Modeling memory B cell responses in a lymphoid organ-chip to evaluate mRNA vaccine boosting

Secondary lymphoid organs (SLO) and tertiary lymphoid structures (TLS) provide the adequate microenvironment for antigen (Ag)-specific B and T lymphocytes to cooperate and develop efficient adaptive immune responses (Victora and Nussenzweig, 2022). Deciphering human immune responses within SLO and TLS has remained challenging. Animal models have provided most of the mechanistic knowledge on the maturation of Ag-specific B cell responses, but they do not fully recapitulate human physiology and, hence, cannot accurately predict immunogenicity in human populations (Mestas and Hughes, 2004). The search for alternative preclinical systems has spurred the development of humanized mouse models, which have yielded valuable insights in terms of pathogenesis, but do not yet develop fully mature SLOs (Chuprin et al., 2023). Fine needle aspirations (FNA) from SLO of human volunteers represent a promising tool as they capture the native lymphoid environment (Röltgen et al., 2022; Havenar-Daughton et al., 2020). However, FNA remain invasive and restrict the evaluation of immune functions due to the collection of a limited number of cells.

In recent years, many efforts have focused on developing human cell-based models of SLO for studying adaptive immune responses in vitro. In a simple implementation, cocultures of human CD4⁺ T cells and B cells were used to demonstrate the key role of T follicular helper cells (Tfh), a specialized CD4⁺ T cell subset, in providing help for B cell maturation in the context of viral infections and autoimmune diseases (Claireaux et al., 2018; Morita et al., 2011). Research then aimed to recreate the three-dimensional (3D) environment typical of SLO, which promotes cell motility and intercellular interactions, thus increasing the chances of encounter between Ag and Ag-specific cells (Lu and Cyster, 2019). Explants of SLO, consisting in blocks of adenoids or tonsils cultivated on collagen sponges, have been used successfully to evaluate HIV replication (Grivel and Margolis, 2009) and the early CD4⁺ T cell recall responses to pertussis toxin (Schmidt et al., 2020). However, tissue explants are not applicable to the study of antibody production due to their short half-lives. More recently, Wagar et al. developed long-term cultures (≥2–3 wk) of dissociated tonsil cells that spontaneously reassembled into lymphoid-like aggregates (Wagar et al., 2021). These tonsil organoids demonstrated class switch recombination and antibody production upon stimulation with several viral vaccines and thus proved useful to compare human immune responses to different vaccination modalities in vitro (Kastenschmidt et al., 2023).

3D cultures based on peripheral blood mononuclear cells (PBMC) have the decisive advantage of allowing the assessment of human immune responses in large cohorts of vaccinated volunteers without the need for invasive lymphoid tissue sampling. A 3D environment mimicking a lymphoid-like architecture can be provided by porous synthetic scaffolds or by a variety of hydrogels that are enriched in extracellular matrix (ECM) components such as collagen or fibrin (Shou et al., 2022; Ozulumba et al., 2023). Researchers who focused on modeling the early steps of antigenic priming developed tissue-like 3D cultures where primed monocytes mature into dendritic cells (DC) upon Ag stimulation and emigrate through an endothelial barrier, before being collected and cocultured with purified T and/or B lymphocytes (Sanchez-Schmitz et al., 2018; Schanen and Drake, 2008). This more physiological approach to DC priming was shown to recapitulate the effect of age on tuberculosis or influenza vaccination, with decreased T and/or B cell responses in cultures from elderly volunteers (Sanchez-Schmitz et al., 2018; Dauner et al., 2017). Several groups have achieved an efficient differentiation of antibody-secreting B cells by bypassing the requirements for CD4⁺ T cell help through the inclusion of CD40L-expressing fibroblasts and cytokines in 3D cultures (Braham et al., 2022; Purwada and Singh, 2017; Unger et al., 2021). The functionalization of a 3D hydrogel by integrin ligands also enhanced antibody secretion upon B cell receptor (BCR) ligation, emphasizing the importance of tissue-dependent signaling for plasmablast (PB) differentiation (Braham et al., 2022; Purwada et al., 2019).

Microfluidic systems that ensure continuous perfusion of 3D cultures further increase lymphocyte adhesion, migration, and differentiation (Ingber, 2022), as physiological levels of shear stress promote integrin function and chemokine production (Tomei et al., 2009; Cinamon et al., 2001). In addition, the continuous renewal of nutrients and controlled oxygenation levels in microfluidic devices enable cultures at high cellular densities that approach those observed in lymphoid organs. Both the increased migration capacity and high cellular density facilitate the encounter of rare Ag-specific CD4⁺ T cells and B cells, highlighting the potential of microfluidic systems for mimicking the germinal center (GC) environment. Using a microfluidic device where purified CD4⁺ T cells and B cells were cocultured within an ECM gel, Goyal et al. showed that fluid perfusion could trigger the self-organization of these cells into lymphoid-like follicles (Goyal et al., 2022). Further, a specific antibody response to an influenza vaccine could be detected in this system after autologous DC addition, without the need for extraneous CD40L or cytokine stimulation. Microfluidics technology thus opens the possibility of evaluating CD4⁺ T cell/B cell interactions in conditions that more closely mimic human physiology and that may better predict humoral responses in vivo.

The COVID-19 pandemic has emphasized the need for preclinical systems that enable a rapid evaluation of humoral responses elicited by candidate vaccines. Due to the successive emergence of SARS-CoV-2 variants and the implementation of various COVID vaccination strategies, the immunologic status of individuals is highly diverse and complex to decipher (Hornsby et al., 2023; Barateau et al., 2023; Reynolds et al., 2022). The possibility to evaluate immune responses at a preclinical level in specific cohorts of vaccinated/infected individuals would clearly facilitate the development of optimal COVID booster vaccines. Hence, we developed a lymphoid organ-chip (LO chip) model adapted to the testing of different COVID vaccine formulations. The LO chip relies on total PBMC embedded in a collagen-based matrix under slow perfusion within a microfluidic chip. This streamlined system promoted a potent amplification of Ag-specific memory B cells and neutralizing antibody production without the requirement for large blood samples or purified immune cell populations. Importantly, the presence of myeloid cells allowed the capture and expression of mRNA vectored by lipid nanoparticles (LNP), enabling the assessment of responses to the new generation of COVID vaccines.

To establish a 3D human lymphoid culture, we used a two-channel microfluidic chip developed by Emulate. The S1 chip is composed of polydimethylsiloxane (PDMS), a polymer that enables gaseous exchange, and is dynamically perfused by the medium through connection to a microfluidic controller. The lower chip channel, devised as a tissue-like compartment, contains human PBMC seeded at a high concentration (6 × 10⁸/ml) within a collagen-based ECM to mimic the high cellular densities achieved in lymphoid tissue. The upper channel is used as a vascular-like compartment to perfuse nutrients and Ag, which can reach the tissue-like compartment by diffusion through a porous membrane (Fig. 1 A). The LO chip model was benchmarked using PBMC from healthy blood donors who had been previously exposed to SARS-CoV-2 and/or to COVID vaccines based on their positive serology for the spike Ag. To determine whether we could induce a spike-specific recall B cell response for these donors, the LO chips were perfused for 6 days with a recombinant spike protein derived from the ancestral SARS-CoV-2 Wuhan strain or with a control Ag, bovine serum albumin (BSA). At the end of the incubation period, the ECM was recovered and digested to allow the analysis of cell phenotype by flow cytometry (see gating strategy in Fig. S1 A).

Upon spike perfusion, the frequency of B cells among total lymphocytes showed only a slight increase from a median of 4.1% in the BSA control condition to 4.5% (Fig. 1 B, left), while the number of B cells remained unchanged (Fig. 1 B, right). In contrast, the frequency of CD27^hi CD38^hi PB among CD19⁺ B cells increased for 16 out of 18 donors tested (example shown in Fig. 1 C, left), with a median PB percentage rising from 0.3% in the control condition to 4.1% in the spike condition (Fig. 1 C, middle). In line with this observation, the number of PB increased ninefold upon spike perfusion (Fig. 1 C, right). We then asked whether these signs of B cell maturation reflected an Ag-specific B cell response. Spike-specific B cells were detected by staining with a fluorescently labeled Wuhan spike protein using the same protein labeled with two distinct fluorochromes to ensure low background staining (Fig. 1 D, left). A marked increase in spike-specific B cells was observed in LO chips from 12 out of 13 donors tested, with a median specific B cell percentage rising from 0.3% to 9.8% of total B cells upon spike stimulation (Fig. 1 D, middle). This corresponded to a median 32.7-fold increase, highlighting the massive amplification of spike-specific B cells after a 6-day culture in the LO chip. Consistently, the median number of spike-specific B cells showed a 35-fold increase (Fig. 1 D, right). The frequency of spike-specific B cells was further enriched in the PB population compared with the total B cell population, reaching a median percentage of 25.5% (Fig. 1 D, middle) and suggesting that a high fraction of antibody-secreting B cells was Ag-specific.

