SARS-CoV-2 mRNA vaccines sensitize tumours to immune checkpoint blockade

Abstract

Immune checkpoint inhibitors (ICIs) extend survival in many patients with cancer but are ineffective in patients without pre-existing immunity^{1,2,3,4,5,6,7,8,9}. Although personalized mRNA cancer vaccines sensitize tumours to ICIs by directing immune attacks against preselected antigens, personalized vaccines are limited by complex and time-intensive manufacturing processes^{10,11,12,13,14}. Here we show that mRNA vaccines targeting SARS-CoV-2 also sensitize tumours to ICIs. In preclinical models, SARS-CoV-2 mRNA vaccines led to a substantial increase in type I interferon, enabling innate immune cells to prime CD8⁺ T cells that target tumour-associated antigens. Concomitant ICI treatment is required for maximal efficacy in immunologically cold tumours, which respond by increasing PD-L1 expression. Similar correlates of vaccination response are found in humans, including increases in type I interferon, myeloid–lymphoid activation in healthy volunteers and PD-L1 expression on tumours. Moreover, receipt of SARS-CoV-2 mRNA vaccines within 100 days of initiating ICI is associated with significantly improved median and three-year overall survival in multiple large retrospective cohorts. This benefit is similar among patients with immunologically cold tumours. Together, these results demonstrate that clinically available mRNA vaccines targeting non-tumour-related antigens are potent immune modulators capable of sensitizing tumours to ICIs.

Main

Although ICIs substantially improve survival in some patients, most patients do not benefit from these therapies^{1,2,3,4,5,6,7,8,9}. These poor responses are attributed to immunosuppressive tumour microenvironments (TMEs) characterized by tolerogenic dendritic cells (DCs), myeloid suppressor cells and regulatory T cells, and may be predicted in some histologies by low intratumoural PD-L1 before immunotherapy initiation. However, there are currently no clinically available methods to improve responses to ICI by modifying the TME. We recently reported that systemic administration of highly immunogenic mRNA nanoparticles induces a viraemia-like cytokine/chemokine response that resets the systemic and intratumoural immune milieu, sensitizing resistant tumours to ICIs^10,11,12,14. Although the personalized mRNA vaccines that we and others are developing remain in clinical evaluation (NCT04573140)¹², COVID-19 mRNA vaccines also induce robust stimulation of cytokine secretion¹⁵, and there are now multiple case reports of patients whose tumours spontaneously resolved after COVID-19 mRNA vaccines^16,17. However, the impact of COVID-19 mRNA vaccines on immune therapy is unknown.

Here we report that the innate immune response to SARS-CoV-2 spike mRNA vaccination resets the cancer immunotherapy cycle and primes adaptive immunity for synergy with ICIs. We found that receipt of a SARS-CoV-2 mRNA vaccine within 100 days of ICI initiation was associated with substantial improvements in overall survival (OS) in patients with non-small cell lung cancer (NSCLC) and melanoma. In preclinical models, we found that this effect required a surge in type-I interferon (IFN) that enhanced antigen-presenting cell (APC) priming of T cells in lymphoid organs. Although tumour cells subvert these primed responses by increasing PD-L1 expression, co-administration of ICIs sustains T cell responses and elicits epitope spreading against tumour-associated antigens. We revealed analogous response correlates for humans receiving COVID-19 mRNA vaccines, including heightened IFNα production, innate/adaptive immune activation and increases in tumour PD-L1 expression. Together, our results demonstrate that clinically available mRNA vaccines targeting non-tumour antigens are potent immune modulators capable of sensitizing tumours to ICIs.

Improved survival with COVID-19 vaccination

To determine whether COVID-19 mRNA vaccines were associated with improved responses to immune checkpoint blockade, we first compared the OS among a cohort of patients with stage III/IV NSCLC treated at The University of Texas MD Anderson Cancer Center (MDACC) between January 2015 and September 2022 (Supplementary Table 1a). We identified 180 patients who received a COVID mRNA vaccine within 100 days of ICI initiation, and 704 patients who were treated with ICI and did not receive a COVID vaccine (Supplementary Table 1a). Of the 180 patients who received an mRNA vaccine within 100 days of ICI initiation, 117 received the BNT162b2 vaccine and 63 received mRNA-1273 (Extended Data Fig. 1a); 24 received a priming dose only, 57 received a booster only, 93 received both a prime and a boost dose, 5 received 2 booster doses, and 1 received a priming dose and 2 booster doses within 100 days of ICI initiation (Extended Data Fig. 1b); 81 received 1 dose of COVID-19 mRNA vaccination within 100 days, 98 received 2 doses, and 1 received 3 doses (Extended Data Fig. 1c). After controlling for 39 covariables with Cox proportional hazards regression, including clinical stage, histology, steroid use, performance status, mutation status, comorbidities and treatment year, we found that receipt of a COVID-19 mRNA vaccine within 100 days of initiation of ICI was associated with significantly improved median OS (20.6 months versus 37.3 months) and 3 year OS (30.8% versus 55.7%, adjusted hazard ratio (HR_adj = 0.51, 95% confidence interval (CI) = 0.37–0.71, P < 0.0001) (Fig. 1a and Supplementary Tables 2 and 3). This survival advantage was similar for patients with stage III unresectable NSCLC (HR_adj = 0.37, 95% CI = 0.16–0.89, P = 0.0268) (Fig. 1b and Supplementary Tables 4 and 5) and stage IV NSCLC (HR_adj = 0.52, 95% CI = 0.37–0.74, P = 0.0002) (Fig. 1c and Supplementary Tables 6 and 7); patients who received mRNA vaccines from either vaccine manufacturer (Extended Data Fig. 1a); and patients who had or had not received a previous COVID-19 mRNA vaccine (Extended Data Fig. 1b). Patients who received two vaccines in the 100 days surrounding initiation of ICI experienced similar OS compared with those who received only one vaccine (Extended Data Fig. 1c). These results were also consistent when considering only those patients whose closest mRNA vaccine was within 100 days before their first ICI (Extended Data Fig. 1d), when narrowing the vaccination window to 50 instead of 100 days (Extended Data Fig. 1e), when restricting to only those patients treated during the pandemic (Extended Data Fig. 1f), after correcting for immortal time bias (Extended Data Fig. 1g) and with propensity score matching (PSM; Extended Data Fig. 1h,i). Patients who received a COVID-19 vaccine within 100 days of chemotherapy (a group that did not include targeted therapies owing to significant heterogeneity and limited patient numbers within different drug cohorts) but did not receive ICI had no detectable survival benefit (Extended Data Fig. 2a). Likewise, patients who received a pneumonia or influenza vaccine within 100 days of initiating ICI (Extended Data Fig. 2b–e) and those with resectable stage III tumours (Extended Data Fig. 2f,g) experienced no improvement in survival.

Fig. 1: COVID-19 mRNA vaccines are associated with improved survival in patients with NSCLC or metastatic melanoma receiving immunotherapy.

a–e, Survival for patients with NSCLC (a–c) or metastatic melanoma (d,e) treated with ICI who received a COVID-19 mRNA vaccine within 100 days of initiating ICI or did not receive a COVID-19 mRNA vaccine. Survival is shown for all patients with NSCLC (a), patients with unresectable stage III NSCLC (b), patients with stage IV NSCLC (c) and patients with metastatic melanoma (d and e). P values and HR_adj were calculated using two-sided Cox proportional hazards regression (Supplementary Tables 3, 5, 7, 9 and 11), including all variables that were significantly associated with survival on univariable analysis (Supplementary Tables 2, 4, 6, 8 and 10). The number of patients at risk at each timepoint is indicated below each graph.

