In vitro recapitulation of intramuscular mRNA vaccination with naive and recall antigens using a human lymphoid follicle chip platform Authors: Yunhao Zhai, Min Wen Ku, Ka Yang, Aditya Patil, Pranav Prabhala, Liqun He, Yuncheng Man, Mizanur Rahman, Kavya Menon, Sven B. Spörri, Supriya Gharpure, Abdul Rahman Isaacs, Shay Ferdosi, Andrzej S. Pitek, Giulietta Maruggi, Sylvie Bertholet, Jessica Firestone, Kambiz Mousavi, Eric Miller, Kate Luisi, Caleb A. Hellman, LaTonya D. Williams, Georgia D. Tomaras, Peng Yin, Steven P. Gygi, Josie McAuliffe, Donald E. Ingber, Girija Goyal Published: 2026 ABSTRACT SummaryPredicting the efficacy and toxicity of mRNA vaccines remains challenging. We describe an in vitro human model that replicates immune responses to lipid nanoparticle (LNP)-based mRNA vaccines upon intramuscular injection. Vaccines are administered to human skeletal myoblasts and antigen-presenting cells (APCs). Non-amplifying mRNA (NAM) vaccines directly induce antigen expression in APCs, whereas self-amplifying mRNA (SAM) vaccines require muscle cell-APC contact. The transfer of APCs and soluble factors to a microfluidic human lymphoid follicle chip (LF Chip), to mimic lymphatic drainage, induces LF expansion, de novo immunoglobulin G (IgG) production against a naive antigen, and cytokine release, with responses varying by LNP type. Similar vaccination of LF chips with Moderna vaccine against SARS-CoV-2 spike recall antigen generates neutralizing antibodies and induces somatic hypermutation (SHM). These studies offer an all-human alternative for evaluating vaccine-induced immunity, limiting the need for non-human primates and accelerating vaccine development. Introduction mRNA vaccines have been essential in controlling the COVID-19 pandemic.1,2 The Moderna and BioNTech mRNA vaccines against COVID-19 comprise uridine-modified, non-replicating mRNAs that encode SARS-CoV-2 spike proteins. Recently, a self-amplifying mRNA (SAM) vaccine against SARS-CoV-2 that provides equivalent protection to non-replicating mRNA vaccines but at 6- to 20-fold lower doses was also approved by Japanese authorities,3,4,5,6,7 further highlighting the potential of mRNA vaccine technology. However, the mechanisms underlying vaccine efficacy and toxicity have been primarily studied in inbred mice and non-human primates (NHPs), which often exhibit immune responses that differ from those seen in humans.8,9 Furthermore, there has been an extreme shortage of NHPs for vaccine testing since the COVID-19 pandemic,10 and testing in animals, particularly in NHPs, also raises major ethical concerns. Although animal models have historically dominated FDA approval processes, the FDA’s new 2.0 approval pathway now expands the use of human in vitro data, including results generated using microfluidic human organ-on-a-chip (Organ Chip) technology, in investigational new drug (IND) applications for regulatory approval.11,12,13 Despite this advance, the field of vaccine development has been slow to adopt human cell-based models. Lymphoid organoids and Organ Chip methods using human peripheral blood mononuclear cells (PBMCs) and tonsil cells have been employed for vaccine testing in academic laboratories14,15,16,17,18; however, these systems primarily focus on immune memory responses to influenza and COVID-19 vaccines, and, most importantly, they have not been able to replicate clinically relevant immune responses that are generated at the injection site. Antigen uptake, distribution, and innate immune responses are all critical factors that must be assessed to define the efficacy of mRNA vaccines.19 However, unlike protein subunit or inactivated virus vaccines, mRNA vaccines require cellular uptake and intracellular translation of the antigen from the injected mRNA. Lipid nanoparticles (LNPs) are commonly used to encapsulate the mRNAs to promote mRNA uptake and antigen expression in human tissues, and they also may serve as adjuvants to enhance immune responses.20,21,22,23 Consequently, there is a pressing need for human-relevant preclinical models that can more accurately study the mechanism of action of LNP-based mRNA vaccines, as well as their efficacy, toxicities, and associated biomarkers. Most mRNA vaccines are administered through intramuscular injection. Monocytes infiltrate the site of injection due to inflammation. Antigen production at the muscle injection site after mRNA vaccine injection quickly peaks after 4–24 h and then gradually wanes. The infiltrating monocytes, monocyte-derived dendritic cells (DCs), and resident DCs are the main antigen-presenting cells (APCs), and the release of pro-inflammatory cytokines such as IL-1β and IL-6 is also important for the vaccine effects.19,24,25,26 In this study, we expanded the functionality of our previously published lymphoid follicle chip (LF Chip),16 which recapitulated germinal center-like features including the formation of LFs composed of T and B cells that expressed activation-induced cytidine deaminase, exhibited plasma cell differentiation, and produced immunoglobulin G (IgG) against influenza antigen upon vaccination, by combining it with a two-dimensional (2D) human intramuscular injection simulation module in which human skeletal myoblast cells (HSMCs) are cultured with APCs, either monocytes or DCs, and then exposed to LNP-based mRNA vaccines in vitro. This module provides a method for detecting the expression of antigens encoded by mRNA vaccines and innate immune responses at the intramuscular vaccination site. By transferring APCs and released factors from the intramuscular injection module to the LF Chip, adaptive immune responses can then be captured by measuring the formation of LFs, cytokine release, and antibody production and somatic hypermutation (SHM) in the LF Chip. Here, we used this method to compare immune responses and changes in protein expression profiles induced by non-amplifying mRNA (NAM) versus SAM vaccines, as well as different LNP formulations. The method successfully modeled antigen presentation, activation of APCs, primary IgG responses, and cytokine responses induced by SAM vaccines against rabies virus glycoprotein (RABV-G), which is a naive antigen for most United States-based donors. SHM and the production of cross-reactive antibodies against various SARS-CoV-2 variants with neutralization activity were also detected when the Moderna bivalent COVID-19 mRNA vaccine was administered to this in vitro intramuscular vaccine model. Results Contact between APCs and HSMCs is required for modeling intramuscular vaccination Monocytes are known to infiltrate vaccine injection sites, take up SAM vaccines, express the antigens, and contribute to transporting them to the lymph nodes. Some monocytes differentiate into DCs during this process.19,21,27,28 To explore modeling of intramuscular vaccination with a naive antigen, we introduced a SAM vaccine encoding RABV-G and green fluorescent protein (GFP) encapsulated in an LNP (SAM-LNP1) into cultures of primary human monocytes or monocyte-derived DCs either when grown alone or co-cultured with HSMCs (Figure 1A). Non-toxic doses (1–100 ng/mL; Figures S1A, S1B, and S1C) of SAM-LNP1 did not drive reporter expression (GFP) in monocyte monocultures (as assessed by flow cytometry), implying that the antigen was not expressed (Figures 1B and S1D). However, when the SAM-LNP1 was added to co-cultures of HSMCs with monocytes (Figures 1B, 1C, S1D, and S1E) or DCs (Figure S2A) to simulate intramuscular vaccination, a dose-dependent increase in GFP expression was detected in both APCs suggesting successful antigen expression. While the GFP expression levels were variable between donors, the highest level of GFP reporter expression was observed at the 100 ng/mL dose of SAM-LNP1 (Figures 1B and S1D); thus, this dose was used in all subsequent studies.Figure 1In vitro model of intramuscular vaccination using SAM-LNP(A) Schematic of the intramuscular vaccination model created with BioRender.com.(B) Quantification of live CD14+ GFP+ cells by flow cytometry in monocyte (Mono) monoculture and in co-culture with HSMCs (Mono+HSMCs) without treatment (“no treatment,” NT) or various doses of SAM-LNP1. 3 technical replicate wells are shown per group for each dose of SAM-LNP.(C) Density dot plots illustrating GFP expression in NT and 100 ng/mL SAM-LNP1-treated representative wells.(D) Cytokine levels detected using a Meso Scale Discovery (MSD) assay, comparing NT and various doses of SAM-LNP1 in Mono and Mono+HSMC conditions. Average fold change (log2) in two SAM-LNP1-treated wells relative to the average of two NT wells is displayed.(E) Fold change in median fluorescence intensity of myeloid activation markers CD80, CD86, and HLA-DR by flow cytometry of 3 treated wells as compared to the average MFI of 3 NT wells of monocyte monoculture.In (B)–(E), results are shown from one immune cell donor co-cultured with one HSMC donor. Similar results were obtained in 2 additional immune donors cultured with the same HSMC donor. Error bars indicate standard deviation (SD). A two-way ANOVA, followed by Tukey’s multiple comparisons test to assess differences between groups, was performed in (B) and (E). In vitro model of intramuscular vaccination using SAM-LNP (A) Schematic of the intramuscular vaccination model created with BioRender.com. (B) Quantification of live CD14+ GFP+ cells by flow cytometry in monocyte (Mono) monoculture and in co-culture with HSMCs (Mono+HSMCs) without treatment (“no treatment,” NT) or various doses of SAM-LNP1. 