In T-dependent B cell responses, immunoglobulin class switch recombination is triggered by B cell/CD4⁺ T cell interactions at the lymphoid follicle border (Roco et al., 2019). To assess this early step in B cell maturation, we measured the frequency of B cells expressing the IgM and IgG immunoglobulin isotypes (Fig. 1 E, left). We observed a shift from a predominantly IgM⁺IgG⁻ B cell population to a mixed population with an increased frequency of IgM⁻IgG⁺ B cells from a median of 6.9% in the BSA condition to 15.4% upon spike stimulation. The acquisition of an IgM⁻IgG⁺ phenotype was more marked in PB with a median of 45.6% cells expressing surface IgG in this population after spike treatment (Fig. 1 E, right). Thus, antigenic stimulation in the LO chip promoted IgG expression in a significant fraction of B cells. The frequency of spike-specific B cells showed a strong correlation with that of PB (Fig. 1 F; R = 0.65, P < 0.0001) and with that of IgM⁻IgG⁺ B cells (Fig. 1 G; R = 0.67, P < 0.0001), suggesting that a coordinated maturation of the B cell response to the spike Ag took place in the LO chip.

To determine whether spike-specific B cell maturation resulted in detectable antibody production, we measured spike-binding IgG antibodies using the S-Flow assay (Grzelak et al., 2020). This cell-based assay, which measures the amount of human IgG bound at the surface of spike-expressing 293-T cells (Fig. 2 A), has the advantage of detecting antibodies specific for the spike in its native conformation and of being more sensitive than a classic ELISA. The production of spike-specific IgG was induced at day 6 for 12 out of 14 donors in spike-stimulated compared with BSA-stimulated LO chips, with an increase in median MFI (mean fluorescent intensity) from 1.7 × 10³ to 60.3 × 10³ (Fig. 2 B; P < 0.001). These antibody measurements were done in the solution recovered from the chip after non-enzymatic ECM digestion, as the antibody concentrations measured in the chip effluent were low (Fig. S2, A, D, and E), suggesting that secreted antibodies were retained in the ECM. As expected, the amount of spike-binding IgG recovered from the ECM showed a strong correlation with the frequency of spike-specific B cells (Fig. 2 C; R = 0.77, P = 0.0002). Spike-specific IgA production was also induced upon spike stimulation in 7 of 14 donors tested (Fig. S2 B; P < 0.01), reinforcing the notion that class switch recombination took place in the LO chip. The levels of spike-specific IgA appeared however lower and more variable than those measured for spike-specific IgG (Fig. S2 C).

We next evaluated the neutralizing activity of the antibodies produced in the LO chip upon spike stimulation. To this goal, we used the S-Fuse assay, which consists of measuring the inhibition of infection in U2OS-ACE2 cells inoculated by SARS-CoV-2 using a GFP-split cell system (Planas et al., 2021). Neutralization activity was determined by the capacity of antibodies to limit infection and syncytia formation, as measured by a decrease in the GFP signal. Neutralizing activity against the SARS-CoV-2 strain D614G (a variant very close to the ancestral strain) could be detected or 5 out of 13 donors tested in the ECM solution diluted at 1:8 (Fig. 2 D, left). Analyses of positive ECM samples showed a dose-dependent response upon serial dilutions (Fig. 2 D, right), with a measured ID₅₀ (dilution inhibiting 50% of infections) in the 6–44 range. Thus, neutralizing antibody production could be achieved in the LO chip even after a short culture duration of 6 days.

We then aimed to further characterize the phenotype of spike-specific B cells induced in the LO chip. In the control BSA condition, we detected a few spike-specific B cells that had a predominant memory phenotype, as defined by the CD27⁺IgD⁻ marker combination (median = 45.0% memory; Fig. S3 A, top). Naive spike-specific B cells, defined as CD27⁻IgD⁺, were detected at a lower frequency (median 22.7% naive) and presumably corresponded to B cell precursors. The frequencies of naive and memory spike specific B cells observed in the control condition at day 6 were compatible with those reported ex vivo (Reyes et al., 2021; Feldman et al., 2021), suggesting that unstimulated culture conditions in the LO chip did not bias B cell subset distribution. Spike stimulation did not change the distribution of memory and naive subsets within non-specific B cells (Fig. S3 A, bottom), suggesting that antigenic stimulation had a limited effect on bystander B cells. In contrast, spike-specific B cells were enriched in the CD27⁺IgD⁻ memory subset after antigenic stimulation (median = 67.8% memory, Fig. S3 A, bottom), reflecting the amplification of specific memory B cells and possibly the acquisition of a memory phenotype by activated naive B cells.

In the context of a recall response, memory B cells can either differentiate rapidly into antibody-secreting PB and plasma cells or adopt an activated memory B cell phenotype able to enter the GC reaction for a further round of B cell response maturation. B cells that differentiate directly into PB/plasma cells tend to have a higher BCR affinity for Ag compared to those that retain a memory phenotype (Victora and Nussenzweig, 2022). To determine whether these features were recapitulated in the LO chip, we analyzed the phenotype of CD19⁺ B cells in more detail by defining five subsets based on CD27 and CD38 expression, as described (Sanz et al., 2019; Avery et al., 2005). A classic distribution of B cell subsets was observed in LO chips, with the presence of double negative (DN) B cells (CD27⁻CD38⁻), memory B cells (CD27^medCD38⁻), activated memory B cells (CD27^medCD38^med), unswitched B cells (CD27⁻CD38^med), and PB (CD27^hiCD38^hi), as shown in Fig. 3 A. Analysis of spike-stimulated chips showed a significant increase in both PB and activated memory B cells in the spike-specific population compared with the non-specific population from the same chip (Fig. 3 B). In contrast, the resting memory B cell subset and the unswitched B cell subset (which preferentially express IgM; see Fig. 3 C, below) declined in the spike-specific population, while the DN subset remained unchanged. Thus, spike-specific B cells transitioned to more mature states and could adopt either an activated memory or a PB cell fate, like the differentiation pathways documented in lymphoid organs.

We then mapped the intensity of spike-PE binding onto the different B cell subsets within the spike-specific population (Fig. 3 C, right plot). For this analysis, an identical number of spike-specific B cells obtained from eight donors were concatenated (Fig. 3 C, left), resulting in a representative spike-binding pattern. Interestingly, the PB subset showed a higher intensity of spike binding as compared to the four other B cell subsets, which suggested a higher avidity of PB for the spike protein. In contrast, expression of the BCR (both IgG and IgM) and the coreceptor CD19 were lower in PB compared with the other four subsets (Fig. 3 C, middle and right plots), consistent with a transition toward an antibody-secreting phenotype. As the higher spike-binding capacity of PB could not be ascribed to increased BCR expression, these observations suggested that the PB that differentiated in the LO chip had a higher affinity for the spike Ag compared to other B cell subsets.

One of the key markers of B cell maturation in lymphoid tissues is the expression of enzyme activation–induced deaminase (AID), which mediates class-switch recombination and somatic hypermutation. Interestingly, AID expression appeared predominantly detected in spike-specific B cells upon antigenic stimulation and remained low in non-specific B cells (ns; Fig. 3 D, top). Of note, we observed a slight but significant increase of fluorescence in specific B cells using an isotype control antibody, which may be due to the predominance of large cells in the Ag-stimulated specific B cell population (P < 0.05, Fig. 3 D, bottom). We then examined the expression of two additional markers associated with the GC reaction. FAS is a proapoptotic receptor that is upregulated on GC B cells and is thought to control the Ag-dependent selection of high-affinity B cells (Butt et al., 2015). CD21, a coreceptor for the BCR, was shown to be downregulated in recent GC graduates primed for plasma cell differentiation (Lau et al., 2017). We noted that spike-specific B cells showed a strong induction of FAS and an equally strong downregulation of CD21 in the spike-stimulated condition, compatible with a GC phenotype (Fig. 3 E). Altogether, these data suggest that the spike-specific B cell response occurring in the LO chip recapitulates several key features associated with the GC reaction, including the differentiation of memory B cells and PB with different BCR affinity ranges, the expression of the AID enzyme, and the generation of a FAS^hi CD21^lo B cell population.