Source Data

We then repeated this analysis in a separate cohort of patients treated with their first round of ICI for metastatic melanoma, including 43 patients who received a COVID-19 mRNA vaccine within 100 days of initiating ICI and 167 who did not receive a COVID-19 vaccine (Supplementary Table 1b). Of the patients who received mRNA vaccines, 21 received BNT162b2 and 22 received mRNA-1273 (Extended Data Fig. 3a); 6 received a priming dose only, 8 received a booster only, 16 received both a prime and a boost dose, and 13 received 2 booster doses during our period of interest, (Extended Data Fig. 3b); 14 received 1 dose of COVID-19 mRNA vaccination within 100 days, and 29 received 2 doses (Extended Data Fig. 3c). After accounting for covariables, including histology, steroid use, performance status, mutation status, comorbidities and treatment year with Cox proportional hazards regression, we found that receipt of a COVID-19 mRNA vaccine within 100 days of initiating ICI was associated with substantially improved OS (median OS, 26.67 months versus unmet; 36-month OS, 44.1% versus 67.6%; HR_adj = 0.37, 95% CI = 0.18–0.74, P = 0.0048) (Fig. 1d and Supplementary Tables 8 and 9) and progression-free survival (PFS) (median PFS, 4.0 months versus 10.3 months; 36-month PFS, 23.7% versus 39.5%; HR_adj = 0.63, 95% CI = 0.40–0.98, P = 0.0383) (Fig. 1e and Supplementary Tables 10 and 11). These effects were again similar for both vaccine manufacturers (Extended Data Fig. 3a), prime and boost vaccines (Extended Data Fig. 3b), patients receiving single or multiple vaccine doses (Extended Data Fig. 3c), when limiting the analysis to only patients who received their vaccine prior to initiating ICI (Extended Data Fig. 3d), when limiting the analysis to only those treated during the pandemic (Extended Data Fig. 3e,f) and when expanding the analysis to include patients on second- or third-line therapy and those without distant metastases (Extended Data Fig. 3g–i). Moreover, the survival advantage was magnified with PSM (OS: HR = 0.44, 95% CI = 0.18–0.77, P = 0.0063; PFS: HR = 0.46, 95% CI = 0.27–0.77, P = 0.0022) (Extended Data Fig. 3j,k).

Type I IFN mediates RNA vaccine immunity

To demonstrate whether effects observed in humans could be modelled in animals, we recreated commercial preparations of COVID mRNA vaccines for administration to tumour-bearing animals in conjunction with ICIs. As SARS-CoV-2 spike protein has been fully sequenced, we synthesized the published mRNA construct used for the Pfizer/BioNTech vaccine (BNT162b2) and validated the fidelity of synthesis based on mRNA size (Extended Data Fig. 4a,b) and the ability to elicit neutralizing antibodies after in vivo administration to animals (Extended Data Fig. 4c). mRNA was encapsulated in lipid nanoparticles (LNPs) and met the specification range delineated for clinical preparations of BNT162b2 based on encapsulation efficiency, size distribution, polydispersity and charge¹⁸ (Extended Data Fig. 4d–g), the latter of which we found could be disproportionately affected by buffer conditions leading to a net positive charge (Extended Data Fig. 4h). We chose mouse B16F0 melanoma and Lewis lung carcinoma (LLC) as models to test immunogenicity and efficacy of spike RNA-LNP vaccines due to the effects of the vaccine in these clinical settings and because both tumour models are poorly responsive to ICIs. We first treated mice with two vaccine doses in conjunction with ICI treatment. We found this regimen to be superior to either monotherapy alone in mice with established B16F0 tumours with tumour volumes of about 80 mm³ (Fig. 2a and Extended Data Fig. 5a), in mice with subcutaneous (s.c.) LLC (Fig. 2b and Extended Data Fig. 5b) and in mice with established s.c. LLC with tumour volumes around 100 mm³ (Fig. 2c and Extended Data Fig. 5c). In mice with established LLC, we identified metastatic lesions in the lungs of untreated mice that were similar for monotherapy but significantly reduced with the combination treatment (Extended Data Fig. 5d,e). We also found that starting RNA-LNPs before ICI produced similar effects relative to concomitant treatment (Extended Data Fig. 5f). We next tested whether RNA-LNPs could reduce the growth of orthotopic intrapulmonary tumours. We implanted LLC cells orthotopically and administered treatment starting on day 3. In this model, the combination of RNA-LNPs and ICI resulted in superior inhibition of tumour growth as measured by lung weights (Extended Data Fig. 5g).

Fig. 2: Spike RNA-LNPs prime anti-cancer immunity in an IFN-I dependent manner.

a, The experimental design and tumour volume of mice inoculated with B16F0 cells. Groups included untreated (UT; n = 7), anti-PD-1 (n = 8), RNA-LNPs (n = 8) and RNA-LNPs + anti-PD-1 (n = 8). mAb, monoclonal antibody. b, The experimental design and tumour volume for mice inoculated with LLC cells. Groups included untreated (n = 8), anti-PD-1 (n = 9), RNA-LNP (n = 9) and RNA-LNPs + anti-PD-1 (n = 9). c, The experimental design and tumour volume measurements for mice inoculated with LLC cells. Groups included untreated (n = 9), anti-PD-1 (n = 10), RNA-LNP (n = 7) and RNA-LNPs + PD-1 (n = 8). d, The experimental design and tumour volume for mice inoculated with B16F0 cells. n = 12 per group. e, IFNα plasma enzyme linked immunosorbent assay (ELISA) from B16F0-tumour-bearing mice (n = 8 per group) 24 h after one RNA-LNP vaccine (day 3). f–i, Cellular phenotyping within 24 h of vaccine 3 (days 3, 6 and 20) of cells from spleens of mice bearing B16F0 tumours (n = 5 per group), including the percentage of activated (CD80⁺CD86⁺) DCs (f) and macrophages (g). h,i, PD-L1 median fluorescence intensity (MFI) on activated mature DCs (h) and macrophages (i). j–o, Characterization of antigen presentation among myeloid cells in tumour draining lymph nodes (tdLNs) and spleens of mice bearing B16F10-ova tumours 24 h after vaccine 2 (days 10 and 13). The overall percentage of CD45⁺ cells that express MHC-II⁺ (j), the percentage of MHC-II⁺ cells presenting SIINFEKL (k), the percentage of SIINFEKL-presenting MHCII⁺ cells that express the activation marker CD86 (l), the percentage of CD45⁺ cells that are MHC-II⁺Ly6C⁺ (m), SIINFEKL⁺MHC-II⁺Ly6C⁺ cells as a percentage of all CD45⁺ cells (n) and Ly6C⁺MHCII⁺ cells as a percentage of total SIINFEKL-presenting cells (o) are shown. n = 4 biologically independent mice per group. Significance was determined using two-way analysis of variance (ANOVA)/mixed-effect analysis with Geisser–Greenhouse correction (a–d) and two-tailed unpaired t-tests (e–o). n indicates the number of biologically independent samples. For the box plots, the whiskers extend to the highest and lowest values, the box limits show the first and third quartiles and the centre line shows the median value. For a–d, data are mean ± s.e.m.

Source Data

We next sought to understand the mechanism by which RNA-LNPs targeting the spike protein mediate these antitumour effects. To do this, we used an early treatment model of B16F0 with or without cytokine-blocking antibodies. In this model, RNA-LNPs and PD-L1 blockade each provided numerical but statistically insignificant survival benefits. However, combination therapy with RNA-LNPs and PD-L1 blockade strongly inhibited tumour growth (Fig. 2d and Extended Data Fig. 6a,b). Although we previously demonstrated a role for IFNα in the response to mRNA vaccines, recent evidence suggests a dominant role for IL-1 signalling in responses to the specific RNA-LNPs targeting the COVID-19 spike protein¹⁵. We therefore evaluated the importance of each pathway with antibodies blocking the IL-1 and IFNα receptors. While blockade of IL-1R had no effect on tumour growth, anti-tumour responses were completely abrogated when blocking type I interferon signalling with IFNAR1 monoclonal antibodies (Fig. 2d). Moreover, direct administration of supraphysiologic doses of type I IFN recapitulated the antitumour effects (Extended Data Fig. 6c). However, stimulation of type I IFN signalling with low molecular mass (LMW) poly(I:C) did not elicit similar immunity (Extended Data Fig. 6d). These results highlight the importance of IFN in driving innate and adaptive immunity, and the ability to manipulate and reset the immune setpoint away from tolerance toward effector immune responses in the presence of ICIs using a commercially available mRNA preparation.

Next, we sought to assess whether other mRNA species would elicit similar antitumour effects. We first modified the protein encoded by the mRNA by replacing the spike mRNA with mRNA encoding the cytomegalovirus antigen pp65, which is overexpressed in human glioma but not in B16F0. We found no significant difference in antitumour activity between these two groups, suggesting that innate immune sensing of the mRNA itself is the primary driver of antitumour activity from mRNA vaccines (Extended Data Fig. 6e,f).

