Lymphatic constraint of germinal centers optimizes protective antibody responses

Summary

Immunization strategies are central to pathogen control, where efficacy relies on antigen distribution and inflammatory context. Lymphatic vessels are necessary conduits for antigen transport that actively remodel in peripheral tissues and draining lymph nodes (LN) in response to their inflammatory context. In the present study, we found that inhibiting viral-induced lymphatic remodeling in skin and LNs prompted germinal center (GC) expansion but, paradoxically, decreased the protective antibody response. Whilst GCs scaled proportionately, larger GCs failed to support proliferative bursts consistent with clonal selection, due to impaired T-B cell interactions. Accordingly, follicle and GC size is evolutionarily conserved across diverse mammals despite increased LN size. Lymphatic vessels controlled GC size through two coordinated processes: lymphatic remodeling in infected skin limited viral transport to the LN, while the resulting permissive environment in the LN supported perifollicular lymphangiogenesis that constrained GC expansion.

Graphical Abstract

eTOC blurb:

Lymphatic vessel networks remodel during infection to regulate antigen and immune cell dynamics, yet whether these changes impact humoral immunity remains unknown. Mudianto et al. demonstrate that lymph node lymphangiogenesis encapsulates activating germinal centers to constrain their size, which enhances clonal selection and protective antibody responses, establishing lymphatic vessels as structural determinants of germinal center fitness.

INTRODUCTION

Conventional immunization strategies remain central to preventing infection and disease, but often fail to generate broadly neutralizing antibodies. Although vaccine design emphasizes antigen and adjuvant selection, the transport and persistence of vaccine components also shape efficacy. The orthopoxvirus vaccinia virus (VACV) generates superior humoral and cellular immunity¹ when delivered by skin scarification rather than by subcutaneous or intramuscular injection². Similarly, slow, sustained immunization enhances T follicular helper (Tfh) cell responses and neutralizing titers of germinal center (GC)-dependent antibodies compared with bolus immunization³. Lipid nanoparticle distribution, expression kinetics, and therapeutic outcomes also depend upon particle size and administration route⁴. Thus, defining endogenous mechanisms that regulate transport may reveal parameters for optimizing current and future vaccines.

Peripheral lymphatic capillaries transport interstitial fluid, molecules, and leukocytes to secondary lymphoid organs⁵ and regulate the distribution of cell-free and cell-bound antigen. Their specialized, discontinuous junctions, termed buttons, permit passive paracellular transport⁶. Within lymph nodes (LN), lymphatic endothelial cells (LECs) support antigen scavenging subcapsular sinus (SCS) macrophages, shape chemokine gradients, and filter molecular species entering the conduit system⁷. However, rather than being static structures, dermal lymphatic capillaries dynamically remodel to restrict passive fluid and virion transport to LNs^8,9. This restriction promotes dendritic cell (DC) migration and CD8⁺ T cell priming, demonstrating that lymphatic remodeling actively shapes anti-viral immune responses^8,9.

High affinity antibodies are generated from germinal centers (GC), spatially organized niches beneath the SCS in which B cells interact with antigen and Tfh cells. Lymph-borne antigen is captured by SCS macrophages and transferred to follicular dendritic cells (FDC)¹⁰, while iterative rounds of mutation and selection expand higher-affinity B cell clones^11,12. Rather than imposing fixed affinity thresholds, GCs enable T cell-dependent competition among available clones^11–13,15. Because small changes in GC initiation, maintenance, and selection can alter clonal competition, infection-induced changes in lymphatic transport and perifollicular lymphangiogenesis^14,16 may influence antibody responses. Here, we test the hypothesis that viral-induced lymphatic remodeling, including peripheral capillary zippering and LN lymphangiogenesis, shape GC dynamics and protective antibody output.

RESULTS

A perifollicular lymphangiogenic response precedes germinal center formation following viral infection

Vaccinia (VACV) infection induces active mechanisms of infection-induced lymphatic remodeling that alter transport and optimize the anti-viral cellular response. Whether these changes in transport also impact on humoral responses⁹ remains unknown. We therefore began to investigate the dynamics of the humoral response within the draining superficial cervical LN following VACV infection by skin scarification. Upon infection by scarification, VACV remains local at the site of infection with peak titers from days 3–5 post infection, does not infect the draining LN, and is cleared by day 15^9,17. Stromal components of the LN undergo remodeling as a function of peripheral tissue inflammation¹⁸, including LN lymphatic vessels. Indeed, in draining LNs we observed lymphatic growth that appeared to invade into the LN cortex encapsulating the activating B cell follicle as early as 5 days post-infection and continued to surround follicles containing active GCs 15 days post-infection (Figure 1A). By pulsing with bromodeoxyuridine (BrdU) at five-day intervals post infection, we observed an early increase in LEC proliferation (Figure S1A and Figure 1B) with LEC numbers accumulating over time (Figure S1B), that preceded fibroblastic reticular cell (FRC) and blood endothelial cell (BEC) expansion (Figure S1C and D) as well as the activation of GC B cells (Figure 1A and C). To investigate the nature of this lymphangiogenic response at higher resolution, we performed single cell RNAseq analysis of sorted LN endothelial cells (CD45⁻CD31⁺) from days 0, 7, 15, and 25 post-infection. We subset on LECs (Prox1⁺Lyve1⁺) (Figure S1E and F) and reclustered to define LEC subsets (Figure 1D). Unbiased clustering identified five LEC clusters¹⁹, including ceiling (Ackr4⁺Foxc2⁺Lyve1⁻), floor (Lyve1^loCd274⁺Madcam1⁺Ccl20⁺), Marco⁺ (Lyve1⁺Marco⁺Cd274⁺) and PTX3⁺ (Lyve1⁺Ptx3⁺Cd274⁻) homeostatic subsets and an inflamed population¹⁹ (e.g. Ccl2, Lgmn, Samhd1, Cxcl9, Ifit1) (Figure S1G and H). These inflamed LECs also exhibited expression of Ptx3 and Marco indicating that likely arose from these homeostatic subsets (Figure S1G). We found that there was enrichment for PTX3⁺ and inflamed LECs relative to other subsets at effector time points, 7–15 days post-infection (Figure 1E and F). PD-L1 and LYVE1 can be used to distinguish four LEC states consistent with the transcriptional clusters observed in single cell RNAseq, including LYVE1⁺PD-L1⁻ (PTX3-like), LYVE1⁺PD-L1⁺ (Marco-like), LYVE1⁻PD-L1⁺ (Floor), and LYVE1⁻PD-L1⁻ (Ceiling) (Figure 1G and S1G). We observed enrichment of both the PTX3⁺ (LYVE1⁺PD-L1⁻) and Marco⁺ (LYVE1⁺PD-L1⁺) LECs relative to the floor and ceiling subsets at day 5 post infection (Figure 1H and I), indicating that these subsets likely contribute to the first proliferative lymphangiogenic response. While previous reports describe perifollicular lymphangiogenic responses in multiple inflammatory contexts^14,16,20, their functional significance remains unknown.

Figure 1. A perifollicular lymphangiogenic response associates with B cell follicle expansion.

(A) Representative images of lymphatic vessels and B cell follicles in draining lymph nodes (dLNs) from wildtype (WT) mice at the indicated times after vaccinia virus (VACV) infection by scarification. LYVE-1, green; B220, magenta; GL-7, cyan. Scale bar, 200 μm. Representative of at least five independent experiments.

(B) Frequency of BrdU⁺ lymph node (LN) lymphatic endothelial cells (LECs; CD45⁻CD31⁺PDPN⁺) after BrdU administration during the indicated 5-day intervals. Each point represents one mouse; one-way ANOVA.

(C) Frequency of GC B cells (CD19⁺B220⁺GL-7⁺CD38⁻) among total B cells (B220⁺) over time after infection. Each point represents one mouse; lines connect means.

(D–F) UMAP projection of Prox1⁺ LN LECs from WT mice at the indicated times after VACV infection and colored by Seurat cluster (D), cluster proportions over time, with days 7 and 15 combined (E), and UMAP density by time point (F). Naive, n = 10 mice; days 7, 15, and 25, n = 5 mice each.

(G–I) Representative flow cytometry plots of LN LECs from naive and day 5 dLNs (G), frequencies of the indicated LEC subsets among total LECs (H), and subset proportions over time (I). Each point represents one mouse. Data are from two independent experiments; two-tailed unpaired Student’s t test in (H).

See also Figure S1.

Lymphatic remodeling constrains the germinal center response and enhances the production of protective antibodies

Prior work has demonstrated that LN lymphangiogenesis depends on VEGFA-induced signaling through VEGFR2 and implicates B cells as a key source for the inflammatory growth cue^16,21. We therefore used the Prox1:Cre^ERT2 to drive the inducible deletion of Vegfr2 (VEGFR2^iProx1) specifically in lymphatic vessels in the adult (Figure S2A and B) to investigate the impact of perifollicular lymphangiogenesis on GC induction and function post infection. Upon Vegfr2 deletion and prior to infection, no changes in general LN structure, lymphatic vessel density (Figure S1C) or follicular and perifollicular stromal cells occurred (Figure S1D and E). After infection, however, and consistent with the literature¹⁶ and our prior work⁸, we found that lymphatic-specific deletion of Vegfr2 was sufficient to blunt the infection-induced lymphangiogenic response (Figure S2F and G) relative to controls (Vegfr2^WT). This decrease in infection-induced proliferation associated with a loss the perifollicular lymphangiogenic response (Figure 2A and B), and the reduction of LYVE1⁺PDL1⁻ and LYVE1⁺PDL1⁺, which include the PTX3^+d and MARCO⁺ LEC subsets, respectively (Figure 2C and D). This reduction in LN lymphangiogenesis led to fewer (Figure 2E) but larger B cell follicles in the first week of infection (Figure 2F), with B cell follicles lacking an encapsulating lymphatic network exhibiting the largest cross-sectional area (Figure 2G). To test whether the enlargement of the B cell follicle resulted from impaired B cell exit from the LN, we transferred congenic, naïve B lymphocytes into infected Vegfr2^WT and Vegfr2^iProx1 and subsequently blocked new entry (αCD62L) to assay retention. We found no difference in transferred B cells retained in the LN (Figure S2H).

Figure 2. VEGFR2-dependent lymphatic remodeling regulates the expansion of the B cell follicle and germinal center response.

(A and B) Representative dLN images (A) and perifollicular lymphatic vessel area (B) in Vegfr2^WT and Vegfr2^iProx1 mice 6–7 days after VACV infection. White box indicates inset. LYVE-1, green; B220, magenta. Scale bar, 200 μm. Each point represents one follicle. Data are from two independent experiments (Vegfr2^WT, n = 8 mice; Vegfr2^iProx1, n = 6 mice); lines indicate means; two-tailed unpaired Student’s t test in (B). Images are representative of at least three independent experiments.

(C and D) Frequency of BrdU⁺ cells in expanding LEC subsets 1 (C) and 2 (D) in Vegfr2^WT and Vegfr2^iProx1 dLNs after BrdU administration from days 0–5. Each point represents one mouse. Data are from two independent experiments (Vegfr2^WT, n = 8 mice; Vegfr2^iProx1, n = 7 mice); lines indicate means; two-tailed unpaired Student’s t test.

(E) Number of follicles per dLN section in Vegfr2^WT and Vegfr2^iProx1 mice at day 7; two-tailed unpaired Student’s t test.

(F and G) B cell follicle area (F) and lymphatic vessel area normalized to follicle area (G) in the dLNs in (B). Each point represents one follicle. Data are from two independent experiments (Vegfr2^WT, n = 8 mice; Vegfr2^iProx1, n = 6 mice); lines indicate means; two-tailed unpaired Student’s t test.