Another feature of SLO is that differentiated PB exit the lymphoid tissue to join the lymph and blood circulation before homing to their supportive niches in the bone marrow or peripheral sites, while activated memory B cells are retained in GC (Cyster and Allen, 2019). We noted that a fraction of cells seeded into the lower channel of the LO chip egressed to the upper channel during culture after spike stimulation. To explore the nature of these migrating cells, we recovered cells separately from the upper and lower channels at the end of the 6-day spike stimulation and quantified them (Fig. 4 A). As expected, the number of cells found in the upper channel increased upon spike stimulation (mean number of lymphocytes: 1,936 for BSA versus 62,230 for spike, n = 3; Fig. 4 A). The overall distribution of major lymphocyte subsets in the upper compartment did not differ markedly in the BSA and spike conditions, with a predominance of CD4⁺ T cells among emigrated cells. However, we noted a relative enrichment in emigrated B cells upon spike stimulation, from 1.5% in the BSA condition to 5.7% in the spike condition (Fig. 4 A, top row). Phenotyping showed that B cells that egressed upon spike stimulation were highly enriched in PB (median = 71.6% PB in upper channel versus 15.5% in lower channel; Fig. 4 B, left graph) and were also enriched in spike-specific cells (median = 64.8% in upper channel versus 35.6 in lower channel; Fig. 4 B, middle graph). Consistently, most of the egressed spike-specific B cells were PB (Fig. 4 B, right graph). These findings indicate that B cell maturation in the LO chip leads to the preferential egress of PB from the lymphoid compartment, thus recapitulating a key feature of SLO.

CD4⁺ T cells help play a key role in the maturation of the B cell response. We asked whether we could detect activated CD4⁺ T cells with B cell helper capacity based on the co-expression of the activation marker CD38 and the costimulatory receptor ICOS (Herati et al., 2017; Law et al., 2021). The CD4⁺ T cell frequency and number in the LO chip did not change upon spike stimulation (Fig. 5 A). In contrast, a marked increase in the frequency and number of CD38⁺ICOS⁺ CD4⁺ T cells was observed in spike-stimulated LO chips at day 6 compared with BSA-stimulated chips (Fig. 5 B). The frequency of CD38⁺ICOS⁺ CD4⁺ T cells correlated with that of PB, suggesting a concomitant induction of CD4⁺ T cell activation and B cell maturation (Fig. 5 C; R = 0.70, P < 0.0001). We restimulated cells recovered from chips with a pool of overlapping peptides spanning the spike Ag to evaluate the specific CD4⁺ T cell response. Significant induction of CD4⁺ T cells producing TNF-α, IFN-γ, or both cytokines was observed in spike-stimulated chips (Fig. 5 D). IL-2 production was also induced in CD4⁺ T cells upon spike stimulation (Fig. 5 E) and was enriched in the subset of cells that also produced IFN-γ and TNF-α (Fig. 5 F), showing the induction of polyfunctional CD4⁺ T cells. Thus, specific CD4⁺ T cells were amplified in parallel to specific B cells in spike-stimulated LO chips.

To investigate whether CD4⁺ T cells and B cells spatially interacted in the LO chip, we analyzed the chips by fluorescence microscopy at day 6. Low magnification imaging of the lower channel revealed that spike stimulation induced the formation of large cell clusters where CD19⁺ B cells colocalized with CD4⁺ T cells, while cells remained more dispersed upon BSA stimulation (Fig. S4, A and B). B cells aggregated at high density within the clusters, while CD4⁺ T cell density appeared more variable. Interestingly, CD4⁺ T cells expressing the costimulatory marker ICOS were highly enriched within the clusters (Fig. S4 A), suggesting the formation of GC-like areas where activated CD4⁺ T cells interacted with B cells. This notion was confirmed by the observation that the proliferation marker Ki67 was confined to the clusters (Fig. S4 B), suggesting that CD4⁺ T cell/B cell interactions promoted cellular proliferation. Analysis of the clusters at higher magnification confirmed the enrichment of CD4⁺ T cells expressing ICOS within these areas (Fig. 5 G). The ICOS⁺ CD4⁺ T cells had an ameboid shape and were enlarged compared with their ICOS⁻ counterparts, consistent with cellular activation in the ICOS⁺ population. The ICOS⁺ CD4⁺ T cells could form tight contacts with one or more B cells, suggestive of immunological synapse formation (Fig. 5 H). Further, spike stimulation induced the secretion of the CXCL13 chemokine, a marker of GC activity (Fig. 5 I) (Havenar-Daughton et al., 2016). Thus, specific Ag stimulation in the LO chip promoted the formation of cellular clusters that recapitulated several features of GC, including enrichment in CD4⁺ T cell/B cell interactions, immune activation, cellular proliferation, and CXCL13 secretion.

To investigate the effect of continuous fluid perfusion on Ag-specific responses, we compared dynamic cultures in the LO chip to static 3D cultures. To this goal, we seeded an equal number of PBMC (10⁷) within an equal amount of ECM gel (17 μl) either in the lower channel of an LO chip or as a drop in a well of a classic tissue culture plate (Fig. 6 A). After stimulation with Ag, we could compare the reactivity of PBMC grown at a similar density in 3D, either in static or dynamic conditions. Of note, we chose not to perform the static culture within a microfluidic chip, as we reasoned that the restricted diffusion of the nutritive medium through the channels would be insufficient to sustain a high-density culture. With our setup, spike stimulation resulted in a limited induction of PB at day 6 for the in-gel culture, while PB induction was significantly higher for the in-chip culture (Fig. 6 B; P < 0.05). We also observed that simple 2D cultures without ECM resulted in even lower PB induction than in-gel cultures, consistent with a beneficial effect of ECM components and/or of a 3D spatial organization (Fig. S3, B and C).

We then evaluated the effect of fluid perfusion on spike-specific responses. The amplification of spike-specific B cells remained undetectable in gel, while it was detected in four out of five donors tested in-chip (Fig. 6 C). Similarly, an increase in the percentage of IgM⁻IgG⁺ B cells upon spike stimulation was detected in-chip, but not in-gel (Fig. 6 D; P < 0.01). Analyses of spike-specific CD4⁺ T cell responses upon peptide restimulation showed that specific CD4⁺ T cells producing TNF-α and/or IFN-γ could be significantly induced both in-gel and in-chip, though the induction reached higher levels in-chip (Fig. 6 E). Thus, B cells appeared to have a more stringent requirement than CD4⁺ T cells for dynamic culture conditions. This notion was further confirmed by the analysis of spike-binding IgG production, as only B cells from dynamic LO chip cultures produced spike-specific antibodies at day 6, while B cells from static 3D cultures failed to do so (Fig. 6 F). These findings show that microfluidic perfusion in the LO chip significantly contributes to the amplification of B and CD4⁺ T cell memory responses.

We then explored the potential of the LO chip to evaluate recall responses to mRNA vaccines, as this new vaccination modality has proved instrumental in the rapid deployment and success of COVID vaccines (Heinz and Stiasny, 2021). Studies in mouse and non-human primate models suggest that mRNA vaccines vectored by LNP are primarily expressed by APC in vivo, and that mRNA expression peaks at early time points (Verbeke et al., 2021; Kranz et al., 2016; Liang et al., 2017). Therefore, we perfused LO chips with the spike-encoding mRNA vaccine mRNA-1273 (Moderna), harvested the chips at an early time point (day 2), and analyzed the status of different APC populations including CD11c⁺ HLA-DR^hi DC, CD14⁺ monocytes, and CD19⁺ B cells (see gating strategy in Fig. S1 B). BSA protein stimulation was kept as a reference in these experiments.

