Complete sets of left and right nostril MTS (professional; N = 88, self; N = 16), NPS (N = 44), left and right nostril MLF (professional; N = 88, self; N = 16), OPS (N = 44) and saliva (professional N = 40; self N = 16) were analyzed with LegendPlex; four saliva samples lacked sufficient volume to perform this assay. Concentrations below the lower limit of detection (LLOD) were assigned a value of half the LLOD while for the total antibody measurements, antibody concentrations above the upper limit of quantitation were assigned a value of twice the maximum per sample type and sampling strategy. Wilcoxon signed-rank tests for paired data were used to compare total antibody concentrations within sample types, the left to right nostrils’ concentration, professionally and self-collected samples and sample differences in RSV PreF-specific and SARS-CoV-2 S1-specific IgG concentrations but not for B. pertussis PT since most samples were below LLOD. Because no difference was observed between nostrils, their mean concentration was used for downstream analysis. Pairwise Spearman’s rank correlation coefficients (ρ) with false discovery rate (FDR) < 0.05 were used to assess relationships between immune markers within and across sample types (N = 36).⁵³ Correlations were performed using the cor.test function in R. For concentration comparisons across professionally collected sample types, NPS served as the reference. Log-transformed immune data were analyzed with a log-linear mixed-effects model using restricted maximum likelihood with the lme function from the ‘nlme’ R package. This included fixed effects for the interaction between sample type and analyte and a random intercept for participant ID to account for repeated measures. Geometric mean ratios (GMRs) and their 95% confidence intervals (CIs) were calculated by exponentiating the model’s fixed effect coefficients, representing population-level predictions based on fixed effects. The R package ‘ggplot2’ was used to visualize the results and ‘factoextra’ was used to perform and visualize Principal Component Analysis (PCA) of scaled and centered cytokine data. To test whether sample types differed in the multivariate space, we performed a PERMANOVA (adonis2 function, ‘vegan’ package) on a Euclidean distance matrix of the scaled cytokine concentrations.

16S rRNA gene sequencing data analysis

16S rRNA gene sequence data analyzed using the DADA2 pipeline as previously described.⁵⁴ Default parameters were maintained, with forward and reverse reads trimmed at 220 nt and 100 nt, respectively, and truncated at a quality score ≤ 5. Error rates were modeled based on ≥1×10⁹ total bases. Chimeric sequences were removed from the resulting amplicon sequence variants (ASVs), and taxonomy was assigned using the RDP classifier with the SILVA database (v138.1).^42,55 ASV counts, taxonomic assignments and sample information were merged into R as a phyloseq object for further analyses (‘phyloseq’ v1.46).⁴⁷ The phyloseq object from DADA2 was split into nasal (MTS, MLF, NPS) and oral (Saliva, OPS) subsets due to expected compositional differences.¹² Before filtering, the nasal dataset comprised 24,661,642 reads across 9,942 taxa, and the oral dataset 14,163,846 reads across 3,371 taxa. Samples with fewer than 6,342 reads were excluded (1/56 saliva, 11/101 MLF, 3/100 MTS, 4/42 NPS). Blank extraction controls were included throughout DNA extraction and library preparation to monitor potential background contamination associated with low-biomass respiratory samples. Blank controls generally yielded substantially lower read counts than biological samples, although a limited number contained detectable reads. Contaminant ASVs were identified separately per niche using the `decontam` package (v1.26.0); isContaminant (threshold = 0.5) against blank controls, removing 288 ASVs (1.78% of reads) from the nasal set and 66 ASVs (0.47% of reads) from the oral set. In addition, we removed taxa belonging to the phyla Cyanobacteria, Chloroflexi or Mitochondria, and those contributing < 0.15% of total abundance. We additionally removed potential contaminating taxa not expected in URT samples.

