Human tonsils were obtained from children aged between two and five years who underwent tonsillectomies between 2014 to 2017 at the Otorhinolaryngological department (ORLD) Karolinska University Hospital Huddinge, Sweden. Written informed consent was obtained from the parents/guardians of the patients. Sample collection was approved by the Swedish Ethical Review Authority (Dnr 2014/1000-31/1 NCT02315911).

Tonsil hypertrophy was scored according to Brodsky grading ( Table 1 ) (27). The group with tonsil hyperplasia (n=13) had significantly enlarged tonsils, size 4, in combination with very severe OSA, defined by an obstructive apnea–hypopnea index (OAHI) ≥ 30. Children with small tonsils (n=6) had tonsil size 2, with the exception of one patient (tonsil size 2.5). All six patients had an OAHI of 11 or less, which was termed “moderate” OSA. The exclusion criteria were the presence of craniofacial abnormality, neuromuscular disease, chromosomal abnormality, previous adenotonsillar surgery, bleeding disorder and cardiopulmonary disease. All 19 children had undergone polysomnography (PSG) overnight in a sleep laboratory at the ORLD to diagnose their degree of OSA before surgery. PSG measures sleep stages and respiratory functions using recordings of electroencephalogram, electro-oculogram, electromyogram, pulse, oronasal airflow, transcutaneous oxygen saturation, respiratory movements (abdomen and thorax), body position, and video and sound recordings. All PSGs were with the EMBLA technology (Flaga Medical, Iceland), and scored manually by the same registered polysomnography technologist according to the scoring rules of the American Academy of Sleep Medicine. The OAHI was calculated from the PSG scoring. There were no differences in age, sex distribution, weight or height between the groups. The children were all normal- or underweighted except for one overweight child, who also had large tonsils. No child was obese. Details of the patient cohort are provided in Table 1 .

Tonsils were cut into small pieces, ground through a 100μm cell strainer using a plunger of a plastic syringe and washed with PBS. Cell suspension was collected, spun down and resulting pellet resuspended in 35ml PBS. The obtained cell suspension was loaded on top of 15ml lymphoprep (Fisher Scientific) and mononuclear cells were isolated using ficoll gradient density centrifugation. Subsequently, mononuclear cells were frozen in FCS + 10% DMSO (Sigma Aldrich) freezing medium and stored in the gas phase of liquid nitrogen.

Frozen mononuclear cells were thawed and washed twice with IMDM + 10% FCS. Cells were stained with fixable viability dye (ThermoFisher) and antibodies targeting surface molecules at room temperature in FACS buffer (PBS + 2% FCS + 2mM EDTA) for 30 minutes whereafter cells were fixed for 10 minutes with 2% paraformaldehyde. Ki67 and T-bet were stained using eBioscience™ FoxP3/Transcription Factor Staining Buffer Set (Thermo Fisher Scientific) according to manufacturer’s instructions. A full list of antibodies used and staining panels can be found in Supplementary Tables 1, 4, 9 . Stained samples were acquired on BD LSR Fortessa™ equipped with 355-, 405-, 488-, 561-, and 639-nm lasers.

FCS3.0 files were analyzed using FlowJo v. 10.6.1 (BD Biosciences). Automated compensation was calculated using single-stained compensation beads (BD Biosciences) on BD FACSDiva software and manually adjusted on FlowJo by looking at pairwise expression of all parameters. Automated cluster analysis and visualization of concatenated cell populations was performed using FlowJo plugins. Cell numbers of the populations of interest were reduced using DownSample (v. 3.3) function and samples belonging to the same patient group were concatenated using all compensated parameters. The cell numbers were adjusted to be equal between the patients within each group and comparable between the two patient groups (small vs. large tonsils) in the final concatenated file. The final cell number was 7 200 and 7 205 cells for ILCs and 18 000 and 18 005 cells for T cells and B cells for small and large tonsil group, respectively. UMAP (v. 2.1) was used for dimensionality reduction and PhenoGraph (v. 1.8) was used to investigate the population heterogeneity within each group and the results were visualized using ClusterExplorer (v. 1.1.6). Default settings were used for UMAP (Euclidean distance function, nearest neighbors = 15 and minimum distance = 0.5) and PhenoGraph (K = 30) algorithms. Parameters specific to the cell type of interest were selected to run UMAP and PhenoGraph and are specified in the Supplementary Tables 1, 4, 9 .