Spike expression was detected using the human IgG1 mAb s102 directly coupled to the AF647 fluorophore (Planchais et al., 2022; Planas et al., 2021), with the control consisting in the unrelated human IgG1 mAb GO53 coupled to the same fluorophore (Fig. 7 A, top and middle rows). Interestingly, mRNA-1273 treatment caused a significant increase in the frequency of spike+ cells among DC and monocytes, but not among B cells nor CD4⁺ T cells (Fig. 7 A, bottom row), suggesting that only myeloid cells were efficiently targeted by the mRNA vaccine. We also noted a limited but significant increase of DC and monocytes labeled with the control GO53 antibody after mRNA-1273 treatment, suggesting that the mRNA-LNP was sufficient to induce activation of APC that increased non-specific antibody binding, consistent with the reported self-adjuvanting effect of this type of vaccine (Alameh et al., 2021). To verify the nature of cells capable of expressing mRNA-LNP in the LO chip, we used mRNA-LNP that coded for the green fluorescent protein (GFP) and that had similar lipid composition and mRNA modifications as those of Moderna mRNA vaccines (Fig. 7 B, top). After 2 days of GFP mRNA-LNP perfusion, expression of GFP could be detected in DC and monocytes, but not in B cells nor CD4⁺ T cells. No GFP expression was detected after perfusion with an mRNA-LNP encoding the control protein ovalbumin (Fig. 7 B, bottom). These findings confirmed that mRNA-LNP was predominantly expressed by cells of the myeloid lineage.

The analysis of activation markers on day 2 after mRNA-1273 perfusion showed a significant increase in HLA-DR expression in DC and a trend for an increase in monocytes, while no increase was detected in B cells (Fig. 7 C). The induction of HLA-DR was more marked in DC and monocytes that expressed the spike (populations labeled as spike+ in Fig. 7 A), suggesting that these cells could be directly activated by mRNA LNP capture and/or expression. The expression of the costimulation markers CD40 and CD86 was also strongly induced in DC and monocytes treated with mRNA-1273, while it was induced to a lower extent in B cells (Fig. 7 D). Further analysis showed a marked enrichment of CD40/CD86 coexpression in the subset of spike+ DC and monocytes (Fig. 7 D, right), confirming that myeloid cells that captured and expressed the mRNA-1273 LNP had the most prominent signs of activation. In addition, spike+ monocytes showed induction of typical DC markers, namely CD11c and DC-SIGN (Fig. 7 E), suggesting that mRNA vaccine capture could trigger monocyte differentiation toward a DC phenotype. Taken together, these findings indicated that the DC and monocytes present in the LO chip were competent at capturing and expressing mRNA-LNP vaccines. Further, these myeloid cells upregulated activation and differentiation markers, suggesting the induction of Ag presentation capacity.

We then analyzed the induction of spike-specific B cell recall responses upon mRNA vaccine stimulation in LO chips harvested at day 6 (Fig. 8 A). Significant induction of PB was observed upon mRNA-1273 stimulation (Fig. 8 B; P < 0.01), though PB frequency appeared lower than that observed with spike protein stimulation (difference not significant). Similarly, spike-specific B cells were induced by mRNA-1273 stimulation (Fig. 8 C; P < 0.01), with a non-significant trend for lower specific B cell frequencies compared to samples stimulated with the spike protein. Interestingly, we observed that the phenotype of amplified specific B cells differed between the two stimulation conditions, with a predominance of PB among the specific B cells induced by the mRNA vaccine (Fig. 8 D; median PB in spike+ cells: 41.8% for mRNA-1273 versus 12.8% for spike, P < 0.05). Note that only samples with a sufficiently high frequency of spike+ cells were kept for this analysis (4 out of 7 samples in each group, with ≥120 spike+ cells) to ensure a robust phenotypic analysis. In line with PB frequency, the production of spike-specific IgG was higher upon the perfusion of the mRNA-based vaccine than of the spike protein (Fig. 8 E; P < 0.05). The production of spike-specific IgA induced by the mRNA vaccine was also significantly higher (Fig. 8 F; P < 0.05), suggesting the induction of class switch recombination. The finding of increased antibody production appeared generalizable to another mRNA vaccine, as shown by a trend for higher spike-specific IgG production upon stimulation by the BNT162b2 vaccine (Pfizer-BioNTech) compared with the spike protein (Fig. 8 G). Taken together, these findings indicated that mRNA vaccine boosting in the LO chip resulted in the induction of an efficient B cell recall response dominated by mature PB.

The SARS-CoV-2 variant that emerged in late 2021, Omicron BA.1 (or B.1.1529), proved highly divergent from previous variants, with ≥32 non-synonymous mutations in the spike. Consequently, Omicron BA.1 was poorly neutralized by sera from individuals infected by previous variants and/or vaccinated against the ancestral Wuhan strain (Planas et al., 2022). To address this public health challenge, bivalent mRNA vaccines encoding both Wuhan and Omicron spikes were engineered and rapidly deployed worldwide (Tartof et al., 2023; Winokur et al., 2023; Chalkias et al., 2023). We set out to evaluate the in vitro boosting capacity of such a bivalent vaccine, by testing B cell responses induced in the LO chip by the mRNA-1273.214 vaccine (Moderna), made of LNP containing equal amounts of Wuhan and Omicron BA.1 spike mRNA. We generated LO chips with PBMC from seven volunteers who donated blood in 2023 and compared their in-chip B cell responses to stimulation with equivalent doses of mRNA-1273 and mRNA-1273.214 (corresponding to 3 µg mRNA coding for Wuhan S or 1.5 + 1.5 µg mRNA coding for Wuhan S and BA.1 S, respectively). For these experiments, the LO chips were cultured for 14 days rather than 6 days to increase the rate of antibody detection (Fig. S2, D and E), while maintaining adequate cell viability (median: 66% at day 6 versus 48% at day 14), enabling the analysis of the specific B cell response by flow cytometry (Fig. 9, A–C).

B cell maturation was induced by both mRNA vaccines to equivalent levels, as measured by the percentage of PB that persisted at day 14 (Fig. 9 B). To measure specific B cells, we focused the analysis on cells that could bind the receptor binding domain (RBD) of the spike, as this highly variable domain better distinguishes Wuhan and BA.1 specificities, while remaining critical for SARS-CoV-2 neutralization (Addetia et al., 2023). After 14 days of culture, the monovalent mRNA-1273 and the bivalent mRNA1273.214 vaccines induced an equivalent percentage of B cells specific to the Wuhan RBD (Fig. 9 C, top row; medians of 1.49% and 2.53% of Wuhan RBD⁺ cells, respectively). More surprisingly, the monovalent and bivalent vaccines also induced an equivalent percentage of B cells specific to the BA.1 RBD (Fig. 9 C, bottom row; medians of 2.08% and 2.47% of BA.1 RBD⁺ cells, respectively). There were individual instances where the bivalent vaccine induced a higher amplification of RBD-specific B cells than the monovalent vaccine (for instance, donor #5674 represented by a red dot), though the reverse situation could also be observed. In all cases, the pattern observed by measuring Wuhan RBD-specific cells (top row) was similar to that observed by measuring BA.1 RBD-specific cells (bottom row). Further analysis confirmed that measurements of B cells specific for the Wuhan and BA.1 RBD showed a strong positive correlation (Fig. 9 D; R = 0.92, P < 0.0001). These findings suggested that most of the specific B cells induced by monovalent or bivalent mRNA vaccine boosting in the LO chip cross-reacted with both RBD, consistent with the notion of immune imprinting (Addetia et al., 2023).

We next analyzed the levels of antibodies induced in the LO chip by the S-Flow assay and observed the induction of equivalent amounts of spike-specific IgG after a monovalent or bivalent vaccine boost (Fig. 9 E). The levels of antibodies detected correlated positively with the frequencies of B cells specific to the Wuhan RBD (R = 0.79, P < 0.001) and to the BA.1 RBD (R = 0.67, P < 0.01) (Fig. 9 F), consistent with direct production of antibodies by the specific B cells amplified in the chips upon boosting. We then evaluated the neutralizing capacity of antibodies produced in the chips towards the SARS-CoV-2 D614G variant, which is very close to the ancestral Wuhan strain (Korber et al., 2020), and the Omicron BA.1 variant. To this goal, we used the S-Fuse assay, which measures the capacity of antibodies to inhibit the fusion of reporter cells infected with the variants of interest. This assay has been previously validated for measuring neutralization capacity toward a wide array of SARS-CoV-2 variants (Planas et al., 2022; Buchrieser et al., 2020). After mRNA-1273 boosting, antibodies neutralizing the D614G strain could be detected in four out of seven donors (Fig. 9 G, middle), as defined by a half-maximal inhibitory dilution above the threshold value (ID₅₀ > 14). These four donors showed a trend for lower neutralization of BA.1, consistent with a boosting of Wuhan-specific responses. After bivalent mRNA-1273.214 boosting, antibodies neutralizing D614G could be detected in three out of seven donors (Fig. 9 G, right). There was no consistent increase in BA.1 neutralization, even though the bivalent vaccine encoded the BA.1 spike. Overall, boosting with the bivalent vaccine did not confer a significant advantage over the monovalent vaccine in inducing BA.1-specific B cells or BA.1 neutralizing antibodies. These findings are in line with results from in vivo vaccination trials pointing to the limited efficacy of a bivalent boost at inducing Omicron-specific B cell responses (Aguilar-Bretones et al., 2023; Addetia et al., 2023; Yisimayi et al., 2023).