3 technical replicate wells are shown per group for each dose of SAM-LNP. (C) Density dot plots illustrating GFP expression in NT and 100 ng/mL SAM-LNP1-treated representative wells. (D) Cytokine levels detected using a Meso Scale Discovery (MSD) assay, comparing NT and various doses of SAM-LNP1 in Mono and Mono+HSMC conditions. Average fold change (log2) in two SAM-LNP1-treated wells relative to the average of two NT wells is displayed. (E) Fold change in median fluorescence intensity of myeloid activation markers CD80, CD86, and HLA-DR by flow cytometry of 3 treated wells as compared to the average MFI of 3 NT wells of monocyte monoculture. In (B)–(E), results are shown from one immune cell donor co-cultured with one HSMC donor. Similar results were obtained in 2 additional immune donors cultured with the same HSMC donor. Error bars indicate standard deviation (SD). A two-way ANOVA, followed by Tukey’s multiple comparisons test to assess differences between groups, was performed in (B) and (E). Upon SAM-LNP1 vaccination, increases in levels of multiple inflammatory cytokines were observed in both monocyte and DC monocultures and in the combined HSMC-APC cultures; however, levels of CCL4, IL-6, and IL-8 were notably higher in the co-culture model when HSMCs were present (Figures 1D and S2B). Although some variation was observed among donors in the magnitude of induction, cells from all donors upregulated multiple inflammatory cytokines that are known to support adaptive immune responses (Figure S2C). SAM-LNP1 treatment also increased the expression of CD86 and HLA-DR, which are markers of monocyte and DC activation (Figures 1E and S2D). However, CD80 was not induced, but CD80 is often upregulated later or under specific conditions, such as prolonged inflammation.22,29 These results highlight the ability of HSMCs to promote antigen expression and activation of APCs and demonstrate the value of the co-culture model. To further investigate the role of HSMCs in antigen expression by APCs, we conducted co-cultures in Transwell (TW) inserts with varying pore sizes (0.4, 3, or 8 μm) placed above an HSMC monolayer (Figure S3A). As expected, both untreated and SAM-LNP1 vaccinated monocytes were able to migrate from the upper well into the lower chamber containing the HSMCs when they were cultured on filters containing 3 or 8 μm pores but not the 0.4 μm pore, and vaccinated cells migrated to a greater degree (Figure S3B). Importantly, only vaccinated monocytes that were able to migrate through the larger pores and come in direct contact with the HSMCs displayed the expression of the GFP-labeled antigen, and no antigen reporter expression was observed in monocytes that remained on top of the filter membrane (in either the untreated or vaccinated cultures) or in monocytes that migrated to wells that did not contain HSMCs (Figures S3C and S3D). These findings indicate that direct cell-cell contact between HSMCs and monocytes is necessary for antigen expression by monocytes vaccinated with SAM-LNP1. Mimicking mRNA vaccine-induced adaptive immune responses using human LF chips To mimic the drainage of the vaccine and migration of monocytes from the vaccine site to the lymph node, we removed the conditioned medium from the co-culture of monocytes and HSMCs and flowed it through the top flow channel of a 2-channel, human LF Chip microfluidic culture device, which has been previously shown to support self-assembly of germinal center-like functions.16 The parallel lower channel that is separated from the top channel by a porous membrane contains autologous B and T cells as well as monocytes isolated from the vaccinated intramuscular injection module embedded at a high lymph node-like density within an extracellular matrix gel (Figure 2A).Figure 2Induction of LFs and anti-RABV-G IgG production by SAM-LNP1 in human LF chips(A) Schematic of the LF Chip created with monocytes and conditioned medium from the intramuscular vaccination-mimicking module created with BioRender.com.(B) Representative 3D confocal microscopic stacks showing pseudo-colored follicles (blue) and cell nuclei (green) present within extracellular matrix (ECM) gels cultured for 4 days within a perfused LF Chip when vaccinated with SAM-LNP1 in the absence or presence of IL-2 and IL-4 (SAM-LNP1 + IL-2&IL-4); scale bars, 100 μm.(C) Quantification of the number (left) and size (right) of LFs in LF chips of one donor based on immunostaining followed by confocal imaging. Each data point represents one field of view (left) or an individual follicle (right); 2 independent chips were analyzed per condition.(D) Anti-RABV-G IgG levels in effluents of no treatment LF chips or chips vaccinated with or without IL-2 and IL-4 and cultured for 14 days were detected using a cell-based assay; 3–4 independent chips were analyzed per condition.(E) Quantification of the number (left) and size (right) of LFs in LF chips from four different donors, with each color representing a different donor. Each data point corresponds to one field of view (left) or an individual follicle (right).(F) Anti-RABV-G IgG levels in effluents of LF chips. Each data point represents one chip, with different symbols indicating chips from three independent donors.(G) Anti-RABV-G IgG levels in effluents of LF chips seeded with naive B + bulk T + monocyte mixture. Each data point represents one chip, with different symbols indicating chips from three independent donors.(H) Heatmap showing average log2 fold changes in cytokine levels in the effluents of three LF chips created using bulk lymphocytes from donors whose prior exposure to rabies was unknown, measured using a Luminex Multiplex Assay at various time points (4, 7, 15, 22, and 28 days post-vaccination) compared to the average levels of each cytokine found at day 1.Representative results from one donor are shown in (B)–(D) and (H), with similar outcomes observed in two donors. Data shown are mean ± SD; (C, left; D; and E) one way ANOVA test followed by Tukey’s multiple comparisons test, (C, right; and E) Mann-Whitney U test, (F and G) Welch’s t test. Induction of LFs and anti-RABV-G IgG production by SAM-LNP1 in human LF chips (A) Schematic of the LF Chip created with monocytes and conditioned medium from the intramuscular vaccination-mimicking module created with BioRender.com. (B) Representative 3D confocal microscopic stacks showing pseudo-colored follicles (blue) and cell nuclei (green) present within extracellular matrix (ECM) gels cultured for 4 days within a perfused LF Chip when vaccinated with SAM-LNP1 in the absence or presence of IL-2 and IL-4 (SAM-LNP1 + IL-2&IL-4); scale bars, 100 μm. (C) Quantification of the number (left) and size (right) of LFs in LF chips of one donor based on immunostaining followed by confocal imaging. Each data point represents one field of view (left) or an individual follicle (right); 2 independent chips were analyzed per condition. (D) Anti-RABV-G IgG levels in effluents of no treatment LF chips or chips vaccinated with or without IL-2 and IL-4 and cultured for 14 days were detected using a cell-based assay; 3–4 independent chips were analyzed per condition. (E) Quantification of the number (left) and size (right) of LFs in LF chips from four different donors, with each color representing a different donor. Each data point corresponds to one field of view (left) or an individual follicle (right). (F) Anti-RABV-G IgG levels in effluents of LF chips. Each data point represents one chip, with different symbols indicating chips from three independent donors. (G) Anti-RABV-G IgG levels in effluents of LF chips seeded with naive B + bulk T + monocyte mixture. Each data point represents one chip, with different symbols indicating chips from three independent donors. (H) Heatmap showing average log2 fold changes in cytokine levels in the effluents of three LF chips created using bulk lymphocytes from donors whose prior exposure to rabies was unknown, measured using a Luminex Multiplex Assay at various time points (4, 7, 15, 22, and 28 days post-vaccination) compared to the average levels of each cytokine found at day 1. Representative results from one donor are shown in (B)–(D) and (H), with similar outcomes observed in two donors. Data shown are mean ± SD; (C, left; D; and E) one way ANOVA test followed by Tukey’s multiple comparisons test, (C, right; and E) Mann-Whitney U test, (F and G) Welch’s t test. When we analyzed multiple LF chips created with cells from a single donor, we found that exposure to conditioned medium and monocytes from SAM-LNP1-vaccinated co-cultures with HSMCs resulted in a significant