As N¹-methyl-pseudouridine is used in place of uridine in COVID-19 mRNA vaccines to reduce innate immune activation, we hypothesized that replacing N¹-methyl-pseudouridine with uridine to activate pattern recognition receptors would provide even more robust innate immune activation and antitumour activity. As expected, replacing N¹-methyl-pseudouridine with uridine in pp65 mRNA resulted in further synergy with ICI (Extended Data Fig. 6e,f). However, replacing N¹-methyl-pseudouridine with uridine in spike mRNA provided only numerical improvement in antitumour response (Extended Data Fig. 6e,f), suggesting that innate immune sensing is a multifactorial process that may be influenced by specific mRNA constructs.

We then completed additional studies to identify how RNA-LNPs elicit antitumour immunity. Previous work has established that RNA-LNPs stimulate type I IFN production by stimulating the intracellular double-stranded RNA (dsRNA) sensor MDA5¹⁹. However, the mechanism by which RNA-LNPs activate this sensor is unclear. To understand this further, we first ruled out a role for dsRNA in our vaccines by measuring the level of dsRNA contamination in our manufactured ssRNA product. We found the dsRNA/ssRNA ratio in our mRNA to be 0.011% (Extended Data Fig. 7a). We then repeated our tumour growth curve incorporating an additional dsRNA-removal protocol²⁰. With this step, we successfully removed all detectable dsRNA contamination (Extended Data Fig. 7a). Importantly, we found no change in antitumour efficacy with the complete elimination of dsRNA using this method (Extended Data Fig. 7b). Finally, we found no abrogation of type I IFN after RNA-LNP administration in mice lacking RIG-I, a critical sensor of dsRNA and activator of type I IFN response (Extended Data Fig. 7c). Together, these data suggest that dsRNA is not a major cause of the antitumour effects in our preclinical models.

We next tested whether the specific LNP construct impacted antitumour immunity with the same mRNA input. We found that anionic lipoplexes formulated as lipid particle aggregates (LPA) do not elicit similar antitumour effects (Extended Data Fig. 7b). Given that previous literature described that higher-order RNA structures may also stimulate MDA5 activation²¹, we hypothesized that RNA-LNPs might create higher order structures. Consistent with this hypothesis, we found that RNA extracted from our RNA-LNPs contained high-molecular-mass secondary structures similar in mass to dsRNA (Extended Data Fig. 7d and Supplementary Fig. 1). Combined with previous reports of MDA5 activation with RNA-LNPs but not with ssRNA alone, these data may suggest a mechanism by which encapsulation with LNPs forms high-molecular-mass secondary structures enabling activation of double-stranded sensing machinery (that is, MDA5) for induction of type I interferons.

RNA vaccines stimulate innate immunity

We next sought to better understand the impact of RNA-LNPs on innate immune cells. IFNα was significantly elevated in both RNA-LNP and combination groups in conjunction with other T helper 1 chemokines (Fig. 2e and Extended Data Fig. 8). This surge in cytokine/chemokine response correlated with an increase in myeloid cell activation in the lymphoid organs of mice treated with RNA-LNP alone or in combination with ICI. After treatment with spike-encoding RNA-LNP, there was precipitous activation of APCs, including DCs, macrophages and Ly6C⁺ cells expressing MHC class II (MHC-II) in lymphoreticular organs, that was abrogated by IFNAR1 blockade (Fig. 2f–i, Extended Data Fig. 9a,b, Supplementary Table 12 and Supplementary Fig. 2). Although this response was dependent on interferon signalling, the magnitude of the response could not be achieved by simply administering systemic IFNα in combination with ICI (Fig. 2f–i and Extended Data Fig. 9a,b).

Myeloid activation extended to the tumour, where receipt of RNA-LNPs was associated with IFNAR1-dependent increases in activated Ly6C⁺ myeloid cells (Extended Data Fig. 9c and Supplementary Fig. 3). To determine whether increased myeloid activation corresponded to enhanced presentation of tumour antigens, we treated mice with B16F10 tumours expressing chicken ovalbumin (B16F10-OVA) at days 10 and 13 and evaluated presentation of ovalbumin on MHC-I using flow cytometry. We found that spike-encoding RNA-LNPs stimulate APCs to present tumour antigens in lymphoid organs in the presence of costimulatory molecules (Fig. 2j–l and Supplementary Fig. 4). Notably, presentation of tumour antigens in the presence of costimulatory molecules was particularly enriched in Ly6C⁺MHC-II⁺ cells (Fig. 2m–o).

RNA vaccines reprogram adaptive immunity

In addition to the precipitous activation of APCs, spike-encoding RNA-LNPs in combination with ICIs also elicited expansion of CD8⁺ T lymphocytes with a marked increase in activation within the effector and effector memory compartments (Fig. 3a,b, Extended Data Fig. 9d, Supplementary Table 13 and Supplementary Fig. 5). Concomitant with these findings, PD-1 expression was increased in T cells and effector/effector memory CD8⁺ cell subsets, underscoring the potent ability of combination therapy to rapidly prime T cells (Extended Data Fig. 9e,f and Supplementary Fig. 6). Overall, these results illustrate a role for systemic immunomodulation in the recruitment of myeloid cells to lymphoid organs for tumor antigen presentation to cognate T cells.

Fig. 3: Spike RNA-LNPs generate tumour-reactive T cells and increase PD-L1 expression on tumour cells.

a,b, The percentage of activated effector (a) and effector memory (b) T cells in the spleens of tumour-bearing mice on day 21 (vaccine days 3, 6, 17) (n = 5 per group). c, The percentage of tetramer⁺ cells of splenic CD8⁺ T cells collected from mice bearing B16F0 tumours on day 21 (vaccination days 14 and 17). Groups include untreated (n = 6), anti-PD-1 (n = 6), RNA-LNPs (n = 5) and RNA-LNPs + anti-PD-1 (n = 7). d, The normalized percentage of AIM⁺ T cells (n = 5 mice per peptide; exceptions are shown in the Supplementary Information) after splenocyte co-culture with overlapping peptide pools. e,f, Representative images (e) and blinded manual counting (n = 4 tumours per group with 4 counts per tumour) (f) of PD-1⁺CD3⁺ cells by immunofluorescence 24 h after vaccine 3 (days 3, 6 and 20) from s.c. tumors of B16F0-tumour-bearing mice treated with or without anti-PD-L1. For e, scale bars, 100 μm. AF647, Alexa Fluor 647. g, The percentage PD-1⁺CD8⁺ cells of CD3⁺ T cells in tumours of B16F0-bearing mice vaccinated with RNA-LNPs (days 14 and 17). Groups included untreated (n = 7), anti-PD-L1 (n = 8), RNA-LNPs (n = 9) and RNA-LNPs + anti-PD-L1 (n = 7). h, Pooled tetramer positivity (%) among CD8⁺ T cells in B16F0 tumours. Groups included untreated (n = 6), anti-PD-1 (n = 6), RNA-LNPs (n = 4) and RNA-LNPs + anti-PD-1 (n = 7) (RNA-LNPs days 14 and 17). i, PD-L1 expression on B16F0 tumour cells (CD45^−FSC-A^high) isolated from mice 24 h after vaccine 3 (days 3, 6 and 17) as determined using flow cytometry. Groups included untreated (n = 4) and RNA-LNPs (n = 5). j,k, Blinded manual counting (n = 6 tumours per group with 4 counts per tumour) (j) and representative images (k) of PD-L1⁺ tumour cells (SOX10) by immunofluorescence 24 h after vaccine 3 (days 3, 6, 21) from B16F0-tumour-bearing mice. For k, scale bars, 50 μm. For j, the circle symbols indicate PBS treatment and the square symbols represent anti-PD-L1 treatment. For f and j, the colours represent individual tumours. Significance was determined using two-tailed unpaired t-tests (a–c and g–j), two-tailed Welch’s t-test (f), and two-tailed Brown–Forsythe and Welch ANOVA, followed by Dunnett’s T3 multiple-comparison test (d). For the box plots, the whiskers extend to the highest and lowest values, the box limits show the first and third quartiles and the centre line shows the median value. n values indicate biologically independent samples unless indicated otherwise.