Statistical analysis was performed using Prism software v. 6 (GraphPad). Statistical difference between data sets was assessed using Mann-Whitney U test. Two-tailed P-values < 0.05 were considered significant.

To address if immunological changes in tonsillar tissue correlate with the tonsil size in pediatric patients, we selected a cohort of 19 children with small tonsils and moderate OSA (n = 6) and with large tonsils and very severe OSA (n = 13) ( Table 1 ), hereafter referred to as small and large tonsils. Three flow cytometric panels of surface and intracellular proteins were designed to identify the main ILC, T-cell and B-cell populations ( Supplementary Tables 1, 4, 9 ). To reduce dimensionality and avoid bias related to manual gating, The FlowJo Plugins UMAP (25) and PhenoGraph (26) were applied to identify and project distinct lymphocyte clusters. First, major cell subsets of interest (Lin^-CD127⁺ ILCs, CD3⁺ T cells, or CD3^-CD19⁺CD20⁺ B cells) were manually gated on the basis of viable cells in all FCS files. Subsequently, viable cell populations from samples belonging to the same patient group were down-sampled and concatenated. Cell numbers were adjusted, so that individual patient samples were equally represented in each group and the final cell number in the concatenated files was comparable between both tonsil size groups according to the workflow presented in Figure 1A . UMAP and PhenoGraph algorithms were run using all compensated parameters specific to the cell type of interest ( Supplementary Tables 1, 4, 9 ). We then analyzed groups of cells projected on the UMAP, followed by PhenoGraph analysis and the projection of the resulting clusters onto the UMAPs ( Figure 1B ; workflow). Based on the defining marker expression and pre-existing knowledge, clusters were then grouped into distinct functional groups. This allowed for a direct unbiased comparison of lymphocyte populations between the two tonsil size groups. The MFI values for each marker and cluster can be found in Supplementary Data 1 .

We first set out to investigate the heterogeneity of ILC populations in the tonsils. For this, we designed a panel of surface markers to identify the main ILC lineages, their maturation/activation status as well as tissue residency phenotype ( Supplementary Table 1 ). ILCs were readily detectable across all samples as Lin^-CD3^-CD127⁺ lymphocytes ( Supplementary Figure 1A ) as previously reported (19).

For both small and large tonsils, UMAP projected the CD127⁺ ILCs in three major groups based on their lineage characteristics: CRTH2⁺ ILC2, CD117⁺ ILC3 including LTi-like cells, and CD161^- ILCs ( Supplementary Figures 1B, C ). Using PhenoGraph analysis, 11 clusters of CD127⁺ ILCs were detected in the small tonsils ( Figures 2A–D ; Supplementary Table 2 ), and 12 in the large tonsils ( Figures 2E–H and Supplementary Table 3 ). In both disease groups, CRTH2⁺ ILC2 were allocated to one cluster, hence making up a highly homogenous cell population, while CD117⁺ ILC and CD161^- ILC groups displayed additional heterogeneity in terms of the markers analyzed ( Figures 2C, G ). In line with previous observations (28, 29), tonsillar ILC2 showed largely uniform expression of CRTH2, CD161, CD25 and KLRG1 ( Figures 2A, E and Supplementary Tables 2, 3 ). Of note, most ILC2 in both small and large tonsils did not express CD69, suggesting that these are circulating and/or less activated cells ( Figures 2A, E and Supplementary Tables 2, 3 ).

