Systems immunology analysis of human immune organoids identifies host-specific correlates of protection to different influenza vaccines

SUMMARY

Vaccines are an essential tool to significantly reduce pathogen-related morbidity and mortality. However, our ability to rationally design vaccines and identify correlates of protection remain limited. Here, we employed an immune organoid approach to capture human adaptive immune response diversity to influenza vaccines and systematically identify host and antigen features linked to vaccine response variability. Our investigation identified established and unique immune signatures correlated with neutralizing antibody responses across seven different influenza vaccines and antigens. Unexpectedly, heightened ex vivo tissue frequencies of Th1 cells emerged as both a predictor and correlate of neutralizing antibody responses to inactivated influenza vaccines (IIVs). Secondary analysis of human public data confirmed that elevated Th1 signatures are associated with antibody responses following in vivo vaccination. These findings demonstrate the utility of human in vitro models for identifying in vivo correlates of protection and establish a role for Th1 functions in influenza vaccination.

eTOC Blurb

Wagoner and Yates et al. leverage tonsil organoids to explain heterogeneity in human immune responses to influenza vaccination. They identify novel predictors (Th1 signatures) of enhanced neutralizing antibody responses and validate this finding using in vivo data, demonstrating the utility of their model for identifying novel correlates of protection.

Graphical Abstract

INTRODUCTION

Vaccines are one of the most successful public health interventions and significantly reduce global morbidity and mortality¹. Nonetheless, developing effective vaccines against complex pathogens remains a major challenge due to an incomplete understanding of how vaccines confer protection^2,3. This knowledge gap is largely attributed to the continued reliance on empirical approaches and narrow scope of correlates of protection (CoP)^4,5. Seasonal influenza vaccines are a prime example of this problem. Empirical approaches have produced a diverse array of influenza antigen formats^6,7, including live attenuated influenza vaccines (LAIVs), inactivated influenza vaccines (IIVs), subunit vaccines, and adjuvanted formulations. Unfortunately, influenza vaccines typically protect no more than 60% of recipients and effectiveness can be as low as 10% in some seasons⁸. Consequently, there is an urgent need to systematically dissect the cellular and molecular mechanisms supporting protective immunity after vaccination and thereby advance rational vaccine design.

It is challenging to identify immune features associated with protective vaccine responses due to the complexity and variability of human immune responses. To address this challenge, systems biology has emerged as a strategy to capture and analyze multimodal datasets from large vaccine cohorts^9–13. These studies typically leverage computational modeling of transcriptional data to pinpoint predictive or correlative immune features^14–20. While systems immunology has advanced our comprehension of the overarching mechanisms governing vaccine response heterogeneity, additional challenges persist^21,22. Most prior investigations have relied on peripheral immune cells as a readout of vaccine response. While practical for sampling purposes, blood inadequately reflects the evolving immune responses within lymphoid tissues²³. Additionally, blood lacks many of the cell types that contribute to the adaptive immune response, including those within germinal centers²⁴. Immune organoids serve as a complementary approach for characterizing immune responses within lymphoid tissues^25,26. These tissue-derived organoids can recapitulate several hallmarks of adaptive immunity and hold numerous advantages, including the capacity to capture human diversity and explore intra-individual vaccine responses that would otherwise be inaccessible. Crucially, organoids provide a controlled environment where perturbations can be conducted to elucidate, test, and validate mechanistic drivers of differences in the vaccine response^25,27.

The overall aim of this study was to elucidate the complex immune mechanisms involved in potentiating productive vaccine responses. Using a systems immunology approach, we characterized the diversity of cellular and humoral host features and explored the intricate dynamics underlying vaccine responses in immune organoids from 100 donors to seven distinct influenza antigens. In doing so, we established the relative influence of a variety of host and antigen factors in predicting influenza neutralizing antibody responses from primary human tonsil tissues. Our investigation unveiled significant inter-donor immune heterogeneity within lymphoid tissues and antigen format-specific differences in the cellular kinetics of the vaccine response. Based on these data, we developed a statistical model that correctly identified host immune traits linked to heightened neutralizing antibody responses and validated their use in predicting the in vivo vaccine response using publicly available immunization data. Our study reveals promising cellular targets for enhancing future influenza vaccine development.

RESULTS

Ex vivo lymphoid tissues are highly heterogeneous in cell composition

The main objective of this study was to investigate how mucosal and lymphoid tissue responses to vaccines are influenced by host and antigen features. To achieve this, we implemented a systems immunology approach combined with a previously described tonsil organoid system²⁵ to uncover predictive and correlative host features associated with adaptive immune responses to influenza antigens. Tissues and demographic data were collected from 100 otherwise healthy individuals undergoing tonsillectomy to establish a diverse donor cohort (Table 1). To characterize inter-individual heterogeneity, we quantified baseline differences in immune features (Figure 1A) including cell composition by flow cytometry (Figure S1 and Table S1), influenza exposure history assessed through circulating antibodies, and demographic factors such as age and sex. We observed considerable baseline variability in ex vivo tissue composition across the cohort (Figures 1B–D). Innate populations, including type 1 and 2 conventional dendritic cells (cDC1s, cDC2s) and plasmacytoid dendritic cells (pDCs) varied as much as 41-, 19-, and 184-fold, respectively among individuals (Figure 1B). Two- to 20-fold differences were common within many lymphocyte populations, including CD4 and CD8 T cell and B cell subsets (Figures 1C,D). Donor age significantly correlated with higher proportions of many innate and T cell populations (Figures S2A,B) and decreased frequencies of most B cell subsets, excluding classical memory (Figure S2C). We further characterized the frequency and phenotypes of B cells specific for A/California 2009 H1N1 virus hemagglutinin (HA) from ex vivo tonsils (Figure 1E). In line with repetitive exposures over time, the proportion of influenza HA-specific (HA+) B cells increased significantly with donor age (Figure 1F) and this expansion featured enriched proportions of HA+ memory and HA+ pre-germinal center (preGC) subsets. Figures 1G, S2D,E). However, variability both within and across age groups remained substantial, indicating major sources of age-independent heterogeneity.

Table 1.