(H and I) Representative three-dimensional reconstructions of GCs (PNA, magenta) and follicular dendritic cells (FDCs; CD35, green) (H) and quantification of GC volume (I) in dLNs 15 days after infection. Scale bar, 200 μm. Each point represents one GC. Data are from one experiment (n = 3 mice per genotype); two-tailed unpaired Student’s t test in (I).

(J–M) Representative images of GC zonal organization (J), GC area (K), light zone area (CD35) relative to total GC area (L), and T follicular helper (Tfh; CD3ε) cell area relative to GC area (M) in dLNs at day 15. LYVE-1, green; GL-7, cyan; CD35, red. Scale bar, 200 μm. Each point represents one GC. Data in (K and L) are from three independent experiments (Vegfr2^WT, n = 8 mice; Vegfr2^iProx1, n = 6 mice); data in (M) are from one experiment (Vegfr2^WT, n = 2 mice; Vegfr2^iProx1, n = 4 mice). Two-tailed unpaired Student’s t test. Images are representative of four independent experiments.

See also Figure S2.

We next asked whether the infection-induced B cell phenotype associated with VEGFR2-dependent lymphatic remodeling impacted GC function. The reduction in B cell follicles seen at early time points persisted, and the GCs that formed in Vegfr2^iProx1 mice were larger 15 days post infection than those from littermate controls (Figure S2I and 2H–K). Despite their size, however, GCs that formed, proportionately scaled their light and dark zones (LZ and DZ, Figure 2J and L and Figure S2M and N), maintained an equal density of follicular helper T cells, (Tfh cells, Figure S2J and 2M), and did not differ in peak activation or overall magnitude of response (Figure S2K and L), though we did note a change in the circularity of the FDC networks (Figure S2O). The increased size of the GC response, however, did not translate to more productive antibody responses. When normalizing to total concentration of circulating immunoglobulin (Ig, Figure 3A), we found a reduction in total VACV-specific antibodies 15 days post-infection (Figure 3B) that was largely the result of the reduction in IgG2b and c isotypes, which are GC-dependent and critical for systemic protection against viral pathogens (Figure 3C)²². This deficiency in anti-VACV IgG2b and c antibody production persisted through memory time points (Figure S3A–D) and provided inferior protection in neutralization assays even when normalized to total VACV IgG concentrations (Figure 3D), indicating that despite their size, GCs induced in mice with impaired lymphatic remodeling were less functional.

Figure 3. VEGFR2-dependent lymphatic remodeling enhances the formation of virus-specific antibodies.

(A–C) Serum total immunoglobulin concentration (A), normalized VACV-specific IgG (B), and VACV-specific IgM, IgG1, IgG2b, and IgG2c (C) in Vegfr2^WT and Vegfr2^iProx1 mice at day 15 after VACV infection. Values in (B and C) are normalized to the median of the corresponding Vegfr2^WT control group. Each point represents one mouse; lines indicate means. Data are from three independent experiments (Vegfr2^WT, n = 8 mice; Vegfr2^iProx1, n = 13 mice); two-tailed unpaired Student’s t test.

(D) In vitro protection conferred by serum collected 40–44 days after infection and normalized to total VACV-specific IgG. Each point represents one mouse. Data are from two independent cohorts; two-tailed unpaired Student’s t test.

(E and F) Viral titers in ears 3 days after VACV rechallenge of immune Vegfr2^WT and Vegfr2^iProx1 mice (E) or WT recipients given serum from Vegfr2^WT or Vegfr2^iProx1 donors before rechallenge (F). Dotted lines indicate the limit of detection. Stacked bars show the proportions of mice with low (<1,000 PFU/mL), medium (1,000–4,500 PFU/mL), or high (>4,500 PFU/mL) titers. Each point represents one mouse; data are from two independent experiments. Lines indicate geometric means in (E) and medians in (F); lognormal Welch’s t test in (E) and two-tailed unpaired Student’s t test in (F).

See also Figure S3.

To test whether this reduction of VACV-specific antibodies had a functional significance in vivo, we performed a homologous rechallenge experiment. Memory Vegfr2^iProx1 and Vegfr2^WT mice were treated with both CD4 and CD8 depleting antibodies to remove protective memory T cells prior to re-infection with VACV on the uninfected, contralateral ear of the mouse (Figure S3F and G). Viral titers were evaluated three days post re-infection, where T cell-independent protection was observed in wildtype mice (Figure S3H). Three days post re-challenge, we found that 3 of 8 Vegfr2^WT mice were completely protected from re-infection, whereas all Vegfr2^iProx1 mice had detectable virus in their skin (Figure 3E). Furthermore, convalescent serum from either Vegfr2^iProx1 or Vegfr2^WT VACV mice was administered to naïve recipients one hour prior to VACV infection (Figure S2H). Three days post-infection, we again observed higher viral titers in the skin of mice receiving Vegfr2^iProx1 serum (22-fold change in median viral load; Figure 3F) compared to serum from Vegfr2^WT donors. We saw no obvious reductions in subcapsular sinus macrophages, marginal reticular cells, and FDCs (Figure S3I and J), that might otherwise implicate inefficient antigen loading of the GC. Therefore, despite the expansion of the GC niche itself, the protective capacity of Vegfr2^iProx1 GC-derived antibodies is inferior to wildtype, demonstrating a direct functional relationship between active lymphatic remodeling, GC size, and anti-viral humoral immunity.

B cell selection is deficient in germinal centers lacking lymphatic constraint

The above data indicated that despite the seemingly normal GC architecture, the efficiency of the GC response was compromised. Therefore, we sought to understand how the dynamics of the GC response might be altered in the context of the Vegfr2^iProx1 mice. To do this, we performed scRNAseq analysis of sorted and GC B cells (B220⁺CD38⁻GL7⁺), plasma cells (PC; B220^loCD138⁺), and Tfh cells (TCRβ⁺CD4⁺PD1⁺CXCR5⁺) pooled from three Vegfr2^WT and three Vegfr2^iProx1 mice 25 days post-infection (Figure S4A). After sub-setting on B cells (Cd19) we performed unbiased clustering and UMAP visualization to identify four clusters^11,23,24, consisting of plasma cells (PC; Sdc1, Xbp1), and GC B cells (GC; Cd19, Bcl6), which included LZ (Cd86, H2-Oa), DZ (Foxo1, Mki67), and pre-memory B cells (Ccr6, Sell) (Figure S4B and C). LZ and DZ cluster annotations were further validated by cell cycle analysis, where DZ cells were cycling (S and G2/M phases) and LZ cells were in G1 arrest, but also showed no statistically significant differences between groups (Figure S4D). Higher resolution clustering revealed eight clusters that highlighted different stages of the LZ to DZ cycle (Figure S4E–H). While there was no differences in the transcriptional profiles of the clusters between groups, proportional analysis showed a significant shift, where GC B cells from Vegfr2^iProx1 were diminished in clusters 5 and 6, which represented the transitory Mki67⁺ LZ/DZ and an early Mki67⁺ proliferating DZ population, respectively (Figure 4A–C). Since the magnitude of B cell activation is known to drive DZ proliferation potential^11,25, a specific reduction of highly activated recent LZ emigrants could result in a deficit in the signals necessary to drive strong proliferation and selection. Therefore, these data prompted us to consider a defect in clonal selection, rather than intrinsic differences in GC B cell transcriptional programming.

Figure 4. Clonal expansion is impaired in the absence of VEGFR2-dependent lymphatic remodeling.

(A–C) High-resolution Seurat clustering of GC B cells with UMAP density by genotype (A), genotype composition of Seurat clusters (B); frequencies of Ki67⁺ LZ/DZ cells (cluster 6) and Ki67⁺ DZ cells (cluster 7) among sequenced B cells (C). Each point represents one mouse; two-tailed unpaired Student’s t test.

(D–G) Representative dLN images (D), GC volume (E), normalized dominance score (NDS) (G), and binned NDS distribution (G) in Aicda-Confetti mice treated with isotype control or VEGFR2-blocking antibody. Labels i and ii identify GCs shown at higher magnification; bracketed values indicate NDS. Scale bars, 200 μm (overview) and 50 μm (magnification). Each point represents one GC. Gray shading indicates NDS > 0.5; dotted lines indicate NDS = 0.5 and 0.75. Data in are from two independent experiments (isotype, n = 4 mice; αVEGFR2, n = 5 mice); two-tailed unpaired Student’s t test. Images are representative of three independent experiments.

(H) NDS in Aid-Confetti mice treated as in (D), 42–44 days after infection and one week after the measurement in (F). Each point represents one GC. n = 2 mice and 2 LNs per group; two-tailed unpaired Student’s t test.

(I and J) NDS (I) and its binned distribution (J) in Aicda-Confetti bone marrow chimeras with Vegfr2^WT or Vegfr2^iProx1 recipients. Each point represents one GC. Data are from three independent experiments (Vegfr2^WT, n = 4 mice; Vegfr2^iProx1, n = 3 mice); two-tailed unpaired Student’s t test in (F).

(K) Representative images of GCs (PNA, magenta) and lymphatic vessels (LYVE-1, green) in dLNs from Aid-Confetti mice treated with isotype control or VEGFR2-blocking antibody. Scale bar, 50 μm. Representative of one experiment (n = 3 mice per group).

(L and M) Numbers of nonsynonymous mutations per cell (L) and ratios of replacement to silent mutations per cell (M). Each point represents one cell; lines indicate means; two-tailed unpaired Student’s t test.

(N) Ratio of nonsynonymous to silent immunoglobulin heavy-chain mutations in individually sorted GC B cells (B220⁺CD38⁻GL-7⁺). Each point represents the mean per-cell ratio for one mouse. Data are from two independent experiments (Vegfr2^WT, n = 7 mice, 278 cells; Vegfr2^iProx1, n = 7 mice, 299 cells); two-tailed unpaired Student’s t test.

See also Figure S4 and S5.

To test the hypothesis that GC selection is compromised in the absence of VEGFR2-dependent lymphatic remodeling, we turned to the use of the Aicda-Confetti multicolor fate-mapping model (Aicda^CreERT2/+.Rosa26^{Confetti/Confetti}), in which the Brainbow cassette can be conditionally activated specifically in GC B cells participating in an ongoing GC reaction. In these mice, GC B cells are stochastically colored at peak GC time points, and the emergence of individual, single-color GCs over time indicates successful clonal selection and emergence of a single, heavily-expanded B cell clone^26,27 (Figure S5A). Aicda-Confetti mice were infected with VACV as previously described and treated with either isotype control or VEGFR2-blocking antibody (DC101), which we previously demonstrated to replicate the Vegfr2^iProx1 phenotype⁸. Here, consistent with our findings in Vegfr2^iProx1 mice, anti-VEGFR2–treated Aicda-Confetti mice showed reduced perifollicular lymphangiogenesis 5 days post-infection (Figure S5B). While in isotype treated controls, we observed the presence of single-color GCs, consistent with strong clonal selection, in VEGFR2-antibody treated mice all GCs maintained multiple colors (Figure 4D) and were larger than controls as measured both by the average radius (Figure S5C) and volume (Figure 4E). We found no difference in GC turnover (entry and exit dynamics) (Figure S5D and E). We quantified selection by normalized dominance score (NDS), which is a measure of clonal dominance within a GC based on color representation and accounts for unlabeled cells within the GC²⁷. An NDS of >0.5 is consistent with clonal selection, whereas an NDS of >0.75 indicates homogenizing selection of a single clone. VEGFR2–antibody treated mice exhibited a marked reduction in NDS compared to controls (Figure 4F and G), generally failing to reach an NDS> 0.5. Though the kinetics of GC B cell expansion are similar in both models (Figure S2K), we extended our analysis window to days 42–44 to see if selection kinetics were simply delayed. Still, we found a significant reduction NDS in treated relative to isotype controls (Figure 4H). These data were validated in bone marrow chimeras, where Aicda-Confetti bone marrow was transferred into either Vegfr2^WT or Vegfr2^iProx1 mice (Figure S5F–H). Again, endpoint NDS indicated an impairment in the ability of individual GCs to undergo clonal selection (Figure S5I and J and 4I and J). Across all experimental conditions, the selected GCs were consistently encapsulated by lymphatic vessels, while large, multi-colored GCs were not (Figure 4K). Finally, to provide further evidence for less effective selection, we sorted and sequenced single GC B cells from Vegfr2^WT and Vegfr2^iProx1. GC B cells in Vegfr2^iProx1 mice, while still undergoing somatic hypermutation (Figure S5K–N), were less likely to select clones with coding mutations that cause amino acid replacements and exert functional impacts on the BCR, and instead had a higher proportion of silent, non-coding mutations (Figure 4L–N). These data together show that the GCs in Vegfr2^iProx1 mice, while structurally normal, were less able to drive competitive clonal selection and thereby exhibited reduced effectiveness of their systemic antibody repertoire.