Using single-cell RNA sequencing (scRNA-seq), Björklund and Forkel et al. have previously described transcriptionally distinct sub-populations of ILC3 in human tonsils, defined by their expression of CD62L and NKp44 (19). CD62L⁺ cells co-expressed CD45RA and delineated a naïve-like subset of ILC3, while NKp44 expression characterized mature, IL-22 producing ILC3. Indeed, the marker distribution pattern of the CD117⁺ ILC group ( Supplementary Figures 1B, C ) demonstrated mutually exclusive expression of CD45RA and NKp44 in both disease groups. PhenoGraph distinguished seven clusters of CD117⁺ ILCs in small and large tonsils alike. Based on existing knowledge (19, 22, 24), the CD117⁺ ILC clusters were combined into three major functional groups: NRP1⁺ LTi-like cells, CD45RA⁺ naïve ILC3-like and NKp44⁺ mature ILC3 ( Figures 2A–H and Supplementary Tables 2, 3 ). Additionally, a separate cluster of CD45RA^-NKp44^- ILC3 was identified in the large-tonsil group, likely representing a transient stage between naïve and mature ILC3 ( Figures 2F, H and Supplementary Table 3 ). While naïve CD45RA⁺ and transient CD45RA^-NKp44^- ILC3 clusters contained equal proportions of CD69^- and CD69⁺ cells, mature NKp44⁺ ILC3 displayed predominantly a tissue-resident or activated CD69⁺ phenotype ( Supplementary Tables 2, 3 ).

PhenoGraph analysis detected several clusters of LTi-like cells ( Figures 2A–H and Supplementary Tables 2, 3 ). The detected heterogeneity was attributed to variable expression of CD56 and NKp44 on the individual LTi-like-cell populations ( Figures 2A, E and Supplementary Tables 2, 3 ). Although some CD69^- cells were present in the LTi-like cell clusters ( Supplementary Tables 2, 3 ), most of the LTi-like cells expressed CD69, highlighting their tissue-resident or activated phenotype ( Figures 2A, E and Supplementary Tables 2, 3 ).

Within the CD161^- ILC group, PhenoGraph identified three separate cell clusters in the small tonsils and four in large tonsils ( Figures 2A–H and Supplementary Tables 2, 3 ). Besides the CD161^-CD117⁺ ILC3-like cells and CD161^-CRTH2^-CD117^- ILC1-like, CD4-expressing CRTH2^-CD117^- ILC1-like cells were identified in both tonsil groups, although this cluster was very small in small tonsils ( Figures 2A–H and Supplementary Tables 2, 3 ). Roan et al. have previously described a distinct population of CD4⁺ ILC1 that was increased in the peripheral circulation of the individuals with systemic sclerosis (30). That population was characterized by the absence of surface TCR and CD3ϵ expression, but showed intracellular expression of CD3ϵ. Tonsillar CD4⁺ CRTH2^-CD117^- ILC1-like cells were largely CD3^-, but contained a subset of CD3^lo cells. This indicates that the CD161^-CD4⁺ ILC cluster might contain a mixture of CD4⁺ ILC1 and highly activated CD4⁺ T cells that have downregulated surface TCR and CD3.

Finally we evaluated the differences in ILC composition between small and large tonsils. Significantly lower percentage of CD127⁺ ILCs out of CD45⁺ cells, as well as among CD3^- lymphocytes, was detected in large as compared to small tonsils ( Figure 2I ). While we observed similar frequencies of ILC2 and ILC1-like cells, a lower frequency of LTi-like cells was found in the concatenated data set from large tonsils ( Supplementary Tables 2, 3 ). The frequency of NRP1⁺ ILC3 out of CD45⁺ cells was indeed significantly lower and tended to be lower as a proportion of total ILC3, in large versus small tonsils ( Figure 2J ).

In summary, we here provide an overview of ILC heterogeneity in human tonsils and identify skewed ILC3 composition in enlarged tonsils.