100 Donor demographics

Donor	Collection Source	Tissue Type	Indication for Surgery	Age (years)	Sex	Race	Ethnicity
CHOC-003	CHOC	Tonsil/Blood	OSA	7	M	White	Not Hispanic/Latino
CHOC-006	CHOC	Tonsil/Blood	OSA	9	M	Other	Not Hispanic/Latino
CHOC-012	CHOC	Tonsil/Blood	OSA	13	F	White	Hispanic/Latino
CHOC-035	CHOC	Tonsil/Blood	Hypertrophy	6	F	Asian	Not Hispanic/Latino
CHOC-037	CHOC	Tonsil/Blood	OSA	13	M	White	Hispanic/Latino
CHOC-041	CHOC	Tonsil/Blood	Hypertrophy	15	M	Other	Hispanic/Latino
CHOC-049	CHOC	Tonsil/Blood	OSA/ Hypertrophy	6	M	Asian	Not Hispanic/Latino
CHOC-050	CHOC	Tonsil/Blood	OSA	7	F	Other	Not Hispanic/Latino
CHOC-055	CHOC	Tonsil/Blood	OSA/ Hypertrophy	4	F	White	Not Hispanic/Latino
CHOC-059	CHOC	Tonsil/Blood	Hypertrophy/sleep disordered breathing	4	M	White	Hispanic/Latino
CHOC-060	CHOC	Tonsil/Blood	Hypertrophy/Sleep disordered breathing	4	M	White	Not Hispanic/Latino
CHOC-061	CHOC	Tonsil/Blood	OSA	5	F	White	Not Hispanic/Latino
CHOC-063	CHOC	Tonsil/Blood	OSA	3	M	White	Not Hispanic/Latino
CHOC-071	CHOC	Tonsil/Blood	Hypertrophy/Sleep disordered breathing	4	F	Unknown	Unknown
CHOC-076	CHOC	Tonsil/Blood	OSA/ Hypertrophy	3	F	White	Not Hispanic/Latino
CHOC-079	CHOC	Tonsil/Blood	OSA/ Hypertrophy	9	F	White	Not Hispanic/Latino
CHOC-080	CHOC	Tonsil/Blood	Hypertrophy/Sleep disordered breathing	4	M	White	Not Hispanic/Latino
CHOC-081	CHOC	Tonsil/Blood	Hypertrophy	7	M	White	Not Hispanic/Latino
CHOC-085	CHOC	Tonsil/Blood	Hypertrophy/Allergic Rhinitis/Sleep Disordered Breathing	11	M	Other	Hispanic/Latino
CHOC-086	CHOC	Tonsil/Blood	OSA	10	F	White	Hispanic/Latino
CHOC-091	CHOC	Tonsil/Blood	Hypertrophy/Sleep disordered breathing	4	M	White	Not Hispanic/Latino
CHOC-092	CHOC	Tonsil/Blood	Hypertrophy	6	F	White	Not Hispanic/Latino
CHOC-098	CHOC	Tonsil/Blood	Hypertrophy	8	F	White	Not Hispanic/Latino
CHOC-103	CHOC	Tonsil/Blood	OSA	15	F	Other	Hispanic/Latino
CHOC-105	CHOC	Tonsil/Blood	Hypertrophy/Sleep disordered breathing	6	F	Other	Hispanic/Latino
CHOC-109	CHOC	Tonsil/Blood	Hypertrophy	8	M	White	Hispanic/Latino
CHOC-112	CHOC	Tonsil/Blood	Hypertrophy	3	M	White	Not Hispanic/Latino
CHOC-115	CHOC	Tonsil/Blood	Hypertrophy/Sleep disordered breathing	12	M	Asian	Not Hispanic/Latino
CHOC-116	CHOC	Tonsil/Blood	OSA	9	M	White	Not Hispanic/Latino
CHOC-123	CHOC	Tonsil/Blood	Chronic Sinusitis/Hypertrophy	16	M	Other	Not Hispanic/Latino
CHOC-124	CHOC	Tonsil/Blood	Hypertrophy, Adenotonsillar/Suspected PFAPA	3	F	White	Not Hispanic/Latino
CHOC-127	CHOC	Tonsil/Blood	Hypertrophy, Adenoid	5	M	White	Not Hispanic/Latino
CHOC-130	CHOC	Tonsil/Blood	Hypertrophy/Sleep Disordered Breathing	7	M	Unknown	Unknown
CHOC-134	CHOC	Tonsil/Blood	OSA	7	M	White	Hispanic/Latino
UCI-002	UCIMC	Tonsil/Blood	Recurrent tonsillitis/Hypertrophy	20	F	White	Not Hispanic/Latino
UCI-004	UCIMC	Tonsil/Blood	Recurrent Tonsillitis/Hypertrophy	31	M	White	Hispanic/Latino
UCI-005	UCIMC	Tonsil	Recurrent Tonsillitis	22	F	White	Middle Eastern
UCI-007	UCIMC	Tonsil	Hypertrophy	18	F	American Indian	Not Hispanic/Latino
UC-008	UCIMC	Tonsil/Blood	Hypertrophy	39	M	White	Not Hispanic/Latino
UCI-010	UCIMC	Tonsil/Blood	Recurrent Tonsillitis	17	F	White	Not Hispanic/Latino
UCI-011	UCIMC	Tonsil/Blood	Hypertrophy	37	M	White	Hispanic/Latino
UCI-012	UCIMC	Tonsil	Recurrent Tonsillitis/Hypertrophy	21	M	White	Not Hispanic/Latino
UCI-013	UCIMC	Tonsil/Blood	Recurrent Tonsillitis/Hypertrophy	23	F	White	Hispanic/Latino
UC-014	UCIMC	Tonsil	Recurrent Tonsillitis/Hypertrophy	28	F	White	Hispanic/Latino
UCI-019	UCIMC	Tonsil/Blood	Hypertrophy	26	F	White	Not Hispanic/Latino
UCI-020	UCIMC	Tonsil/Blood	Recurrent Tonsillitis	23	F	White	Not Hispanic/Latino
UCI-021	UCIMC	Tonsil/Blood	Recurrent Tonsillitis	21	F	White	Not Hispanic/Latino
UC-022	UCIMC	Tonsil/Blood	Recurrent Tonsillitis/Hypertrophy	22	F	White	Not Hispanic/Latino
UCI-023	UCIMC	Tonsil/Blood	Recurrent Tonsillitis	22	M	Asian	Hispanic/Latino
UCI-024	UCIMC	Tonsil	Hypertrophy	13	F	White	Hispanic/Latino
UCI-025	UCIMC	Tonsil/Blood	Recurrent Tonsillitis	23	F	White	Not Hispanic/Latino
UCI-026	UCIMC	Tonsil/Blood	Recurrent Tonsillitis	19	M	White	Not Hispanic/Latino
UCI-027	UCIMC	Tonsil/Blood	Recurrent Tonsillitis	25	F	White	Not Hispanic/Latino
UCI-028	UCIMC	Tonsil/Blood	Recurrent Tonsillitis/Hypertrophy	34	F	Asian	Not Hispanic/Latino
UCI-029	UCIMC	Tonsil/Blood	Recurrent Tonsillitis/Hypertrophy	26	M	White	Hispanic/Latino
UCI-030	UCIMC	Tonsil/Blood	Recurrent Tonsillitis	19	F	Asian	Not Hispanic/Latino
UCI-031	UCIMC	Tonsil	Recurrent Tonsillitis/Hypertrophy	44	F	White	Hispanic/Latino
UCI-032	UCIMC	Tonsil/Blood	Recurrent Tonsillitis	17	F	White	Not Hispanic/Latino
UCI-033	UCIMC	Tonsil/Blood	Recurrent Tonsillitis/Hypertrophy	42	F	Asian	Not Hispanic/Latino
UCI-034	UCIMC	Tonsil/Blood	Recurrent Tonsillitis	19	F	Black	Not Hispanic/Latino
UCI-035	UCIMC	Tonsil/Blood	Hypertrophy	35	M	White	Hispanic/Latino
UCI-037	UCIMC	Tonsil	Hypertrophy	52	M	White	Not Hispanic/Latino
UCI-038	UCIMC	Tonsil	Recurrent Tonsillitis/Hypertrophy	20	F	White	Hispanic/Latino
UCI-039	UCIMC	Tonsil	Recurrent Tonsillitis	45	F	Asian	Not Hispanic/Latino
UCI-040	UCIMC	Tonsil/Blood	Recurrent Tonsillitis	23	F	White	Not Hispanic/Latino
UCI-041	UCIMC	Tonsil/Blood	Recurrent Tonsillitis	25	F	White	Not Hispanic/Latino
UCI-043	UCIMC	Tonsil	Recurrent Tonsillitis/Hypertrophy	25	F	White	Hispanic/Latino
UCI-045	UCIMC	Tonsil/Blood	Recurrent Tonsillitis/Hypertrophy	33	F	White	Not Hispanic/Latino
WD-828	CHTN	Tonsil	a	2	M	White	Not Hispanic/Latino
WD-572	CHTN	Tonsil	a	14	M	White	Not Hispanic/Latino
WD-573	CHTN	Tonsil	a	11	F	White	Not Hispanic/Latino
WD-818	CHTN	Tonsil	a	2	F	Black	Not Hispanic/Latino
WD-962	CHTN	Tonsil	a	11	F	White	Not Hispanic/Latino
WD-965	CHTN	Tonsil	a	2	F	White	Not Hispanic/Latino
WD-636	CHTN	Tonsil	a	2	M	White	Not Hispanic/Latino
WD-840	CHTN	Tonsil	a	13	M	White	Not Hispanic/Latino
WD-490	CHTN	Tonsil	a	6	M	White	Not Hispanic/Latino
WD-901	CHTN	Tonsil	a	7	F	White	Not Hispanic/Latino
WD-938	CHTN	Tonsil	a	12	M	White	Not Hispanic/Latino
WD-999	CHTN	Tonsil	a	4	M	White	Not Hispanic/Latino
WD-951	CHTN	Tonsil	a	4	F	White	Not Hispanic/Latino
WD-319	CHTN	Tonsil	a	15	F	White	Not Hispanic/Latino
WD-320	CHTN	Tonsil	a	6	F	White	Not Hispanic/Latino
WD-201	CHTN	Tonsil	a	3	F	White	Not Hispanic/Latino
WD-275	CHTN	Tonsil	a	12	F	American Indian	Not Hispanic/Latino
WD-276	CHTN	Tonsil	a	5	F	White	Not Hispanic/Latino
WD-277	CHTN	Tonsil	a	3	F	Black	Not Hispanic/Latino
WD-700	CHTN	Tonsil	a	6	M	White	Not Hispanic/Latino
WD-701	CHTN	Tonsil	a	6	F	White	Not Hispanic/Latino
WD-132	CHTN	Tonsil	a	12	F	White	Not Hispanic/Latino
WD-133	CHTN	Tonsil	a	15	F	White	Not Hispanic/Latino
WD-252	CHTN	Tonsil	a	4	F	American Indian	Not Hispanic/Latino
WD-307	CHTN	Tonsil	a	3	M	White	Not Hispanic/Latino
WD-308	CHTN	Tonsil	a	6	M	White	Not Hispanic/Latino
WD-419	CHTN	Tonsil	a	13	M	Black	Not Hispanic/Latino
WD-420	CHTN	Tonsil	a	4	F	Black	Not Hispanic/Latino
WD-481	CHTN	Tonsil	a	7	M	Unknown	Not Hispanic/Latino
WD-641	CHTN	Tonsil	a	6	F	White	Not Hispanic/Latino
WD-049	CHTN	Tonsil	a	5	M	White	Not Hispanic/Latino
WD-050	CHTN	Tonsil	a	4	M	Black	Not Hispanic/Latino

UCIMC: University of California Irvine Medical Center

CHOC: Children’s Hospital of Orange County

OSA: Obstructive sleep apnea

PFAPA: Periodic Fever, Aphthous Stomatitis, Pharyngitis, Adenitis

^aMost surgeries were performed for hypertrophy, but donor-specific data were not available

Figure 1. Inter-individual variability in lymphoid in tissue immune composition and pre-existing antibodies.

(A) Experimental design and workflow. Created with Biorender.com.

(B-D) Cell composition of ex vivo tonsil innate (B), T cells (C), and B cell (D) populations across 100 individuals. Frequencies were determined by flow cytometry. Representative flow cytometry plots display gating schemes for notable B and T cell populations.

(E) Representative flow cytometry plots displaying ex vivo frequencies of HA+ B cells of total B cells (left) and overlaid over major B cell phenotypes (middle). Frequencies of HA+ B cell phenotypes (% of Live) from 100 donors (right).

(F) Correlation assessing relationship between donor age (years) and frequencies of total HA+ B cells (% of Live). Statistics were determined using Spearman’s rank correlation coefficient.

(G) Frequencies of HA+ B cell phenotypes (% Total HA+ B cells) separated by age range.

(H) Donor plasma antibodies (n=53) as measured by high throughput protein microarray (left) and quantification of median MFI of influenza-specific IgG values for seasonal influenza subtypes and non-hemagglutinin influenza proteins in donor plasma (right). Median fluorescence intensity (MFI) represents antibody binding magnitude. Column dendrograms represent unbiased grouping of donors based on antibody profile similarity. Rows represent individual influenza proteins manually grouped by influenza subtype and protein.

Boxplots display median, with hinges indicating the first and third quartiles. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001; Kruskal-Wallis tests were performed to assess differences among groups and Mann-Whitney U tests were used to calculate significance values (G and H).

See also Figures S1, S2 and Tables 1, S1 and S2.