Germinal center size is optimized to promote T-B cell interactions for clonal selection

The picture emerging from our data was that the failed humoral immunity observed in mice deficient in VEGFR2-dependent lymphatic remodeling depended upon early spatial regulation of the B cell follicle and subsequent GC volume. We therefore hypothesized that the size of the GC structure could optimize B cell dynamics and clonal selection. In agreement with this, GC size negatively correlated with the NDS score even in wildtype mice (radius, μm; Figure 5A) with GCs larger than 160 μm never reaching color dominance (Figure 5B). To mechanistically explore the impact of GC volume on selection dynamics we turned to an in silico approach using a previously developed model of GC selection²⁸. Here, we systematically increased the number of founder cells within the model, proportionately scaling all other parameters. We modeled four GC follicle sizes with increasing reaction sphere radii of 160, 208, 240, and 320 μm, which represented the range in GC sizes present in our VACV data (Figure 5A). The model predicted a reduction in clonal dominance (the percent of GC B cells from the same clone per GC; Figure 5C) and mean GC B cell affinity (Figure 5D) as the GC reaction radius rose above 160 μm. The model further predicted reduced Myc expression (Figure 5E) in GC B cells from larger GCs, despite the scaling of both FDC and Tfh cell networks, indicating a deficiency in the signaling that is required to promote B cell selection and regulate extent of DZ proliferation^29,30. Consistently, we observed reductions in Myc expression, the expression of Myc-target genes, and expression of the related cell-cycle regulator, Ccnd3, in LZ populations from Vegfr2^iProx1 mice relative to controls, particularly in clusters 5 and 6 which were reduced in Vegfr2^iProx1 mice with associated failures in clonal selection (Figure 5F and G and S6A–E).

Figure 5. An optimal germinal center size promotes affinity-based clonal selection.

(A and B) Relationship between GC radius and NDS in VACV-infected Aid-Confetti mice receiving isotype control (A), and NDS stratified by GC radius (<160 or >160 μm) (B). Each point represents one GC. Gray shading indicates NDS > 0.5; dotted lines indicate a radius of 160 μm in (A) and NDS = 0.5 in (B). Data are from two independent experiments (n = 7 mice). Simple linear regression with 95% confidence interval in (A); two-tailed unpaired Student’s t test in (B).

(C–E) Computational modeling of GC dynamics across the indicated GC radii: percentage of the dominant clone, calculated for GCs containing at least 200 cells (C), mean GC B cell affinity (D), and mean Myc level (E), as functions of time after infection or GC initiation.

(F and G) Z-scored Myc expression in B cells from Figure 4, separated by genotype, with boxes indicating Ki67⁺ DZ and Ki67⁺ LZ/DZ clusters (F), and relative expression (log₂ fold change) of Myc, Myc target genes, and Ccnd3 across GC B cell clusters by genotype (G). Numerical p values are shown in (G); Mann–Whitney U test.

(H and I) Representative H&E-stained human axillary and feline mesenteric LNs (H) and GC radii across mammalian species (I). Scale bar, 200 μm. Images are representative of one LN from each of at least three individuals. Each point in (I) represents one GC; data are from at least three LNs per species. Dotted line indicates the mean GC radius (159.63 μm).

(J) GC radius in mouse dLNs 15 days after VACV infection (MuVACV) and human LN biopsies after hepatitis B vaccination (HuHBV). Each point represents one GC (n = 3 mice; n = 4 patients); two-tailed unpaired Student’s t test.

See also Figure S6

The inferior performance of GCs above a radius of 160 μm prompted us to ask whether GC size was maintained in large animals. To answer this question, we collected archived LN tissues from cats, dogs, pigs, cows, horse, and humans, and quantified the average GC radius. Consistent with our modeling data, we found that GCs across all species exhibited an optimal mean radius of approximately 160μm (Figure 5H and I) with GC number rather than individual GC size appearing to scale in larger LNs. This species conservation of GC size was particularly obvious when comparing VACV-induced GCs in mouse to human GCs induced following vaccination against hepatitis B virus (Figure 5J). These data thereby not only support evolutionary conservation of GC size, but importantly, that the lymphatic vasculature plays an active role in setting these size constraints.

Large GC volumes reduce the efficiency of follicular helper T cell interactions

While in silico modeling data paired with scRNAseq analyses identified a deficiency in signaling associated with clonal selection, why this would be the case was less clear. GC dynamics and B cell clonal expansion depends upon the appropriate and timely acquisition of T cell help^12,31–33. While Tfh cell density was preserved in larger GCs, increasing the physical distance of the search space any given B cell would have to traverse to find cognate T cell help, would be expected to lower the efficiency of engaging in successive productive T cell-B cell entanglements over the same fixed timeframe. Indeed, when we turned back to the mathematical model, the predicted failure to reach clonal dominance associated with an increase in the ‘free’ Tfh cell percentage as a function of GC size (Figure 6A), and fewer individual T cell-B cell interactions expected per selected B cell (Figure 6B), indicating a decrease in both the likelihood and magnitude of timely T cell help. Again, predictions from the mathematical model matched our experimental data. Cells sorted from Vegfr2^WT and Vegfr2^iProx1 mice 25 days post-infection were subset on cells expressing both Cd3e and Cd4 and re-clustered to reveal seven subclusters (Figure S6F). Cluster annotations defined cytotoxic CD4⁺ (CD4⁺-CTL), cycling CD4⁺, Ifng⁺ CD4⁺, regulatory T cells (Treg), Bach2-expressing CD4⁺, and Tfh cell clusters, either expressing Il4 or expressing Il21^34–36. PCA analysis indicated few changes in expression across genotypes within each cluster (Figure S6G). However, detailed analysis of the Tfh cell subsets (Cd4⁺Cxcr5⁺Pdcd1⁺Bcl6⁺Icos⁺; Figure S6H–J) defined three Tfh cell clusters (Figure 6C), and a reduced proportion of Il21^high expressing Tfh cells in Vegfr2iProx1 mice (Figure 6D and E and S6K and L). Differential expression and pathway analysis revealed a significant reduction in gene programs associated with Tfh cell function, including MHCII antigen presentation, cell cycle and proliferation, and TCR signaling³⁷ in Tfh cells taken from Vegfr2^iProx1 mice (Figure 6F and S6M). In addition to its feedforward effects on activation of the GC Tfh cell population, IL-21 also acts directly on B cells, augmenting positive selection signals, DZ transition, and increased number of divisions in the DZ³⁸.

Figure 6. Large GC volumes reduce the efficiency of follicular helper T cell interactions.

(A and B) Computational modeling of Tfh–B cell interactions across GC radii: percentage of unengaged Tfh cells over time (A) and mean T–B cell interactions per B cell per day or in total (B).

(C–E) UMAP projection of sorted Tfh cells (TCRβ⁺CD4⁺PD-1⁺CXCR5⁺) at day 21, annotated as IL-21⁺ Tfh cells or either of two IL-4-expressing Tfh populations (C); UMAP density by genotype (D); and frequencies of IL-21⁺ and IL-4⁺ Tfh cells and their ratio (E). n = 3 mice per genotype. Each point in (E) represents one mouse; two-tailed unpaired Student’s t test.

(F) Normalized enrichment scores for MHC class II antigen presentation, cell-cycle, and TCR-signaling gene signatures in Vegfr2^iProx1 versus Vegfr2^WT Tfh cells from (M). Numerical p values were calculated by permutation testing and adjusted using the Benjamini–Hochberg method.

(G–I) Representative histogram of biotin⁺ Tfh cells detected using LIPSTIC^AID in dLNs from isotype- or αVEGFR2-treated mice at day 15 (G), and frequencies of LIPSTIC^hi Tfh cells (H) and biotin median fluorescence intensity (I) at days 15 and 21. Data are from two independent experiments at each time point (per experiment: isotype, n = 2 mice; αVEGFR2, n = 3 mice); two-tailed unpaired Student’s t test in (H and I).

(J) Model illustrating how GC niche size regulates T–B cell interactions and clonal expansion.

See also Figure S6.

To specifically validate the prediction that GC B cells less efficiently engaged Tfh cells in large volume GCs we used the uLIPSTIC in vivo proximity-labeling system, where GC B cells were donors (Aicda-CreERT2; Rosa26^uLIPSTIC, LIPSTIC^AICDA). LIPSTIC^AICDA mice were infected and treated with either isotype or VEGFR2-blocking antibody. At peak GC time points we administered the LIPSTIC substrate (LPETG-biotin) intradermally in the ear and evaluated labeling of Tfh cells (Figure S6N and O). With this strategy, we achieved >99% efficiency of donor GC B cell loading (Figure S6Q). Similar donor loading together with a consistent ratio of GC B cells and Tfh cells (Figure S6P and R) between the two groups, ensured that LIPSTIC labeling reflected acceptor contact and not donor-side differences. In this context, we found that Tfh cells were robustly labelled by GC B cells (Figure 6G). Importantly, however, the percent of Tfh cells that were highly labelled (Figure 6H) and the MFI of all labeled Tfh cells (Figure 6I) was reduced in mice treated with VEGFR2-blocking antibody. Taken together, these data support a model where GC niche size is an essential determinant of clonal expansion; where oversizing the GC impairs efficiency of T cell-B cell engagement, thus reducing positive selection and inertial GC B cell cycling in the DZ^39,40 (Figure 6J).

Viral dissemination to LNs impairs perifollicular lymphangiogenesis and germinal center selection.

The implications of these data are that lymphatic remodeling dynamics optimize clonal selection during GC formation in a two-phase process. Specifically, peripheral lymphatic capillary zippering functions to restrict viral transport to the LN⁸, and perifollicular lymphangiogenesis actively encapsulates activating B cell follicles to create an ideal anatomical niche for GC selection. What remained unclear, however, is whether this selection phenotype could be simply explained by the presence of virus in the draining LN that is observed in Vegfr2^iProx1 mice⁸. We therefore last sought to understand the extent to which virus itself was sufficient for the observed phenotype in order to uncouple the relative effects of peripheral and LN lymphatic remodeling.