Next, we sought out to examine the composition of adaptive lymphocytes in the tonsillar tissues. A panel of surface markers was selected to dissect B- and T-cell populations and address their activation, maturation and functional status ( Supplementary Table 4 ). T cells were manually gated as live CD45⁺CD3⁺ lymphocytes ( Supplementary Figure 2A ). As small and large tonsils showed a comparable composition of CD3⁺ T cells and the main T cell lineages including T_FH cells ( Figures 3A–C and Supplementary Figure 2A ) we proceeded to assess the heterogeneity of T cells in the two tonsil groups ( Figures 3D–K ).

UMAP projected the T cells in five main groups in both small and large tonsils: CD45RA⁺ and CD45RA^- CD4⁺, CD4⁺CD8⁺ double positive (DP), CD4^-CD8^- double negative (DN) and CD8⁺ T cells ( Supplementary Figure 2 ). Interestingly, DP and DN T cells were located closer to CD4⁺ T cells than to CD8⁺ T cells in both patient groups, with DP T cells projected close to CD45RA^- and DN to CD45RA⁺ CD4⁺ T cells. CD8⁺ T cells were projected furthest from other T-cell subsets, with CD45RA^- and CD45RA⁺ CD8⁺ cells located more closely to each other as compared to CD45RA^- and CD45RA⁺ CD4⁺ cells. This may be attributed to the inclusion of T_FH-centric surface markers during the flow-cytometric panel design, as this would allow a more in-depth analysis of CD4⁺ T-cell populations.

PhenoGraph algorithm identified 13 and 15 clusters of T cells in small and large tonsils, respectively ( Figures 3D–K and Supplementary Tables 5, 6 ). Based on the surface marker expression ( Figures 3D, H and Supplementary Tables 5, 6 ), the clusters identified were categorized into 7 major functional groups: naïve, memory and T_FH CD4⁺ cells, naïve and effector CD8⁺ cells, as well as DN and DP T cells ( Figures 3E–G, I–K ). Identified clusters were superimposed onto the UMAP described above ( Figures 3F, G, J, K ). In both small and large tonsils, the CD45RA⁺CD4⁺ T-cell group consisted of a homogenous naïve cell cluster that was defined by the expression of CD45RA and CCR7 ( Figures 3D, F, G, H, J, K and Supplementary Tables 5, 6 ).

The CD45RA^-CD4⁺ T-cell groups contained 7 and 8 cell clusters in small and large tonsils, respectively, representing functional groups of T_FH cells and memory helper T cells. The T_FH-cell clusters showed variable expression of the defining markers ICOS, PD-1 and CXCR5 and were largely CCR7^-. Moreover, the distinct T_FH clusters differed in their expression of IL-2 receptor α chain, CD25 ( Figures 3D, H and Supplementary Tables 5, 6 ). Interestingly, DP T cells displayed a phenotype highly similar to T_FH cells, characterized by CXCR5, PD1 and ICOS expression, but without increased size, thus unlikely representing cell doublets ( Figures 3D, H and Supplementary Tables 5, 6 ). Three and five non-T_FH memory helper T-cell clusters were identified in the small and large tonsils, respectively. This difference was not due to an enrichment of memory CD4⁺ T cells in large tonsils, as no differences in the frequencies of CD45RA^-CD4⁺ T cells or T_FH cells were detected between the groups at the patient level ( Figure 3C and Supplementary Figure 2E ).

DN T cells contained two clusters in the small and large tonsils ( Figures 3D–K ). Both clusters expressed CD27, were CCR7^- and did not express any T_FH-related surface markers. These clusters differed from each other in their expression of CD38 and CD45RA ( Figures 3D–K ), as well as PD1 in the large tonsils ( Figures 3H–K ).