increase in the number of LFs within the chips, 4 days after vaccination (Figures 2B, 2C, and S4A). This time point was chosen due to past work where follicle formation peaked at 4 days.16 Next, we assessed whether the SAM-LNP1-vaccinated LF chips were able to generate antigen-specific, anti-RABV-G IgG by testing whether the chip outflows collected 14 days after vaccination could bind to RABV-G-expressing 293T cells (Figure S4B). This time point was chosen based on past work where recall responses were evident as early as day 9. We postulated that primary immune responses would take longer. We did not observe significant induction of IgG specific to RABV-G (Figures 2D and S4C). As class switching can take a few days and survival of primary lymphocytes, particularly naive lymphocytes, is poor in vitro, we supplemented the chip culture medium with IL-2 and IL-4, which are known to support T and B cell proliferation and differentiation.30,31 Addition of IL-2 and IL-4 to culture medium in SAM-LNP1-vaccinated chips resulted in increased levels of anti-rabies IgG (Figures 2D and S4C), although it did not affect follicle number or size (Figures 2C and S4A) as compared to SAM-LNP1 vaccination alone. Thus, in all subsequent experiments, we used IL-2- and IL-4-containing medium for LF Chip culture in both vaccinated chips, and chips that received no further treatment or vaccine are referred to as no treatment chips. Vaccination of these chips created with cells from 4 different human donors with SAM-LNP1 induced an increase in the number of LFs but no change in size of LFs in vaccinated chips (Figure 2E). When we quantified anti-RABV-G IgG levels in unvaccinated and SAM-LNP1-vaccinated chips created with cells from 3 donors, we found that some vaccinated chips from each donor had anti-RABV-G IgG above baseline, while others did not (Figure 2F). Given the high diversity of the human immune repertoire, the variation between chips from the same donor is likely influenced by the presence and persistence of rare anti-RABV-G B cells. Indeed, when the B cell repertoire (BCR nucleotide sequences) from 2 different donors’ chips was compared, there was only ∼5% overlap (Figures S5A and S5B). We further validated that this protocol can induce IgG responses to a naive antigen by creating chips from cells obtained from naive donors provided by a commercial supplier (i.e., individuals with no prior exposure to rabies antigens through vaccination or infection). As an additional control, we created these chips with purified naive B cells instead of bulk B cells by depleting CD27+ memory B cells. We found that anti-RABV-G IgG was induced in SAM-LNP1-vaccinated chips created from the cells of 3 such naive donors (Figures 2G and S4D). Finally, when we compared the levels of 18 different cytokines in the outflows of the LF chips created using donors with bulk lymphocytes from donors with unknown exposure to rabies, at several times during 28 days of culture, we found that cytokines known to support the germinal center response and B cell differentiation, such as IL-5, IL-6, CCL4, and IFN-γ,19,20,22,32 increased during the first 2 weeks, followed by a subsequent plateau (Figure 2H). SAM-LNP1 and non-amplifying mRNA-LNP vaccines elicit distinct immune responses NAM vaccines, such as the Moderna and Pfizer COVID vaccines, may employ different immunological mechanisms than SAMs and hence have different efficacies and toxicities. When we treated HSMC co-cultures with monocytes or DCs with a NAM encoding an RABV-G antigen that was encapsulated in the same LNP (NAM-LNP1), we found that, unlike the SAM-LNP1, it showed higher antigen expression in DCs compared to monocytes 1 day after exposure in co-culture with HSMCs (Figure 3A). Moreover, the overall expression induced by NAM-LNP1 was significantly lower than with SAM-LNP1 (not shown) even when using a 10-fold higher dose (1,000 ng/mL), likely due to its inability to amplify the number of mRNA copies. LNPs can affect tissue tropism and reactogenicity of the vaccine cargo. Thus, we also tested another LNP (LNP2) and found that the same NAM vaccine formulated in LNP1 or LNP2 exhibited different levels of antigen presentation in DCs (Figure 3B).Figure 3Induction of anti-RABV-G IgG and cytokine release by non-replicating mRNA(A) Quantification of monocytes and DCs after a 1-day co-culture with HSMC and 1,000 ng/mL of NAM-LNP1, assessed using immunofluorescence staining and detected by high-throughput imaging. Each bar represents the mean fluorescence intensity (MFI) of 18 fields of view in 5 independent wells of cells from 2 donors.(B) Quantification of live RABV-G++ DCs, by flow cytometry, 1 day after exposure to either 1,000 ng/mL of NAM-LNP1 or NAM-LNP2; each bar represents 3 biological replicates.(C) Anti-RABV-G IgG levels were detected in the effluents of LF chips cultured for 14 days using a cell-based assay; each data point represents one chip.(D) Heatmap showing average log2 fold changes of two technical repeats in cytokine levels measured via an MSDa Meso Scale Discovery assay in supernatants from HSMC-DC co-culture, 1 day after treatment with SAM-LNP1, NAM-LNP1, and NAM-LNP2 as compared to the empty LNP control.(E) Fold change in IL-6, IL-8, and CXCL10 levels measured via a Meso Scale Discovery (MSD) assay in supernatants from LF chips, 1, 7, and 14 days after treatment with SAM-LNP1, NAM-LNP1, and NAM-LNP2 as compared to the empty LNP control. Each bar represents three individual chips. Data shown are mean ± SD.Representative results from one donor are shown in (A)–(E), with similar outcomes observed in two donors. (A) Mann-Whitney U test; (B and C) ordinary one-way ANOVA followed by Tukey’s multiple comparisons test; (E) two-way ANOVA followed by Tukey’s multiple comparisons test. Error bars indicate SD. Induction of anti-RABV-G IgG and cytokine release by non-replicating mRNA (A) Quantification of monocytes and DCs after a 1-day co-culture with HSMC and 1,000 ng/mL of NAM-LNP1, assessed using immunofluorescence staining and detected by high-throughput imaging. Each bar represents the mean fluorescence intensity (MFI) of 18 fields of view in 5 independent wells of cells from 2 donors. (B) Quantification of live RABV-G++ DCs, by flow cytometry, 1 day after exposure to either 1,000 ng/mL of NAM-LNP1 or NAM-LNP2; each bar represents 3 biological replicates. (C) Anti-RABV-G IgG levels were detected in the effluents of LF chips cultured for 14 days using a cell-based assay; each data point represents one chip. (D) Heatmap showing average log2 fold changes of two technical repeats in cytokine levels measured via an MSDa Meso Scale Discovery assay in supernatants from HSMC-DC co-culture, 1 day after treatment with SAM-LNP1, NAM-LNP1, and NAM-LNP2 as compared to the empty LNP control. (E) Fold change in IL-6, IL-8, and CXCL10 levels measured via a Meso Scale Discovery (MSD) assay in supernatants from LF chips, 1, 7, and 14 days after treatment with SAM-LNP1, NAM-LNP1, and NAM-LNP2 as compared to the empty LNP control. Each bar represents three individual chips. Data shown are mean ± SD. Representative results from one donor are shown in (A)–(E), with similar outcomes observed in two donors. (A) Mann-Whitney U test; (B and C) ordinary one-way ANOVA followed by Tukey’s multiple comparisons test; (E) two-way ANOVA followed by Tukey’s multiple comparisons test. Error bars indicate SD. To test whether SAM-LNP1 vaccination (100 ng/mL) using DCs as APCs would be sufficient to induce anti-RABV-G IgG, we transferred the DCs from 2 donors and the outflows from the HSMC co-cultures to 3–4 LF chips per donor. We detected anti-RABV-G IgG in the effluents of two-thirds of the chips when SAM-LNP1 was tested using DCs as the APC. NAM-LNP1 and NAM-LNP2 (1,000 ng/mL) also induced the production of anti-RABV-G IgG compared to empty LNPs, although the levels were modest compared to those induced by SAM-LNP1 (Figures 3C and S4E). The induction of inflammatory cytokines by NAM-LNP1 and NAM-LNP2 was also significantly lower compared to SAM-LNP1, both during the first day of 2D culture and after 14 days in LF chips (Figures 3D and 3E). We then performed quantitative mass spectrometry to compare the proteomic responses induced by NAM-LNP1 (1,000 ng/mL) versus SAM-LNP1 (100 ng/mL) in immune cells extracted from vaccinated LF chips from one donor. Proteomic analysis revealed that SAM-LNP1 caused distinct and broader perturbations in the host proteome compared to NAM-LNP1 (Figure 4A). Eight proteins were downregulated by both SAM and NAM. Among these were H1.4 and H1.5 that are critical for chromatin compaction and which have been reported to influence immune cell differentiation by modulating multiple signaling pathways.33,34 There were no common upregulated proteins between SAM and NAM-treated LF chips, suggesting significant differences in the pathways altered by SAM and NAM vaccines.Figure 4Different proteomic responses induced by self-amplifying mRNA and non-amplifying mRNA vaccines(A) Venn diagram showing the number of overlapping vs. distinct protein perturbations induced by SAM-LNP1 and NAM-LNP1.