To first understand the impact of intact virus we compared mice infected by scarification to a cohort infected by intradermal injection of VACV at the same dose. While scarification induces zippering of peripheral lymphatic capillaries, which restrains viral transport to the LN, intradermal injection rapidly forces virus into naïve lymphatic capillaries through high interstitial fluid pressures, leading to direct infection of the LN (Figure S7A)⁸. No differences in peak skin titers are observed when comparing infections by these two routes⁸. Five days following infection, LNs draining skin infected by intradermal infection demonstrated a loss of perifollicular lymphatic structures (Figure 7A–B) and an increase in B cell follicle area (Figure 7C), consistent with the phenotype observed in Vegfr2^iProx1 mice. These early changes in lymphatic and B cell dynamics consistently translated to a loss of GC efficiency. When evaluating GC time points in Aicda-Confetti mice (Figure 7D), GCs induced by intradermal infections were larger in volume (Figure 7E) and impaired in their ability to select relative to GCs forming in LNs draining skin infected by scarification (Figure 7F and G). These data indicate that the presence of virus in the LN, regulated by VEGFR2-dependent peripheral lymphatic zippering, is central to the observed phenotype and could suggest that perifollicular lymphangiogenesis is a readout of productive B cell activation rather than an active player in constraining or optimizing the response.

Figure 7. Viral dissemination to LNs reduces germinal center efficiency by inhibiting perifollicular lymphangiogenesis.

(A-C) Representative draining lymph nodes (dLN) images (A) and quantification of perifollicular lymphatic vessel (B) and B cell follicle (C) area 5 days after vaccinia virus (VACV) infection by scarification (SS) or intradermal (ID) injection. White box indicates inset. Green, LYVE-1; magenta, B220. Scale bars = 200 μm. Each point represents one follicle. Data are from two independent experiments (n=6 mice, SS; n=8 mice, ID); two-tailed Student’s t-test.

(D-G) Representative dLN images (D), germinal center (GC) volume (E), normalized dominance score (G) in from Aicda-Confetti mice infected by SS or ID injection. Scale bars = 200 μm. Each point represents one GC. Gray shading indicates NDS>0.5; dotted lines indicate NDS=0.5 and 0.75. Data are from two independent experiments (n=4 mice per group); two-tailed unpaired Student’s t test in (E).

(H and I) Feature plot showing expression of tip-like lymphatic vessel gene module in LN lymphatic endothelial cells (LECs) over time after VACV infection by SS (H) and violin plots of the indicated genes across LEC clusters (I).

(J-L) Representative dLN images (J) and quantification of perifollicular lymphatic vessel (K) and B cell follicle (L) area 5 days after ID infection in Ifnar^WT (IR^WT) and Ifnar^iProx1 (IR^iProx1) mice. White box indicates inset. Green, LYVE-1; magenta, B220. Scale bars = 200 μm. Each point represents one follicle. Data are from two independent experiments (n = 6 mice, Ifnar^WT; n = 7 mice, Ifnar^Prox1); two-tailed unpaired Student’s t test.

(M-O) Representative dLN images (M) and quantification of perifollicular lymphatic vessel (N) and B cell follicle (O) area 5 days after ID infection in Ifnar^iProx1 (IR^Prox1) treated with VEGFR2-blocking antibody (αVR2) or isotype control (IgG1). White box indicates inset. Green, LYVE-1; magenta, B220. Scale bars = 200 μm. Each point is one follicle. Data are from one experiment, n=3 mice per group; two-tailed Student’s t-test.

Please also see Figure S7.

Lymphangiogenesis, however, while driven by growth factor signaling (e.g. VEGFR2 and VEGFR3) is also sensitive to negative regulation in inflammatory microenvironments. Both type I and II IFNs are potent inhibitors of LN^41,42 and tumor-associated lymphangiogenesis⁴³. We recently found that in the absence of IFN-dependent control, proliferating LECs acquired a highly invasive transcriptional state⁴³. This transcriptional state is defined by high expression of Ptx3, Mrc1, Ccl2, and Ackr2 and a module of genes associated with a tip-like cell state. It was therefore notable that the LEC population that expanded in VACV-draining LNs following infection by scarification was the PTX3⁺ LEC subset, which shares a similar transcriptional program. We therefore projected a module score based on a tip-like signature that is enriched in proliferative and invasive tumor-associated LECs⁴³, and found that this signature was particularly elevated in the PTX3 parent and inflamed populations at days 7–15 post infection (Figure 7H). Furthermore, while all LEC subsets exhibited similar expression of pro-growth receptors, Kdr and Flt4 (VEGFR2 and VEGFR3, respectively), and that the expression of Vegfa and Vegfc in the LN and specifically B cells in response to infection (Figure S7B and C), there was variable expression of the type I interferon (IFN) receptor (Ifnar1) and the beta chain of the type II IFN receptor (Ifngr2) transcripts, with only the PTX3⁺ LEC subset appearing to be poised for response to both types of IFN in the LN (Figure 7I). Based on these data, we hypothesized that lymphatic sensitivity to growth factor cues could be regulated by the presence of virus in the LN by way of sensing an altered inflammatory environment. Therefore, rather than the loss of lymphangiogenesis seen in Vegfr2^iProx1 mice being driven by the absence of a pro-growth signal, the failure to proliferate could be a consequence of local virus-induced IFNs. We therefore postulated that if we could protect LN lymphatic vessels from an inhibitory, anti-viral IFN signal, we could determine if perifollicular lymphangiogenesis was sufficient to rescue GC control– even in the presence of abundant LN virus. To test this hypothesis, we repeated intradermal VACV infections in mice with lymphatic-specific loss of the type I IFN receptor (Ifnar^1iProx1). Importantly, lymphatic-specific loss of Ifnar1 did not impact peak viral titers in the ear following intradermal infection (Figure S7D). Although, again, intradermal infection failed to trigger perifollicular lymphatic growth in Ifnar1^WT controls, in mice whose LN lymphatic vessels were insensitive to type I IFNs (Ifnar1^iProx1) we saw robust perifollicular lymphangiogenesis (Figure 7J and K), and the subsequent rescue of B cell follicle constraint (Figure 7L). Despite this, however, VEGFR2 inhibition in this setting did not impair perifollicular lymphangiogenesis, indicating that VEGFR2 is not absolutely required for LN lymphangiogenesis and that the main effect of its deletion in our model is on peripheral lymphatic remodeling and viral dissemination (Figure 7M–O). Collectively, these data indicate that lymphatic vessels directly optimize GC dynamics, first by regulating the flow of virus and viral antigens to the LN, which in turn preserves a LN microenvironment favorable to the induction of a perifollicular lymphangiogenic response. Perifollicular lymphangiogenesis then actively constrains GC size to facilitate efficient positive selection by T cells and optimal clonal expansion.

Discussion

Despite the critical role of functional lymphatic transport for the delivery of antigen to draining LNs, whether and how lymphatic vessels actively regulate transport and downstream adaptive immune responses remains poorly understood. We previously found that lymphatic vessels constrain tissue efflux following infection, sequestering virus at the infection site while optimizing antigen presentation and CD8 T cell priming⁸. Here we investigated how viral-induced lymphatic remodeling affects humoral immunity.

Our model identifies lymphatic transport to the LN as a determinant of B cell follicle size and GC dynamics, which may depend in part on changes in antigen dynamics. Peripheral lymphatic capillaries, SCS macrophages, and LN conduits filter peripheral antigens for handoff to B cells¹⁰. Our data indicate that infection-induced cues alter these size exclusion properties at the level of the peripheral lymphatic network^8,44, shifting antigen delivery from passive to active transport. Whether lymphatic zippering places a complete block on soluble antigen transport, or if antigen transport to B cells could be enhanced by dynamin-dependent vesicular transcytosis^45,46 remains unclear. Still, DCs can also capture and re-expose viral particles⁴⁷ and immune complexes² enabling cell-mediated antigen transport to B cells^48,49. This may be particularly important because GC B cells are largely agnostic to soluble antigen stimulation^50,51 using mechanical force to extract antigen from FDCs during affinity discrimination^50,52. A shift toward active transport may further delay the delivery of antigen to FDCs over time, resembling the effects of slow vaccine dosing^3,53–55.

High-affinity antibody production depends on competition among GC B cells for antigen and T cell help. Despite becoming oversized, GCs lacking lymphatic constraint had normal FDC and Tfh cell zonation, V(D)J mutations, and GC B and T cell transcriptional profiles, arguing against direct LEC effects on lymphocytes. Instead, we propose that GC niche size, set in part by lymphangiogenesis, governs clonal selection. GC expansion was associated with impaired accumulation of T cell help and a shift from Il21⁺ to Il4⁺ Tfh cells. IL-21⁺ Tfh cells are considered highly activated, and regulate B cell selection, proliferation, cyclic re-entry and clonal expansion, whereas IL-4⁺ Tfh cells are associated with plasma cell fate decisions^23,34,35,56. Accordingly, clonal bursts were absent in Vegfr2^iProx1 GCs, accompanied by reduced selection of productive mutations, affinity maturation, and protective antibody production.

Our data show that viral entry into the LN inhibits lymphangiogenesis through local type I IFN production; lymphatic-specific Ifnar1 deletion restores both lymphatic growth and GC constraint. Similarly, systemic IFNs restrict stromal remodeling, LN lymphangiogenesis, and protective B cell responses during acute LCMV infection⁵⁷. Thus, LN virus indirectly increases GC size by preventing formation of a bounding lymphatic perimeter. PTX3⁺ LECs likely compose this boundary and exhibit a transcriptional profile that may suggest active mechanisms of B cell engagement. PTX3 released by neutrophils binds splenic marginal zone B cells and promotes IgM and IgG production⁵⁸; acts as a fibroblast growth factor (FGF)-trap and regulates proliferationl⁵⁹; and sequesters cell remnants from APCs⁶⁰, suggesting several potential mechanisms.

The signal initiating perifollicular lymphangiogenesis remains uncertain. B cells can produce VEGFA and induce LN lymphangiogenesis after vaccination¹⁶, while B cell-specific VEGFA, or expansion of the B cell compartment²¹ increases lymphatic vessel networks. Immune complexes similarly induce LN lymphangiogenesis through induction of VEGFA in SCS macrophages and DCs²⁰, which may suggest that early antigen capture initiates concordant lymphatic growth. Consistent with prior work¹⁶, our data indicates that Vegfa transcription is increased in whole LNs and B cells following infection, however, VEGFR2 is not required for LN lymphangiogenesis in the absence of type I IFN signaling. We therefore suggest that the peripheral lymphatic vasculature critically regulates the LN inflammatory context, where the balance between pro- and anti- growth signals shapes LN lymphatic dynamics^41–43 and GC responses.

Together, our data define a lymphatic regulated size limit for productive GCs that is conserved across mammals despite differences in LN size. Natural GCs vary substantially in size, making monoclonality a relatively rare ‘jackpot’ event across infection, immunization, autoimmunity, and mucosal GCs^26,27,61,62. Although extensive clonal selection produces highly-specific, high-affinity antibodies, both primary and memory responses benefit from balancing affinity with breadth. Active lymphatic control of transport and compartmentalization may therefore be an innate determinant of host-pathogen responses and a target for optimizing vaccine biodistribution.

Limitations of the Study

Although our data demonstrate lymphatic regulation of the GC response to VACV, how perifollicular lymphangiogenesis constrains GC size remains unknown. GC fitness may reflect direct and indirect effects of lymphatic remodeling on chemokine gradients, antigen distribution, and stromal networks that require further resolution. Moreover, VACV is a large, complex, enveloped virus; whether other viruses and bacteria are similarly constrained or bypass these mechanisms remains unclear. Determining whether evolved or engineered lymphatic evasion alters protective antibody responses will require further investigation.

Resource Availability

Lead contact:

Further information and questions should be directed to Amanda W. Lund, (amanda.lund@nyulanone.org).

Materials Availability:

The study did not generate new unique reagents.

Data and code availability:

Sequencing data is publicly available at GSE313874 and GSE344762. The full code, parameter files and documentation for the mathematical modeling are available on Zenodo⁶³. Raw files will be made available upon request to the corresponding author. All other data are available in the main text or the supplemental information.