Source Data

To confirm that spike RNA-LNPs were mediating the expansion of tumour reactive T cells, we isolated CD8⁺ cells from spleens of treated mice and stained them with tetramers targeting peptides with high predicted binding affinity from six melanoma-associated antigens: GP-100 EGSRNQDWL, GP-100 KVPRNQDWL, claudin 6 (CLDN6), survivin, WT1 and Trp2. We found that combination therapy with RNA-LNPs and ICIs stimulated expansion of tetramer reactive T cells targeting each of these antigens (Fig. 3c and Supplementary Fig. 7). To confirm that these T cells were truly tumour reactive, we used an activation-inducible marker (AIM) assay^22,23. In this assay, antigen-reactive T cells cultured ex vivo with overlapping peptide pools from the same tumour antigens are identified through tandem expression of the AIMs CD69 and 4-1BB (Supplementary Fig. 8 and Supplementary Table 14). Using this approach, we detected a substantial increase in peptide-reactive (AIM⁺) CD8⁺ T cells from mice treated with the combination of RNA-LNPs and ICI, further supporting tumour reactivity (Fig. 3d). These data confirm that spike RNA-LNPs prime the immune response for activation, presentation and recognition of tumour-associated antigens in a manner that can be significantly expanded through concomitant treatment with ICI.

RNA vaccines induce T cell infiltration

We next evaluated the T cell compartment in treated tumours. In mice with B16F0 tumours, we identified substantial infiltration of PD-1⁺CD8⁺ T cells using both immunofluorescence (Fig. 3e,f, Supplementary Fig. 6 and Supplementary Table 15) and flow cytometry (Fig. 3g). Importantly, we found that treatment with RNA-LNPs and ICIs increased PD-1 expression on total tumour-infiltrating CD8⁺ cells by greater than twentyfold (2.39% versus 51.363%, P < 0.0001) (Fig. 3f), and that PD-1⁺CD8⁺ T cells dominated the total CD3⁺ T cell compartment in treated mice while representing only a small minority of total CD3⁺ T cells in untreated mice (5.69% versus 60.6%, P < 0.001) (Fig. 3g). We next evaluated the antigen specificity of these cells. As there were many fewer CD8⁺ T cells in tumours compared with in spleens after enriching CD8⁺ cells, we pooled the six tetramers described above for a single pan-tetramer stain containing all six targets. Using this approach, we found that CD8⁺ tumour-infiltrating lymphocytes from mice treated with RNA-LNPs and ICIs were twice as likely to be tetramer reactive compared with non-RNA-LNP controls (3.10% versus 7.98%, P = 0.0229) (Fig. 3h and Supplementary Fig. 7).

Commensurate with the increase in tumour-reactive T lymphocytes in the TME, we found that spike-encoding RNA-LNPs significantly increased PD-L1 expression on tumour cells (Fig. 3i–k, Extended Data Fig. 9g,h and Supplementary Figs. 9 and 10). Blockade of IFNα signalling abrogated PD-L1 expression, confirming its importance in initiating the immunotherapy response (Extended Data Fig. 9g,h). Together, these data suggest that spike RNA-LNPs stimulate the production and infiltration of activated, tumour-reactive CD8⁺ T cells that overcome compensatory expression of PD-1 and PD-L1 in the presence of ICIs.

COVID-19 vaccines shape immunity in humans

As there are species-specific differences in how humans and mice respond to mRNA, we next sought to confirm that the pathways identified in mouse models are relevant in humans. To do this, we collected blood and plasma samples from five healthy volunteers at the baseline and at 6 h, 24 h, 7 days and 14  days after receipt of mRNA-1273 Spikevax Monovalent XBB.1.5 (COVID-19 mRNA vaccine, 2023–2024 formulation, 50 μg mRNA) (Fig. 4a). We first evaluated plasma from these volunteers using the NULISA-Seq Inflammation Panel (Alamar Biosciences)—a multiplex assay designed to sensitively detect over 250 immune-related cytokines. As IFNα was found to be essential for antitumour activity in mice, we were particularly interested in the kinetics of this protein at 24 h. Notably, not only was IFNα upregulated, but it was the most upregulated cytokine at any timepoint, increasing by an average of around 280-fold relative to the baseline to a final plasma concentration between 1 and 10 pg ml^−1 (Fig. 4b,c). More broadly, IL-6 and IFNγ were the only cytokines that were significantly elevated at 6 h after immunization (Fig. 4d). By contrast, a multitude of inflammatory cytokines surged at 24 h after immunization, including IFNα, IFNγ, IFNω and IFNγ-inducible protein-10 (CXCL10) (Fig. 4d and Supplementary Fig. 11). Although IL-1 was not found to be significantly increased at any timepoint, IL-1R antagonist (IL1RN) was significantly elevated at 24 h, suggesting a compensatory response to a surge in IL-1 (Fig. 4d). As expected, cytokine responses to COVID-19 mRNA vaccination were short-lived in healthy individuals, with all cytokines returning to the baseline levels by 7 days (Fig. 4d).

Fig. 4: COVID-19 mRNA vaccines generate a surge in IFNα, innate immune activation and adaptive immunity in humans.

a, Schematic of the experimental design in which blood was drawn from five healthy individuals at baseline and 6 h, 24 h, 7 days and 14 days after Spikevax (mRNA-1273) COVID-19 mRNA immunization. b,c, Individual datapoints highlighting changes in expression of IFNα from baseline to 24 h for each of five healthy volunteers. Data are expressed as the fold change measured using the NULISAseq Inflammation Panel (b). The concentration was also measured separately with NULISAseq absolute quantification (AQ) (c). d, Dynamic expression of the cytokines that are significantly elevated at 24 h at 6 h, 24 h, 7 days and 14 days after COVID-19 mRNA vaccination. Significant variables were defined as those with P < 0.05 and a log₂-transformed fold change with an absolute value of greater than 0.5 after linear modelling with fixed effects. Adjusted P values were calculated using moderated two-tailed t-tests with false-discovery rate (FDR) correction for multiple testing. e,f, PD-L1 expression on circulating myeloid cells (CD3^−CD19^−CD56^−CD11b⁺) (n = 5) (e) and DCs (CD3^−CD19^−CD56^−CD11c⁺MHC-II⁺) (n = 5) (f) at 6 h, 24 h and 7 days after immunization. g,h, Activation of natural killer cells (CD56⁺; n = 5) (g), and T cells expressed as numbers of CD69⁺ cells of CD8⁺CD3⁺ cells (n = 5) (h) at 6 h, 24 h, 7 days and 14 days after immunization. Data are mean ± s.e.m. P values were calculated using two-tailed paired t-tests.

Source Data

To evaluate the impact of COVID-19 mRNA-induced antiviral cytokines on immune cells, we next evaluated the phenotypes of circulating myeloid cells using flow cytometry. Consistent with our findings in mouse models, we found that mRNA immunization drove innate immune activation exemplified by increased expression of PD-L1 on circulating CD11b⁺ myeloid cells and CD11c⁺ DCs (Fig. 4e,f, Supplementary Table 16 and Supplementary Fig. 12). Vaccination was also associated with activation of natural killer cells exemplified by a doubling of expression of IL-2Rα (also known as CD25) on CD56^high cells (Fig. 4g) and circulating T cells exhibited by a doubling of expression of CD69 (Fig. 4h, Supplementary Table 17 and Supplementary Fig. 13). Phenotypes for all cells normalized by 7 days, which may be a result of the lack of target in these healthy participants or lymphoreticular localization of activated T cells for polarization into memory cells.

We next repeated this study in 11 healthy volunteers who received the Comirnaty vaccine (BNT162b2, COVID-19 mRNA vaccine, 2024–2025 formula, with 30 μg mRNA) (Extended Data Fig. 10a–h). Although we found similar changes to the cytokine profile overall, the magnitude of the increase in type I IFN and innate immune activation by flow cytometry was significantly reduced with BNT162b2, which contains less mRNA relative to mRNA-1273 (Extended Data Fig. 10b–i).