(B) Gene Ontology analysis highlights biological processes predominantly regulated by SAM-LNP1 compared to the empty LNP1.(C) Heatmap illustrating the top differentially expressed proteins, labeled by gene locus name, showcasing the extent of regulation by the treatments. Each data point represents three technical replicates from one independent chip, and 3–4 chips were analyzed per condition. Different proteomic responses induced by self-amplifying mRNA and non-amplifying mRNA vaccines (A) Venn diagram showing the number of overlapping vs. distinct protein perturbations induced by SAM-LNP1 and NAM-LNP1. (B) Gene Ontology analysis highlights biological processes predominantly regulated by SAM-LNP1 compared to the empty LNP1. (C) Heatmap illustrating the top differentially expressed proteins, labeled by gene locus name, showcasing the extent of regulation by the treatments. Each data point represents three technical replicates from one independent chip, and 3–4 chips were analyzed per condition. SAM-LNP1 predominantly affected proteins involved in antiviral defense and cytokine production regulation (Figure 4B), strongly upregulating interferon response proteins, particularly members of the antiviral oligoadenylate synthase (OAS)35 and interferon-induced proteins with tetratricopeptide repeats (IFIT)36 families. This is consistent with cellular sensing of replicating RNA. While OAS protein expression is variable, IFIT proteins are typically expressed at low levels in healthy cells and are strongly induced upon interferon exposure or viral infection. Similarly, myxovirus resistance (MX) proteins, which are antiviral proteins induced by viral infection,37 were significantly upregulated by SAM-LNP1. Additionally, SAM vaccines induced the expression of C-C motif chemokine receptor 2 (CCR2), which plays a key role in the maturation of APCs38 and their trafficking to draining lymph nodes39 via the expression of C-C motif ligand 2 (CCL2, also known as monocyte chemoattractant protein 1, MCP-1). IL-2, crucial for the early priming and activation of T cells,40,41 was also significantly induced by SAM vaccines. In contrast, NAM-LNP1 induced only modest proteomic changes compared to the empty LNP group (Figure 4C), consistent with the significantly lower mRNA levels42 in the NAM versus SAM groups. Notably, the proteins upregulated by NAM-LNP1 are closely associated with T cell responses. Among these, two proteins involved in T cell activation showed increased abundance. Kinase suppressor of Ras 1 (KSR1), a scaffold protein, plays a crucial role in regulating the activation threshold of extracellular signal-regulated kinase (ERK) in T cells,43,44 and Unc-13 Homolog D (UNC13D) is essential for mediating the release of cytolytic granules from T cells45 (Figure S6). These results provide a basis for designing future experiments with LF chips made with cells from a larger number of donors to understand if they are observed in all vaccinees or those of a specific age, sex, or ethnicity. SHM induced by mRNA-LNP vaccines To study the recall responses of mRNA vaccines in this LF Chip-based intramuscular vaccination platform, we used the Moderna Spikevax Bivalent Original and Omicron BA.4/BA.5 vaccine and conducted experiments using both DCs and monocytes from vaccinated HSMC co-cultures, mixed with autologous lymphocytes and perfused with the HSMC-APC co-culture conditioned medium supplemented with IL-2 and IL-4. This vaccine was chosen because induction of neutralizing antibodies has been well documented in individuals receiving this vaccine, and SHMs often require multiple vaccine exposures.46 In multiple LF chips created from the cells of 4 different donors, vaccination using the Moderna mRNA vaccine (500 ng/mL) induced robust IgG responses against the SARS-CoV-2 Wuhan strain, as well as the Omicron BA.4 and BA.5 variants. The IgG titer in unvaccinated and vaccinated chips against the Wuhan spike was 3–10 times higher than the titer against the BA.4 and BA.5 spikes (Figure 5A), which may reflect the wide distribution of the monovalent vaccines against the Wuhan strain within Massachusetts where our immune cell donor pool is located. To directly assess specificity, we also measured antibody titers against an unrelated antigen (HIV-1 gp120 protein, BJOX002_D11 gp120) on the same batch of chip outflows. As expected for a specific response, gp120 antibody levels remained at background across all chips, indicating that the antibody response we detected was specific for the SARS-CoV-2 antigens (Figure S7). We also verified the neutralization efficiency of the antibodies produced by the LF chips from 2 donors (Figure 5B) by testing the ability of the chip outflows to block infection of ACE2-expressing 293T cells by a pseudovirus expressing the Wuhan strain spike protein.Figure 5Induction of spike-specific neutralizing antibodies and somatic hypermutation in LF chips following vaccination with Moderna bivalent COVID-19 vaccine(A) Moderna bivalent COVID-19 vaccine was administered at 500 ng/mL (M500). Day 9 LF Chip outflows post-vaccination (M500) and no treatment (NT) were tested to determine COVID-19 spike-specific IgG concentrations using the Meso Scale Discovery (MSD) V-PLEX SARS-CoV-2 Panel 27 (IgG) Kit; each point represents one chip.(B) Neutralization by outflows from each no treatment or vaccinated chip created from the cells of one individual donor is depicted. 0% neutralization is defined by the luciferase signal in cells infected with SARS-CoV-2 pseudovirus using fresh medium controls, and 100% neutralization corresponds to the signal in uninfected cells; each line represents one chip. Similar results were obtained using cells from a 2nd donor.(C) BCR repertoire analysis on LF chips from two donors. Total templates, unique sequences, mutated templates, and total mutations of DNA templates from cells recovered from both vaccinated (M500) and no treatment chips from two donors are shown.(D) Comparison of SHM counts of top 10 V-D-J combinations from both vaccinated (M500) and no treatment chips from both donors.(E) Changes in average SHM in vaccinated samples compared to unvaccinated samples across both donors.Data shown are mean ± SD, (A) two-way ANOVA followed by Tukey’s multiple comparisons test, (D) Wilcoxon matched pairs signed rank test. Induction of spike-specific neutralizing antibodies and somatic hypermutation in LF chips following vaccination with Moderna bivalent COVID-19 vaccine (A) Moderna bivalent COVID-19 vaccine was administered at 500 ng/mL (M500). Day 9 LF Chip outflows post-vaccination (M500) and no treatment (NT) were tested to determine COVID-19 spike-specific IgG concentrations using the Meso Scale Discovery (MSD) V-PLEX SARS-CoV-2 Panel 27 (IgG) Kit; each point represents one chip. (B) Neutralization by outflows from each no treatment or vaccinated chip created from the cells of one individual donor is depicted. 0% neutralization is defined by the luciferase signal in cells infected with SARS-CoV-2 pseudovirus using fresh medium controls, and 100% neutralization corresponds to the signal in uninfected cells; each line represents one chip. Similar results were obtained using cells from a 2nd donor. (C) BCR repertoire analysis on LF chips from two donors. Total templates, unique sequences, mutated templates, and total mutations of DNA templates from cells recovered from both vaccinated (M500) and no treatment chips from two donors are shown. (D) Comparison of SHM counts of top 10 V-D-J combinations from both vaccinated (M500) and no treatment chips from both donors. (E) Changes in average SHM in vaccinated samples compared to unvaccinated samples across both donors. Data shown are mean ± SD, (A) two-way ANOVA followed by Tukey’s multiple comparisons test, (D) Wilcoxon matched pairs signed rank test. To understand whether the variable neutralization efficiency of the induced antibodies in the outflow from different chips and donors (Figure 5B) reflects a lack of B cell activation and maturation or the detection limits of our workflow, we sequenced the CDR3 loop of the IgH locus from cells isolated from the LF chips from 2 donors 14 days after vaccination. Due to the diversity of the human immune repertoire and the limited overlap between chips from the same donor, sequences from 3 chips for each donor were pooled for further analysis; over 100,000 unique sequences were obtained per donor (Figure 5C). Both donors showed an increase in the total number of B cell sequences, unique sequences, B cells with SHMs, and total SHMs upon vaccination. Donor 2, which produced higher levels of neutralizing antibodies, had more B cells with existing mutations, and yet a significant increase in SHM was still observed upon vaccination; vaccination did not alter V, D, or J gene usage (data not shown). The top 10 V-D-J combinations had >100 mutations in the LF Chip even without vaccination, which indicates the baseline diversity of the donor BCR repertoire, but vaccination resulted in a 5%–10% increase in SHMs in donor 2 (Figure 5D), and a similar trend was observed in donor 1. We found that vaccinated chips had a higher average level of SHMs as clone size increased (Figure 5E), a pattern evident in both donors, and donor 2 had larger clones than donor 1. To look at these effects in greater detail, we compared sequences from specific, highly expanded V-D-J combinations between vaccinated and unvaccinated chips. As shown for a representative V-D-J combination from donor 2 (Figure 6), vaccination increased the number of mutated sequences and the number of mutations in expanded clones, confirming that the LF Chip displays SHM upon stimulation. The BCR sequences of 3 out of the top 5 expanded clones show highly significant homology with known anti-SARS-CoV-2 antibody sequences, as shown for a representative clone in Figure S8.Figure 6Clonal expansion and somatic hypermutation in LF chipsClonal sequences of BCR containing the V-D-J genes IGHV01-69, IGHD04-23∗01, and IGHJ06-01∗02 were analyzed using ClustalW for alignment. Phylogenetic analysis was subsequently visualized with iTol, generating the depicted dendrograms.