STAR Methods

Mice, Housing, and Animal Procedures

All animal procedures were approved and performed in accordance with institutional guidelines for the care and use of laboratory animals at the New York University Langone Health. Mice were maintained under specific pathogen–free conditions in facilities at NYU, with a 12-hour light/12-hour dark cycle at a constant temperature (20–23 °C) and given ad libitum access to normal chow diet and water.

The following mouse strains were used: C57BL/6J (stock no. 000664), Ifnar1^fl/fl (stock no.028256), and Vegfr2^fl/fl (stock no. 018977) were purchased from The Jackson Laboratory. Prox-1:CreERT2 mice⁶⁴ were provided by V.H. Engelhard (University of Virginia, Charlottesville, VA) in agreement with T. Makinen (Uppsala University, Uppsala, Sweden) and bred to the indicated floxed alleles in house. Aicda^CreERT2 (Aicda^{tm1.1(cre/ERT2)Crey})⁶⁵ and Rosa26^{Confetti 66} (collectively referred to as Aicda-Confetti) and Aicda^CreERT2 Rosa26^{uLIPSTIC 67} were provided by G. Victoria (Rockefeller University, New York, NY) and bred in house. For Prox1CreERT2 mice, Cre recombination was induced by administering 2 mg tamoxifen (Sigma T5648) dissolved in corn oil (Sigma, C8267) at 20 mg ml⁻¹ via intraperitoneal injection (i.p.) for five consecutive days, followed by a one-week rest period before experiments. For AICDA-confetti mice, Cre recombination was induced by two gavages of 10 mg tamoxifen dissolved in corn oil at 50 mg ml⁻¹, 2 days apart.

VEGFR2 blocking antibody (InVivoMAb DC101, Bio X Cell, A250834; 200 μg) was administered intraperitoneally on days 0 and 3. BrdU (Fisher Scientific, H2726006) 2.0 mg/ mouse was injected i.p. on the day of infection and provided in drinking water (0.8 mg/ mL) for five days. For in vivo experiments, age- and sex-matched mice (8–14 weeks old) were used, with at least 3 mice per group. Each experiment was independently repeated at least twice. Littermate controls, tamoxifen-induced controls, or isotype antibody (InVivoMAB anti IgG1, BE0088) controls were used in all experimental designs.

VACV expansion and infection

Vaccinia virus (VACV) was propagated in BSC-40 cells using standard procedures^68,69. For cutaneous infection by scarification, mice were inoculated on the dorsal ear pinna by 25 punctures with a 29-gauge needle following topical application of 5 × 10⁶ PFU VACV in 10 μl PBS. For intradermal infection, 5 × 10⁶ PFU VACV in 10 μl PBS was injected into the ear tip using a Hamilton syringe.

Immunofluorescence microscopy

Tissues were fixed in 1% paraformaldehyde (Thermo Scientific, AAJ19943K2) for 24 hours at 4°C and transferred to 15% sucrose (Sigma, S9378) (overnight at 4°C) followed by 30% sucrose (overnight at 4°C). Tissues were then embedded in O.C.T. Compound (Fisher Scientific, 23-730-571) and frozen in 2-methylbutane (Sigma, 320404) in liquid nitrogen. Whole lymph nodes were cryosectioned at 10–30 μm thickness using a Leica Cryotome, and sections containing GCs at comparable depths from the LN surface were selected for staining. LNs are sliced from top to bottom and when selecting which LN slice to analyze, the first section in which B cell follicles are restricted to the periphery of the LN is chosen. Tissue sections were blocked and permeabilized using 2.5% BSA (Fisher Scientific BP1600), 0.1% Triton X (Sigma, X100), and 0.3 M glycine (Sigma, G8898) in PBS for 30 minutes at RT. Sections were stained with primary antibodies in blocking buffer overnight at 4C. Primary antibodies were purchased from ReliaTech, eBioscience, BioLegend, BD Biosciences, and Abcam: LYVE-1 (103-PA50S), Ki-67 (SolA15), B220 (RA3-6B2), CD3e (145 2C11), CD19 (6D5), CD35 (8C12), PDPN (8.1.1), MadCAM-1 (MECA-367), and CD169 (EPR27102-11).

Slides were washed with PBS two times before stained with secondary antibodies in blocking buffer for 30 minutes at RT. Secondary antibodies were purchased from Invitrogen: anti-Rabbit IgG (H+L) AF647, anti-Rat IgG (H+L) AF647, anti-Rat IgG (H+L) AF546, anti-hamster IgG AF488, and Streptavidin AF546 (Invitrogen). After three washes, slides were mounted with Prolong Diamond Antifade (Invitrogen, P36961) or Vectashield (Vector Laboratories, H-1700) mounting media on 1.5 mm coverslips and imaged on Keyence BX-X810 microscope after curing.

Hematoxylin and eosin (H&E) staining

A sub-cohort of four healthy adult males (ages 35–50) enrolled in a Hepatitis B (HepB) vaccine study at NYU Langone Health (NCT04674462) was analyzed for germinal center size. Two participants with prior HepB vaccination received a single booster dose, while two vaccine-naïve participants received a standard three-dose vaccine series administered at approximately 0, 1, and 6 months after the initial visit. Core needle biopsies were performed 10–21 days after booster or final vaccine dose. Ultrasound-guided biopsies of ipsilateral axillary lymph nodes were performed with an 18-gauge needle by NYU Langone Interventional Radiology. Biopsied cores were immediately placed in RPMI and transported to the NYU Center for Biospecimen Research and Development (CBRD). One core from each participant was formalin-fixed and paraffin-embedded for histologic analysis.

A retrospective search was conducted in the archives of the NYU CBRD (RRID:SCR_017930) for human lymph nodes. A similar search was performed from the Section of Anatomic Pathology, New York State Animal Health Diagnostic Center, Cornell University College of Veterinary Medicine for LNs from other nonhuman, mammalian samples. The keywords “lymph node” and “hyperplastic” were used, yielding three unique cases each from canine, feline, bovine, equine, and porcine samples. All tissues were processed using standard histologic H&E staining protocols, performed by the NYU Langone Experimental Pathology Core for human samples and the Cornell University College of Veterinary Medicine for mammalian samples. Slides were digitally scanned using a Leica AT2 scanner (human samples) or a Ventana DP200 slide scanner (mammalian samples) and analyzed for the presence of germinal centers.

Image analysis

Perifollicular lymphatic vessel (LYVE-1) area was measured by an ImageJ macro pipeline. First, B follicle masks were manually drawn. From there, the macro generates a band with a total 180-micron thickness outside of the drawn B follicle mask and measures LYVE1+ area within the band (perifollicular lymphatic vessel area).

GC diameter from different mammalian species were measured using Animal Health Diagnostic Center Annotation Tools (Cornell University) or ImageJ software. The GC diameter was measured across two perpendicular axes and averaged.

Quantification of selection in AICDA-Confetti mice

Color dominance in AICDA-Confetti GCs was determined in 3D datasets reconstructed using ImageJ software and the Bio-formats plugin. Cells of each color or color combination were counted manually in two or more Z-planes, at least 20μm apart, using the Cell Counter plugin on ImageJ. The normalized dominance score (NDS), a measure of how single-colored a given GC is considering both colored and uncolored cells, was calculated using previously published parameters²⁶. Briefly, for each GC, we calculated the proportion of Confetti-labeled cells by determining the density of fluorescent cells per 100 μm² in anatomically defined DZ regions where cell distribution is homogeneous. The proportion of labeled cells was then multiplied by the fraction of cells expressing the dominant Confetti color combination to establish the true dominance of that color within the entire GC. Fully recombined GCs have a cell density very close to 1, making an NDS of 1 a good approximation of 100% GC occupancy by the dominant color.

Multiphoton imaging

Superficial cervical dLNs were excised and prepared for imaging as described previously²⁶. Briefly, adipose tissue was removed under a dissecting light microscope and LNs were mounted in PBS between two coverslips sealed with vacuum grease. Imaging was performed on an Olympus multiphoton FVMPE-RS with a MaiTai DeepSee tunable Ti:Sapphire laser (Spectra-Physics) set to 930 nm, a 25X/1.05 XLPLN25XWM objective lens with zoom set to a pixel size of 0.8 um, narrow band blue/green/red bandpass filters, and FluoView software.

VACV plaque assay

Ears and LNs were homogenized in 0.5–1 ml RPMI 1640 (HyClone SH30027.FS) supplemented with 1% heat-inactivated FBS using a handheld tissue homogenizer. Homogenates underwent three freeze–thaw cycles with vortexing, followed by serial dilution. Diluted samples were applied to BSC-40 cell monolayers and incubated for 48–72 h under an agarose gel overlay. After overlay gel removal, plaques were visualized by counterstaining with 1% crystal violet (Sigma, C6158) in PBS containing 2% ethanol, and plaque-forming units (PFU) were quantified as previously described⁷⁰.

Neutralization Assay

BSC-40 cells were seeded into 6-well plates and cultured for 24 hours to approximately 90–100% confluency. Serum from mice infected with Vaccinia virus (as previously described) underwent two-step normalization. First, samples were normalized for total IgG concentration, followed by normalization for VACV-specific IgG. The normalized serum samples were serially diluted and incubated with 10⁻⁷ PFU VacV for 1–2 hours at 37°C to allow antibody-mediated neutralization. Following incubation, 200ul of the virus-serum mixtures were added to BSC-40 cell monolayers and incubated for 1 hour at 37°C with gentle rocking every 15 minutes. Afterward, the cells were overlAicda with 2 mL of 1% agarose and MEM supplemented with 2% FBS. Plates were incubated at 37°C for 72 hours. Plaques were quantified as above.

3D lymph node staining, imaging, and analysis

Superficial cervical dLNs were excised and thoroughly removed of any fat. The following protocol for 3D lymph node staining was adapted from⁷¹. Lymph nodes were fixed with 4% PFA for at least 1 hr at 4°C. For methanol pretreatment, fixed LNs were washed sequentially in 50% methanol (in PBS) for 1 hr, 100% methanol for 1 hr, 50% methanol for 1 hr, PBS for 1 hr twice, and PBS with 0.2% Triton X-100 for 1 hr twice at RT. LNs were left in PBS with 0.2% Triton X-100, 20% DMSO, and 0.3 M glycine overnight at RT then blocked in PBS with 0.2% Triton X-100, 10% DMSO, and 6% donkey serum or goat serum or anti-CD16/CD32 (Fc block) 1:400 overnight at RT. LNs were washed in PBS with 0.2% Tween-20 and 10 μg/ mL heparin for 1 hr twice at RT before incubating with primary antibody (Rat anti-CD35) in PBS with 0.2% Tween-20, 10 μg/ mL heparin, 5% DMSO, and 3% donkey or goat serum or Fc block 1:100 for 3 days at RT. LNs were then washed at least 5 times over the course of a day at RT and further incubated with a secondary antibody mix of Donkey anti-Rat Alexa Fluor 488 and PNA-Cy3 antibodies in staining buffer at 1:100 for 3 more days at RT. LNs were again washed at least 5 times over the course of a day at RT, then incubated with 50% methanol for 1 hr and 100% methanol for 1 hr three times. LNs were cleared with a mixture of 1-part benzyl alcohol: 2-parts benzyl benzoate (BABB) overnight at RT and imaged on a Lightsheet (Zeiss Z.1) microscope. Image analysis was performed with Imaris (version 11.1.0).