Examination of individual neutralization curves showed various case scenarios (Fig. 9 H). Some donors showed equivalent responses to monovalent and bivalent boosts (left; compare red and orange curves), other donors showed a better response to the monovalent vaccine (middle), pointing to a bias toward ancestral strain Ag, and one donor showed a better response to the bivalent vaccine (right), suggesting recent exposure to Omicron Ag. The LO chip system could thus capture individual variability in responses to an mRNA vaccine boost, reflecting the complexity of the immunological landscape in a given population.

We developed a microfluidic-based system that recapitulates several important features of a recall B cell response in lymphoid tissue, including a massive amplification of Ag-specific memory B cells, a concomitant induction of a specific CD4⁺ T cell response, the spontaneous formation of cellular clusters enriched in CD4⁺ T cell/B cell interactions, the induction of Ig class switch recombination, the maturation of PB with a high affinity for the stimulating Ag, and the spontaneous emigration of these PB from the lymphoid tissue–like compartment. Further, spike stimulation in the LO chip induced the production of specific antibodies in sufficient amounts to evaluate their neutralizing capacity, a key point in developing a preclinical system for vaccine evaluation. The LO chip was designed to include the myeloid cell population naturally present in PBMC, which enabled the capture and expression of mRNA-LNP vaccines. Proof of concept experiments showed that a single bivalent mRNA vaccine boost in the chip was insufficient to shift the specificity of neutralizing antibodies toward the Omicron variant, consistent with in vivo vaccination trials. The LO chip thus represents a novel preclinical model suitable for the evaluation of human B cell responses upon mRNA vaccine boosting. As setting up an LO chip requires only a 10 million human PBMC sample, this system should prove useful to evaluate responses to candidate booster vaccines in a variety of populations that differ in immune competency and infection or vaccination history.

Comparing perfused LO chips to 3D static cultures highlighted the importance of a continuous fluid flow for the maturation of B cell responses. PB differentiation and antibody production occurred early in the LO chip after Ag boosting (day 6), with kinetics similar to that observed in vivo (Wrammert et al., 2008). In contrast, PB showed only a limited increase, and specific antibodies were undetectable in static 2D and 3D cultures. CD4⁺ T cell responses appeared less dependent on fluid flow, as cytokine-producing spike-specific CD4⁺ T cells were significantly increased in the static 3D gel culture, though at lower levels than in the LO chip. The particularly high metabolic demands on PB that ramp up antibody secretion (Boothby et al., 2022) may make these cells more dependent on the continuous supply of nutrients and oxygen provided by microfluidic devices. Another factor promoting B cell maturation may be the continuous perfusion of Ag, as studies in animal models show that slow Ag delivery over several days is superior to bolus Ag injection to sustain GC activity and antibody maturation (Lee et al., 2022). The optimized environment provided in the microfluidic chip may account for the early detection of antibody production (day 6), while 9–12 days are usually required in static 2D cultures, which in addition require stimulation with superantigens rather than classic Ag (Claireaux et al., 2018; Morita et al., 2011). It was noteworthy that PB differentiated in the LO chip bound the spike Ag more efficiently than activated memory B cells from the same chip, as measured by flow cytometry labeling. This higher Ag binding capacity could not be accounted for by an avidity effect, as IgG/IgM surface expression was decreased rather than increased in PB. This suggested that the higher Ag binding capacity of PB resulted from a higher affinity of their surface Ig, which mimicked the enhanced affinity of PB that differentiates within GC in vivo (Victora and Nussenzweig, 2022; Cyster and Allen, 2019). The fact that PB differentiated in the LO chip preferentially egressed from the tissue-like compartment to join the perfused compartment also recapitulated an important aspect of B cell maturation in vivo (Lu and Cyster, 2019). The migratory behavior of PB in the LO chip may also have practical applications for monoclonal antibody production, as B cells harvested from the upper chip compartment are enriched in high-affinity Ag-specific cells, which should facilitate the cloning of BCR of interest.

A CD4⁺ T cell/B cell dialogue was established in the LO chip upon Ag perfusion, as indicated by the concomitant induction of CD4⁺ T cell and B cell–specific responses, and the positive correlation between PB induction and ICOS⁺ CD38⁺ CD4⁺ T cell frequencies. The spontaneous organization of CD4⁺ T cell/B cell clusters upon Ag perfusion highlighted a direct interaction between the two cell types, with tight contacts suggestive of immunological synapse formation. Consistent with this notion, the CD4⁺ T cells found within the clusters had an activated phenotype, based on large size and ameboid morphology, and expressed the costimulatory molecule ICOS, a marker upregulated in Tfh cells (Ma et al., 2012), and were involved in driving B cell maturation (Liu et al., 2015). The CD4⁺ T cells within the clusters were thus likely engaged in providing help to B cells, similar to their function in GC. The induction of CXCL13 chemokine expression, a well-established marker of GC Tfh activity (Havenar-Daughton et al., 2016), further supports this notion.

Ag perfusion in the LO chip induced signs associated with a GC reaction, such as the induction of the proapoptotic FAS receptor in Ag-activated B cells, which can make these cells susceptible to Ag-dependent selection (Butt et al., 2015). The expression of the AID enzyme also suggests that somatic hypermutation can take place in these activated B cells. However, we have yet to demonstrate B cell affinity maturation in the LO chip, which remains a key objective for future studies. Of note, we did not detect a segregation between a light zone enriched in activated CD4⁺ T cells and a dark zone enriched in proliferating B cells within the clusters. Therefore, the LO chip recapitulates some important features of GC, but not all of them, and may be further improved, for instance by the inclusion of stromal cells that would structure lymphoid tissue territories through the secretion of chemokines (Lu and Cyster, 2019). The current LO chip model retains the advantage of providing CD4⁺ T cell help in an Ag-specific fashion in contrast to systems requiring superantigen stimulation, CD40L-expressing fibroblasts, or exogenous cytokine addition (Polini et al., 2019; Ozulumba et al., 2023; Shou et al., 2022). Further, the LO chip does not show spontaneous T cell or B cell activation in the absence of a cognate Ag and thus appears well-suited to the combined evaluation of Ag-specific T and B memory responses from human donors.

We could also monitor the activation of myeloid cells in LO chips. By harvesting the chips at day 2, we documented the early induction of activation (HLA-DR) and costimulation (CD40, CD86) markers on DC and monocytes after mRNA vaccine treatment. Labeling with an anti-spike antibody showed these two cell types were primarily responsible for expressing the mRNA-encoded spike, consistent with in vitro and in vivo studies of mRNA-LNP delivery (Buschmann et al., 2021; Liang et al., 2017). The early activation of these APCs was compatible with the notion that mRNA-LNP particles possess an intrinsic adjuvant effect (Heinz and Stiasny, 2021; Alameh et al., 2021). Interestingly, monocytes in mRNA-1273 treated chips showed signs of differentiation towards a DC phenotype, as indicated by the induction of the CD11c and DC-SIGN markers. This suggested that mRNA-LNP treatment could promote a myeloid differentiation pathway conducive to Ag presentation. The preferential upregulation of HLA-DR and costimulatory markers in myeloid cells that expressed the spike after mRNA-1273 treatment also supported the idea of an improved Ag presentation capacity in cells that had captured the vaccine. The presence of PBMC-derived myeloid cells and their differentiation into functional APC represents an asset of the LO chip system as it enables the evaluation of new generation mRNA vaccines in contrast to systems that rely on purified B cells and CD4⁺ T cells. The presence of other leukocyte subsets may also be relevant. For instance, a recent report suggests that natural killer (NK) cells may limit the expression of mRNA vaccines by killing spike-expressing cells through an Fc-dependent mechanism if spike-specific antibodies preexist in the circulation (Dangi et al., 2023). The impacts of preexisting antibodies not only on NK cell activity but also on the breadth and affinity of the memory B cell response (Schaefer-Babajew et al., 2023) remains to be tested in the LO chip, which should be feasible by perfusing autologous serum and/or spike-specific monoclonal antibodies. Overall, the LO chip system opens the possibility of modulating leukocyte populations and circulating antibody concentrations to better model parameters involved in the efficiency of recall responses to mRNA vaccines.