To further define differences in immune history, we characterized plasma circulating antibodies using a custom protein microarray containing influenza HA and non-HA proteins (Table S2) spanning human and zoonotic strains from 1918 to the present (Figures 1H, S2F,G). While notable overlap was observed across all age ranges for all antigens, HA-specific IgG and IgA magnitude and breadth were elevated with increasing age, whereas IgM antibodies were highest among donors aged 6-10 years (Figures 1H, S2F,G). Collectively, we demonstrate that both age dependent and independent factors contribute to the substantial heterogeneity in pre-existing antibodies in plasma and cell composition in lymphoid tissues.

Vaccine format and season influences the magnitude and diversity of vaccine response in tonsil organoids

Previously, we demonstrated that distinct immune responses are induced in adult-derived tonsil organoids stimulated by inactivated influenza vaccines (IIV) compared to live-attenuated influenza vaccines (LAIV)²⁷. We next investigated how differences in antigen format and seasonal formulation impact the kinetics, magnitude, and quality of the cellular response to influenza antigens in donors across the age span. We stimulated tonsil organoids from 100 donors, encompassing both pediatric and adult samples, with numerous influenza antigens produced over multiple vaccine seasons (Table S3), including LAIVs (Flumist 19/20, Flumist 21/22), IIVs (Fluzone 19/20, Fluarix 21/22), an adjuvanted IIV (FluAD 21/22), a subunit vaccine (Flucelvax 21/22), and a wild-type H1N1 virus (A/Cal/07/2009) (Figure 2A). Following stimulation on day 0, organoids were harvested and analyzed by flow cytometry on days 4, 7, 10, and 14 (Figure 2A). Significant differences were observed both between antigen formats at the same time point, and within the same antigen across different timepoints (Tables S4 and S5). Within T cells, live antigens (LAIVs and A/Cal 2009) induced robust Th1 differentiation by day 7, which remained elevated for the remainder of the culture period (Figure 2B), and elicited higher frequencies of memory CD8s (Figure S3A). In contrast, T cells stimulated by non-live antigens (IIVs and subunit) supported primarily Th2 phenotypes (Figure 2B) at frequencies similar to unstimulated controls. IIV stimulated organoids contained higher proportions of T regulatory (Treg) cells early in culture (day 4) compared to live antigens, while T follicular helper cells (Tfh) were not statistically different between stimulation conditions throughout the cultures. The antigen-specific B cell response showed unique kinetics and endpoint phenotypes depending on vaccine format. Live antigens were the most stimulatory in organoids and followed similar B cell kinetics to that observed previously in a smaller cohort of adult donors²⁷. B cell kinetics differed between IIVs and LAIVs, but profiles were generally similar between antigens of the same format. FluAD 21/22, which contains MF59 adjuvant²⁸, elicited a small but significant increase in early B cell activation and preGC B cell formation compared to IIVs (Figure 2C and Table S4). HA+ B cells often showed more pronounced cell differentiation kinetics, similar to those observed in total B cells. (Figure 2D). Notable differences were primarily observed following LAIV stimulation, which led to greater activation of HA+ memory cells and enhanced differentiation of HA+ preGC cells into mature plasmablasts (PBs).

Figure 2. Vaccine format and season influences the magnitude and diversity of vaccine response in tonsil organoids.

(A) Experimental design and workflow. Created with Biorender.com.

(B-D) Longitudinal experiments in organoids displaying changes in the kinetics and frequencies of T cell (B), B cell (C) and HA+ B cell (D) subsets. B and T cell data was collected on days 4, 7, 10, and 14 following influenza antigen stimulation; HA+ B cell data was collected on days 7, 10, and 14. Colors represent vaccine/antigen stimulants. Unstimulated (Unstim) organoids were used as negative controls.

(E) H1-virus neutralization by day 14 organoid culture supernatants.

(F) Neutralization data was quantified by area under the curve (AUC) and then transformed (1-AUC) to reflect positive values for increased neutralization.

(G) Summary antibody data from day 14 organoid culture supernatants analyzed by protein microarray separated by protein type and virus source.

(H) Proportion of organoid neutralizing high responders determined by a manual cutoff of 0.75 (1-AUC).

(I) Proportion of organoid responders to vaccine-associated viral antigens in each age range to either IIV 19/20 or LAIV 19/20 vaccine. High responders are defined as donors with median MFIs two standard deviations above the unstimulated median for each antigen group.

See also Figure S3 and Tables S3, S4, and S5.

Neutralizing antibodies (nAbs) against influenza HA^29,30 are widely accepted as the current gold standard CoP following vaccination^31–37. Therefore, we measured nAb secretion as a functional readout of organoid response (Figure S3B). Area-under-the-curve (AUC) analysis was used to quantify the presence of nAbs against A/California 2009 (Figure 2E). All antigens induced significant increases in nAbs compared to unstimulated controls (Figures 2F and S3C). Stronger neutralizing responses were elicited by antigens associated with enhanced B cell activation and differentiation into preGC, GC, and PB phenotypes compared to unstimulated conditions. This was particularly evident within the HA+ B cell subsets (Figures 2C,D). Independent of influenza season, LAIV elicited significantly greater neutralizing responses compared to other vaccine types (Figures 2F and S3C).

To complement this functional analysis, we quantified the influence of antigen format on antibody magnitude and breadth (Figures 2G and S3D). As expected, organoids stimulated by A/Cal 2009 H1N1 virus elicited antibodies primarily against itself and other H1N1 antigens, with minimal heterosubtypic cross-reactivity. Despite their quadrivalent formulations, similar trends in antibody breadth were observed for IIV and Flucelvax stimulated organoids, albeit with lower magnitude. LAIV-stimulated organoids demonstrated superior antibody breadth compared to all other influenza antigens (Figure 2G and S3D). Notably, despite nearly identical trends in kinetics and neutralization, vaccine season did play a role in antibody diversity. While post-2009 H1N1 antibody responses were similar in both LAIV seasons tested, LAIV 19/20 elicited higher antibody responses against pre-2009 H1N1 viruses and H3N2 antigens, while LAIV 21/22 stimulated higher influenza B antibodies (Figures 2G). Flucelvax, the subunit vaccine, elicited the lowest overall B cell response in organoids and thus was removed from subsequent experiments. Lastly, live-attenuated antigens greatly enhanced nAb production (Figure 2H) and antibody breadth (Figure 2I) in organoids derived from individuals of all ages, while IIV primarily elicited antibody responses from adult (18+) organoids. Collectively, these data demonstrate that influenza vaccine format and season significantly influence the kinetics of the cellular response and the magnitude, breadth, and functional quality of elicited antibodies.

Neutralizing responses to different vaccine formats are associated with shared and unique cellular signatures

Given the differences we observed in baseline host immune composition (Figure 1) and distinct immune responses to various influenza antigens (Figure 2), we next aimed to explore how interactions between these two factors might explain or predict disparities in vaccine response and identify potential CoP. To determine the extent of host influence on the influenza response, we stratified donors into high or low responders based on their nAb response, independently for each antigen (Figure 3A). A Jaccard similarity index analysis performed across all antigens revealed that high responders to one vaccine were more likely to be high responders to other vaccine formulations; the same was true for low responders (Figure 3B). Comparing the same vaccine format across different influenza seasons yielded similar results, though the overlap between nAb responder groups was further reduced (Figure 3C). Thus, we hypothesized that host-specific cellular factors play a central role in explaining response heterogeneity. To identify these features, we conducted a correlation analysis between cell frequencies and organoid nAb responses. In line with previous blood-based assays¹⁵, strong positive correlations were observed between organoid HA+ PBs and nAbs for all influenza antigens (Figure 3D). In contrast, poor nAb levels correlated with higher frequencies of HA+ preGC B cells for all vaccine formats and for LAIVs, with HA+ GC cells. These associations were present across all time points and were similarly reflected in A/Cal 2009 stimulated cultures, but with reduced significance (Figure S4A). We then used our nAb response classification (Figure 3A) to identify what cellular features from organoid responses could accurately classify donors as nAb responders or non-responders. In line with our previous findings, high nAb responders generated significantly greater frequencies of HA+ PBs at all time points, whereas low responders were more likely to have a pre-GC phenotype (Figure 3E). Changes in HA+ naive and memory proportions were not correlated with high or low response (Figure S4B). Together, these data show that there are predictable B cell differentiation features that correlate with a productive influenza vaccine response, independent of antigen format.

Figure 3. Neutralizing responses to different vaccine formats are associated with shared and unique cellular signatures.

(A) Cut-off neutralization values for the classification of donors as high neutralizing (right) and low neutralizing (left) responders.

(B and C) Donor overlap was quantified by the Jaccard index within and between high (Hi) and low (Lo) response quartiles across all seven vaccine stimulations (B) or grouped by antigen modality (C).

(D) Heatmap of neutralization data correlated against day 14 frequencies of HA+ B cell subsets (% total HA+ B cells); color represents either positive (red) or negative (blue) correlations. Statistics were determined using spearman–s rank correlation coefficient.

(E) Comparison in the kinetics of activated HA+ B cell subsets between high (solid) and low (dashed) neutralizing donors previously defined in (3A).

(F) Heatmap of neutralization data correlated against day 4, 7, 10, and 14 frequencies of select B cell (% B cells) and T cell (% T cell) subsets. Statistics were determined using spearman’s rank correlation coefficient.

(G-I) Comparison in the kinetics of notable cell populations between high (solid) and low (dashed) neutralizing donors defined in (3A).

For all comparisons: *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001; Kruskal-Wallis tests were performed to assess differences among groups and Mann-Whitney U tests were used to calculate significance values (E, G-I).

See also Figure S4.