Flow cytometry

To generate single cell suspensions for lymphocyte analysis, LNs were mechanically dissociated using disposable micropestles (Axygen, PES-15-B-SI) in 100 μL PBS supplemented with 0.5% BSA and 1 mM EDTA (Sigma, EDS-500G) (FACS buffer). For analysis of lymph node stromal cells (LNSC), pooled LNs were minced with surgical scissors and digested sequentially in digestion medium (I) containing collagenase IV (1 mg/ mL; Worthington Biochemical, LS004186) and DNase I (>80 U/ mL; Roche, 04536282001), followed by digestion in medium (II) containing collagenase D (1 mg/mL; Sigma, 11088866001) and DNase I (>50 U/ mL). In all cases, single cell suspensions were filtered through 70-μm strainers, stained with antibody panels diluted in FACS buffer supplemented with Fc-block for 20 min at 4°C, and fixed in 1% paraformaldehyde (PFA) at 4°C overnight prior to flow cytometry analysis. Antibodies were purchased from BioLegend, eBioscience, BD Biosciences, and Tonbo and included: Ghost Dye Red 780 and BV510 for dead cell exclusion, CD8 (53-6.7), CD45 (104), CD4 (GK1.5), VEGFR2 (AVAS12), LYVE-1 (ALY7), CD274 (MIH5), TCR α/β (H57-597), CD3 (17A2), CD138 (281-2), GL7 (GL7), CD31 (MEC13.3), Pdpn (8.1.1), CD38 (HB-7 or 90), PD-1(29F.1A12), CXCR4 (12G5), B220 (RA3-6B2), Fas (Jo2).

Intracellular BrdU staining was performed following surface staining of single-cell suspensions, after which cells were fixed in BD Cytofix/Cytoperm buffer (BD Biosciences, 554722) on ice for 15–30 minutes. Fixed cells were washed in BD perm/wash buffer (BD Biosciences 554723) and incubated on ice in BD Cytoperm Plus buffer (BD Biosciences 561651) for 10 min. Cells were refixed for 5 min and treated with DNAse (300 μg/ mL) for 1.5 h at 37°C and then stained with anti-BrdU (BioLegend Bu20a FITC, 1:25) for 20 min at RT. Data were acquired on a BD Biosciences Symphony A5 flow cytometer and analyzed using FlowJo software (Treestar Inc.).

ELISA

Total immunoglobulin (Ig) in serum from each mouse were quantified and normalized to ensure equal loading for detection of VACV–specific antibodies. Serum Ig concentrations were measured by ELISA using high-binding microplates (Corning, 3690) coated with Goat anti-mouse Ig (1:1000, SouthernBiotech, 1010-01). Serum samples were serially diluted four times, starting at 1:500 with a 4X dilution factor. A standard curve was generated on each plate using serial dilutions of purified total Ig standards, starting at 200 ng/ mL and serially diluted eight times starting at 2X dilution factor. Isotype-specific detection antibodies were used to detect IgG (Invitrogen, cat # 62-6520), IgM (Invitrogen, cat# PA1-84383), IgG2c (Southern Biotech, cat# 1077-05) and IgG2b (Southern Biotech, cat# 1090-05). Sample Ig concentrations were interpolated from the standard curve, and equal amounts of total Ig from each sample were used for subsequent VACV-specific ELISAs. Vaccinia virus–specific ELISAs were then performed using high-binding 96-well plates coated with inactivated lysates from VACV (WR strain)–infected BSC-40 cells (1:800 dilution). Plates were blocked with 1% BSA in PBS for 1 h at room temperature, then incubated overnight at 4 °C with equal amounts of total Ig from each serum sample. Bound antibodies were detected using horseradish peroxidase (HRP)–conjugated goat anti-mouse IgG (Jackson ImmunoResearch, 115-035-003), IgM (115-035-020), IgG2c (115-007-188) +IgG2b (115-007-187). In all assays, ELISAs were developed with TMB (slow kinetic form, Sigma). Optical density was measured at 450 nm with a Flexstation 3 Multi-Mode Microplate Reader.

Bone marrow chimera generation

8–10-weeks old recipient mice received two doses of whole-body radiation (2 × 450 rads, at least 4 h apart) using an x-ray irradiator. Bone marrow was isolated from hind limbs of AICDA-Confetti mice and 5–10 × 10⁶ cells injected retro-orbitally into recipients. Mice were maintained on 2 mg/ mL Ampicillin (Cayman Chemical Company, 14417) antibiotic water changed twice per week for 6 weeks. Mice were rested for a further two weeks without antibiotics prior to performing experimental procedures.

Single cell RNA-sequencing

For endothelial cell (EC) isolation, cervical lymph nodes from 10 naïve female C57BL/6J mice and draining LNs from 5 VACV-infected female C57BL/6J mice were pooled separately. The tissues were then processed into single-cell suspensions as described in the “Flow cytometry” section. Cells were stained with anti-CD31 APC, anti-PDPN PE-Cy7, anti-CD45 PE, and DAPI for live vs. dead discrimination. Endothelial cells (CD31⁺) were enriched via positive selection using anti-APC Microbeads (Miltenyi Biotec, 130-090-855) and passed through LS magnetic columns (Miltenyi Biotec, 130-042-401) following the manufacturer’s protocol.

For GC B cells, PCs, and Tfh cell isolation, single-cell suspensions from 3 VEGFR2^WT and 3 VEGFR2^iProx1 dLNs were processed as separate samples as described in the “Flow Cytometry” section. Prior to sorting, lymphocytes were stained with anti-TCRβ APC-Cy7, anti-CD138 BV650, anti-B220 BUV395, anti-GL7 FITC, anti-CD38 BUV737, anti-CD4 PE, and anti–PD-1 BV785 (Fig. S4). Each mouse sample was further labeled with antibody hashtag oligonucleotide TotalSeq C (BioLegend).

Stained and enriched cells were sorted into RPMI supplemented with 2% FBS using a FACS Aria III (BD Biosciences; 100 μm nozzle for EC or 70 μm nozzle for lymphocytes; <3,000 cells/ second) following sorting strategy: ECs (CD45⁻CD31⁺), GC B cells (TCRβ⁻ B220⁺GL7⁺CD38⁻), PCs (TCRβ⁻B220^mid-loCD138⁺), and Tfh cells (TCRβ⁺CD4⁺PD-1⁺CXCR5⁺) (see Fig. S4). Sorted cells were pelleted, concentrated, and immediately loaded onto the 10x Genomics Chromium platform to generate single-cell gel beads in emulsions. For EC, libraries were prepared using the Chromium Next GEM Single Cell 3′ v3.1 Reagent Kit (16 rxns, PN-1000121; 10x Genomics) according to the manufacturer’s instructions and sequenced on a NovaSeq 6000 (Illumina).

For lymphocytes, libraries were prepared using the Single-cell 5’ Gene Expression kit (10X Genomics) according to the manufacturer’s instructions and sequenced on a NovaSeq X Plus (Illumina). Pooled VEGFR2^WT samples (29,750 cells) and VEGFR2^iProx1 samples (50,400 cells) were loaded into separate lanes. Library preparation and scRNA-sequencing were performed at the NYU Langone Health Genome Technology Center. The demultiplexing, barcoded processing, gene counting, and aggregation were made using the Cell Ranger software (v9) and aligned to the mouse genome (mm10).

Single-Cell RNA-seq analysis

Analysis of single-cell RNA sequencing was performed at NYULMC HPC UltraViolet. Data were processed and visualized using the Seurat (v5.3.0) package in R ((v4.4.2 LEC data sets) and (R4.4.0 lymphocytes data sets)). Quality control filtering was applied to exclude cells with fewer than 100 or more than 10,000 detected genes, >10% (LECs) >5% (lymphocytes) mitochondrial reads, >20% ribosomal protein genes (lymphocytes only), or >25% ribosomal RNA genes. EC datasets from multiple time points post VACV infection were integrated using the Harmony package⁷². The optimal number of principal components was determined using the ElbowPlot function. Cells were clustered using shared nearest-neighbor (SNN) modularity optimization (FindNeighbors and FindClusters) at a resolution of 0.7 for LECs, 0.5 for B cells, and 0.1 for T cells. Gene expression was then normalized using LogNormalize (scale factor = 10,000) and log transformed. LEC identity was confirmed based on the expression of Prox1, and LEC subsets were further classified based on the expression of Ackr4, Foxc2, Lyve1, Ptx3, Marco, Cd274, Madcam1, Ccl21¹⁹. Identity of GC B cells, plasma cells, and Tfh cell was confirmed based on the expression of the surface used for their initial sorting. Gene signature scores were calculated using AddModuleScore. Cell cycle phase assignment (G0–G1 vs. S–G2–M) was performed using the CellCycleScoring function and the cc.genes.updated.2019 dataset provided in Seurat. For replicate-level analyses of lymphocyte datasets, principal component analysis (PCA) was performed using SingleCellExperiment (v1.18.1) after applying a variance-stabilizing transformation with DESeq2 (v1.46.0). Batch effects were corrected using the removeBatchEffect function from limma (v3.62.2), and data visualization was performed with ggplot2.

Gene set and pathway enrichment analysis

Gene set enrichment analysis (GSEA) was performed using fgsea (v1.32.4) with gene sets derived using escape v 2.5.5 from the “C2:CP:REACTOME” category for Mus musculus. Gene sets were generated from significantly differentially expressed genes (DEG) identified by comparing the same cell type across different genetic mouse lines with minimum size of 15. DEGs were ranked by average log2 fold change to generate ranked gene lists, which were subsequently sorted in descending order. Enrichment results were visualized using the enrichplot (v 1.26.6) and ggplot2(v 3.5.2).

Immunoglobulin sequencing

Single GC B cells were index-sorted into 96-well plates containing 5 μl of TCL buffer (Qiagen) supplemented with 1% β-mercaptoethanol^16,37. RNA was reverse-transcribed using oligo-dT as in^16,19. PCR primers were used as described previously²⁷. Pooled PCR products were then purified using SPRI beads (0.7x volume ratio), gel purified and sequenced with a 600-cycle Reagent kit v3 on the Illumina Miseq platform.

Immunoglobulin sequence analysis

Single-cell Igh analyses were carried out as described²⁷. Raw, paired-end sequences were merged at the overlapping regions using PANDASeq (v 2.11) for full amplicon reconstruction, then processed with the FASTX toolkit. Only sequences with high counts for each single cell were analyzed. For V-(D)-J gene rearrangement annotation, Ig sequences obtained were submitted to HighV-QUEST (v. 1.9.5) and Vbase2 databases, choosing the assignment yielding the lowest number of somatic mutations in case of discrepancy. Sequences were then processed in the R environment to group cells with common V_H/J_H genes and same CDR_H3 length into clonal lineages when CDR_H3 nucleotide identity was ≥ 75%. Thereafter, downstream mutational and clonality analyses were performed in Microsoft Excel.

In silico germinal center modeling

We adapted a previously published simulator of the germinal center reaction to test the implications of follicle enlargement on the selection process. A comprehensive description of the simulator is provided in²⁸ (MiXed-model). Modifications made to the original model are described below.

Space representation.

The GC reaction space is represented as a three-dimensional discretized rectangular lattice with lattice constant of Δx = 5μm. Within the lattice, a spherical volume of radius r defines the effective region in which cells are distributed and move. The sphere is conceptually divided into two zones, representing the DZ and LZ. Each lattice node can be occupied by a single cell only.

Space shape for antibodies.

Antibodies are represented on a four (d = 4) dimensional shape space⁷³. The shape space is restricted to a size of 10 positions per dimension, thus, only considering antibodies with a minimum affinity to the antigen. The optimal clone φ* is positioned in the center of the shape space. A position on the shape space φ is attributed to each B cell. The 1-Norm with respect to the optimal clone ${‖\phi -{\phi }^{*}‖}_1={\sum }_{i=1}^d \left|\left({\phi }_i-{\phi }_i^{*}\right)\right|$, i.e., the minimum number of mutations required to reach the optimal clone, is used as a measure for the antigen binding probability. The binding probability is calculated from the Gaussian distribution with width Γ = 2.8⁷⁴:

$$b\left(\phi ,{\phi }^{*}\right)=\text{exp}\left(-\frac{{‖\phi -{\phi }^{*}‖}_1^2}{{\Gamma }^2}\right)$$

Founder cells.