RNA vaccines amplify PD-L1 in NSCLC

As infiltration of antigen-specific T cells in tumours is known to be associated with increases in PD-L1 expression on tumour cells and was correlated with antitumour response in our preclinical models, we hypothesized that patients who received a COVID-19 mRNA vaccine would exhibit higher PD-L1 expression on their tumours. To test this hypothesis, we assembled two cohorts. We first evaluated 2,315 pathology reports from patients with NSCLC with biopsies reporting tumour proportion score (TPS), separating patients into three groups based on the timing between their biopsy and their most recent mRNA vaccine (Fig. 5a). We found that patients who had received a COVID-19 mRNA vaccine less than 100 days before biopsy exhibited a 24% increase in mean TPS of PD-L1 compared with patients who had not received any COVID-19 mRNA vaccines before biopsy (31% versus 25%, P = 0.0450) and a 41% increase in mean TPS relative to patients who received an mRNA vaccine 100 or more days before biopsy (31% versus 22%, P = 0.0099) (Fig. 5b and Extended Data Fig. 10j). As a TPS of 50% is a clinically important threshold to determine whether patients with NSCLC are eligible for single-agent immunotherapy instead of chemoimmunotherapy, we next evaluated these data as a binary outcome around this threshold. Notably, we found that patients who received a COVID-19 mRNA vaccine were 29% more likely to meet or exceed the 50% TPS threshold over unvaccinated patients (36% versus 28%, P = 0.0295) (Fig. 5c), suggesting that mRNA vaccines have a sufficient impact on TPS to modify treatment decisions. By contrast, pre-biopsy influenza and pneumonia vaccines were not associated with TPS changes (Fig. 5d).

Fig. 5: COVID-19 mRNA vaccines are associated with increased PD-L1 expression on tumours and improved clinical outcomes across a broad set of tumour histologies.

a, Schematic of patients with NSCLC biopsies documenting PD-L1 TPS. b, TPS stratified by COVID-19 mRNA vaccination timing. c, The distribution of samples with TPS ≥ 50%. d, TPS stratified by influenza (left) or pneumonia (right) vaccination timing. e, Schematic of biopsies documenting TPS or combined positive score (CPS) of PD-L1 at our institution (January 2020 to October 2023). f, Primary tumour locations from this diverse cohort. g, TPS in the tissue-agnostic cohort stratified by COVID-19 mRNA immunization timing. h, TPS stratified by timing of influenza vaccination. P values were calculated using two-tailed unpaired t-tests (b, d and h), two-tailed unpaired t-tests with Welch’s correction for unequal variance (g) and two-sided Fisher’s exact test evaluating the likelihood of TPS greater than 50% (c). The violin plots show the distribution of data with individual datapoints included. i, Survival of patients in the tissue-agnostic cohort treated with ICI who received any COVID-19 vaccine within 100 days of initiating ICI or did not receive any COVID-19 vaccine. j, Survival of patients in i stratified by receipt of COVID-19 vaccine before ICI. k, Survival for patients in i who started ICI in the pandemic era (since 2 September 2020, 100 days before mRNA vaccine approval). Survival analyses in the tissue-agnostic cohort were not limited to only those patients with a clear TPS value. l–o, The OS for patients with metastatic stage IV NSCLC treated with ICI who received a COVID-19 mRNA vaccine within 100 days of initiating ICI or did not receive a COVID-19 vaccine who had baseline PD-L1 expression at baseline biopsy TPS < 1% (l), 1–49.9% (m) or ≥50% (n). To evaluate the impact of vaccination in each clinical setting, patients were excluded if they received a COVID-19 mRNA vaccine before their biopsy. o, OS of unvaccinated patients with stage IV NSCLC stratified by era of ICI start who had baseline TPS < 1% at biopsy. P values and HRs were calculated using log-rank (Mantel–Cox, two-sided) tests (i–o).

Source Data

RNA vaccines amplify tumour PD-L1 broadly

To expand our findings beyond NSCLC and melanoma, we assembled a separate cohort including all patients at our quaternary referral centre with pathology reports including the term ‘PD-L1’ in a four-year period including the pandemic era (Fig. 5e). Together, we identified 5,317 unique pathology reports from January 2020 to October 2023 with the term ‘PD-L1’ including 2,831 reporting TPS from a diverse array of patients representing a variety of primary sites and histologies (Fig. 5f). In this cohort, receipt of a COVID-19 mRNA vaccine within 100 days before biopsy was associated with a 37% increase in TPS (13.3% versus 9.7%, P = 0.0364) (Fig. 5g and Extended Data Fig. 10k), similar to the effect seen in patients with NSCLC. As in the NSCLC cohort, influenza vaccines were not associated with similar increases in TPS (Fig. 5h). We next repeated the survival analysis in this broad patient cohort, including patients at our institution with a biopsy for PD-L1 during our dates of interest who received ICI (n = 888). In this cohort, patients who received any COVID-19 vaccine within 100 days of initiating ICI experienced significantly improved survival relative to their unvaccinated peers (P = 0.0038, HR = 0.73, 95% CI = 0.60–0.90) (Fig. 5i). This effect was consistent when limited to patients who received their vaccine within 100 days before the start of ICI (P = 0.0311, HR = 0.76, 95% CI = 0.60–0.96) (Fig. 5j) and patients who started their ICI during the pandemic era (P = 0.0056, HR = 0.74, 95% CI = 0.60–0.91) (Fig. 5k).

RNA vaccines boost ICIs in cold tumours

We next sought to evaluate whether vaccination could restore immune sensitivity in patients with immunologically ‘cold’ tumors. We used pre-vaccine TPS as a surrogate for immune sensitivity, as NSCLC patients with TPS <1% have reduced benefit from ICIs compared to patients with baseline TPS >1%. Among patients with stage IV NSCLC and baseline TPS <1%, those who received a COVID-19 mRNA vaccine within 100 days of initiating ICI exhibited OS similar to that of patients with baseline TPS >1%, suggesting restored sensitivity to ICIs (Fig. 5l). In addition, the association between COVID-19 mRNA vaccination and OS was similar for patients with TPS <1% relative to those patients with a TPS at biopsy of 1–49.9% or ≥50% (Fig. 5l–n). This effect was not explained by changes in patient management during the pandemic period, as patients with stage IV NSCLC with TPS < 1% at biopsy who did not receive a vaccine had similar outcomes before and during the pandemic era (Fig. 5o).

Together, these data suggest a model in which mRNA vaccines targeting non-tumour-related antigens stimulate robust antitumour immune responses that sensitize tumors to ICIs (Extended Data Fig. 11). mRNA vaccines first stimulate a surge in antiviral cytokines, including IFNα, that drive systemic innate immune activation. Tumour-resident innate immune cells activated by this cytokine surge prime T cells, which become activated and infiltrate tumours. Although tumour cells evade attack by upregulating PD-L1 expression, combination with ICI enables COVID-19 mRNA vaccines to overcome this compensatory response, eliciting tumour regression and improved survival.

Discussion

Immunotherapy promises to deliver systemic anti-cancer therapy with long-term memory preventing recurrence. However, immunotherapy with ICIs relies on pre-existing anti-cancer immunity, which is absent in most patients. mRNA vaccines have recently emerged as a promising strategy to generate anti-cancer immunity to magnify the effects of ICIs^24,25,26,27. Here we used mRNA vaccines targeting the COVID-19 spike protein to demonstrate a mechanism by which mRNA vaccines improve survival in combination with ICI, even when the mRNA does not encode tumour antigens. Spike RNA-LNPs elicit body-wide APC activation, leading to the expansion of highly activated tumour specific T cells. When combined with ICIs, these T cells mediate tumour regression. As personalized neoantigen vaccines require considerable manufacturing time, off-the-shelf RNA-LNPs targeting tumour-associated or even infectious disease antigens may represent widely available, low-cost alternatives for patients waiting for personalized neoantigen vaccines or in settings in which personalized neoantigen vaccines are not available.

Although the local injection of RNA-LNPs for infectious disease or personalized cancer vaccines does not aim to directly modify the tumour immune microenvironment, we show here that even local injection of RNA-LNPs containing N¹-methyl-pseudouridine to minimize innate immune activation produces sufficient stimulus to reprogram TMEs and sensitize tumours to ICI. Moreover, we find that augmenting innate immune activation by replacing N¹-methyl-pseudouridine-modified mRNA with unmodified mRNA may further enhance the antitumour effects of this approach. The surprising effectiveness of this approach may explain the relative success of mRNA vaccines compared with other methodologies that also generate T cells reactive against targeted neoepitopes^{10,11,12,13,28,29}.

We show that COVID-19 mRNA vaccines increase PD-L1 expression on tumours, rendering immunologically cold tumours sensitive to ICIs. These results reveal that the timing of a routine immunization might influence treatment trajectories, and provide an over-the-counter means to overcome intrinsic ICI resistance. More broadly, they establish systemic innate immune modulation as a strategy to sensitize tumours to ICIs, and position off-the-shelf RNA therapeutics targeting infectious disease antigens as universal modulators of antitumour immunity.