Within the CD8⁺ T-cell group, two clusters were identified in small tonsils and three in large tonsils ( Figures 3D–K ). While naïve CD8⁺ T cells represented a separate cluster in both groups, PhenoGraph identified CD45RA^- memory and CD45RA⁺CCR7^- TEMRA cells as one cluster in patients with small tonsils ( Figures 3D–K and Supplementary Tables 5, 6 ). These differences were not reflected in frequency changes of the described cell subsets between the two patient groups ( Supplementary Figure 2F ) and could simply be attributed to the uniformity of surface marker expression therein.

Taken together, automated clustering projection and analysis using UMAP and PhenoGraph, respectively, allowed unbiased identification of T-cell diversity in small and large tonsils from pediatric patients with OSA, which could not be readily identified using manual gating.

To assess the composition of tonsillar B cells in pediatric patients with OSA, B cells were manually gated based on their CD19 and CD20 expression ( Supplementary Figures 2A, 3A ). As previously reported (31), no CD20^-CD38^hi plasma cells were detected in tonsillar tissue, as all CD38^hiCD3^- cells displayed low CD20 expression ( Supplementary Figure 3B ). Patients with large tonsils demonstrated a tendency for higher frequency of total B cells than patients with small tonsils ( Figure 4A ). This difference was statistically significant, when only the CD3^- population was accounted for.

B-cells were projected in four main groups in UMAPs generated from small tonsils: CD38^hiCD20^lo cells (likely representing plasmablasts), CD38⁺CD20^hi germinal center (GC), CD27⁺IgM^- memory and IgM⁺ B cells ( Supplementary Figure 3C ). In large tonsils an additional cell group of CD27⁺IgM⁺ cells was detectable, that localized between the IgM⁺ and CD27⁺IgM^- cell clusters ( Supplementary Figure 3D ).

13 and 16 clusters of B cells were identified by PhenoGraph in small and large tonsils, respectively ( Figures 4B–I ). Based on the surface marker expression ( Figures 4B, F and Supplementary Tables 7, 8 ), B-cell clusters were categorized into five functional groups: plasmablasts, GC B cells, naïve and memory B cells, as well as IgM⁺CD27⁺ B cells ( Figures 4C, E, G, I ). In both, small and large tonsils, plasmablasts were represented by one homogenous cluster of cells, defined by their CD20^loCD27^hiCD38^hi phenotype ( Figures 4B–I and Supplementary Tables 7, 8 ). Additionally, plasmablasts did not express IgM and were largely CXCR5^-. The GC B-cell functional group contained three clusters that showed variance in the expression intensity of CD38, CD27, CXCR5, CD19 and/or CD20 in small as well as large tonsils ( Figures 4B–I and Supplementary Tables 7, 8 ). This phenotypic variability might reflect different maturation stages of GC B cells in the process of plasma-cell, or memory B-cell development. Specifically, the expression intensity of the chemokine receptor CXCR5 might be indicative of the migration status of the given cluster through B-cell follicles (32).

IgM^- memory B cells and IgM⁺CD27⁺ B cells, each represented by two clusters, showed high similarity between small and large tonsils ( Figures 4B–I and Supplementary Tables 7, 8 ). While we lack additional markers to fully dissect the IgM⁺CD27⁺ B cell clusters, they showed differential expression of CD38 suggesting that one of these clusters might be IgM⁺ GC B cells or marginal zone B cells (33) while the other may represent IgM⁺ memory B cells (34). In both small and large tonsils, the CD27^-IgM^- memory B-cell clusters displayed lower expression of CXCR5 than their CD27⁺ counterparts ( Figures 4B, F and Supplementary Tables 7, 8 ). When we compared the frequency of CD27⁺ B cells in small and large tonsils on the individual patient level, we detected a significantly higher percentage of these cells in small tonsils ( Figure 4J ; Supplementary Figure 3E ). The difference observed could be contributed to conventional memory B cells, as frequencies of IgM⁺CD27⁺ memory B cells were the same in both patient groups ( Figure 4J ).