The model is initialized with a fixed number of Tfh cells, FDCs, and stromal cells. B cells enter the GC continuously at a defined rate per hour. For each simulated GC radius r, the total numbers of founder cells and the B cell influx rate are scaled by a volumetric factor (r⁄r_ref)³, where r_ref = 160μm is the reference radius corresponding to the wildtype condition. Approximate scaled values for each condition are as follows: for r160, Tfh cells, 140; FDCs, 200; stromal cells, 300; rate of B cel influx/ hour, 2.0; for r208, Tfh cells, 308; FDCs, 440; stromal cells, 661; rate of B cel influx/ hour, 4.4; for r240, Tfh cells, 473; FDCs, 676; stromal cells, 1013; rate of B cel influx/ hour, 6.76; for r320, Tfh cells, 1120; FDCs, 1600; stromal cells, 2400; rate of B cell influx/ hour, 16.0. Tfh cells are randomly distributed on the lattice and occupy a single node each. Stromal cells are restricted to the DZ, while FDCs are restricted to the LZ. FDCs are represented by their soma occupying one node and six dendrites that extend 40μm from the soma. Dendrites are treated as transparent for B-cell or Tfh-cell migration, while remaining accessible for B-cell interactions. Newly arriving B cells are randomly placed on unoccupied lattice nodes and assigned a shape space position, randomly selected from a predefined set of 100 positions.

Chemokine distribution.

Two chemokines, CXCL12 and CXCL13, are included in the model. CXCL12 is produced by stromal cells in the DZ at a rate of 400nMol/ hour per stromal cell, and CXCL13 by FDCs in the LZ at a rate of 10nMol/hour per FDC. As FDCs and stromal cells are immobile and constant in number within each simulation, the corresponding chemokine distributions were pre-calculated once for each considered GC radius r, and the resulting steady-state distributions were used throughout the simulation. GC B cell sensitivity to chemokine concentrations is held constant throughout.

Implementation and data analysis.

The simulation code was implemented in C++11 (v8.3). Simulation outputs were collected and processed in R (v3.5.2) for analysis and figure generation. The full code, parameter files and documentation are available on Zenodo⁶³.

Statistical analysis

Statistical analysis was performed using GraphPad Prism 10 software. Comparisons between two groups were conducted using unpaired Student’s t tests (parametric). The proper sample size was determined on an ad hoc basis drawing on prior experience. Gene set and significant pathways were identified by adjusting p-values to account for multiple testing using the Benjamini-Hochberg (BH) method. Statistical analysis of sequencing data was performed with pairwise Wilcoxon rank test and Bonferroni correction in R. P values are reported as exact numeric values.

LIPSTIC

Biotin–aminohexanoic acid–LPETGS, carboxy-terminal amide, at 95% purity (biotin–LPTEG), was a gift from the Victora Lab. Stock solutions are prepared in PBS at 20 mM. For uLIPSTIC labelling in cervical LNs, biotin–LPETG was injected intradermally into the ear (10 μl of 5 mM solution in PBS) one time and cervical LNs were collected after 20 min. Mice were briefly anaesthetized with isoflurane during each injection. Single-cell suspensions were prepared as previously described.

B cell Trafficking

Mice were infected in one ear with VACV through scarification as described above. Spleen and LNs from CD45.1 mice were harvested and CD43 dynabeads (Invitrogen, 11422D) used to isolate B cells. 7.5 million CD45.1⁺ B cells were adoptively transferred into mice five days post infection. 12 hours after the transfer, 100μg of CD62L (Bio X Cell, clone:BE0021) was given I.P. 24hrs after CD62L treatment, the mice were sacrificed and superficial cervical LNs excised. The cells were prepped for flow cytometry as described above and the ratio between live cell and CD45.1⁺ cell count remaining in the lymph node measured.

qPCR

LNs were excised and single cell suspensions were prepared. Cell number from each LN was counted and ~5 million cells from the VACV-infected nodes and ~500,00 cells from naïve LNs saved for RNA isolation. From the remaining suspension, CD43 dynabeads (Invitrogen, 11422D) were used to isolate B cells for B cell-specific RNA isolation. RNA isolation was done using the RNeasy Mini Kit (Qiagen, 74104). cDNA was prepared using the Maxima First Strand cDNA Synthesis Kit for RT-qPCR (Thermo, K1642). qPCR was done using 2X Universal SYBR Green Fast qPCR Mix (AB clonal, RK21203). Primers are VEGF-A (FW: 5’-CTGCTGTAACGATGAAGCCCTG, RV: 5’-GCTGTAGGAAGCTCATCTCTCC) and VEGF-C (FW: 5’-CCTGAATCCTGGGAAATGTGCC, RV: 5’-CGATTCGCACACGGTCTTCTGT