(A) Dendrogram of sequences from no treatment chips.(B) Dendrogram of sequences from chips vaccinated with 500 ng/mL Moderna bivalent COVID-19 vaccine. Colors on the outer ring indicate the number of mutations per sequence. Branch labels represent the size of each clone, and the numbers within the ring serve as arbitrary sequence identifiers. Clonal expansion and somatic hypermutation in LF chips Clonal sequences of BCR containing the V-D-J genes IGHV01-69, IGHD04-23∗01, and IGHJ06-01∗02 were analyzed using ClustalW for alignment. Phylogenetic analysis was subsequently visualized with iTol, generating the depicted dendrograms. (A) Dendrogram of sequences from no treatment chips. (B) Dendrogram of sequences from chips vaccinated with 500 ng/mL Moderna bivalent COVID-19 vaccine. Colors on the outer ring indicate the number of mutations per sequence. Branch labels represent the size of each clone, and the numbers within the ring serve as arbitrary sequence identifiers. Discussion The efficacy of mRNA vaccines is closely tied to the mRNA format, delivery vehicle, and route of administration.19,47,48 In this study, we developed an in vitro model mimicking the intramuscular injection site and combined it with LF chips to create an experimental platform that provides a detailed view of mRNA vaccine uptake and immune responses in human lymph nodes and at the intramuscular vaccine site. We observed that the expression of the antigen encoded by the SAM vaccine in APCs depends on co-culture with muscle cells and the formation of direct cell-cell contacts. This co-culture also enhanced the activation of APCs in response to mRNA vaccination, as indicated by increased expression of surface markers (e.g., CD86 and HLA-DR) and the release of pro-inflammatory cytokines (e.g., IL-6 and IL-8). IL-6, a key regulator of T helper cell subsets,49,50,51 was specifically induced by SAM vaccines when APCs were co-cultured with muscle cells. These findings emphasize that the non-immune cells present at the vaccine site can affect the efficacy of vaccines and merit study as previously noted by others.19,46,52,53,54 In contrast, NAM vaccines can be taken up directly by monocytes and DCs in the absence of muscle cells; however, the LNP component affects cytokine release and antigen expression.17,21,52,55 To detect adaptive immune responses induced by mRNA vaccines, we mixed APCs from the intramuscular vaccination simulation module with autologous T and B cells to create an LF Chip model of a draining lymph node, leveraging our published method.16 The conditioned medium from the intramuscular vaccination module containing released soluble factors was also transferred to the top channel of the LF Chip. Notably, we found that SAM vaccines induce LF formation, and the addition of IL-2 and IL-4 enhances antigen-specific antibody production, although it does not further alter the number or size of the LFs. These LF chips can be maintained for up to 28 days in culture, allowing for the long-term study of cytokine release induced by mRNA vaccines. The secreted cytokines we detected, including IL-1α, IL-1β, IL-6, and CXCL10, are consistent with findings from previous in vivo studies of SAM vaccines.20,22,23,32 Our experiments with rabies vaccines, whose antigen most of the population is naive to, were carried out in chips with bulk lymphocytes from randomly chosen donors as well as in chips containing only naive B cells from rabies naive donors. The results of these studies demonstrate that mimicking intramuscular vaccination and subsequent use of the LF Chip supports IgG generation even when challenged with naive antigens, a capability lacking in other systems that utilize blood-derived cells.17,18 It should be noted that our study of vaccine-induced naive immune responses is limited to RABV-G, which is an immunogenic protein. Further studies will be required to validate that this in vitro immune platform can mount similar responses against other naive antigens. Comparing multiple LNP delivery vehicles and RNA payload types opens opportunities for in vitro comparative studies to identify optimal combinations of delivery vehicles and mRNA formats. Our on-chip study with the cells from 2 different donors suggests that in humans also, SAM vaccines can induce similar immune responses at doses 10 times lower than those required for NAM vaccines as previously reported in mice.56 Interestingly, while the cytokine induction patterns of the three mRNA vaccines tested were similar, SAM vaccines triggered significantly more robust cytokine responses and antigen-specific antibodies than NAM vaccines. This may be related to our finding from proteomic analysis of the chips from one donor, which revealed that SAM and NAM vaccines used in this study activate distinct patterns of protein expression. Of the proteins altered differently by SAM and NAM vaccines, only 8 overlapped between the 2 vaccine formats, with 83 unique proteins induced by the SAM and 60 by NAM vaccines. SAM vaccines appeared to increase expression of proteins involved in more robust viral defense and cytokine regulation pathways. To our knowledge, this is the first reported comparison of human immune cell proteome changes induced by different mRNA vaccine formats, and we anticipate that further studies of vaccine-induced signaling pathways will provide valuable insights for future vaccine design. In the future, similar studies in LF chips from multiple donors could confirm whether our initial findings are relevant broadly to vaccines of specific age, sex, or ethnicity. We also tested the Moderna bivalent COVID-19 mRNA vaccine to investigate memory responses and booster effects using our method for modeling intramuscular vaccination. Since most of our immune cell donors had likely received a COVID-19 mRNA vaccine due to state policies in Massachusetts, we anticipated observing memory immune responses against the SARS-CoV-2 antigen contained with this vaccine in cells from these donors. Our results were consistent with previous findings, showing that the bivalent COVID-19 mRNA vaccine induced 3–10 times more antibodies against the Wuhan strain compared to the Omicron BA.4 and BA.5 variants.57,58 Additionally, the neutralizing efficacy of the antibodies produced by the LF Chip platform was confirmed, and SHM analysis revealed that the cultured human B cells underwent SHM within the chip environment. This is also the first report of inducing SHM in blood-derived B cells in vitro with a clinical mRNA vaccine. In conclusion, this study establishes a robust method for interrogating human immune responses to mRNA-LNP vaccination at both the vaccine injection site and within lymphoid tissues. In contrast to previous methods that primarily focused on memory responses,15,16,17,18 this preclinical model is capable of evaluating both de novo and memory human B cell responses to multiple mRNA vaccine formats. We also characterized the dynamics of cytokine release and antibody production associated with different mRNA vaccine modalities, which potentially can be further leveraged to profile in vivo immune responses during clinical trials. While this study primarily focused on cytokine release and B cell responses, future work could expand on these findings by investigating antigen-specific T cell responses. Additionally, this system could be used to study the roles of other immune cells during vaccination, such as mast cells, plasmacytoid DCs, and natural killer cells. Further refinements could include adding stromal cells to the intramuscular module to more accurately represent the complexity of the injection site, as fibroblasts and other stromal cell types play a crucial role in antigen uptake and innate immune responses.59,60 Most importantly, these data demonstrate that this experimental model of human vaccination offers an exciting alternative to NHP studies in future vaccine development programs. Limitations of the study The intramuscular model presented in the manuscript includes skeletal muscle cells and APCs, but additional stromal populations at the injection site and draining lymph node are absent, which may contribute to additional antigen uptake and innate sensing. Donor-to-donor and sex-based variability may have an impact on the outcomes, and the sample size (number of donors and vaccines) needs to be increased for further detailed investigation. Resource availability Lead contact Further information, resources, and materials should be requested from the lead contact, Dr. Girija Goyal (girija.goyal@wyss.harvard.edu). Materials availability This study did not produce new unique reagents. The rabies glycoprotein encoding SAM and NAM vaccines and the two LNP formulations they were delivered in are proprietary. All other materials including cells were acquired from commercial sources, pharmacies, and biobanks as described. Data and code availability The mass spectrometry proteomics data have been deposited at the ProteomeXchange Consortium and are publicly available as of the date of publication (accession number ProteomeXchange: PXD056699). This paper does not report original code. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Acknowledgments The studies on SAM and NAM vaccines were sponsored by GlaxoSmithKline Biologicals SA. The studies with Moderna vaccines and on somatic hypermutation were supported by funding from the 10.13039/100000865Bill and Melinda Gates Foundation (INV-002274 to G.G. and D.E.I.) and the in-kind Women in Science Award from 10.13039/100016796Adaptive Biotechnologies (to G.G.). The development of the LF Chip was also supported by funding from the 10.13039/100000865Bill and Melinda Gates Foundation (INV-002164 to D.E.I.), the 10.13039/100012399Biomedical Advanced Research and Development Authority (contract 75A50121C00075 to D.E.I.), and the Biomedical Advanced Research and Development Authority-U.S. Food and Drug Administration (contract 75A50123D00004 to D.E.I.). SARS-CoV-2-specific IgG antibody quantification in organ-chip outflows was supported by funding from Gates Foundation (INV-060822 to G.D.T.), as well as by a Wyss Technology Development Fellowship (to Y.Z.). The authors would like to thank Harvard University Health Services for sharing the unused portion of the Moderna vaccine from opened vials to enable the studies on the production of neutralizing antibodies and somatic hypermutation. Author contributions Y.Z., J.M., D.E.I., and G.G. conceived the study. J.M., D.E.I., and G.G. oversaw the study. Y.Z., M.W.K., K.Y., L.H., S.F., A.S.P., G.M., S.B., K.L., C.A.H., L.D.W., G.D.T., P.Y., S.P.G., J.M., D.E.I., and G.G. participated in experiment design and/or discussions. Y.Z., M.W.K., K.Y., A.P., P.P., L.H., Y.M., M.R., K.M., S.B.S., S.G., A.R.I., J.F., K.M., E.M., C.A.H., and L.D.W. performed the experiments and data collection. Y.Z., M.W.K., K.Y., D.E.I., and G.G. analyzed the data, prepared the figures and tables, and wrote the manuscript with input from all authors. D.E.I. and G.G. reviewed and edited the manuscript. All authors approved the final version. Declaration of interests D.E.I. is a founder, board member, scientific advisory board chair, and equity holder in Emulate, Inc.; D.E.I. and G.G. are co-inventors on relevant patent applications. S.F., A.S.P., G.M., S.B., J.F., K.M., E.M., K.L., and J.M. are employees of the GSK group of companies and may own GSK shares and/or restricted GSK shares. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERAntibodiesBUV395 Mouse anti-human CD14 (clone MφP9)BD BiosciencesCat#563561; RRID:AB_2744288Brilliant Violet 421™ anti-human CD1c (clone L161)BioLegendCat#331526; RRID:AB_10962909APC anti-human HLA-DR AntibodyBioLegendCat#361610; RRID:AB_2563200PerCP/Cyanine5.5 anti-human CD80 AntibodyBioLegendCat#305231; RRID:AB_2566490Brilliant Violet 605™ anti-human CD86 AntibodyBioLegendCat#305430; RRID:AB_2563824APC anti-human IgG Fc AntibodyBioLegendCat#410712; RRID: AB_2565790Mouse anti-Rabies virus glycoprotein G (clone 24-3F-10)Millipore/Merck (Millipore Corp)Cat#MAB8727; RRID:AB_571110Goat anti-Mouse IgG (H + L) Cross-Adsorbed Secondary Antibody, Alexa Fluor 647Invitrogen (Thermo Fisher Scientific)Cat#A-21235; RRID:AB_2535804Biological samplesDe-identified human patient-derived apheresis collarsCrimson Biomaterials Collection Core Facility (Harvard University)IRB protocol 22470Chemicals, peptides, and recombinant proteinsSkGM-2 Skeletal Muscle Cell Growth Medium-2 BulletKit®LonzaCat#CC-3245Lymphoprep™ density gradient mediumSTEMCELL TechnologiesCat#07801Recovery™ Cell Culture Freezing MediumGibco (Thermo Fisher Scientific)Cat#12648-010Human GM-CSF (research grade)Miltenyi BiotecCat#130-095-372Trypsin-EDTA (0.05%), phenol redGibco (Thermo Fisher Scientific)Cat#25300054Cell Recovery SolutionCorningCat#354253Pierce Protease and Phosphatase InhibitorThermo FisherCat# A32961(3-Aminopropyl)trimethoxysilane (APTMS)Sigma-Aldrich (Merck)Cat#281778Corning® Matrigel® Basement Membrane MatrixCorningCat#356234Corning® Collagen I, High Concentration, Rat TailCorningCat#354249Hoechst nucleic acid stain (solution)Invitrogen (Thermo Fisher Scientific)Cat#H3570ViaKrome 808 Fixable Viability DyeBeckman CoulterCat#C36628Human Interleukin-2 (IL-2)Miltenyi Biotec130-097-742Human Interleukin-4 (IL-4)Miltenyi Biotec130-093-922Critical commercial assaysEasySep™ Human B Cell Enrichment KitSTEMCELL TechnologiesCat#19054EasySep™ Human naive B Cell Enrichment KitSTEMCELL TechnologiesCat#17254EasySep™ Human T cell Isolation KitSTEMCELL TechnologiesCat#17951EasySep™ Human Monocyte Enrichment Kit without CD16 DepletionSTEMCELL TechnologiesCat#19058V-PLEX SARS-CoV-2 (Spike) Panel 27 (IgG) KitMeso Scale Discovery (MSD)Cat#K15437UDeposited dataMass spectrometry proteomics datasetProteomeXchange ConsortiumProteomeXchange: PXD056699Experimental models: Cell linesHSMM – Human Skeletal Muscle Myoblasts (HSMCs)LonzaCat#CC-2580293T cellATCCCat # CRL-3216Software and algorithmsFlowJo (v10)FlowJo LLCRRID:SCR_008520Imaris (v9.3)Bitplane (Oxford Instruments)RRID:SCR_007370GraphPad Prism (v10.1.0)GraphPad SoftwareRRID:SCR_002798Sapphire softwareAzure BiosystemsV1.12.0921.0MetaXpressMolecular Devices6.5.3.427immunoSEQAdaptive BiotechnologiesRRID: SCR_014709ClusterProfiler (R package)Bioconductor/R communityN/AImageJFiji/ImageJv 2.14.0Discovery Workbench 4.0Meso Scale Discoveryv4.0Bio-Plex Manager softwareBio-Radv6.0ComplexHeatmap (R package)Bioconductor/R communityN/AEnhancedVolcano (R package)Bioconductor/R communityN/AiTOLhttps://itol.embl.deN/AClustalWhttps://www.genome.jp/tools-bin/clustalwN/ABioRenderBioRender.comN/AimmunoSEQ AnalyzerAdaptive BiotechnologiesN/AComethttps://uwpr.github.io/Comet/N/AOtherChip-S1 Stretchable ChipEmulate Inc.Cat#10231-2ZOE-ORB instruments (Organ-Chip culture + hub modules)Emulate Inc.CM1CytoFLEX LX flow cytometerBeckman CoulterN/ALeica SP5 confocal microscopeLeicaN/AMeso Sector S 600 microplate readerMeso Scale Discovery (MSD)N/AMSD microplate reader model 1300Meso Scale Discovery (MSD)N/A REAGENT or RESOURCESOURCEIDENTIFIERAntibodiesBUV395 Mouse anti-human CD14 (clone MφP9)BD BiosciencesCat#563561; RRID:AB_2744288Brilliant Violet 421™ anti-human CD1c (clone L161)BioLegendCat#331526; RRID:AB_10962909APC anti-human HLA-DR AntibodyBioLegendCat#361610; RRID:AB_2563200PerCP/Cyanine5.5 anti-human CD80 AntibodyBioLegendCat#305231; RRID:AB_2566490Brilliant Violet 605™ anti-human CD86 AntibodyBioLegendCat#305430; RRID:AB_2563824APC anti-human IgG Fc AntibodyBioLegendCat#410712; RRID: AB_2565790Mouse anti-Rabies virus glycoprotein G (clone 24-3F-10)Millipore/Merck (Millipore Corp)Cat#MAB8727; RRID:AB_571110Goat anti-Mouse IgG (H + L) Cross-Adsorbed Secondary Antibody, Alexa Fluor 647Invitrogen (Thermo Fisher Scientific)Cat#A-21235; RRID:AB_2535804Biological samplesDe-identified human patient-derived apheresis collarsCrimson Biomaterials Collection Core Facility (Harvard University)IRB protocol 22470Chemicals, peptides, and recombinant proteinsSkGM-2 Skeletal Muscle Cell Growth Medium-2 BulletKit®LonzaCat#CC-3245Lymphoprep™ density gradient mediumSTEMCELL TechnologiesCat#07801Recovery™ Cell Culture Freezing MediumGibco (Thermo Fisher Scientific)Cat#12648-010Human GM-CSF (research grade)Miltenyi BiotecCat#130-095-372Trypsin-EDTA (0.05%), phenol redGibco (Thermo Fisher Scientific)Cat#25300054Cell Recovery SolutionCorningCat#354253Pierce Protease and Phosphatase InhibitorThermo FisherCat# A32961(3-Aminopropyl)trimethoxysilane (APTMS)Sigma-Aldrich (Merck)Cat#281778Corning® Matrigel® Basement Membrane MatrixCorningCat#356234Corning® Collagen I, High Concentration, Rat TailCorningCat#354249Hoechst nucleic acid stain (solution)Invitrogen (Thermo Fisher Scientific)Cat#H3570ViaKrome 808 Fixable Viability DyeBeckman CoulterCat#C36628Human Interleukin-2 (IL-2)Miltenyi Biotec130-097-742Human Interleukin-4 (IL-4)Miltenyi Biotec130-093-922Critical commercial assaysEasySep™ Human B Cell Enrichment KitSTEMCELL TechnologiesCat#19054EasySep™ Human naive B Cell Enrichment KitSTEMCELL TechnologiesCat#17254EasySep™ Human T cell Isolation