PhenoGraph identified five naïve B-cell clusters in small tonsils vs. 8 clusters in large tonsils ( Figures 4B–I and Supplementary Tables 7, 8 ). In terms of population size, naïve B-cells were more prominently represented in the large tonsils than in the small ones ( Figures 4C, G ). Moreover, through manual gating we identified an enrichment of CD27^- B cells among total CD45⁺ cells in pediatric patients with large tonsils ( Figure 4K ). However, the frequencies of IgM⁺ cells within the CD27^- B-cell populations were the same between the two patient groups ( Supplementary Figure 3F ).

In summary, we identified enrichment of IgM⁺CD27^- naïve B-cell populations in large tonsils from pediatric patients with very severe OSA. This is consistent with the observed increase in frequency of total B cells and CD27^- B cells in this patient group.

A unique subset of CD27^- B cells harboring memory potential was described in chronic inflammation and suggested to be driven by infections as well as autoimmunity (35–38). These tissue-like memory or atypical memory B cells were characterized by the expression of Fc receptor-like (FcRL) proteins FcRL4 and/or FcRL5, CD11c and T-bet (35, 38–40). To investigate the atypical B cell populations in relation to our finding on increased CD27^- B cells in large tonsils, we developed a flow-cytometric panel including the defining markers as well as Ki67 ( Supplementary Table 9 ). Concatenated group-specific samples were generated from total B-cell populations ( Supplementary Figure 4A ). Based on the markers analyzed, two main B-cell groups were observed in UMAPs generated from large and small tonsils: CD38^hiCD20^lo plasmablasts and other B cells that were projected in close proximity to each other ( Supplementary Figures 4B, C ).

PhenoGraph detected 11 and 13 clusters in small and large tonsils, respectively ( Figures 5A–H ). These clusters could be attributed to five major functional groups: plasmablasts, GC B cells, CD27^- B cells, CD27⁺ B cells and atypical memory B cells ( Figures 5B, D, F, H ). The atypical memory B cells were characterized by co-expression of FcRL4, FcRL5 and CD11c, and were largely CD27^- ( Figures 5A, E and Supplementary Tables 10, 11 ). In small tonsils, atypical memory B cells comprised a single population that was heterogeneous in terms of T-bet expression, with 30% of cells being positive for the transcription factor ( Figures 5A–D , I and Supplementary Table 10 ). In large tonsils, we found both CD27^- and CD27⁺ atypical B-cell clusters (#3 and 9, respectively), where the CD27⁺ cluster also had increased expression of CD11c ( Figures 5E-H and Supplementary Table 11 ) and might represent a transient step in the development of atypical memory B cells. Similar to their counterparts in small tonsils, atypical memory B cells in large tonsils were diverse in terms of T-bet expression, with 25% of cells being T-bet⁺ ( Figure 5I ). While in small tonsils higher T-bet expression intensity appeared to correlate with increased surface CD11c expression on atypical memory B cells, this correlation was not as prominent in large tonsils ( Figure 5I ).

To compare the frequency of atypical memory B cells between both disease groups on an individual patient level we manually gated CD27^-CD21^-FCRL4⁺FCRL5⁺CD11c^lo cells ( Supplementary Figure 4A ). Although atypical memory B cells were significantly increased among CD27^- B cells in patients with small tonsils, the increase was not observed among total CD45⁺ cells ( Figure 5J ).

Based on the CD27 and CD21 expression, four different B cell subsets have been previously described: CD27⁺CD21^hi resting memory, CD27⁺CD21^lo activated memory, CD27^-CD21^lo tissue-like memory and CD27^-CD21⁺ naïve B cells (41, 42). In line with our observations described above ( Figure 4K and Supplementary Figure 3E ), we observed a strong enrichment of CD27^-CD21^hi naïve B-cell population in patients with large tonsils ( Figure 5K ). This was not associated with increased Ki67 expression ( Figure 5L ) thus, is unlikely to be a consequence of increased proliferation of naïve B cells in patients with large tonsils and very severe OSA.