These findings define a role for widely available vaccines for enhancing the efficacy of cancer immunotherapy. Although we focus on a single therapeutic due to its wide availability, these data could pave the way for other universal mRNA therapeutics specifically designed to reset patient immune systems for enhanced response to immunotherapy.

Methods

Retrospective studies

We completed a non-interventional, retrospective review of patient data using the MDACC electronic health record system, which contains a record of the patients who are treated at the primary campus of MDACC, a large quaternary cancer hospital in Houston, Texas. The chart review for this study involved three groups of patients: (1) patients with tumour biopsies confirming stage III or stage IV NSCLC between January 2017 and September 2022; (2) patients with melanoma of any stage who received single- or multi-agent immune checkpoint blockade between January 2019 and December 2022; and (3) a tissue-agnostic cohort, which included all patients with pathology results for PD-L1 from January 2020 to October 2023 at our institution across a wide range of histologies. This study was approved by the MD Anderson Cancer Center institutional review board. Informed consent was waived due to the retrospective and de-identified nature of the data. The data-collection cut-off was 1 September 2024; data analysis was performed from 1 September 2024 to 29 July 2025.

In the NSCLC dataset, patient information was collected regarding patient demographics (such as age at immunotherapy start, sex, ethnicity), primary tumour histology, clinical stage, known tumour mutations (such as those in EGFR, KRAS, HER2, ALK, MET, p53 and RET), metastatic burden at immunotherapy start (brain, liver), Eastern Cooperative Oncology Group (ECOG) performance status (PS; range, 0–5) near the initiation of immunotherapy, radiation therapy in the time around immunotherapy start, chemotherapy history, immunodeficiency, comorbidities (heart disease, kidney disease, liver disease, respiratory disease), history of other primary tumours, steroid use around immunotherapy initiation, date of last follow-up, date of death, date of first recurrence or progression, ICI agent names and start dates, COVID-19, influenza and pneumococcal vaccination dates, and TPS. Recorded PD-L1 expression as reported below was required for inclusion in the survival analysis for NSCLC. For patients with multiple biopsies, the PD-L1 expression recorded from the closest biopsy to ICI start was used for analysis.

In the melanoma dataset, patient information was collected regarding patient demographics (including age at immunotherapy start, sex, ethnicity), primary tumour histology, clinical stage, known tumour mutations (such as those in EGFR, KRAS, HER2, ALK, MET, p53 and RET), metastatic burden at immunotherapy start (for example, brain, liver), PS (range, 0–5) at the initiation of immunotherapy, chemotherapy history, immunodeficiency, comorbidities (for example, heart disease, kidney disease, liver disease, respiratory disease), other primary tumour data, steroid use around COVID-19 vaccination and immunotherapy initiation, date of last follow-up, date of death, date of first recurrence/progression, ICI agent names and start dates, and COVID-19 vaccination dates. Although our dataset did not include specific vaccine formulations administered to each patient, vaccine formulations administered during the study period included the original monovalent mRNA-1273 vaccine from Moderna (100 µg mRNA prime, 50 µg mRNA booster) released on 18 December 2020; the bivalent Moderna vaccine targeting the original strain and Omicron BA.4/BA.5 (50 µg mRNA) released on 1 September 2022; the original monovalent vaccine from Pfizer/BioNTech (30 µg mRNA prime and booster) released on 11 December 2020; and the Pfizer/BioNTech bivalent formulation (30 µg mRNA) released on 31 August 2022.

Patients were separated into two groups: (1) patients who received a COVID-19 mRNA vaccination within 100 days of ICI start; and (2) patients who did not receive a COVID-19 vaccination. Survival analysis was performed using these groups, with subanalysis involving staging of the tumour, brand of mRNA vaccine, number of doses of the COVID-19 vaccine, location of metastases and cycle of immunotherapy.

For patients who received COVID-19 mRNA vaccination in both the NSCLC and melanoma datasets, OS was calculated as the time between the date of immunotherapy start closest to the mRNA vaccination date, and the last follow-up date or date of death. For patients who did not receive COVID-19 mRNA vaccination, OS was calculated as the time between the initiation of their first ICI start and the date of death or last follow-up. For patients who received COVID-19 mRNA vaccination, PFS was calculated as the time between the initiation of ICI closest to mRNA vaccination and the first incidence of either pathology-confirmed recurrence or imaging-confirmed progression, whichever occurred earlier, that was declared progression in their primary medical oncologist’s clinical notes. For patients who did not receive COVID-19 mRNA vaccination, PFS was calculated as the time between their first ICI start date and clinician-confirmed progression as described above. Patients who progressed before the receipt of mRNA vaccination were included in the vaccination group for this analysis. Kaplan–Meier curves were generated using GraphPad Prism.

Cox proportional hazards regression

For Cox proportional hazards regression, time-dependent variables were defined as described in the figure captions. Continuous/numbered variables (such as age, BMI, PD-L1 expression, ECOG and treatment year) were retained as numeric. Binary and categorical variables (for example, stage, gender, mutational status, comorbidities) were converted into factors. For each variable, we constructed individual Cox proportional hazards models. HRs, 95% CIs (Wald) and P values were extracted from model summaries. Variables with P < 0.05 in univariable analysis were considered for multivariable modelling. Multivariable Cox proportional hazard regressions were similarly generated with significant variables from univariable analysis. For multivariable Cox proportional hazards regressions, patients with missing values were excluded from analyses.

Categories with fewer than five cases are reported for completeness but were considered not statistically meaningful due to convergence and insufficient data for reliable inference. Variables with significant relationships with survival were included in multivariable analyses (MVAs), and those significant after MVA (P < 0.05) were included in PSM. Certain variables (for example, steroid use within 1 month of vaccine) represented subgroups of the treatment cohort by definition. These were evaluated descriptively and in univariable models but were not included as covariables in MVAs, as they do not represent baseline confounders and were not significant on univariable analysis.

We assessed the proportional hazards assumption in multivariate models by evaluating Schoenfeld residuals and found no major violations.

Imputation and PSM

A small number of patients in the NSCLC dataset did not have documented ECOG scores, which were an important predictor of outcomes in Cox proportional hazards regression. BMI had missingness in initial univariable analyses but, because it was not required for PSM, imputation was not performed to address the missingness. Multivariable logistic regression and ridge regression were used to evaluate associations between ECOG missingness and clinically relevant pretreatment covariables. Ridge regression was selected due to multicollinearity between important variables that could not be excluded from the model. Variables not included in models predicting ECOG missingness were vaccination status, steroid usage within a timeframe of vaccination or ICI, PD-L1, concurrent chemotherapy, immunotherapy agent and BMI (as mentioned above). Subsequent ridge regression analysis suggested that ECOG scores in this analysis had a high probability of exhibiting missingness at random.

Given the small sample size in the stage III NSCLC dataset treatment group, multiple imputation was performed to estimate missing ECOG values using chained equations and the R package MICE, where predictive mean matching (PMM) (with set.seed(2025)) was used to create five imputed datasets. Variables used for imputation were chosen based on large absolute ridge regression coefficients (|β| > 0.3) and clinical relevance to ensure that prediction of ECOG missingness occurred with accuracy. After selection using regression coefficients, age and previous cycles of systemic therapy were included due to clinical importance. To facilitate PSM while maintaining power in patients with stage III NSCLC, the mode of the five imputed ECOG values was used as a single value for each patient (n = 5). This approach was necessary to minimize variance inflation and avoid diluting treatment effects in a very small sample. As PMM is random, small deviations in variable definition, factor levels and row ordering may result in slightly different imputed values. However, sensitivity analyses with slightly different mode selection in the case of ties, different seeds and an increased number of datasets (n = 20 or n = 10 rather than n = 5) revealed similar outcomes (P < 0.05) after subsequent matching. The plausibility of imputed values was assessed using density plots, which demonstrated appropriate range and overlap between observed and imputed ECOG values.