We then expanded our investigation into other T and B cell populations to search for cellular CoP (Figures S4C and S4D). Several cell populations positively correlated with nAbs across all influenza antigens, while others were uniquely associated with a particular vaccine format. Dark zone (DZ) GC B cell proportions correlated with a strong IIV response, while CD138+ PBs correlated with good LAIV responses. At later time points (days 10 and 14), additional associations were identified between nAb responses and antiviral (e.g., Th1, CD8 T cell) subsets (Figure 3F). A shift from Th2 to Th1 CD4 T cell phenotypes was significantly associated with stronger responses to inactivated antigens. Analysis of the same cell subsets using our high vs. low responder classifications similarly showed elevated frequencies of DZ GC, CD138+ PB, and Th1 cells for their respective antigens (Figures 3G–I). Together, our analyses confirm HA+ PBs as a CoP to all influenza antigens tested in lymphoid tissue-derived organoids and identify additional B and T cell subsets associated with successful nAb responses for different influenza antigen formats.

High baseline Th1 frequencies predict neutralizing antibody responses to inactivated vaccines

We next investigated host predictors of influenza vaccine response. Age is a well-known predictor of vaccine response¹⁴ and on average, age was the most influential demographic feature we measured, accounting for 8.9% of the variance in the organoid immune response (Figure 4A). However, all socio-demographic variables together explained only 16% of the variance in the organoid responses. To identify important cellular features that could explain the remaining variance in nAb response, we leveraged a machine learning approach (Figure 4B). We focused on generating models that could accurately predict high vs. low nAb responders (as defined in Figure 3A). In addition to generating a prediction model for each individual vaccine (Figure S5A), we also produced two “combined” models, which aggregated data per vaccine format - one for IIVs and one for LAIVs (Figure 4C). For the combined LAIV model, the Area Under the Receiver Operating Characteristic Curve (AUROC) was 0.61 (CI 0.46 - 0.76), and the AUROC for the combined IIV model was 0.96 (CI 0.92 - 0.99) (Figure 4C). Although LAIV was better at stimulating nAb responses in organoids, these values indicate that our ability to discriminate responders from non-responders was much better for IIV-treated organoids. Based on this, as well as the fact that IIV is more broadly administered in clinical settings, we chose to continue our prediction analysis on IIV antigens.

Figure 4. High baseline Th1 frequencies predict neutralizing antibody responses to inactivated vaccines.

(A) A linear model was used to determine the proportion of variance explained in the neutralizing antibody response by demographic variables across seven vaccine stimulations, residual refers to the variance that could not be explained by those variables. Error bars depict standard error and bar height is the mean.

(B) The pipeline used for data preprocessing and predictive model development.

(C) A logistic regression model was trained to differentiate between high and low responders. The classification performance was evaluated using nested leave-one-out cross-validation (LOOCV). The ROC curve is displayed for both the IIV (n=65) and LAIV (n=60) combined modality model.

(D) Response classes are identified via hierarchical clustering of predictor SHAP values for the IIV modality model.

(E) SHAP summary plot of the top 10 cellular predictors for the combined IIV modality model. Predictors are ordered on the y-axis (from top) by their importance to the model (mean absolute SHAP value). Positive and negative SHAP values indicate an increase and decrease in probability of responder class membership, respectively. Color is representative of baseline cell frequencies (% of Live).

(F) Ex vivo cell frequency for Th1, cDC1, Treg, and memory CD8 subsets (% of Live) in donors classified as either high or low nAb responders.

*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001; Mann-Whitney U tests were used to calculate significance values.

See also Figure S5.

To evaluate the output of our combined IIV modality model, we performed supervised clustering of the SHapley Additive explanations (SHAP) values (Figures 4D), which allowed us to better understand the relative importance of the various cell populations in predicting nAb responses. SHAP clustering grouped donors based on their IIV response class, reinforcing the notion that there are conserved host factors that potentiate nAb responses to IIV antigens (Figure 4D). To identify the top host features that predicted IIV responses, mean absolute SHAP values were used to assess and rank the relative importance of each cellular population identified in the combined model (Figure 4E). As expected, two of the top three cell subsets identified by the model were influenza-specific B cell subsets. Additional predictors included Th1s, memory CD8 T cells, Tregs, and cDC1s, where elevated ex vivo frequencies of these populations were associated with higher nAb responders (Figure 4F). Of particular interest were Th1 cells, which the model identified as a top shared cellular feature predictive of IIV-induced nAb responses (Figure 4E). Indeed, higher frequencies of Th1s at baseline were among the top 10 predictive cellular features for each of the IIV antigens (Figures S5B–D). Together, high frequencies of Th1 cells were identified as both a correlate (Figure 3) and predictor of enhanced nAb response following stimulation with IIV antigens in organoids.

Th1 cells contribute to humoral responses by enhancing early B cell activation

We previously showed that memory CD4 T cells were critical for supporting humoral responses to IIV and less so for LAIV^25,27. Based on our IIV model, we reasoned that Th1 cells may be the putative subset responsible for supporting a productive antibody response. Supporting this, organoids from high IIV responders secreted more Th1-associated cytokines compared to low responders (Figures 5A and S6A). In responders, we detected an early and significant increase in IL-2 secretion; IL-8 and GM-CSF were also slightly (but not significantly) increased. In contrast, anti-inflammatory IL-10 was not different between responders and non-responders. To further validate the functional role of Th1 cells on the development of humoral responses, we selectively depleted them (CD45RA− CXCR5− CXCR3+ CCR4− CCR6−) from four of our highest IIV responding donors. Organoids were prepared in which Th1 cells were either depleted or reconstituted at their original levels, then stimulated with IIV or LAIV (Figure 5B). Flow cytometry analysis on day 10 confirmed that Th1 frequencies remained low in depleted organoids after culture (Figure 5C). To understand the effect of Th1 depletion on nAb responses, we assessed changes in B cell activation and differentiation. Although donor tissue exhaustion prior to these experiments limited our sample sizes, we found that Th1 depletion generally led to reduced early B cell activation (Figure 5D) within IIV-stimulated cultures. This was characterized by increased frequencies of naive and memory subsets and a corresponding decrease in the proportion of preGC cells. Additionally, IIV-stimulated cultures produced generally fewer PBs following Th1 depletion, while PB differentiation in LAIV-stimulated cultures was independent of Th1 cells (Figure 5D). Th1 depletion diminished the production of influenza-specific antibodies (Figure 5E) and attenuated neutralization function (Figure 5F) in response to both IIV antigens, but not LAIV.

Figure 5. Th1 cells contribute to humoral responses by enhancing early B cell activation.

(A) Secreted cytokine analysis of day 4 organoid supernatants stimulated with IIV 21/22 from n=24 donors (8 adults and 4 children for each response class). Donor response class was defined by IIV 21/22 neutralization values.

(B) Experimental design and workflow for Th1 depletion. Created in Biorender.com

(C) Frequencies of Th1 cells on day 10 from Th1 depleted or remixed (WT) organoid cultures.

(D) Frequencies of major B cell populations (% of Total B cells) on day 10.

(E and F) Antibody magnitude and function from day 10 organoid supernatants were determined through semi-quantitative ELISA measuring total influenza-specific (IgA/M/G) antibodies (E) and microneutralization (F).

(G) Plots of select analytes from Luminex analysis performed on day 10 organoid supernatants from either WT or Th1 depleted cultures. Shapes are representative of different donors.

(H) Frequencies of major B cell populations (% of Total B cells) on day 10 organoids (n=12) following Th1-associated cytokine supplementation. Significance reflects comparisons to the no supplementation control within each stimulation condition.

Boxplots display median, with hinges indicating the first and third quartiles. For all comparisons *p < 0.05, **p < 0.01, ***p < 0.001; Kruskal-Wallis tests were performed to assess differences among groups and Mann-Whitney U tests with multiple hypothesis correction were used to calculate significance values (A and H). Due to limited sample sizes for statistics, calculated p values are shown for Th1 depletion analyses (D-G).

See also Figure S6.

Next, we investigated putative mechanisms through which Th1 cells may contribute to antibody development. Given that Th1 cells do not have a clearly defined role in providing B cell help within the GC^38,39, and our earlier observation that high IIV responders produced elevated levels of Th1-associated cytokines (Figure 5A), we hypothesized that Th1 cells might influence antibody responses through a cytokine-driven mechanism. Multiplex cytokine analysis on supernatants from Th1-depleted organoids revealed that Th1 depletion led to reduced levels of many innate and Th1-associated cytokines, including IFNg, IL-12, and TNFa, compared to Th1-reconstituted organoids (Figures 5G and S6B). To test whether cytokines associated with Th1 differentiation or secretion could promote IIV responses, organoids were supplemented with IL-12 and/or IFNg on day 0 and subsequently cultured for 10 days. Th1-associated cytokines increased B cell activation, as evidenced by increased pre-GC B cell differentiation and reduced naive and memory frequencies (Figure 5H). However, this increased activation did not result in a significant increase in the magnitude or function of the resulting antibodies (data not shown), suggesting that cytokine production may be only one mechanism through which Th1 cells contribute to promoting humoral immunity. Together, these data highlight our model’s ability to identify key cellular predictors of vaccine responses from tonsil organoids and reveal an unexpected role for Th1 cells in supporting humoral responses to inactivated, but not live, influenza antigens.

Th1 gene signatures predict IIV responses to in vivo influenza vaccination

To investigate whether Th1 cells may be a clinically relevant predictor of vaccine response in peripheral blood in vivo, we re-analyzed publicly available gene expression data from the Immune Signatures⁴⁰ data source. These analyses included nine IIV studies, encompassing five influenza seasons and 144 donors aged 50 and younger stratified into vaccine response groups. An eight-gene Th1 representative signature was constructed based on previously defined Th1 differentially expressed genes⁴¹. Transcriptomic activity of this gene signature was significantly upregulated in high vaccine responders compared to moderate and low responders (Figure 6A). Additionally, subjects were partitioned by age, sex, and Th1 enrichment score. Here, transcriptomic activity of the eight-gene signature positively correlated with vaccine response across ages and sexes, with the exception of females aged 30 and older, where moderate expressors showed a modestly stronger mean enrichment with responses compared to the high expressors (Figure 6B). Given that the Th1 expression signature correlated with vaccine response, we next assessed the gene expression similarities and differences between low and high responders to vaccination. These analyses highlighted co-expression differences within the eight-gene Th1 signature based on IIV response class. Only one gene pair, (PZD8 ~ BCR) was consistently correlated in both low and high responders (Figure 6C). In low responders, IRF8 and BCR gene expression were significantly negatively correlated (Figure 6C). In high responders, PMAIP1 expression was significantly positively correlated with IRF8, IFNG, and PDZ8 (Figure 6D). In summary, we found that greater baseline Th1 cell signatures distinguished high from moderate and low vaccine responders across ages and sexes in vivo, and highlighted specific gene relationships driving these differences.