Immune escape is recognized as a key parameter driving the successive waves of SARS-CoV-2 variants worldwide (Tuekprakhon et al., 2022; Harvey et al., 2021), spurring the development of variant-specific booster vaccines. The worldwide spread of the highly divergent Omicron variant starting from late 2021 led to the development of bivalent mRNA vaccines encoding the ancestral Wuhan spike and an Omicron BA.1 or BA4/5 spike, matching the variants circulating in early and late 2022, respectively (Meijers et al., 2023). The bivalent vaccines were rapidly deployed and could boost the induction of SARS-CoV-2 neutralizing antibodies, ensuring protection from severe COVID (Winokur et al., 2023; Chalkias et al., 2022). However, it soon emerged that, in most studies, the titers of antibodies specific to Omicron were not preferentially increased by the bivalent vaccine as compared with those induced by the monovalent Wuhan vaccine (Collier et al., 2023; Wang et al., 2023). There is now convergent evidence that immune imprinting is limiting the induction and efficiency of Omicron-specific responses in individuals with preexisting B cell memory to previous SARS-CoV-2 variants, a phenomenon that can be explained by the preferential amplification of preexisting memory B cells upon vaccine boosting (Sokal et al., 2023; Aguilar-Bretones et al., 2023; Yang et al., 2023; Addetia et al., 2023). Given these complex immunological interactions, we set out to model monovalent and bivalent mRNA vaccine boosting in the LO chip. The comparison of responses induced by the original Wuhan vaccine mRNA-1273 and the bivalent Wuhan/BA.1 vaccine mRNA-1273-214 revealed only moderate differences, with an equivalent frequency of PB induction and an equivalently high level of spike-specific IgG production. A limitation of the study was that only IgG specific for the Wuhan strain was detected in the S-Flow assay. However, we were able to use strain-specific reagents to measure the frequency of RBD-specific B cells and found a comparable amplification of B cells specific for the Wuhan RBD by the two vaccines, and also, less expectedly, a comparable induction of B cells specific for the BA.1 RBD. The tight correlation between the frequencies of B cells specific for the Wuhan and BA.1 RBD (R = 0.92, P < 0.0001) suggested that most of these B cells were cross-reactive to the two strains. Thus, the LO chip system appeared to recapitulate the phenomenon of immune imprinting observed in vivo, with a preferential expansion of cross-reactive B cells upon vaccine boosting. Recent findings suggest that repeated boosting with an Omicron-derived monovalent mRNA vaccine can overcome immune imprinting by the ancestral strain (Yisimayi et al., 2023), a notion that could be tested in the LO chip in future studies.

The production of spike-specific antibodies at day 14 in the LO chip proved sufficient to measure their neutralizing capacity, which is an asset of the LO chip system, by opening the possibility of an in vitro evaluation of vaccine efficacy. The induction of SARS-CoV-2 neutralizing antibodies could be detected in six out of seven donors tested, though with different individual patterns. The monovalent vaccine tended to induce higher neutralizing antibody levels against the ancestral D614G strain than the BA.1 strain. In contrast, the bivalent vaccine did not induce a higher neutralization of BA.1 than of D614G. Further, comparing the two vaccines, we did not detect an increased neutralization of BA.1 after bivalent boosting compared with monovalent boosting, compatible with in vivo data (Collier et al., 2023; Wang et al., 2023), and reinforcing the notion that immune imprinting limited the emergence of Omicron-specific neutralizing antibodies. Examination of individual neutralization curves revealed different case scenarios, with donors responding equivalently well to the two vaccines, donors responding more efficiently to the monovalent vaccine, suggestive of strong imprinting, and one donor responding better to the bivalent vaccine, possibly due to a recent infection with an Omicron variant. This analysis illustrates the diversity of immunological histories in the population and the resulting individual variability in vaccine responses. In the face of such variability, the LO chip can provide a useful preclinical system to evaluate the capacity of candidate vaccines to induce neutralizing antibodies against current SARS-CoV-2 variants in diverse human populations.

In conclusion, we developed a versatile LO chip model suitable for the preclinical evaluation of human recall responses to different vaccine formulations, including the new-generation mRNA vaccines. The LO chip recapitulates the early activation of myeloid cells upon Ag capture and the later activation of Ag-specific CD4⁺ T cell and B cells, thus capturing the multifaceted aspects of a recall response. The formation of CD4⁺ T cell/B cell clusters and the differentiation and emigration of PB mimicked important features of B cell maturation within lymphoid tissues. Further, the microfluidic perfusion promoted a massive amplification of Ag-specific memory B cells and PB, enabling the detection of secreted antibodies and the evaluation of their neutralizing capacity. This approach confirmed the efficiency of mRNA vaccines at boosting antibody responses and also the lack of advantage of a bivalent over a monovalent vaccine boost. The LO chip thus represents a streamlined 3D organ model applicable to the evaluation of vaccine boosters in diverse human populations, which should be an asset in the face of a rapidly evolving SARS-CoV-2 pandemic.

All blood samples were obtained from healthy volunteers who donated blood at the Etablissement Français du Sang (EFS), the national blood bank. All participants provided written consent to donate blood for research. Anonymized blood samples were transferred to Institut Pasteur under a collaboration agreement with EFS (C CPSL UNT—N°18/EFS/041). Research using EFS blood samples was authorized by the Risk Prevention Service of Institut Pasteur (n°HS 2021-24921). PBMC were isolated from buffy coats or cytapheresis collars via density gradient centrifugation on lymphocyte separation medium (#CMSMSL01-01; Eurobio) and were resuspended in fetal bovine serum (FBS; #P30-3306; PAN Biotech) containing 10% dimethyl sulfoxide (DMSO) (#226827; Sigma-Aldrich) before cryopreservation in liquid nitrogen. For culture, PBMC were thawed and resuspended in a complete medium consisting in RPMI-1640 glutaMAX (#61870036; Gibco) supplemented with 100 U/ml penicillin/streptomycin (#15140122; Gibco), 10 mM HEPES (#15630056; Gibco), 1 mM sodium pyruvate (#15630056; Gibco), 0.1 mM non-essential amino acids (#11140050; Gibco), 0.05 mM 2-mercaptoethanol (#31350010; Gibco), and 10% heat-inactivated human AB serum (#201021334; Institut de Biotechnologies). PBMC were rested for a minimum of 2 h in a complete medium supplemented with 1 µg/ml of DNAse from bovine pancreas (#58409400; Thermo Fisher Scientific) prior to culture. Plasma collected during PBMC preparation was used to screen donors for the presence of SARS-CoV-2 spike-specific IgG using the S-Flow assay (see below). Only the donors seropositive for SARS-CoV-2 were included in the study.

Two-compartment S1 chips from Emulate were used throughout the study. S1 chips are composed of PDMS, a transparent and gas-permeable elastomeric polymer, and consist of two parallel channels separated by a membrane with 7-µm pores. Fluid flow can be controlled independently in each channel through a Zoe culture module connected to an Orb hub module (Emulate), with continuous medium perfusion ensuring the renewal of essential nutrients and enabling primary cell culture at high density. Components secreted by cultured cells can be collected in the two reservoirs connected to the outlet of each channel and contained within each chip holder, or Pod (Emulate). Prior to culture, the inner PDMS surface of the chip channels was activated according to the manufacturer guidelines to enable proper ECM attachment. The chips were then coated at 4°C overnight with a solution of 30 µg/ml of type I collagen (#354236; Corning) and 100 µg/ml Matrigel (#354234; Corning) diluted in phosphate-buffered saline (PBS) (#14190-094; Gibco). Before cell seeding, the chips were incubated for at least 1 h in a culture incubator at 37°C, and each channel was then washed twice with 200 μl of complete medium. Just prior to cell seeding, the medium in each channel was thoroughly aspirated.