After imputation, PSM was performed using the R MatchIt package. A logistic regression model was used to estimate propensity scores, predicting treatment assignment of ‘vaccine within 100 days’ or ‘no vaccine’ with covariables based on significant factors associated with survival from MVA, including original or imputed ECOG scores. Across all propensity score analyses, nearest-neighbour matching was performed on the previously estimated propensity scores to identify a balanced cohort. In the case of stage III unresectable NSCLC PSM, nearest-neighbour PSM was performed with a caliper of 0.1 and a 2:1 ratio of control patients to treated patients in stage III UR NSCLC to improve covariable balance and maximize statistical power. Sensitivity analysis with removal of ECOG from the PSM model revealed similar trends.

In stage IV NSCLC, patients with missing ECOG were excluded due to sufficient remaining sample size. PSM was once again performed with nearest-neighbour matching, with a caliper of 0.1 and a 1:1 ratio of control patients to treated patients. Sensitivity analyses with removal of ECOG from PSM rather than imputation and with the same methods as reported in stage III NSCLC (with the same caliper of 0.1 but with a ratio of 1:1 rather than 2:1) also each revealed statistically significant results.

No imputation was performed in the melanoma cohort due to completeness of variables. The melanoma PSM analyses were run with nearest-neighbour matching with a caliper of 0.1 and a 1:1 ratio of control patients to treated patients.

Across all matched cohorts, absolute standardized mean differences (|SMD|) of pscores were consistently less than 0.05 after matching, indicating excellent balance between groups. Balance diagnostics were confirmed visually using jitter plots and histograms of propensity scores.

Survival analysis

The Kaplan–Meier method was used to estimate survival distributions for the matched groups. Log-rank tests were conducted to assess the differences between treatment groups. GraphPad Prism was used to visualize Kaplan–Meier curves, 95% CIs and at-risk numbers at 0, 10, 20, 30 and 40 months, including after extraction of propensity score-matched data from R. Comparisons between groups in primary datasets (Fig. 1) were reported as adjusted HRs calculated with Cox proportional hazard regression. Comparisons between groups in subgroup analyses were reported as unadjusted HRs calculated with log-rank tests to preserve robustness. P values are reported for primary survival analyses.

In settings in which Kaplan–Meier curves crossed, the restricted mean survival time was calculated using area under the curve analysis of the absolute differences between arms at 12 and 24 months.

All statistical analyses were conducted in R v.4.4.2 (2024-10-31) and GraphPad Prism.

Quantification of PD-L1 expression in biopsy samples

For patients in both the tissue-agnostic cohort and the NSCLC dataset, biopsy date, pathology reports, histological information and diagnosis codes were collected for patients who had received a biopsy with pathology evaluating PD-L1 at MD Anderson (from January 2020 to October 2023 for the tissue-agnostic cohort and August 2016 to August 2022 for the NSCLC cohort). PD-L1 on tumour cells (TPS) and PD-L1 on tumour cells and immune cells together (CPS) were obtained from pathology reports. Data were curated according to the following principles: (1) simple quantitative values were replicated exactly; (2) TPS values of “<1%” were reported as 0%; (3) for patients with a CPS of 0% or CPS <1% with no TPS value, TPS was reported as 0%, as CPS of 0 or <1% implies a lack of expression of PD-L1 on both tumour cells and myeloid cells; (4) TPS values between 1.1 and 49.9 reported as “<x%” or “>x%” were excluded due to lack of interpretability; (5) TPS values of >50% were recorded as 50% and TPS values >60% were recorded as 60%, and so on, as these values indicate high PD-L1 expression. For ranges of PD-L1 expression (for example, 0–5%), the median of the numbers in the range was taken as an approximate estimation of PD-L1 on tumours. Replicate patient biopsies (repeat biopsies performed on the same patient within 2 months of a previous biopsy) were excluded, and the earliest biopsy was included.

Separately, survival was analysed in the tumour-agnostic dataset with any COVID-19 mRNA vaccine within 100 days of ICI compared with patients who had no recorded history of a COVID-19 vaccine. Survival analysis was performed as listed in the NSCLC dataset for all patients in this dataset who were treated with ICI in this cohort. TPS values were not required for inclusion.

For group-wise analysis, patients in these cohorts were grouped into (1) those who received their closest COVID-19 mRNA vaccination less than 100 days before tissue biopsy; (2) patients who received their closest COVID-19 mRNA vaccine more than or exactly 100 days before tissue biopsy; and (3) patients who did not receive a COVID-19 mRNA vaccination before biopsy. Patients who received non-mRNA-based COVID-19 vaccines within 100 days before biopsy were included in the no-vaccine group.

Healthy human participants

Studies in healthy volunteers were approved by the MD Anderson Institutional Review Board. Informed consent was obtained from each healthy participant before plasma and peripheral blood mononuclear cells were collected and stored at baseline, 6 h, 24 h and 48 h, or baseline, 6 h, 24 h, 7 days and 14 days after COVID-19 mRNA vaccination, depending on the formulation used. BNT162b2 (COVID-19 mRNA vaccine, 2024–2025 formula, with 30 μg mRNA) and mRNA-1273 (Spikevax Monovalent XBB.1.5, COVID-19 mRNA vaccine, 2023–2024 formulation, with 50 μg mRNA) were used for the first and second timeframes, respectively. Frozen plasma was thawed and analysed using the NULISAseq Inflammation Panel 250 by Alamar Biosciences. Absolute quantification of IFNα concentrations was performed using the NULISAseq Inflammation Panel AQ. Statistical analysis was performed using R, in which we evaluated significance using a linear model with fixed effects based on the time from vaccine and controlling for patient ID (paired), including accommodation of biological variance between the baselines using the limma package’s duplicateCorrelation function. P values were calculated using moderated t-tests with FDR correction for multiple testing. For comparisons between Pfizer and Moderna vaccines, we used a linear modelling approach (limma) to assess the log₂-transformed fold change in cytokine concentrations per-patient from the baseline to 6 h and 24 h. Moderated t-tests with empirical Bayes shrinkage were applied, and P values were corrected using the Benjamini–Hochberg FDR method. Contrasts were constructed to directly compare Moderna versus Pfizer for each timepoint. Cytokines were visualized if they demonstrated significant differences between Moderna and Pfizer from the baseline to 6 h or 24 h (absolute value of log₂-transformed fold change greater than 0.5 and FDR-corrected P < 0.05). Mean expression values per condition were normalized to the baseline and visualized. Volcano plots were generated using R. Heat maps and graphs of cytokines with significant changes in relative expression in the plasma at 24 h were generated in GraphPad Prism.

Human flow cytometry

Peripheral blood mononuclear cells were thawed in a 37 °C water bath and immediately transferred into a 50 ml tube containing 9 ml of complete RPMI medium (RPMI + 10% FBS) at a ratio of 1 part cells to 9 parts medium. Cells were centrifuged at 400g for 10 min, and the supernatant was decanted. The cell pellet was resuspended in 1 ml of PBS, and incubated with live/dead marker (1 µl per 1,000 µl of cell suspension, containing 1–10 million cells per ml) for 30 min at 4 °C. After incubation, 3 ml of PBS + 2% FBS was added, and the cells were centrifuged again at 400g for 8 min and the supernatant was decanted. Cells were then pre-incubated with 5 µl of Human TruStain FcX (BioLegend, 422302) per 100 µl of cell suspension for 5 min at room temperature. The cells were then washed with PBS and centrifuged at 400g for 6 min. After decanting the supernatant, 10 µl of Brilliant buffer was added to each tube, and the mixture was allowed to sit for 5 min before addition of antibodies targeting extracellular markers and the cells were incubated for 20 min at room temperature in the dark. After incubation, cells were washed and fixed in the dark for 45 min at 4 °C with 500 µl of fixation solution (BD). Cells were then permeabilized with 2 ml permeabilization buffer. Intracellular stains were added in permeabilization buffer, and cells were incubated for approximately 40 min at room temperature. After incubation, cells were washed with permeabilization buffer and resuspended in 200 µl of PBS + 2% FBS. Analysis was then conducted using Cytek Aurora Spectral Flow Cytometer.

Preclinical experiments

All mouse experiments and procedures were approved by the University of Florida or the University of Texas MD Anderson Institutional Animal Care and Use Committee (IACUC). Mice (aged 4–10 weeks) were maintained at 21 ± 1 °C and 35% humidity under a 14 h–10 h light–dark cycle. Mice were randomized prior to treatment. Tumour measurements were performed blinded to the treatment group. Euthanasia was performed by CO₂ inhalation followed by cervical dislocation in accordance with approved protocols. Humane end points included tumour ulceration.