Figure 6. Th1 gene signatures predict IIV responses to in vivo influenza vaccination.

(A) Sample-level enrichment analysis (SLEA) based on an eight-gene Th1 signature. Low, moderate, and high classes are derived from maxRBA provided from the original source studies.

(B) Mean maxRBA across Th1 SLEA z-score terciles in males and females younger and older than 30. Values within cells indicate the number of individuals in each group.

(C and D) Gene co-expression in the eight-gene Th1 signature for low (C) and high (D) vaccine responders. For all statistics: *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001; SLEA significance values were calculated with a Mann-Whitney U test and corrected for multiple hypothesis testing using the Holm method (A). Gene Correlation statistics were determined using biweight midcorrelation and a two-sided t-test (C and D).

DISCUSSION

There is a complex relationship between host and antigen features in potentiating and explaining variation in human vaccine responses. Our investigation uncovered substantial diversity in cell composition within lymphoid tissue in healthy individuals that could not be fully explained by demographic factors. Building upon previous findings⁴², which compared the influence of two influenza vaccine antigen formats on adult organoids, this study demonstrated the superior performance of live-attenuated vaccines across all age groups. Furthermore, antigen format significantly influenced the kinetics and magnitude of the influenza-specific antibody response, independent of seasonal formulation changes. Through computational modeling and in vitro validation studies, we identified Th1 frequencies as a predictor and correlate of neutralizing antibodies in response to IIV but not LAIV. We were then able to validate this finding using in vivo human vaccination data, underscoring the translational potential of lymphoid organoids for experimentation in an entirely human system.

Vaccine-induced influenza immunity is less effective in very young⁴³ (<5 years) and older⁴⁴ (65+) individuals. Age-related alterations in cell composition and function are most commonly cited to explain their suboptimal responses^31,45,46. However, the degree to which these immune alterations exist within the broader healthy population and their potential impact on vaccine response diversity has been less explored⁴⁷. Here, we identified extensive baseline variability in the ex vivo frequencies of various immune populations within human tonsils. In line with prior studies^48,49, there were significant associations between age and increased frequencies of innate and memory B and T cell populations. Consistent with increased exposure events, influenza-specific HA+ B cell frequencies correlated with donor age^45,50. In addition to demographic variables, immune cellular heterogeneity can result from a combination of intrinsic^51–54, heritable^55,56 and environmental factors^57,58. While age emerged as the most influential socio-demographic feature we measured to explain variance in organoid vaccine responses, most of the variance (84%) remained unexplained. Thus, age alone is not sufficient to fully explain vaccine response heterogeneity.

In the United States, there are a large number of FDA-approved influenza vaccines⁵⁹, encompassing live-attenuated, split, subunit, and recombinant formats. While most influenza viruses are cultivated in chicken eggs, recombinant proteins can be generated from alternative sources, including insect⁶⁰ and mammalian⁶¹ expression systems. These various production methods introduce significant differences in the final product, including structural differences⁶² among vaccine formats and inconsistent protein and nucleic acid content between seasons⁶³, which could contribute to variation in vaccine immunogenicity^27,64. Using the organoid system, we demonstrated the pivotal role of antigen format in enhancing immune responses and uniquely shaping the kinetics, magnitude, and breadth of antibody responses and T cell phenotypes. In general, inactivated vaccine formats sustained Th2 programs, while live antigens elicited robust Th1 responses^65–67. Although Th1 cells are not conventionally associated with influenza vaccine responses, a Th1 profile is desirable as part of the antiviral response⁶⁶. Previously, we demonstrated that LAIV triggered more robust humoral responses compared to IIV in adult organoids²⁷. Building upon this, our current study reveals that this effect is retained across influenza seasonal reformulations and that LAIV consistently induced broader and more functional antibody responses across donors of all age groups. The functional benefit of LAIV on tonsil organoids was particularly notable for donors under five years of age, who displayed limited response to IIV. We speculate this is due to low frequencies of influenza-specific memory B and T cells, which appear to be critical for a robust IIV response²⁷. Collectively, our findings emphasize the complexity of interactions between host and antigen factors, but point to emerging strategies to predictably identify critical interactions using robustly powered datasets from a well-controlled system.

Identifying host features associated with productive vaccine responses remains a longstanding objective in systems vaccinology⁶⁸. Computational modeling has emerged as a powerful tool to tackle this problem¹³. Nearly a decade ago, modeling efforts from human peripheral blood data identified that variation in temporally stable immune metrics could predict the influenza vaccine response¹⁴. Subsequently, this approach has been instrumental in defining immune signatures for specific vaccines and even in identifying human innate immune “endotypes” that broadly predict vaccine responsiveness¹⁶. While these findings offer critical insights into human vaccine responses, their reliance on blood measurements represents a persistent limitation in interpreting and understanding the ongoing adaptive immune response. To complement these studies, data sourced from immune tissues actively responding to vaccine-relevant antigens is needed. Tonsil organoids present one possible solution to this issue and can replicate many of the hallmarks of lymphoid tissue adaptive immunity. Here, we combined organoids and computational modeling to identify specific host features associated with productive influenza vaccine responses. Increased frequencies of DZ B cells were linked to enhanced neutralizing antibody responses across all antigens, while elevated Th1 cells served as both a predictor and correlate of productive humoral responses following stimulation with IIV, but not LAIV. The potential role for Th1 cells in enhancing humoral responses is supported by previous modeling studies that associated inflammatory gene signatures with vaccine responders^14,16. Additionally, neutralizing antibodies were correlated with increased interferon signaling, and interferon-related genes were associated with IL-2+ CD4 T cells¹⁴, suggesting that CD4 T cells with Th1-like functions may contribute to predicting the antibody response. These findings provide one possible explanation for the historical suboptimal performance of IIV in children^69–71 since Th1 frequencies are notoriously low early in life^72–74.

Despite strong LAIV immunogenicity in organoids for most donors, our models could not identify baseline predictors of LAIV response. This finding aligns with in vivo studies that have similarly struggled to identify a consistent CoP for LAIV^75–77. Since LAIV is a live virus, it is plausible multiple cellular redundancies exist to ensure effective responses, which would hinder the identification of individual cell predictors. LAIV may also be more effective at generating plasmablasts through both germinal center-dependent and -independent pathways⁷⁸. Consequently, the absence of a few key immune features that define vaccine responsiveness likely contributes to the challenge of identifying predictors or CoPs for LAIV. In contrast, IIV responses rely on the presence of defined cell populations, such as memory T and B cells, to elicit successful antibody responses^27,50, which can simplify the task of identifying cellular predictors.

This study identified elevated ex vivo Th1 cell frequencies as both a predictor and correlate of neutralizing antibody responses to IIV antigens. While their role in providing B cell help remains unclear^38,39, we demonstrate that Th1 cells can influence B cell activation, either directly or indirectly. Despite a relatively low number of samples in our depletion studies, our cytokine and supplementation data clearly suggest that cytokine-driven mechanisms are likely involved in this process. Th1s secrete IFNg, which is a crucial factor in B cell differentiation and GC formation in both mice and humans⁷⁹. Moreover, Th1-associated IFNg and IL-21 facilitate protective antibody responses in mice lacking Tfh cells⁸⁰. In line with these studies, Th1 depletion in organoids reduced early B cell activation and attenuated antibody responses following IIV stimulation. While IFNg and IL-12 supplementation was sufficient to enhance B cell activation, they were not sufficient to enhance antibody magnitude or function. It may be that B cells require direct interactions⁸¹ with Th1s or that Th1s indirectly support humoral responses by priming other T cell subsets^82,83 that contribute to further B cell maturation. Though not deeply investigated in this study, our model also identified high frequencies of memory CD8 T cells and cDC1s as top predictive features of IIV responders. These populations may serve as additional sources of IFNg or enhance its production through IL-12 signaling and Th1 skewing. Further, IL-12 secreted by cDC1s may act directly on B cells to promote extrafollicular responses⁸⁴. Together, our data and prior literature point to a signaling axis between Th1 and B cells to support productive humoral responses.

In summary, we highlighted the complex interactions between host and antigen features in dictating the development of successful immune responses and demonstrated that these interactions could be predicted in a well-controlled system. The insights gained from our computational modeling hold profound implications for elucidating the differences driving inter-individual vaccine responses. By identifying robust predictors and correlates of protection associated with successful influenza vaccine response, we have expanded our understanding of several immune mechanisms underlying vaccine efficacy. These findings can be leveraged to target specific cellular mechanisms, complement the currently inefficient empirical methodologies used, and enable pragmatic vaccine design. Additionally, these models could contribute to personalized vaccination strategies by tailoring vaccine formulations to specific populations, such as different age groups, and thereby enhance overall vaccination efficacy at the population level.