The LO chip design consists of a high-density 3D human PBMC culture within an ECM gel in the lower channel of an S1 chip, while medium and Ag are fluxed in the upper channel of the chip. For LO chip seeding, PBMC were resuspended at the concentration of 580 million cells/ml in an ECM composed of half collagen I solution and half degassed complete medium to achieve a final concentration of 1.5 mg/ml of type I collagen. The initial type I collagen solution at 3 mg/ml was prepared by mixing rat tail type I collagen (#354236; Corning), sterile water, DPBS 10×, and an adequate amount of 1 N hydroxide sodium to reach pH = 7.4 (#655104; Sigma-Aldrich), allowing gelation. The complete medium was degassed using aspiration in a Steriflip-HV cup (#SE1M003M00; Merck). All reagents and PBMC were maintained on ice before seeding. 17 μl of the viscous cell/ECM solution containing a total of 10 million PBMC was then introduced through the inlet port to fill the lower channel of the chip. The cell/ECM solution was allowed to gel for 30 min in a cell culture incubator at 37°C and 5% CO₂. After gelation, a thin disc-shaped plastic membrane was inserted to block the inlet port of the gel-filled channel as a measure to prevent bubble formation within the ECM. The disc forms a fluid- and air-tight barrier between the chip inlet port and Pod inlet reservoir, while still allowing pressure to be applied via the chip lower channel outlet. The outlet reservoir was filled with 1 ml of complete medium to avoid drying of the ECM. The chips were perfused through the upper channel inlet with 30 μl/h of complete medium, as programmed on the Orb controller (Emulate), with the chip culture module placed in an incubator maintained at 37°C and 5% CO₂. At day 0, the inlet reservoir connected to the upper channel was filled with 3 ml of complete medium supplemented with Ag. Every 3 days, the perfused medium was renewed with a 1:1 mix of effluent medium and fresh complete medium. The inclusion of an effluent medium ensured the maintenance of the cytokine milieu induced by antigenic stimulation and continuous perfusion of the Ag.

The protein Ag tested included the control protein BSA (#AM2616; Invitrogen) at 1 µg/ml and the spike protein of the SARS-CoV-2 Wuhan strain (#NR-53937; BEI Resources) at 1 µg/ml. The mRNA-based LNP vaccines used included the monovalent Wuhan vaccine mRNA-1273 (Moderna), the bivalent Wuhan/Omicron BA.1 vaccine mRNA-1273.214 (Moderna), and the monovalent Wuhan vaccine BTN162b2 (Pfizer), all tested at 3 µg/ml. Control LNP were engineered with a “Moderna-like” lipid composition by the Oz Biosciences company and contained mRNA encoding for the control protein ovalbumin (OVA-LNP) or for the GFP (GFP-LNP). Control LNP was used at a 3 µg/ml concentration, similar to the one used for mRNA vaccines.

2D or 3D classical cultures were performed to evaluate antigenic responses in conditions devoid of fluid flow. For static 2D cultures, 10 million PBMC were resuspended in 1 ml of complete medium in a 24-well plate. For static 3D cultures, 10 million PBMC were resuspended in an ECM droplet at the same concentration used in the LO chip (10 million cells/17 μl), and the cell/ECM gel droplet was deposited in 1 ml of complete medium in a 24-well plate. For stimulation, Ag (BSA, spike, or mRNA vaccine) was added at the same concentration as the one used in the LO chip.

At each time point (day 2, or day 6, or day 14), effluent from the upper outlet reservoir was collected, aliquoted, and stored at −20°C. Chips were disconnected from the pods, and upper channels were washed twice with 100 μl PBS introduced in the inlet port, while the outlet port was connected to an empty 200 μl tip. The collected cells were either pooled with those recovered from the lower channel or analyzed separately to characterize cell migration from the lower to the upper channel. To harvest cells from the lower channel, the ECM had first to be digested. To do so, all the chip ports were connected with empty tips, except the upper inlet port, which was manually perfused with 100 μl of Cell Recovery Medium (#354253; Corning) or of Cultrex Organoid Digestion solution (#3700-100-01; R&D Systems). Chips were incubated for 45 min at 4°C to allow diffusion of the digestion solution from the upper to the lower channel. After incubation, the lower inlet port was connected to a new tip with 80 μl of cold PBS that was manually perfused through the lower channel to flush the cells and ECM remnants out of the chip for collection. The lower channel was further washed twice with 80 μl of cold PBS, and all the washes were pooled prior to centrifugation at 300 g for 5 min. The pelleted cells were resuspended in PBS for immunostaining while the supernatants from ECM digestion were collected and stored at −20°C.

For the phenotyping of myeloid cells, the cells collected from LO chips at day 2 were resuspended in cold PBS, incubated for 10 min at 4°C with a live/dead fixable dye (#L10119; Invitrogen), and then stained for 30 min at 4°C in PBA buffer (PBS, 0.5% BSA, 2 mM EDTA) in the presence of Fc Block (1:50 dilution; BD-Biosciences), using the following antibody combination: CD19, CD3, CD4, CD14, DC-SIGN, CD11c, HLA-DR, CD40, CD86, and anti-spike Ab s102-AF647 (provided by Cyril Planchais and Hugo Mouquet, Institut Pasteur, Paris, France) (Planchais et al., 2022). Cells were then washed twice in PBS and fixed with a Cytofix/cytoperm kit (554655; BD Biosciences) for 20 min at 4°C. Intracellular staining for anti-spike Ab s102 was done for 30 min at 4°C to detect intracellular spike. The irrelevant mGO53 human IgG1 was used as an isotypic control. After staining, cells were washed twice in PBS, fixed in a 4% paraformaldehyde (PFA) in PBS solution for 20 min at room temperature (RT), and resuspended in cold PBS.

For detection of spike-specific or RBD-specific B cells, recombinant biotinylated spike (Wuhan spike #130-127-682; Miltenyi; Omicron BA.1 spike1#30-130-417) or biotinylated RBD (Wuhan RBD #130-129-570; Miltenyi; Omicron BA.1 RBD #130-130-419) were first coupled with AF647-streptavidin (S32357; Invitrogen) and PE-streptavidin (130-106-790; Miltenyi) at a 5:1 molar ratio in cold PBA buffer for a minimum of 15 min. PBMC were stained with the fluorescent RBD or spike at 3 µg/ml for 45 min in cold PBA buffer. After two PBS washes, cells were incubated for 10 min at 4°C with a live/dead fixable dye (#L34963; Invitrogen) in PBS and then stained in PBA with Fc Block (1:50 dilution) using the following combination of antibodies: CD3, CD4, CD19, CD27, CD38, IgG and IgM, IgD, CD21, FAS. Cells were washed twice in PBS, fixed in a 4% PFA in PBS solution, and resuspended in cold PBS.

For the intracellular detection of the AID enzyme, samples were fixed after surface labeling with 100 μl of Foxp3 fixation buffer (#00-5523-00; eBioscience, Thermo Fisher Scientific) for 30 min at 4°C. Samples were permeabilized with 150 μl of Foxp3 permeabilization buffer and then centrifuged at 300 g for 3 min. After removing the supernatant, samples were stained with 100 μl of an antibody mix containing the anti-AID antibody (Table S1) diluted in Foxp3 permeabilization buffer and incubated for 30 min at 4°C. The samples were washed twice with Foxp3 permeabilization buffer, resuspended in 250 μl of PBS, and analyzed on a flow cytometer.

For detection of spike-specific CD4⁺ T cells, the harvested cells were restimulated using a pool of overlapping Wuhan spike peptides (130-126-701; Miltenyi) at 2 µg/ml for 16 h in the presence of brefeldin A (#400602; Biolegend) at 5 µg/ml. To detect cytokine secretion by CD4⁺ T cells, cells were resuspended in cold PBS, incubated with a live/dead fixable dye (#L10119; Invitrogen) for 10 min at 4°C, and surface stained for CD3 and CD4 in PBA buffer with Fc Block (1:50 dilution) for 20 min at 4°C. Cells were washed twice and then fixed and permeabilized with a Cytofix/Cytoperm kit (554655; BD Biosciences) for 20 min at 4°C. Intracellular staining for IFN-γ, TNF-α, and IL-2 was done for 30 min at 4°C. Cells were centrifuged and resuspended in Fix/Perm buffer and stored at 4°C prior to cytometry analysis.

Flow cytometry analysis was carried out using an Attune NxT flow cytometer (Thermo Fisher Scientific) for most experiments or using an ID7000 spectral cell analyzer (Sony) for the CD21/FAS and AID labeling experiments. Data were analyzed using the FlowJo V10 software (Flowjo, LLC). The antibody references and dilution factors used are listed in Table S1.