KitSTEMCELL TechnologiesCat#17951EasySep™ Human Monocyte Enrichment Kit without CD16 DepletionSTEMCELL TechnologiesCat#19058V-PLEX SARS-CoV-2 (Spike) Panel 27 (IgG) KitMeso Scale Discovery (MSD)Cat#K15437UDeposited dataMass spectrometry proteomics datasetProteomeXchange ConsortiumProteomeXchange: PXD056699Experimental models: Cell linesHSMM – Human Skeletal Muscle Myoblasts (HSMCs)LonzaCat#CC-2580293T cellATCCCat # CRL-3216Software and algorithmsFlowJo (v10)FlowJo LLCRRID:SCR_008520Imaris (v9.3)Bitplane (Oxford Instruments)RRID:SCR_007370GraphPad Prism (v10.1.0)GraphPad SoftwareRRID:SCR_002798Sapphire softwareAzure BiosystemsV1.12.0921.0MetaXpressMolecular Devices6.5.3.427immunoSEQAdaptive BiotechnologiesRRID: SCR_014709ClusterProfiler (R package)Bioconductor/R communityN/AImageJFiji/ImageJv 2.14.0Discovery Workbench 4.0Meso Scale Discoveryv4.0Bio-Plex Manager softwareBio-Radv6.0ComplexHeatmap (R package)Bioconductor/R communityN/AEnhancedVolcano (R package)Bioconductor/R communityN/AiTOLhttps://itol.embl.deN/AClustalWhttps://www.genome.jp/tools-bin/clustalwN/ABioRenderBioRender.comN/AimmunoSEQ AnalyzerAdaptive BiotechnologiesN/AComethttps://uwpr.github.io/Comet/N/AOtherChip-S1 Stretchable ChipEmulate Inc.Cat#10231-2ZOE-ORB instruments (Organ-Chip culture + hub modules)Emulate Inc.CM1CytoFLEX LX flow cytometerBeckman CoulterN/ALeica SP5 confocal microscopeLeicaN/AMeso Sector S 600 microplate readerMeso Scale Discovery (MSD)N/AMSD microplate reader model 1300Meso Scale Discovery (MSD)N/A Experimental model and study participant details Expansion of HSMCs HSMCs (Lonza, CC-2580) were rapidly thawed in a water bath and resuspended in Skeletal Muscle Growth Medium-2 (SkGM-2, Lonza, CC-3245). Cells were seeded in a T-150 flask for expansion. Confluency and morphology were monitored daily using brightfield microscopy, and media were changed every three days. Cells were used for vaccination experiments upon reaching 70–80% confluency. Peripheral blood mononuclear cells (PBMC) isolation and DC differentiation De-identified human patient-derived apheresis collars were obtained from the Crimson Biomaterials Collection Core Facility, under Institutional Review Board approval from Harvard University (protocol 22470). To confirm that the SAM-LNP vaccination against rabies represented a naive immune response, we obtained human leukopaks from Sanguine Biosciences, Inc., from donors who self-reported no history of rabies infection or vaccination. Seronegativity was verified using our anti-rabies antibody detection assay (Figure S4D). PBMCs were isolated by density centrifugation using Lymphoprep (StemCell Technologies, 07801), followed by magnetic bead-based negative selection for B cells (StemCell Technologies, 19054), naive B cells (StemCell Technologies, 17254), T cells (17951), and monocytes (19058). For cryopreservation, cells were frozen in Recovery Cell Culture Freezing Medium (Gibco, 12648-010) at 20 million cells/mL. Monocytes were differentiated into DCs by culturing in complete RPMI with 400 ng/mL GM-CSF (Mitenyi Biotec, 130-095-372) and 250 ng/mL IL-4 for 5–6 days, with 50% medium replacement every 2 days. Method details Rabies Virus Glycoprotein and enhanced green fluorescent protein (eGFP) mRNA preparation The antigen sequence of the full-length Rabies Virus Glycoprotein is based on the Pasteur strain (GenBank accession number: AAA47218.1). The Open Reading Frame (ORF) was optimized for expression in mammalian cells and cloned into SAM or mRNA backbones using standard molecular biology techniques. In the SAM construct, the Rabies G gene was cloned in the first position and the eGFP (GenBank AAB02572) second, separated by the IRES from Enterovirus 71 (EV71). In vitro RNA transcription, purification, and LNP formulation were performed as described previously.61,62,63 Tri-link prepared conventional mRNA with the same Rabies Virus Glycoprotein sequence, incorporating n-1-methyl-pseudouridine in place of uridine. LNP formulation preparation Lipid and RNA components were prepared in ethanol and citrate buffer, respectively. LNPs encapsulating RNA were formulated by lipid nanoprecipitation using a modified ethanol dilution process. After the formation of the LNP, the LNP was buffer exchanged using a desalting column into a holding buffer. A sucrose excipient was added to the LNPs for final storage. The two LNP formulations tested in this study differ in the ionizable lipid component; all other constituents—including PEGylated lipid, sterol lipid, and phospholipid—were held constant. The rationale for comparing these formulations was to evaluate the impact of different ionizable lipids on delivery efficiency and expression. The specific identity and phenotype of the ionizable lipids is proprietary to GSK and cannot be disclosed, consistent with practices in previously published studies.52,61,63 Co-culture of muscle cells and Antigen-Presenting Cells (APCs) and vaccination Skeletal muscle cells were trypsinized using 0.05% trypsin-EDTA (Gibco, 25300054), resuspended in SkGM-2 medium, and seeded at a density of 120,000 cells per well in a 24-well plate. Cells were incubated overnight to allow for adherence. Monocytes were thawed, or cultured dendritic cells (DCs) were collected and resuspended at a concentration of 0.5 million cells/mL, and 0.5 million monocytes or DCs were added to the muscle cell wells along with the indicated mRNA vaccine. Human LF chip Two-channel microfluidic Organ Chip devices and ZOE-ORB instruments were obtained from Emulate Inc. The Organ Chips were plasma-activated and seeded as previously described.16 Briefly, chips were perfused with 1% 3-Aminopropyltrimethoxysilane (Sigma, 281778) in ethanol for 1 h, followed by washing the channels with ethanol. Chips were incubated in an 80°C oven overnight. For muscle-APC co-culture experiments, the APCs were picked up from the coculture by pipetting gently, counted, and then suspended in RPMI complete medium, stocked on ice, and subjected to flow cytometry staining to verify the activation and GFP expression or used immediately for seeding together with autologous T and B cells. For all RABV-G mRNA vaccine experiments except Figure 2G, APCs were mixed with bulk T and bulk B cells at a 5 : 47.5: 47.5 ratio to achieve a final concentration of 1.5 × 10ˆ8 cells/mL in complete RPMI supplemented with Matrigel (60%; Corning, 356234) and type I collagen (15%; Corning, 354249). For COVID-19 mRNA vaccine experiments, DCs and monocytes from muscle co-cultures were mixed with T and B cells in a 1:1:49:49 ratio. A 15 μL aliquot of the cellularized ECM mixture was introduced into the bottom channel of the 2-channel Organ Chip. After allowing the matrix to gel for 30 min at 37°C, the top channel was filled with 1 part conditioned medium from the 2D intramuscular module, and 2 parts fresh medium supplemented with IL-2 (Mitenyi Biotec, 130-097-742) and IL-4 (Mitenyi Biotec, 130-093-922) for experiments with RABV-G mRNA vaccines. For experiments with COVID-19 mRNA vaccines, a 1:1:8 ratio of DC co-culture medium, monocyte co-culture medium, and fresh media (containing IL-2 and IL-4) was used. The chips were perfused with indicated media at 40 μL/h using ZOE instruments in a 37°C incubator. Medium was recirculated every 2–3 days. On days 7, 14 and 21, 50% of the medium was replaced with fresh medium containing IL-2 and IL-4. In Figure 2G, the protocol described above was followed with two key exceptions: only naive B cells were used and the donors’ sera were tested to confirm that they were naive to RABV-G by demonstrating an absence of anti-RABV-G IgG. Immunostaining and imaging Chips were fixed by adding 200 μL of 4% paraformaldehyde to one side and allowing 30-50 μL to flow out. Both ports were then plugged, and an additional 200 μL was added. After 1 h of rocking at room temperature to facilitate back and forth flow of paraformaldehyde solution in the channel, chips were washed twice with PBS for 15 min. Nuclear staining was performed using Hoechst (1:1000 dilution, Invitrogen, H3570) for 1 h, followed by three 15-min PBS washes. Follicles were defined based on the presence of a closely bound cluster of cells, visualized using Hoechst nuclear stain, which allowed clear differentiation of follicular cell clusters from surrounding tissue. Imaging was performed on a Leica confocal microscope using a 25× water immersion objective (1× zoom, 100 μm pinhole size, 400 Hz, 512 × 512 resolution). Z-stacks were acquired with a 1 μm step size from just below the bottom of the bottom channel to just above the membrane to capture the entire 3D structure of the follicles. To minimize experimenter bias, fields of view were pre-defined systematically: for each chip, 2–3 representative stacks were collected from standardized regions to avoid preferential selection of areas with higher follicle density. Follicle quantification was performed using Imaris software (Bitplane, version 9.3). Images were first corrected for signal attenuation across Z-stacks by measuring mean fluorescence intensity at the top and bottom slices and