Limitations of study

Lymph nodes are a complex environment that dynamically change throughout the course of an antigen-specific response. Because the tonsil organoid model is an in vitro system, we recognize several inherent limitations. Static culture conditions prevent us from capturing immune composition changes that can occur in vivo during infection, such as the migration of antigen-presenting cells and lymphocytes in and out of the lymph node. Further, the effect of potential viral replication may not be fully captured by the model due to the low frequencies of major cell types involved in influenza viral replication within the organoids. We show that LAIV induces significantly greater immunogenicity than other antigen formats in tonsil organoids. However, LAIV’s in vivo route of delivery (intranasal) is distinct from the other vaccine formats included in this study, which are delivered intramuscularly. While our findings show that LAIV can effectively stimulate organoid responses, differences in the route of administration compared to in vivo may in part explain why we were unable to identify strong predictors of response to this vaccine format. In contrast, antigen from inactivated vaccines can reach the draining lymph nodes from the injection site via both migrating DCs and as soluble antigen^85,86. Therefore, analysis of IIV responses in the organoid model may be a more accurate representation of how this vaccine interacts with immune cells in vivo.

RESOURCE AVAILABILITY

Lead Contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Lisa Wagar (lwagar@hs.uci.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• Complete flow cytometry and neutralization datasets have been deposited at ImmPort as SDY2947. The accession number is also listed in the key resources table.

• This paper analyzes existing, publicly available data, accessible through the Immune Signatures data resource. The accession numbers for these datasets are listed in the key resources table.

• All original code has been deposited at Zenodo at [https://zenodo.org/records/13334292] and is publicly available as of the date of publication.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Human TruStain FcX	Biolegend	Cat # 422302; RRID: AB_2818986
Anti-Human CCR4 (L291H4) BV421	Biolegend	Cat # 359414; RRID: AB_2562435
Anti-Human CCR6 (G034E3) BV421	Biolegend	Cat # 353408; RRID: AB_2561356
Anti-Human CCR6 (G034E3) BV650	Biolegend	Cat # 353426; RRID: AB_2563869
Anti-Human CCR7 (G043H7) PE/Dazzle594	Biolegend	Cat # 353236; RRID: AB_2563641
Anti-Human CD11c (3.9) Alexa Fluor 700	Biolegend	Cat # 301648; RRID: AB_2819923
Anti-Human CD123 (6H6) BV650	Biolegend	Cat # 306020; RRID: AB_2563827
Anti-Human CD127 (A019D5) BV605	Biolegend	Cat # 351334; RRID: AB_2562022
Anti-Human CD13 (WM-15) FITC	eBioscience	Cat # 11-0138-42; RRID: AB_11043278
Anti-Human CD138 (MI15) FITC	Biolegend	Cat # 356508; RRID: AB_2561882
Anti-Human CD141 (M80) APC	Biolegend	Cat # 344106; RRID: AB_10899578
Anti-Human CD19 (HIB19) BV650	Biolegend	Cat # 302238; RRID: AB_2562097
Anti-Human CD19 (HIB19) PE	Biolegend	Cat # 302208; RRID: AB_314238
Anti-Human CD19 (HIB19) PerCP/Cy5.5	Biolegend	Cat # 302230; RRID: AB_2073119
Anti-Human CD1c (L161) PE/Cy7	Biolegend	Cat # 331516; RRID: AB_2275574
Anti-Human CD21 (Bu32) Alexa Fluor 700	Biolegend	Cat # 354918; RRID: AB_2750239
Anti-Human CD25 (BC96) APC/Cy7	Biolegend	Cat # 302614; RRID: AB_314284
Anti-Human CD27 (O323) BV711	Biolegend	Cat # 302834; RRID: AB_2563809
Anti-Human CD27 (O323) PE/Cy7	Biolegend	Cat # 302838; RRID: AB_2561919
Anti-Human CD3 (HIT3a) Alexa Fluor 700	Biolegend	Cat # 300324; RRID: AB_493739
Anti-Human CD3 (OKT3) APC/Cy7	Biolegend	Cat # 317342; RRID: AB_2563410
Anti-Human CD3 (OKT3) BV605	Biolegend	Cat # 317322; RRID: AB_2561911
Anti-Human CD3 (HIT3a) PE/Cy5	Biolegend	Cat # 300310; RRID: AB_314046
Anti-Human CD38 (HIT2) PE	Biolegend	Cat # 303506; RRID: AB_314358
Anti-Human CD38 (HIT2) PE/Dazzle594	Biolegend	Cat # 303538; RRID: AB_2564105
Anti-Human CD39 (A1) BV785	Biolegend	Cat # 328240; RRID: AB_2814191
Anti-Human CD4 (OKT4) PE/Cy5	Biolegend	Cat # 317412; RRID; AB_571957
Anti-Human CD4 (OKT4) PE/Cy7	Biolegend	Cat # 317414; RRID: AB_571959
Anti-Human CD45 (2D1) PerCP/Cy5.5	Biolegend	Cat # 368504; RRID: AB_2566352
Anti-Human CD45RA (HI100) BV785	Biolegend	Cat # 304140; RRID: AB_2563816
Anti-Human CD45RA (HI100) FITC	BD	Cat # 555488; RRID: AB_395879
Anti-Human CD45RB (MEM-55) PE	Biolegend	Cat # 310204; RRID: AB_314807
Anti-Human CD73 (AD2) BV605	Biolegend	Cat # 344023; RRID: AB_2650973
Anti-Human CD8 (RPA-T8) APC/Cy7	Biolegend	Cat # 301016; RRID: AB_314134
Anti-Human CD8 (SK1) FITC	Biolegend	Cat # 344704; RRID: AB_1877178
Anti-Human CD83 (HB15e) APC/Cy7	Biolegend	Cat # 305330; RRID: AB_2566393
Anti-Human CXCR3 (G025H7) BV711	Biolegend	Cat # 353732; RRID: AB_2563533
Anti-Human CXCR3 (G025H7) PE	Biolegend	Cat # 353706; RRID: AB_10962912
Anti-Human CXCR4 (12G5) BV421	Biolegend	Cat # 306518; RRID: AB_11146018
Anti-Human CXCR5 (J252D4) APC	Biolegend	Cat # 356908; RRID: AB_2561817
Anti-Human CXCR5 (J252D4) FITC	Biolegend	Cat # 356914; RRID: AB_2561896
Anti-Human CXCR5 (J252D4) PE/Cy7	Biolegend	Cat # 356924; RRID: AB_2562355
Anti-Human HLA-DR (L243) Alexa Fluor 700	Biolegend	Cat # 307626; RRID: AB_493771
Anti-Human HLA-DR (L243) BV570	Biolegend	Cat # 307638; RRID: AB_2650882
Anti-Human HLA-DR (L243) Pacific Blue	Biolegend	Cat # 307633; RRID: AB_1595444
Anti-Human HLA-DR (L243) PerCP/Cy5.5	Biolegend	Cat # 307630; RRID: AB_893567
Anti-Human IgD (IA6-2) APC	Biolegend	Cat # 348222; RRID: AB_2561595
Anti-Human IgD (IA6-2) APC/Cy7	Biolegend	Cat # 348218; RRID: AB_11203722
Anti-Human IgD (IA6-2) BV785	Biolegend	Cat # 348242; RRID: AB_2629809
Anti-Human PD-1 (EH12.2H7) Alexa Fluor 700	Biolegend	Cat # 329952; RRID: AB_2566364
Anti-Human PD-1 (NAT105) APC	Biolegend	Cat # 367406; RRID: AB_2566067
Anti-Human PD-1 (EH12.2H7) Pacific Blue	Biolegend	Cat # 329916; RRID: AB_2283437
Streptavidin APC conjugate	eBioscience	Cat # 17-4317-82
Streptavidin PE conjugate	eBioscience	Cat # 12-4317-87
Goat Anti-Human IgG Qdot 800 (custom conjugate)	Invitrogen	N/A
Goat Anti-Human IgA Qdot 588 (custom conjugate)	Invitrogen	N/A
Goat Anti-Human IgM QDot 655 (custom conjugate)	Invitrogen	N/A
Anti-Mouse IgG gamma Horseradish peroxidase (HRP)	SeraCare KPL	Cat # 5220-0339
Anti-Influenza A Antibody, nucleoprotein, clone A1	Millipore Sigma	Cat # MAB8257; RRID: AB_95231
Anti-Influenza A Antibody, nucleoprotein, clone A3	Millipore Sigma	Cat # MAB8258; RRID: AB_95232
Goat Anti-Human IgG+IgM+IgA H&L (HRP)	Abcam	Cat # ab102420; RRID: AB_10712551
Bacterial and virus strains
A/California/07/2009 H1N1 virus	Wagar Lab	N/A
A/California/07/09 X A/Puerto Rico/8/1934 reassortant H1N1 virus	BEI	Cat # NR-44004
Biological samples
Healthy Human Tonsils	University of California, Irvine Medical Center; Children’s Hospital of Orange County; Cooperative Human Tissue Network	Table 1
Fetal Bovine Serum, heat inactivated	R&D Systems	Cat # S11550
Chemicals, peptides, and recombinant proteins
Bovine Serum Albumin (BSA)	Fisher Scientific	Cat # BP9700100
Ham’s F12 medium	Gibco	Cat # 31765035
Normocin	InvivoGen	Cat # ant-nr-1
Penicillin-Streptomycin	Gibco	Cat # 15140122
Antibiotic-Antimycotic	Gibco	Cat # 15240096
Lymphoprep	Stemcell	Cat # 7851
DMSO	Millipore Sigma	Cat # D4540-100ML
RPMI1640 with glutamax	Gibco	Cat # 61870036
Nonessential amino acids	Gibco	Cat # 11140050
Sodium pyruvate	Gibco	Cat # 11360-070
Insulin, selenium, transferrin supplement	Gibco	Cat # 41400045
Recombinant human BAFF	Biolegend	Cat # 559602
Recombinant Human IL-12 (p70) (carrier-free)	Biolegend	Cat # 573004
Recombinant Human IFN-γ (carrier-free)	Biolegend	Cat # 570204
Influenza vaccine (LAIV FluMist® Quadrivalent) 2019-2020	medImmune	Cat # NDC 66017-306-01
Influenza vaccine (LAIV FluMist® Quadrivalent) 2021-2022	medImmune	Cat # NDC 66019-310-10
Influenza vaccine (IIV, Fluzone® Quadrivalent) 2019-2020	Sanofi Pasteur	Cat # NDC 49281-419-88
Influenza vaccine (IIV, Fluarix Quadrivalent) 2021-2022	GlaxoSmithKline	Cat # NDC 58160-887-52
Influenza vaccine (IIV, FluAD® Quadrivalent) 2021-2022	Seqirus	Cat # NDC 70461-121-04
Influenza vaccine (Flucelvax® Quadrivalent) 2021-2022	Seqirus	Cat # NDC 70461-323-04
PBS	Gibco	Cat # 10010-023
Sodium azide	Thermo Fisher	Cat # 71448-16
EDTA	Invitrogen	Cat # 15-575-020
Sodium bicarbonate	Millipore Sigma	Cat # S5761
Sodium carbonate	Millipore Sigma	Cat # 223530
TPCK-treated trypsin	Worthington Biochemical	Cat # NC9783694
Eagle’s Minimum Essential Medium (EMEM)	ATCC	Cat # 30-2003
Triton X-100	Thermo Scientific	Cat # 85111
3,3′,5,5′-tetramethylbenzidine (TMB) peroxidase substrate (SureBlue)	Seracare	Cat # 5120-007
Tween-20	Sigma	Cat # P1379-250ML
Hemagglutinin A/California/07/2009 (H1N1)	Immune Technology	Cat # IT-003-SW12ΔTMp
Microarray influenza proteins	See Table S2 for a list of microarray proteins	NA
Critical commercial assays
Zombie Aqua™ Fixable Viability Kit	Biolegend	Cat # 423101
EZ-link micro-NHS-PEG4-biotinylation kit	Thermo Scientific	Cat # 21955
Custom Premix Human Cyto Panel A 36 Plex luminex assay	Millipore Sigma	Cat # HCYTA-60K-36C
Deposited data
Flow Cytometry	This paper; ImmPort	Accession # SDY2947
Microneutralization	This paper; ImmPort	Accession # SDY2947
Raw Data (Used for Figure 6)	ImmPort	Accession # SDY404; SDY400; SDY269; SDY224; SDY1119; SDY270; SDY63; SDY61; SDY56
Experimental models: Cell lines
Madin-Darby Canine Kidney (MDCK) cells	ATCC	Cat # CCL-34; RRID: CVCL_0422
Software and algorithms
R V4.3.1	R-Project	RRID:SCR_001905
ggplot2 v3.4.3	R Package	RRID:SCR_014601
reshape2 v1.4.4	R Package	RRID:SCR_022679
FlowJo v10.8.1	FlowJo	RRID:SCR_008520
cyCombine v0.2.15	R Package	https://github.com/biosurf/cyCombine
xPONENT	Luminex LTG	RRID:SCR_025653
Scikit-learn	Scikit-learn	RRID:SCR_002577
SHAP v0.41	GitHub	RRID:SCR_021362
ssGSEA v4	GenePattern	RRID:SCR_003201
WGCNA v1.71	R Package	RRID:SCR_003302
rstatix v0.7.2	R Package	RRID:SCR_021240
relaimpo v2.2-6	R Package	https://cran.r-project.org/web/packages/relaimpo/index.html
SoftMax Pro v7.1	Molecular Devices	RRID:SCR_014240
BioRender	Biorender	RRID:SCR_018361
Custom code for data analysis	GitHub	https://doi.org/10.5281/zenodo.13334292
Other
96-Well Clear Ultra Low Attachment Microplates	Corning	Cat # 7007
Costar Assay plate 96 well, high-binding	Corning	Cat # 3361