The chemokine CXCL13 was measured in LO chip effluent using a human CXCL13 ELISA assay (#DY801; Quantikine, R&D Systems) following the manufacturer’s instructions.

IgG and IgA antibodies specific to the SARS-CoV-2 spike were detected by the S-Flow assay, which measures antibody binding to spike-expressing HEK 293T cells, as previously described. This assay was shown to have a 100% specificity (95% confidence interval [CI]: 98.5–100%) and 99.2% sensitivity (95% CI: 97.69–99.78%) for COVID-19 patient sera and outperform ELISA assays in terms of sensitivity (Grzelak et al., 2020). The spike-expressing cells (293T-S) were generated by transducing HEK 293T cells (ATCC CRL-3216) with a lentivector expressing a codon-optimized Wuhan SARS-CoV-2 spike protein (GenBank: QHD43416.1). Control 293T cells were transduced with an empty lentivector to assess background staining. The transduced cells were selected with 2.5 µg/ml of puromycin. To perform the S-Flow assay, 5 × 10⁴ 293T-S cells were plated in a 96-well round bottom plate. 50 μl of patient serum diluted 1:300 in MACS buffer (Miltenyi Biotech) or 50 μl of ECM chip extract (undiluted at day 6, or diluted 1:6 at day 14) were added to the cells, and the mix was incubated for 1 h at 4°C. The cells were then washed in PBS and stained with a mix of secondary antibodies anti-human IgG-Fc-AF647 (1:600) and anti-human IgA-Alpha-Chain-AF488 (1:200) (Thermo Fisher Scientific) for 30 min at 4°C. Cells were fixed for 10 min in 4% PFA and were acquired on an Attune NxT flow cytometer (Life Technologies). Results were analyzed with the FlowJo V10 software. For each sample, the background signal was measured in control 293T cells lacking S and subtracted to define the specific signal. The MFI of IgG and IgA binding was reported as it was found to provide a quantitative measurement of the levels of SARS-CoV-2 spike-specific antibodies (Planas et al., 2021; Grzelak et al., 2020).

For in-chip immunofluorescence, cells were stained by relying on antibody diffusion from the upper channel to the lower channel. The LO chips were emptied of medium and fixed by filling the upper channel through the inlet port with 4% PFA in PBS 1×, while the other ports were connected to empty tips. The chips were incubated for 30 min at RT before aspirating the fixative solution and washing the chips three times with PBS for 5 min. The upper channels were then filled with 100 μl of a solution containing primary antibodies diluted in PBS with 1% FBS, 0.1% Triton-X100, and 1:20 Fc Block (BD Biosciences). The following primary antibodies were used: CD4-AF488 (557695; BD Biosciences, RPA-T4, 1/25), CD19-AF561 (505-0199-42; eBioscience, HIB19, 1/25), ICOS (rabbit 89601; Cell Signaling Technology, D1KT2, 1/100) Ki67 (rabbit MA5 14520; Thermo Fisher Scientific, 1/200). After three washes with 200 μl of PBS added through the upper inlet port, the chips were incubated with the secondary antibody goat anti-rabbit IgG-AF647 (A32733; Invitrogen, polyclonal, dilution 1:200) and the nuclear dye Hoechst (H3570; Life Technologies, 1:1,000) for 4 h at 4°C in the dark. After three PBS washes, the chips were analyzed using a spinning disc confocal microscope (Ti2E; Nikon, Yokogawa, CSU W1) using a 40× objective with a long working distance to visualize the whole depth of the chip’s lower channel.

Antibody neutralization was measured by the inhibition of cell fusion in a GFP-split cell system. U2OS-ACE2 GFP1–10 and U2OS-ACE2-GFP11 cells, also termed S-Fuse cells, become GFP⁺ when they fuse together upon productive infection by SARS-CoV-2 (Buchrieser et al., 2020). The S-Fuse cells tested negative for mycoplasma. S-Fuse cells were mixed (at a 1:1 GFP1-10/GFP11 ratio) and plated at 8 × 10³ cells per well in a μClear 96-well plate (Greiner Bio-One). The tested SARS-CoV-2 strains were incubated with serially diluted chip effluent or supernatants from ECM digestion for 20 min at RT and the mixture was then added to S-Fuse cells. 18 h later, cells were fixed with 2% PFA (#15714-S; Electron microscopy), washed in PBS, and stained with Hoechst at a 1:1,000 dilution (H3570; Invitrogen). Images were acquired automatically using an Opera Phenix high-content confocal microscope (PerkinElmer). The GFP⁺ area and the number of nuclei per well were quantified using the Harmony software (PerkinElmer). The percentage of neutralization was computed using the number of GFP⁺ syncytia as a value with the following formula: 100 × [1 − (value with antibodies − value in “non-infected”)/(value in “no antibodies” − value in “non-infected”)]. The neutralizing activity of each sample was expressed as the ID₅₀. Of note, we previously validated the S-Fuse assay by showing a strong correlation between neutralization titers obtained with the S-Fuse reporter assay and a pseudovirus neutralization assay (Sterlin et al., 2021; Planas et al., 2021).

Statistics were computed with the GraphPad Prism v10.1.1 software. The non-parametric Mann–Whitney test and the Wilcoxon matched-pairs rank test were used to compare groups. Horizontal bars in graphs represent median values. Correlations were analyzed by simple linear regressions with the associated Pearson correlation coefficient R reported. ID₅₀ values were obtained after a non-linear curve fit using a four-parameter logistic regression model in Prism. P values lower than 0.05 were considered statistically significant. The nature of the statistical tests used is reported in the figure legends.

The supplementary material contains four figures and one table: Fig. S1 shows cell gating strategies. Fig. S2 shows localization and kinetics of antibody production in the LO chip. Fig. S3 shows characterization of spike-specific B cells induced in LO chips and static cultures. Fig. S4 shows that spike perfusion induces spatial organization and proliferation of lymphocytes in the LO chip. Table S1 shows references of the antibodies used in the study and is provided at the end of the PDF.

All data needed to evaluate the conclusions in the paper are present in the paper and/or the supplementary materials.

We thank Sophie Novault at the Flow Cytometry Platform of Institut Pasteur (IP) for advice on cytometry data acquisition; Cyril Planchais and Hugo Mouquet at the Humoral Immunology Unit of IP for anti-spike antibodies; Florence Guivel-Benhassine and Françoise Porrot at the Virus & Immunity Unit of IP for generating viral stocks; Blanca Liliana Perlaza and Marie-Noëlle Ungeheuer at the ICAREB biobank facility of IP for access to human cells; Céline Fichot at the Medical Center of IP and Catherine Perves at Hospital Cochin for providing mRNA vaccines; Mathieu Claireaux and Rogier Sanders at the Amsterdam University Medical Center for providing purified spike protein for test experiments. The following reagent was obtained from BEI Resources: recombinant spike protein from the SARS-CoV-2 Wuhan strain (#NR-53937).

This work was supported by Emulate (contract S-RD21002 to L.A. Chakrabarti); the COROCHIP project funded by the Pasteur COVID-19 RP call (to L.A. Chakrabarti and S. Gobaa); the Fondation de France (PR-166156 project, L.A. Chakrabarti); the Urgence COVID-19 Fundraising Campaign of Institut Pasteur (PFR7 project; L.A. Chakrabarti); the French Agency for AIDS and Emerging Diseases Research (ANRS-MIE) (ECTZ213626 project; L.A. Chakrabarti); and the Institut Carnot Pasteur Microbe & Santé (S. Gobaa).

Author contributions: R. Jeger-Madiot: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Validation, Visualization, Writing—original draft, Writing—review & editing, D. Planas: Investigation, Methodology, Resources, I. Staropoli: Investigation, H. Debarnot: Investigation, Writing—review & editing, J. Kervevan: Investigation, H. Mary: Investigation, Methodology, Resources, Visualization, Writing—review & editing, C. Collina: Investigation, B.F. Fonseca: Investigation, Resources, Writing—review & editing, R. Robinot: Investigation, Writing—review & editing, S. Gellenoncourt: Investigation, Writing—review & editing, O. Schwartz: Supervision, L. Ewart: Funding acquisition, Methodology, Validation, Writing—review & editing, M. Bscheider: Conceptualization, Funding acquisition, Writing—review & editing, S. Gobaa: Methodology, Resources, Supervision, L.A. Chakrabarti: Conceptualization, Formal analysis, Funding acquisition, Project administration, Supervision, Validation, Visualization, Writing—original draft, Writing—review & editing.