applying attenuation correction. Surface rendering was performed with 1 μm surface detail, and follicles were identified using a voxel threshold of 15,000, sphericity ≥0.25, and ellipticity ≤0.3. Manual threshold adjustments, if needed, were uniformly applied to all images within a session. This ensured consistent follicle identification across datasets. To prevent double-counting of follicle-like structures, the entire z stack for each field was analyzed, ensuring that each follicle was counted once based on its 3D rendered surface rather than 2D projections. The size of each field of view corresponded to the physical dimensions at 25× magnification with the above settings were then exported to an excel file for statistical analysis. For imaging vaccinated HSMC in Figure S1C, HSMC vaccinated for 24 h in a 96-well plate were imaged with a Sapphire Biomolecular Imager (Azure Biosystems) and analyzed using Sapphire software (V1.12.0921.0). In Figure 4A, co-cultured vaccinated APCs were immunostained with anti-Rabies-G (Millipore Corp, MAB8727, 1:1000) and goat anti-mouse IgG (H&L) Cross-Adsorbed Secondary Antibody (Invitrogen, A-21235, 1:2000). Images were taken with an ImageXpress Micro4 (Molecular Devices) and analyzed using MetaXpress (6.5.3.427) and ImageJ software (v 2.14.0). Extraction of live cells from organ chips The gel from the bottom channel of the LF chip was collected in Cell Recovery Solution (Corning, 354253). The collection tube was centrifuged at 300g for 5 min, and the supernatant was aspirated. The pellet was resuspended in 1 mL of fresh Cell Recovery Solution and incubated on ice for 1 h. Following incubation, the sample was centrifuged again at 300g for 5 min, the supernatant was aspirated, and the cells were prepared for downstream applications. Flow cytometry Cell viability was assessed using ViaKrome 808 Fixable Viability Dye (Beckman Coulter, C36628, 1:100 dilution). Cells were stained with antibodies for CD14 (BD 563561), CD1c (BioLegend 331526), HLA-DR (BioLegend 361610), CD80 (BioLegend 305231), CD86 (BioLegend 305430), anti-Rabies-G (Millipore Corp, MAB8727) and goat anti-mouse IgG (H&L) Cross-Adsorbed Secondary Antibody (Invitrogen, A-21235). Data were collected on a CytoFlex LX (Beckman Coulter) and analyzed using FlowJo V10 software (FlowJo LLC). Cell-based assay for anti-rabies antibody detection 293T cells (1 x 10ˆ6) were seeded in 6-well plates in DMEM with 10% FBS and 1% P/S, 48 h prior to transfection. Cells were transfected with 50 ng/mL of SAM-RABV-G-GFP-LNPs or empty LNPs (control) and incubated in 2 mL of serum-free DMEM. We used the SAM vaccine to make RABV-G expressing 293T. Cells were incubated with outflows to be tested for anti-RABV-G antibodies and then cells stained with ViaKrome 808 dye and anti-human IgG APC (Biolegend 410712). Differential staining with anti-human IgG of control vs. RABV-G expressing 293T indicated the presence of anti-RABV-G antibodies. Specific binding was calculated as shown in Figure S4B. Proteome analysis by mass spectrometry The mass spectrometry-based multiplexed proteomic analysis was performed as previously described.64 Briefly, human immune cells were cultured for two weeks at LF chips with different mRNA vaccines. The cells were then lysed in 8 M urea and 200 mM EPPS pH 8.5 with Pierce Protease and Phosphatase Inhibitor (Thermo Fisher, A32961). Protein concentration was measured by BCA assay. ∼20 μg proteome in 20 μL was treated sequentially with 20 mM TCEP and 200 mM IAA for 15 min and 30 min, respectively. 3 μL SP3 beads (1:1 mixture of hydrophobic and hydrophilic type, 50 mg/mL, (Cat. #45152105050250 and Cat. #65152105050250) and 30 μL ethanol supplemented with 20 mM DTT were added to the lysate. SP3-bind proteome was washed twice with 80% Ethanol followed by sequential digestion with 0.2 μg Lys-C and 0.2 μg trypsin for 2 h and overnight at 37°C, respectively. The digested peptides were added 10 μL acetonitrile and labeled with 4 μL TMTpro reagent for 1 h before combining all samples into one single multiplexed sample and desalting through Sep-Pak C18 column. The eluent was dried and then fractionated by basic-pH HPLC for total 24 fractions. Both fractionated proteome samples were analyzed by liquid chromatography-tandem mass spectrometry (LC-MS/MS) on an Orbitrap Eclipse mass spectrometer coupled to a Proxeon NanoLC-1200 UHPLC50 and FAIMS Pro device. Data were collected alternating between a set of three FAIMS compensation voltages (CVs) at −40, −60, and −80 V was used. MS1 scans were collected in the Orbitrap with a resolution setting of 60 K for proteome, a mass range of 400–1600 m/z, an AGC at 100%, and a maximum injection time of 50 ms. MS2 scans were acquired in Top Speed mode with a cycle time of 1 s. Peptide precursors were selected and fragmented using HCD with a collision energy of 36. MS2 scans were collected in the Orbitrap with a resolution of 50K, a fixed scan range of 110–2000 m/z, and a 500% AGC with a maximum injection time of 86 ms. Dynamic exclusion was set to 90 s with a mass tolerance of ±10 ppm. The mass spectra were processed using a Comet-based pipeline and searches were performed using a 50-ppm precursor ion tolerance for the analysis of the total protein levels.65 The product ion tolerance was set to 0.9 Da. Tandem mass tags on lysine residues and peptide amino termini (+304.2071 Da) and carbamidomethylation of cysteine residues (+57.021 Da), were set as static modifications, whereas oxidation of methionine residues (+15.995 Da) was set as a variable modification. Peptide-spectrum matches were adjusted to a 1% false-discovery rate.66 Proteins were quantified by summing the reporter-ion counts across all matching peptide-spectrum matches, as described previously. Gene Ontology analysis and graphs was generated using R packages ClusterProfiler,67 Complexed Heatmap and Enhanced Volcano. Proteomic data availability The mass spectrometry data have been deposited at the ProteomeXchange Consortium and are publicly available. The accession number is ProteomeXchange: PXD056699. (The reviewer access can be provided if required during submission.). Immunoglobulin and cytokine quantification To measure immunoglobulins targeting multiple SARS-CoV-2 spike protein variants in the day 9 outflow from both unvaccinated and vaccinated LF chips, the Meso Scale Discovery (MSD) V-PLEX SARS-CoV-2 Spike Panel 27 (IgG) Kit (K15437U) was used, with data acquired on an MSD Meso Sector S 600 microplate reader and analyzed with Discovery Workbench 4.0 software. The MSD V-PLEX SARS-CoV-2 Spike Panel 27 (IgG) Kit included 96-well plates immobilized with Wuhan, BA.4, and BA.5 spike antigens (data reported in Figure 6A), along with a beta variant (B.1.351), delta variant (B.1.617.2; AY.4 Alt Seq 2) and additional omicron sub-lineage spike proteins (BA.2, BA.2.12.1, BA.2 + L452M, BA.2 + L452R, and BA.3; data not shown), in addition to a reference standard and three controls with known levels of antibodies against antigens in the kit. Assays were performed following the MSD-provided, kit-specific protocol, testing LF chip outflows in duplicate and including CH65 IgG monoclonal antibody (mAb) (influenza receptor binding site-specific mAb) as a negative control. Data from LF chip outflow dilutions (1:10, 1:70, 1:490) falling nearest to the midpoint of the accurate range of the standard curve were selected for reporting. Values chosen were back calculated against the standard curve and reported as arbitrary units per milliliter (AU/mL), accounting for dilution factor. The cytokine levels in the media were assessed using a custom MSD kit (as presented in Figures 1D, 3D, 3E, and S2B) and read using an MSD microplate reader model 1300 alongside the Discovery Workbench 4.0 software. Customized Luminex Multiplex Assay kits from Thermo Fisher Scientific were employed for the cytokine concentrations in the effluents depicted in Figure 2G. Analyte concentrations were determined on a Bio-Plex 3D suspension array system, and the standard curve fitting and concentration calculations were conducted using Bio-Plex Manager software (Bio-Rad, version 6.0). BCR nucleotide sequences analysis Human immune cells were cultured on LF chips for two weeks with 500 ng/mL of the Moderna bivalent COVID-19 mRNA vaccine. Following this, cells were extracted from the chips and sent to Adaptive Biotechnologies for sequencing of the CDR3 loop of IgH. The data was then analyzed and visualized using the immunoSEQ Analyzer, Clustal W,68 iTOL69 and pivot charts in Prism 10. Quantification and statistical analysis All experiments (excluding proteomics) were performed at least twice, with at least three technical replicates for each group in each experiment. Data are presented as mean values ±standard deviation (SD). Graphing and statistical comparisons were carried out using GraphPad Prism software (v10.1.0). Statistical significance for comparisons between two groups was determined using an unpaired t test. A P-value of <0.05 was considered statistically significant, with exact P-values labeled in all figures.