STAR METHODS

Experimental model and study participant details

Acquisition of donor samples

Tonsils from 100 consented individuals undergoing surgery for obstructive sleep apnea/hypertrophy, recurrent tonsillitis, or both were collectively acquired through three sources: University of California Medical Center (UCIMC), Children’s Hospital of Orange County (CHOC), and Cooperative Human Tissue Network (CHTN). Autologous blood was collected the same day as tonsillectomy from UCIMC and CHOC donors when available. For samples acquired from UCIMC and CHOC, all samples were collected in accordance with the University of California, Irvine Institutional Review Board (IRB). Ethics approval was granted by the University of California Irvine IRB (protocol #2020-6075) and all participants provided written informed consent. Samples collected through CHTN were deemed not human subject’s research. In this study, the participant cohort consisted of tissues acquired from both male and female individuals aged 2-52 years (Table 1). Overall, tonsil tissue was healthy in appearance.

Tissue processing and organoid generation

Tonsil tissues were processed and organoid cultures were prepared following previously established protocols^25,26. To summarize, post-surgery tonsil tissues were collected in an antimicrobial solution consisting of Ham’s F12 medium (Gibco) supplemented with 2% heat-inactivated fetal bovine serum (FBS), and 2x Antibiotic-Antimycotic (Anti-Anti, Gibco) for 30-60 minutes at 4°C to allow sufficient time for tissue decontamination. Tonsils were then briefly rinsed with PBS and mechanically dissociated followed by gradient centrifugation (Lymphoprep, Stemcell) to remove cellular debris. Samples were cryopreserved in FBS supplemented with 10% dimethyl sulfoxide (DMSO) and stored in liquid nitrogen until further use. Plasma samples were collected from blood following centrifugation at room temperature and stored at −80°C. For organoid generation, cryopreserved cells were thawed, counted, and plated at a final density of 7.5 x10⁶/ml (200 μl final volume) in ultra-low attachment plates (Corning). Organoids were plated and fed using complete organoid media comprised of RPMI1640 with glutamax (Gibco), 10% FBS, 1x nonessential amino acids (Gibco), 1x sodium pyruvate (Gibco), 1x Antibiotic-Antimycotic (Gibco), 1x Normocin (InvivoGen), 1x insulin, selenium, transferrin supplement (Gibco), and 0.5 μg/ml recombinant human BAFF (prepared in-house from Expi293 cells). For organoid stimulation, influenza antigen dosing was previously optimized based on the induction of B cell activation and Ab production while ensuring organoid viability²⁷. Specifically, the following amounts of influenza antigens were added to immune organoids at the time of culture setup: Live attenuated influenza vaccines (LAIV, FluMist^® Quadrivalent 2019/20, FluMist^® Quadrivalent 2021/22) - 1:2,000 final dilution; Inactivated influenza vaccines (IIV, Fluzone^® Quadrivalent 2019/20, Fluarix^® Quadrivalent 2021/22, FluAD^® Quadrivalent 2021/22) - 1:10,000 final dilution; Subunit influenza vaccine (Flucelvax^® Quadrivalent 2021/22) - 1:10,000 final dilution; A/California/07/2009 H1N1 virus - 2.5 hemagglutination units (HAU), or approximately 2.5 x 10⁵ to 2.5 x 10⁶ plaque forming units (PFU), per culture (Table S3). Cultures were incubated at 37°C with 5% CO₂ with humidity and maintained by replenishing media every other day by exchanging 100μl of culture media with complete organoid media.

Method details

Flow cytometry and Th1 depletion

To harvest organoids for flow cytometric analysis, cells were resuspended and washed with Staining Buffer consisting of 1x PBS, 3% FBS, 2mM EDTA, and 0.05% sodium azide (NaN₃). Cells were then stained for 30 minutes on ice and in the dark following the addition of antibody cocktails (Table S1). All antibody solutions were diluted using Staining Buffer and supplemented with Zombie Aqua live/dead stain (1/200; Biolegend) and Human TruStain FcX (1/50; Biolegend) to prevent nonspecific antibody binding. After incubation, cells were washed with Staining Buffer to remove unbound antibodies and data were collected using an ACEA NovoCyte Quanteon (Agilent) flow cytometer. Cell sorting was performed using a BD FACSAria Fusion instrument. Analysis of cytometry data was performed using FlowJo. All antibody products were acquired from Biolegend unless otherwise specified. For antigen-specific B cell staining, cells were initially labeled for 30 minutes on ice with (2μg/ml) biotinylated A/California/07/2009 (H1N1) hemagglutinin (HA) protein (Immune Technology), before washing three times with Staining Buffer to remove unbound HA. Samples were then stained with streptavidin-conjugated fluorophores and additional lineage/phenotype defining antibodies. Biotinylated HA was generated in house using the EZ-link micro-NHS-PEG4-biotinylation kit (Thermo Scientific) following manufacturer’s guidelines.

Protein microarray

To assess the diversity and cross-reactivity of influenza-specific antibodies in both donor plasma and culture supernatants we utilized a high throughput protein microarray consisting of 168 purified influenza proteins (Table S2). Microarrays were carried out as previously described^87,88. Briefly, human plasma was diluted 1/100, while day 14 culture supernatants were diluted 1/5 in protein array blocking buffer (GVS), followed by overnight incubation with gentle agitation at 4°C. After washing with Tris-buffered saline (TBS) containing 0.05% Tween 20 (T-TBS), the arrays underwent a 1.5-hour incubation with a mixture of custom conjugated goat anti-human IgG Qdot 800 (1/400, Invitrogen), goat anti-human IgA Qdot 588 (1/200, Invitrogen), and goat anti-human IgM QDot 655 (1/400, Invitrogen) in blocking buffer. Subsequent washes were performed three times with T-TBS and three times with TBS, followed by rinsing in water and air drying. Image acquisition and intensity quantification were conducted using the ArrayCAM imager and software (Grace Bio-Labs). Data normalization was carried out by employing a collection of established methods^89,90. In summary, control spots underwent normalization using a quantile-based method, followed by calculating the sum of these control spots. Subsequently, a rescaling factor was determined for each sample by dividing the sum of normalized control spots by the sum of its respective control spots. This factor was then applied to rescale the reactivity of each spot. The specificities and cross-reactivity of flu-specific antibodies were analyzed using R v4.3.1.