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-Sialoadhesin/CD169	Abcam	Cat#ab312840; Clone: EPR27102-11
APC anti-mouse Podoplanin	Biolegend	Cat#127416; Clone: 8.1.1; RRID: AB_2629802
APC anti-mouse MAdCAM-1	Biolegend	Cat#120712; Clone: MECA-367; RRID: AB_2629562
PE anti-mouse/human CD45R/B220	Biolegend	Cat#103208; Clone: RA3-6B2; RRID: AB_312993
Rat anti-CD35	Produced in house	Clone: 8C12
Peanut Agglutinin, Cy3	Vector labs	Cat#CL-1073-1; RRID: AB_2336640
BUV737 Rat Anti-Mouse CD38	BD Biosciences	Cat#741748; Clone: 90/CD38; RRID: AB_2871114
Cd95 Hamster anti Mouse, Brilliant Violet 421	BD Biosciences	Cat#562633; Clone: Jo2; RRID: AB_2737690
Brilliant Violet 785 anti-mouse/human CD45R/B220	Biolegend	Cat#103246; Clone: RA3-6B2; RRID: AB_2563256
TCR beta Armenian Hamster anti Mouse, FITC	eBioscience	Cat#11-5961-82; Clone: H57-597; RRID: AB_465323
Anti-CD4 Rat Monoclonal Antibody (PerCP)	Biolegend	Cat#100538; Clone: RM4-5; RRID: AB_893325
PE anti-mouse CD45.1	Biolegend	Cat#110707; Clone: A20; RRID: AB_313496
Alexa Fluor 647 anti-mouse CD45.2	Biolegend	Cat#109818; Clone: 104; RRID: AB_492870
RY703 Rat Anti-Mouse CD19	BD Biosciences	Cat#571790; Clone: 1D3
Rat monoclonal anti-B220 BUV395	BD Biosciences	Cat#563793; Clone: RA3-6B2; RRID: AB_2738427
Armenian hamster monoclonal anti-TCR beta BUV737	Invitrogen	Cat#367-5961-82; Clone: H57-597; RRID: AB_2896026
Rat monoclonal anti-CD4 BV785	BioLegend	Cat#100551; Clone: RM4-5; RRID: AB_11218992
Rat monoclonal anti-CD8a BV605	BD Biosciences	Cat#563152; Clone: 53-6.7; RRID: AB_2738030
Mouse monoclonal anti-CD90.1 (Thy-1.1) BV421	Biolegend	Cat#202529; Clone: OX-7; RRID: AB_10899572
Rat monoclonal anti-CD185 (CXCR5) BV650	Biolegend	Cat#145517; Clone: L138D7; RRID: AB_2562453
Rat anti-mouse CD38 BB700	BD Biosciences	Cat#742132; Clone: 90/CD38; RRID: AB_2871393
Rat monoclonal anti-DYKDDDDK (FLAG) APC	BioLegend	Cat#637308; Clone: L5; RRID: AB_2561497
Mouse monoclonal anti-biotin PE	Miltenyi Biotec	Cat#130-090-756; Clone: Bio3-18E7; RRID: AB_244258
Hamster Anti-Mouse CD95 PE-Cy7	BD Biosciences	Cat#557653; Clone: Jo2; RRID: AB_396768
CD184 (CXCR4) Monoclonal Antibody APC	eBioscience	Cat#17-9999-42; Clone: 12G5; RRID: AB_1724113
Rat monoclonal anti-CD279 (PD-1) APC-Cy7	BioLegend	Cat#135224; Clone: 29F.1A12; RRID: AB_2563523
BUV395 Mouse Anti-Mouse CD45.2	BD Biosciences	Cat#564616; Clone: 104; RRID: AB_2738867
Brilliant Violet 421™ anti-mouse CD31 Antibody	BioLegend	Cat#102423; Clone: 390; RRID: AB_2562186
PE/Cyanine7 anti-mouse Podoplanin Antibody	BioLegend	Cat#127411; Clone: 8.1.1; RRID: AB_10613294
PE anti-mouse CD309 (VEGFR2, Flk-1) Antibody	BioLegend	Cat#136403; Clone: Avas12; RRID: AB_1967093
APC BrdU Monoclonal Antibody	BioLegend	Cat#339808; Clone: BU20A; RRID: AB_10895898
Biotin LYVE1 Monoclonal Antibody	eBioscience	Cat#13-0443-82; Clone: ALY7; RRID: AB_1724157
BV711 Rat Anti-Mouse CD274	BD Biosciences	Cat#563369; Clone: MIH5; RRID: AB_2738163
APC-eFluor™ 780 TCR beta Monoclonal Antibody	eBioscience	Cat#47-5961-82; Clone: H57-597; RRID: AB_1272173
BV711 Rat Anti-Mouse CD4	BD Biosciences	Cat#563726; Clone: RM4-5; RRID: AB_2738389
PE/Cyanine7 anti-mouse CD8b Antibody	BioLegend	Cat#126616; Clone: YTS156.7.7; RRID: AB_2562777
PE anti-mouse CD3 Antibody	BioLegend	Cat#100205; Clone: 17A2; RRID: AB_312662
APC anti-mouse CD8a Antibody	BioLegend	Cat#100711; Clone: 53-6.7; RRID: AB_312750
BV650 Rat Anti-Mouse CD138	BD Biosciences	Cat#564068; Clone: 281-2; RRID: AB_2738574
BUV395 Rat Anti-Mouse CD45R/B220	BD Biosciences	Cat#563793; Clone: RA3-6B2; RRID: AB_2738427
FITC Rat Anti-Mouse T- and B-Cell Activation Antigen	BD Biosciences	Cat#553666; Clone: GL-7; RRID: AB_394981
PE CD4 Monoclonal Antibody	eBioscience	Cat#12-0041-82; Clone: GK1.5; RRID: AB_465506
Brilliant Violet 785™ anti-mouse CD279 (PD-1) Antibody	BioLegend	Cat#135225; Clone: 29F.1A12; RRID: AB_2563680
APC anti-mouse CD185 (CXCR5) Antibody	BioLegend	Cat#145505; Clone: L138D7; RRID: AB_2561969
Brilliant Violet 711™ anti-mouse/human CD45R/B220 Antibody	BioLegend	Cat#103255; Clone: RA3-6B2; RRID: AB_2563491
PerCP-Cy™5.5 Rat Anti-Mouse CD38	BD Biosciences	Cat#562770; Clone: 90/CD38; RRID: AB_2737782
Purified anti-mouse CD16/32 Antibody (Fc Block)	BioLegend	Cat#101301; Clone: 93; RRID: AB_312800
InVivoMab anti-mouse L-Selectin (CD62L)	Bio X Cell	Cat#BE0021; RRID: AB_1107665
InVivoMAb anti-mouse VEGFR-2	Bio X Cell	Cat#BE0060; RRID: AB_1107766
InVivoMAb rat IgG1 isotype control	Bio X Cell	Cat#BE0088; RRID: AB_1107775
Rabbit Anti-Mouse Lyve-1	ReliaTech	Cat#103-PA50; RRID: AB_2783787
Ki-67 Monoclonal Antibody	eBioscience	Cat#14-5698-82; Clone: SolA15; RRID: AB_10854564
CD45R (B220) Monoclonal Antibody	eBioScience	Cat#14-0452-82; Clone: RA3-6B2; RRID: AB_467254
CD3e Monoclonal Antibody	eBioScience	Cat#16-0031-82; Clone: 145-2C11; RRID: AB_468847
PE anti-mouse CD19 Antibody	BioLegend	Cat#115508; Clone: 6D5; RRID: AB_313643
Biotin Rat Anti-Mouse CD35	BD Biosciences	Cat#553816; Clone: 8C12; RRID: AB_395068
Alexa Fluor™ 647, Goat anti-Rabbit IgG (H+L) Cross-Adsorbed Secondary Antibody	Invitrogen	Cat#A-21244; RRID: AB_2535812
Alexa Fluor™ 647, Goat anti-Rat IgG (H+L) Cross-Adsorbed Secondary Antibody	Invitrogen	Cat#A-21247; RRID: AB_141778
Alexa Fluor™ 546, Goat anti-Rat IgG (H+L) Cross-Adsorbed Secondary Antibody	Invitrogen	Cat#A-11081; RRID: AB_2534125
Alexa Fluor™ 488, Goat anti-Syrian Hamster IgG (H+L) Cross-Adsorbed Secondary Antibody	Invitrogen	Cat#A-21110; RRID: AB_2535759
Donkey anti-Rat IgG (H+L) Alexa Fluor™ 488	Invitrogen	Cat#A-21208; RRID: AB_2535794
Donkey anti Rabbit IgG (H+L) Alexa Fluor 488	Invitrogen	Cat#A-21206; RRID: AB_2535792
Donkey anti Rabbit IgG (H+L) Alexa Fluor 546	Invitrogen	Cat#A10040; RRID:AB_2534016
Goat anti-mouse Ig, Human ads-HRP	Southern Biotech	Cat#1010-05; RRID: AB_2728714
Goat Anti-Mouse Ig, Human ads-UNLB	Southern Biotech	Cat#1010-01; RRID: AB_2794121
Goat anti-Mouse IgG (H+L) Secondary Antibody, HRP	Invitrogen	Cat#62-6520; RRID:AB_2533947
Goat anti-Mouse IgM Secondary Antibody, HRP	Invitrogen	Cat#PA1-84383; RRID:AB_934032
Goat Anti-Mouse IgG2c, Human ads-HRP	Southern Biotech	Cat#1077-05; RRID: AB_2794452
Goat Anti-Mouse IgG2b, Human ads-HRP	Southern Biotech	Cat#1090-05; RRID: AB_2794521
Peroxidase AffiniPure® Goat Anti-Mouse IgG (H+L)	Jackson ImmunoResearch	Cat#115-035-003; RRID:AB_10015289
Peroxidase AffiniPure® Goat Anti-Mouse IgM, μ chain specific	Jackson ImmunoResearch	Cat#115-035-020; RRID:AB_2338502
AffiniPure® Fab Fragment Goat Anti-Mouse IgG2c Fcγ Fragment Specific	Jackson ImmunoResearch	Cat#115-007-188; RRID:AB_2632501
AffiniPure® Fab Fragment Goat Anti-Mouse IgG2b, Fcγ fragment specific	Jackson ImmunoResearch	Cat#115-007-187; RRID:AB_2632500
Anti-APC MicroBeads	Miltenyi Biotec	Cat#130-090-855; RRID:AB_244367
TotalSeq™-C0311 anti-mouse Hashtag 11 Antibody	BioLegend	Cat#155881; Clone: M1/42; 30-F11; RRID:AB_2924482
TotalSeq™-C0312 anti-mouse Hashtag 12 Antibody	BioLegend	Cat#155883; Clone: M1/42; 30-F11; RRID:AB_2924483
TotalSeq™-C0313 anti-mouse Hashtag 13 Antibody	BioLegend	Cat#155885; Clone: M1/42; 30-F11; RRID:AB_2922483
TotalSeq™-C0314 anti-mouse Hashtag 14 Antibody	BioLegend	Cat#155887 Clone: M1/42; 30-F11; RRID:AB_2922484
TotalSeq™-C0315 anti-mouse Hashtag 15 Antibody	BioLegend	Cat#155889 Clone: M1/42; 30-F11; RRID:AB_2924484
Bacterial and virus strains
Vaccinia Virus	Churchill et al.8	Oregon Health and Science University
Chemicals, peptides, and recombinant proteins
Normal Donkey Serum	Jackson ImmunoResearch	Cat#017-000-121
Normal Goat Serum	Jackson ImmunoResearch	Cat#005-000-121
Tween-20	Bio-Rad	Cat#1706531
Triton X-100	Sigma Aldrich	Cat#X100-100ML
Dimethyl sulfoxide (DMSO), Molecular Biology Grade, Liquid, ≥99.9%	Sigma Aldrich	Cat#D8418-50ML
Benzyl benzoate, ReagentPlus®, ≥99.0%	Sigma Aldrich	Cat#B6630-250ML
Benzyl alcohol, anhydrous, 99.8%	Sigma Aldrich	Cat#305197-100ML
Heparin sodium salt, Grade I-A, Powder, From porcine intestinal mucosa, ≥180 USP units/mg	Sigma Aldrich	Cat#H3393-10KU
Pierce™ Methanol-free Formaldehyde Ampules, 16% Formaldehyde (w/v)	Thermo Scientific	Cat#28908
Biotin-aminohexanoic acid-LPETGS, carboxy-terminal amide, at 95% purity (biotin-LPTEG)	Gift from Victora Lab	LifeTein (custom synthesis)
Dynabeads™ Mouse CD43 (Untouched™ B Cells)	Invitrogen	Cat#11422D
Tissue-Tek® O.C.T. Compound, Tissue Tek Oct Compound, 4 oz	Sakura Finetek USA Inc	Cat#4583
Corn oil	Sigma Aldrich	Cat#C8267-500ML
Tamoxifen	Sigma Aldrich	Cat#T5648
5-Bromo-2’-deoxyuridine, 99%	Fisher Scientific	Cat#AAH27260MD
Paraformaldehyde Solution, 4% in PBS	Thermo Scientific	Cat#J19943.K2
Sucrose	Sigma Aldrich	Cat#S9378
2-Methylbutane	Sigma Aldrich	Cat#320404
Bovine Serum Albumin, Heat Shock Treated	Fisher Scientific	Cat#BP1600-100
Glycine	Sigma Aldrich	Cat#G8898
ProLong™ Diamond Antifade Mountant	Invitrogen	Cat#P36961
VECTASHIELD Vibrance® Antifade Mounting Medium	Vector Laboratories	Cat#H-1700-2
DAPI solution (1 mg/mL), 1 mL	Invitrogen	Cat#62248
RPMI 1640 Media	Cytiva	Cat#SH30027FS
Crystal Violet	Sigma Aldrich	Cat#C6158
Ethylenediaminetetraacetic acid	Sigma Aldrich	Cat#EDS
Collagenase, Type 4	Worthington Biochemical	Cat#LS004186
DNase I recombinant	Roche	Cat#4536282001
Collagenase D	Millipore Sigma	Cat#11088866001
Ghost Dye Red 780	Tonbo Bioscience	Cat#SKU 13-0865-T100
Ghost Dye BV 510	Tonbo Bioscience	Cat#SKU 13-0870-T100
Fixation and Permeabilization Solution	BD Biosciences	Cat#554722
Perm/wash buffer	BD Biosciences	Cat#554723
Cytoperm plus buffer	BD Biosciences	Cat#561651
3,3,5,5-Tetramethylbenzidine	Sigma Aldrich	Cat#T4319
Ampicillin (sodium salt)	Cayman Chemical	Cat#14417
Streptavidin Alexa Fluor™ 647 conjugate	Invitrogen	Cat#S21374; RRID:AB_2336066
Streptavidin Alexa Fluor™ 546 conjugate	Invitrogen	Cat#S11225; RRID:AB_2532130
Critical commercial assays
2X Universal SYBR Green Fast qPCR Mix	AbClonal	Cat#RK21203
Maxima First Strand cDNA Synthesis Kit for RT-qPCR	Thermo Scientific	Cat#K1642
RNeasy mini kit	Qiagen	Cat#74104
Deposited data
Raw and processed scRNA-seq data (mouse ECs)	This paper	GEO: GSE313874
Raw and processed scRNA-seq data (mouse lymphocytes)	This paper	GEO: GSE344762
Experimental models: Cell lines
BSC-40	ATCC	CRL-2761 RRID: CVCL_3656
Experimental models: Organisms/strains
Mouse: C57BL/6J	Jackson Laboratories	Strain #: 000664; RRID:IMSR_JAX:000664
Mouse: Vegfr2fl/fl, Prox-1:CreERT2	Churchill et al.8
Mouse: AicdaCreERT2/+.Rosa26Confetti/Confetti	Nowosad et al.27
Mouse: CD45.1 (C57BL/6J-Ptprcem6Lutzy/J (JAXBoy))	Jackson Laboratories	Strain #: 033076; RRID:IMSR_JAX:033076
Mouse: universal LIPSTIC (C57BL/6-Gt(ROSA)26Sortm1(CAG-Thy1,-srta)Vic/J (Rosa26uLIPSTIC))	Nakandakari-Higa et al.67	Gift from Victora lab
Oligonucleotides
Primers for VEGF-A: FW: 5’- CTGCTGTAACGATGAAGCCCTG RV: 5’-GCTGTAGGAAGCTCATCTCTCC	Karakousi et al.43	Custom synthesized by IDT
Primers for VEGF-C: FW: 5’- CCTGAATCCTGGGAAATGTGCC RV: 5’- CGATTCGCACACGGTCTTCTGT	Karakousi et al.43	Custom synthesized by IDT
Primers for GAPDH: FW: 5’-AGGTCGGTGTGAACGGATTTG-3’ RV: 5’-TGTAGACCATGTAGTTGAGGTCA-3’	This paper	Custom synthesized by IDT
Software and algorithms
Imaris (version 11.0.1)	Oxford Instruments	https://imaris.oxinst.com/; RRID: SCR_007370
ImageJ	National Institutes of Health, NY, USA	https://imagej.net/; RRID: SCR_003070
StepOnePlus™ Real-Time PCR System	Applied Biosystems	Cat#43-766-00
FlowJo (version 10.10.0)	FlowJo	http://flowjo.com/; RRID: SCR_008520
Prism 10	Graphpad	https://www.graphpad.com; RRID: SCR_002798
Germinal center reaction simulation and mathematical modeling	Schips et al.63	https://zenodo.org/records/17494281
R version	R Core Team	https://www.r-project.org; RRID: SCR_001905
Other
Lightsheet (Zeiss Z.1)	Zeiss	NYU Langone Microscopy Core
Stellaris 8	Leica	NYU Langone Microscopy Core
Micropestles	Axygen	Cat#PES-15-B-SI
High-binding microplates	Corning	Cat#3690
LS magnetic columns	Miltenyi Biotec	Cat#130-042-401

Highlights:

• Lymphatic remodeling constrains germinal center size during viral infection

• Enlarged germinal centers show reduced Tfh-B cell interaction and clonal selection

• Viral spread to lymph nodes suppresses perifollicular lymphangiogenesis via interferon.

• The optimal germinal center size observed in mice is conserved in large mammals.