Alpha diversity (Shannon, Simpson) was used to assess microbial richness and evenness and beta diversity to compare community composition. Pairwise comparisons were done using Wilcoxon signed-rank or paired t-tests, depending on whether the data followed a normal distribution or not, with FDR correction (significant at FDR < 0.05). Bray–Curtis dissimilarity was calculated on transformed compositional data (‘microbiome’ package). Distance matrices were generated using ‘phyloseq’, and pairwise PERMANOVA was performed with pairwiseAdonis. Ordination was visualized via principal coordinates analysis (PCoA) with 95% confidence ellipses. Although all 44 participants provided all five sample types (NPS, MTS, MLF, OPS, and saliva), some samples did not pass sequencing quality control. For analyses requiring comparisons across all five sample types, we therefore restricted to participants with sequencing data available for every type (N = 22), ensuring fully paired within-subject comparisons. For pairwise comparisons between sample types, larger subsets of participants were available and used accordingly. For comparisons between left and right nostril samples (MLF; 33 left and 38 right, MTS; 40 each), we randomly subsampled the larger group nostril samples without replacement to match the number of samples in the smaller group, and retained only these balanced sets for diversity analyses. For diversity comparisons across sample types, when both left and right nostril samples were available for MTS and MLF, we randomly selected one nostril per participant so that each individual contributed only one MTS and one MLF sample.

Associations between microbes and immune markers

Associations between the relative abundance of each ASV and immune markers were tested using non-parametric STAR Methods implemented in the ‘coin’ R package per sample type/niche, as previously described.^48,56 Samples that were positive for one or more respiratory pathogens (assessed using the RespiFinder assay) were excluded, resulting in 39 MLF, 32 MTS, 29 NPS, 41 OPS and 34 saliva samples available for testing. For MTS, immune marker levels obtained from the Olink platform were additionally included. All tests were stratified by sex, and p-values were estimated by permutation. Given the exploratory nature and large number of microbe-immune comparisons, multiple testing correction was applied using the FDR, with associations considered significant at FDR < 0.20 to balance false positives and false negatives. To confirm robustness of the findings, analyses were repeated with any positive RespiFinder PCR result included as an additional stratification variable alongside sex (Figure S9).

Multi-omics factor analysis

To integrate microbiome, cytokine, and total antibody datasets, we applied MOFA (R package ‘MOFA2’).⁴⁹ MOFA identifies a set of latent factors that capture shared and domain-specific patterns across data types. Each factor is a weighted combination of features (e.g., genera, cytokines, total antibodies) and samples receive factor scores that summarize how strongly they express that pattern. This approach reduces dimensionality, highlights coordinated variation across domains, and enables interpretation of which features drive differences between sample types. Only professionally collected samples were used (40 MLF, 40 MTS, 36 NPS, 44 OPS and 35 saliva). For the microbiome, amplicon sequence variants were aggregated at the genus level, filtered to retain genera present in >10% of samples, center log-ratio (CLR)-transformed and subsequently z-scored. Cytokine (LegendPlex) and total antibody concentrations were log10-transformed and then z-scored. For each data domain, replicate nostril samples were collapsed to the participant–site level by averaging. MOFA models were trained using default training settings. The resulting latent factors were interpreted by inspecting variance explained per domain, factor associations with metadata, sample clustering in factor space, and the highest-loading features within each domain. Redundancy analysis (RDA) and distance-based RDA (dbRDA) were additionally performed to confirm MOFA findings.

Confirmatory integration by constrained redundancy analyses (RDA/dbRDA)

Redundancy analysis (RDA) was performed on a concatenated feature matrix including microbiome (genus-level, Hellinger-transformed and z-scored), cytokines (log10-transformed and z-scored), and antibodies (log10-transformed and z-scored), constrained by Sample Type while controlling for subject ID, sex, age range, and for PCR positive results. In parallel, distance-based RDA (dbRDA) was run on a Bray–Curtis dissimilarity matrix of the microbiome constrained by Sample Type and immune features, with the same covariates conditioned. Prior to concatenation, each data block (microbiome, cytokines, total antibodies) was variance-scaled to unit total variance, such that all modalities contributed equally to the analysis. Significance of models and terms was assessed using 999 permutations with subject ID as a blocking factor using the function anova.cca from the ‘vegan’ R package. Both RDA and dbRDA confirmed that anatomical compartment was the dominant source of variation in the joint dataset, consistent with the MOFA analysis. The top microbial and immune features contributing to the constrained ordinations overlapped with those identified by MOFA (e.g., CXCL8, IFN-γ, Corynebacterium, Streptococcus), supporting the robustness of these findings.