ELISA

High-binding assay plates (Corning) were prepared by coating with Fluarix^® Quadrivalent 2021/22 (GlaxoSmithKline) at a final concentration of 2 μg/ml (1/60) in 100mM sodium carbonate/bicarbonate ELISA coating buffer and left overnight at 4°C. Wells were then washed once with a wash buffer consisting of 0.05% Tween-20 in PBS (PBS-T) followed by a 2-hour incubation in PBS with 1% BSA at room temperature to prevent nonspecific binding. Plates were washed once with PBS-T before adding 100μL of day 10 culture supernatants, diluted with PBS (1/40), followed by a one-hour incubation at room temperature. After thoroughly washing wells three times with PBS-T, bound antibodies were then detected using 100μL horseradish peroxidase-conjugated anti-human secondary antibody (diluted 1/7500 in PBS-T) specific to IgM/IgG/IgA (Abcam) for one hour. Finally, wells were thoroughly washed three times with PBS-T to remove unbound secondary antibodies and treated with TMB substrate and 1N HCl. Flu-specific antibody quantities were determined by measuring the A450 using Cytation5 as a semi-quantitative assessment of antibody concentrations.

Microneutralization

Microneutralization assays (MN) in this study were conducted with adaptations from previously outlined protocols⁹¹. Madin-Darby canine kidney cells (MDCKs, ATCC) were cultivated in Eagle’s minimum essential medium (EMEM, ATCC) supplemented with 1% Anti-Anti and 10% heat-inactivated FBS and incubated at 37°C with 5% CO₂. Only early passaged cells were used for MN assays and subculturing of cells occurred upon reaching 80–85% confluency. One day prior to the assay, MDCK cells were subcultured into flat-bottomed 96-well plates at a density of 1.1 × 10⁵ cells/ml in 100μL (1.1 x 10⁴ cells/well). Organoid culture supernatants were prepared at 100μl and diluted (1/2.5) in virus growth media (VGM) consisting of serum-free EMEM supplemented with 0.6% bovine serum albumin (BSA, Sigma-Aldrich) and 1 μg/mL N-p-Tosyl-l-phenylalanine chloromethyl ketone (TPCK)-treated trypsin (Worthington Biochemical) and then serially diluted (two-fold) in VGM, 50μl were left over in all wells. The infectious A/California/07/09 X A/Puerto Rico/8/1934 reassortant H1N1 virus (BEI NR-44004) was diluted to 50 TCID50 per 50 μL in VGM and subsequently added to the serially diluted supernatants, followed by an incubation period of 1 hour at 37°C and 5% CO₂. Replicate control samples consisting of 100μl diluted virus only or 100μl VGM only were also prepared. After incubation, the media from MDCK monolayers was replaced with the serum-virus mixtures and further incubated for 1 hour. Following this, the serum-virus mixtures were substituted with 200 μL of VGM supplemented with 2% FBS, and cells were incubated for 48 hours at 37°C. Post-incubation, media was removed and cells were fixed in 4% paraformaldehyde (PFA) in PBS for 30 minutes, washed once in PBS, and then permeabilized in 0.1% PBS/Triton X-100 (PBS-T) at room temperature for 15 minutes. Next, cells were washed twice with PBS and blocked using 200μl of blocking buffer (3% BSA) in PBS for 1 hour at room temperature. Influenza virus nucleoprotein (NP) was detected utilizing an equal mixture of anti-NP mAbs (Millipore Cat Nos. MAB 8257 and MAB 8258) diluted (1/1000) in blocking buffer, followed by horseradish peroxidase (HRP)-conjugated anti-mouse IgG (KPL, Cat. No. 074-1802) diluted at 1:3000 in blocking buffer. Plates were developed in TMB peroxidase substrate and reactions were halted using 1N HCl. Finally, assays were quantified in an ELISA plate reader at 450 nm using SoftMax Pro 7.1 software.

Organoids cytokine supplementation

Tonsil organoids were generated from intermediate IIV responders, including both children (n=6; ages 4-6) and adults (n=6; ages 19-23), defined by an IIV 21/22 neutralization value (1-AUC) between 0.65 and 0.75. On day 0, organoids were stimulated with IIV 21/22 alone or in combination with recombinant human IL-12 (100 ng/ml), IFNγ (100 ng/ml), or both cytokines. Cytokines were added only at culture initiation. Every other day, 100 μl of culture media was exchanged with fresh complete organoid media. Organoids were harvested and stained on day 10, following standard procedures as described above.

Luminex

Day 10 organoid culture supernatants from Th1 depletion experiments as well as Day 4 IIV 21/22 supernatants from the 100 donor cohort were interrogated using a custom 36-plex human cytokine Luminex assay (Millipore Sigma). Supernatants were diluted 1/2 and plated in duplicate. Plates were processed per manufacturer’s instructions in a 96-well plate format on the MAGPIX multiplexing system (Luminex Corp). Standard curves were then generated using five-parameter logistic regression on the xPONENT software (Luminex Corp).

Neutralizing response donor classification

Donors were defined as low, intermediate, or high responders to a vaccine stimulation based on microneutralization response at day 14. Thresholds were selected based on the AUC (area under the curve) of this response. Donors with a 1-AUC greater than or equal to 0.75 were classified as high responders while donors less than or equal to 0.65 were classified as low responders. Resulting sample sizes for high responders: LAIV 19/20 (n=53), LAIV 21/22 (n=50), A/Cal 2009 (n=17), IIV 19/20 (n=19), IIV 21/22 (n=12), FluAD 21/22 (n=9); sample sizes for low responders: LAIV 19/20 (n=34), LAIV 21/22 (n=23), A/Cal 2009 (n=74), IIV 19/20 (n=66), IIV 21/22 (n=65), FluAD 21/22 (n=71). Intermediate donors between 0.65 and 0.75 were not included in the predictive modeling. For the vaccine modality-specific models, a donor was required to be present in both the high responder and low responder classes for all antigens of a given format (IIV or LAIV) to be considered a high responder or low responder, respectively. Sample sizes for high responders: LAIV (n=41), IIV (n=9); sample sizes for low responders LAIV (n=19), IIV (n=56).

Quantification and statistical analysis

Cytometry data preprocessing

Day 0,4,7,10 and 14 flow cytometry data was preprocessed with the CyCombine v0.2.15 batch correction pipeline⁹². Briefly, the flow cytometry data was compensated, arcsinh cofactors were empirically determined for each marker, and the arcsinh transformation was applied. Next, CyCombine was run with an 8x8 grid size and batch effect reduction was evaluated by inspecting marker histograms across batches. After correction, manual gating was performed with FlowJo and cell composition for each subset was exported as a percent of live cells. The cell composition data was then aggregated for all donors and scaled with StandardScaler function from scikit-learn v1.1.3⁹³.

Model training and evaluation

For every vaccine stimulation we used penalized L2 logistic regression along with Leave-one-out cross-validation (LOOCV), where one sample is used as the test set and the remaining samples are used as the training set. For each train-test iteration, StandardScaler was fitted on the training data, which was then applied to both the training and test data for transformation. Within each training split, five-fold cross-validation was used to tune model hyperparameters with GridSearchCV. Model hyperparameters that were optimized included the regularization strength and the maximum number of iterations. Next, predictions were made on the test set using the logistic regression model that was optimized through grid search. To understand the contribution of each feature, SHAP v0.41 values (Shapley Additive exPlanations) were computed with KernelExplainer for each test set⁹⁴. Models were evaluated with the Area under the Receiver Operating Characteristic (AUROC) metric. Refer to https://doi.org/10.5281/zenodo.13334292 for predictive modeling code.

Th1 in vivo analysis

We used the expression and endpoint data available in the Immune Signatures⁴⁰ data source to investigate Th1 signatures in vivo. Data was filtered based on several parameters to include subjects which had received the inactivated influenza vaccine, were part of the young cohort (age < 50), had gene expression data derived from PBMC at baseline, and had a HAI readout. To construct our eight-gene Th1 cell signature we correlated (biweight midcorrelation) “Th1 up” genes⁴¹ against maximum residual after baseline adjustment (maxRBA), which is a measure of the relative strength of a donor’s antibody response that is independent of pre-existing antibody titer⁹⁵. Next, we used ssGSEA as previously described¹⁶ to generate enrichment scores for the eight-gene signature across all 144 subjects. Co-expression analysis was performed for low and high vaccine responders with biweight midcorrelation as implemented in WGCNA v1.71⁹⁶.

Statistical analysis

All statistical analyses performed in this study were calculated in R v4.3.1 using Rstatix⁹⁷. Proportion of variance explained for each demographic variable was determined with relaimpo v2.2-6⁹⁸. For performing pairwise analyses (cell frequency comparisons, cell kinetics experiments, microneutralization assay, protein microarray, and in vivo Th1 SLEA z-score) we first performed kruskal-wallis tests to identify significant groups followed by two-sided Wilcoxon–Mann–Whitney tests, paired and unpaired when appropriate, and multiple hypothesis corrections (Benjamini-Hochberg). Spearman’s rank correlation tests were used for correlation analyses. Significance for in vivo correlations were determined using biweight midcorrelation and a two-sided t-test. No statistical tests were performed for experiments where four or less donors were used due to limited sample size.

Supplementary Material

1

Document S1. Figures S1-6 and Tables S1 and S3

2

Table S2. Microarray influenza protein list, related to STAR Methods

3

Table S4. Cellular kinetics statistics antigen vs antigen, related to Figures 2B–D

4

Table S5. Cellular kinetics statistics harvest vs harvest, related to Figures 2B–D

HIGHLIGHTS.

• Human tonsils exhibit age-independent cellular heterogeneity

• Host and antigen factors both shape antibody magnitude and quality

• Modeling identified Th1s as a predictor and correlate of organoid IIV response

• Higher pre-vaccination Th1 signatures predict enhanced IIV responses in humans