C57BL/6 and Rigi-knockout mice (C57BL/6NJ-Rigiem1(IMPC)J/Mmjax) were purchased from The Jackson Laboratory (046070-JAX) and bred in house. Tumour-bearing mice were implanted s.c. with 50,000 B16F0 melanoma cells, 1 million B16F10-OVA melanoma cells or 200,000 LLC cells in the right flank. Orthotopic LLC models were implemented by implanting 100,000 LLC cells into the left lung of C57BL6 mice through direct injection below the 9^th rib. A mix of male and female mice was used for experiments with B16F0, and male mice were used for experiments with LLC. All mice were vaccinated intramuscularly with 25 μg per dose of RNA-LNP or RNA-LPA (mRNA fraction)¹². We administered anti-mouse IFNα receptor (aIFNAR1, Bio X Cell, BE0241) and anti-mouse IL-1R (Bio X Cell, BE0256) antibodies intraperitoneally at an initial dose of 500 μg per mouse for the first dose, followed by a maintenance dose of 250 μg per mouse twice a week for the remaining treatment period. Anti-PD-L1 (Bio X Cell, BE0101) checkpoint inhibitor was administered at 400 μg per mouse for the initial dose, followed by a maintenance dose of 200 μg per mouse twice a week for the remaining treatment period. Anti-PD-1 (Bio X Cell, BE0146) checkpoint inhibitor was administered at an initial dose of 400 μg per mouse, followed by a maintenance dose of 200 μg per mouse. LMW poly(I:C) (InvivoGen, tlrl-picw) was administered intramuscularly with 25 μg per mice for two doses. Tumours were measured at a frequency of three times a week starting on day 8 until more than 20% of mice reached the end point. Mice were euthanized after reaching the humane end point.

mRNA

SARS-CoV-2 spike coding sequence with K986P and V987P mutations was inserted into a pGEM-4Z backbone downstream of the T7 promoter with previously published UTRs and poly(A) signal analogous to the mRNA in BNT162b2^30,31. The 5′ SpeI restriction site was removed from the sequence to allow for restriction of the poly(A) tail. The T7 promoter was changed to have an AGG initiator sequence by site-directed mutagenesis (NEB, E0554) according to the company’s recommended protocol. Plasmids were grown in NEB5a competent Escherichia coli and purified using the RNeasy Maxi kit (Qiagen, 75162) and sequenced using whole-plasmid sequencing by Genewiz. Plasmids were restricted by using 2 U μg^−1 SpeI HF (NEB, R3133L) for 2 h at 37 °C followed by DNA precipitation. mRNA was synthesized using the mMESSAGE mMACHINE T7 mRNA Kit with CleanCap Reagent AG (Thermo Fisher Scientific, A57620). Spike mRNA was made using CleanCap reagent AG (3′ OMe) (Trilink, N-7413-5) and N¹-methylpseudouridine-5′-triphosphate (Trilink, N-1081-10). PP65 mRNA was capped using ARCA (NEB, S1411L) followed by treatment with mRNA cap 2′-O-methyltransferase (NEB, M0366L). The in vitro transcription reaction was performed at 20 °C for 10 h. DNA was removed according to the kit instructions and RNA was purified using the RNeasy maxi kit (Qiagen, 75162). RNA was eluted in purified RNase-free water and stored at −80°C until use. mRNA was checked for quality using an Agilent TapeStation 4150. After thawing, RNA was diluted to 100 µg ml^−1 in RNase-free water followed by incubation at room temperature or 72 °C for 3 min. Then 100 ng of RNA or 1 µl of ladder was added to 5 µl of RNA ScreenTape sample buffer (Agilent, 5067-5577) and mixed at 2,000 rpm for 1 min. The samples were run using an RNA ScreenTape (Agilent, 5067-5576). For dsRNA removal, the method described previously was followed²⁰. In brief, mRNA was precipitated and reconstituted in chromatography buffer. Cellulose fibres (Sigma-Aldrich, C6288-100G) suspended in chromatography buffer were added to Nucleospin filter units (Macherey-Nagel, 740606) followed by washing. Up to 500 μg of RNA per filter unit was added to cellulose and incubated by rapid mixing for 30 min. RNA was recovered by centrifugation, and incubation was repeated using a second column containing cellulose fibres. RNA was then filtered through a 0.45 µm syringe filter (Pall, 4604) to remove any cellulose particulate before being precipitated and reconstituted in RNase-free water²⁰.

Fabrication of LNPs

Before complexation, 0.5 M of RNase-free citrate buffer (pH 3.75) (Teknova, Custom order) was added to the RNA for a final concentration of 0.1 M. ALC-0159 (Avanti, 880155 P), ALC-0315 (Avanti, 890900 O), cholesterol (Sigma-Aldrich, C8667) and DSPC (Avanti, 850365P) were reconstituted in 100% ethanol at a ratio of 1.7:47.5:40.8:10. Lipids and RNA were mixed at a 3:1 FRR on a Nanonsemblr Ignite (Precision NanoSystems, now part of Cytiva) for an N/P ratio of 6. RNA-LNPs were dialysed overnight at 4 °C with two buffer exchanges at the 3 h and 6 h timepoints using a 3.5 kDA dialysis cassette (Thermo Fisher Scientific, A52967). After dialysis, LNPs were filtered through the 0.2 µM Supor EX ECV filter (Pall, KS2ECV2S). LNPs were concentrated using 30 kDA Amicon filters (Millipore, UFC903024) by centrifugation at 2,000g. The final LNP formulation had sucrose dissolved in PBS added for a final concentration of 12%. Anionic RNA-LPAs were made as previously described¹². In brief, DOTAP liposomes 2.5 mg ml^−1 were mixed with mRNA at a 1:1 mass to mass ratio and complexed for 15 min at room temperature before administration.

RNA concentration was determined in LNPs using the RiboGreen assay (Invitrogen, R11490) using the BioTek Cytation 3 plate reader. The encapsulation efficiency was determined by comparing readings to LNPs in TE buffer versus LNPs in 1% Triton X-100. LNPs were diluted up to 500 μg ml^−1 with PBS before injection. Empty LNPs were used at a volume consistent with the dose of RNA-LNPs fabricated using the same steps and same amounts of lipids. Fluorescence values were plotted in a standard curve with an extrapolation factor set to 1.1. The values from wells with Triton X-100 (total mRNA) and TE buffer (free mRNA) was entered into the following equation: EE% = [1 – (Free RNA/Total RNA)] × 100%. Total mRNA RNA-LNPs were also loaded onto a 1% agarose gel made in 1× TAE buffer (Quality Biological, 351-008-131) with GelRed (Biotium, 41003-T). Then 500 ng of RNA was loaded into LNPs and free RNA was mixed with 6× SDS-free (NEB, B7025S) loading dye and run at 90 V for 45 min. Particle size was acquired using a Malvern Zetasizer Ultra and NanoSight’s NS300. Before particle-size, PDI, concentration and surface ζ potential analysis, LNPs were diluted 100-fold with HyClone HyPure water (Cytiva, SH30538.02). They were and run in triplicate at 25 °C on the Zetasizer. Orthogonal particle-size distribution and concentration measurements were carried out using Malvern’s NanoSight NS300 NTA equipment. Each LNP sample was diluted 500-fold with PBS and pumped through the equipment at constant speed for five captures with optimized camera settings. An optimized detection threshold was set for analysis. mRNA was extracted from LNPs by dissolving 750 μg ml^−1 with 5 volumes of ethanol or 2 volumes of isopropanol followed by 2 washes with cold 70% ethanol. RNA pellets were dissolved in RNase-free H₂O before analysis using the TapeStation.

Plasma collection

Mice were bled through the retroorbital route using heparinized capillaries (Thermo Fisher Scientific, 22-362566) collecting a maximum of 200 µl into EDTA coated tubes (Thermo Fisher Scientific, NC9414041). Whole blood was centrifuged at 1,200g for 15 min. Plasma was collected and stored at −80 °C until use.

Processing tumours for flow cytometry analysis

Tumours were dissected from euthanized mice with the external fibrous sac left intact. Tumours were bisected using a scalpel and half of each sample was placed into a GentleMacs column GentleMACS C Tube (130-096-334) with an enzyme mix containing 10 mg ml^−1 of collagenase (Sigma-Aldrich, C5138-1G), 1 mg ml^−1 of hyaluronidase (StemCell Technologies, 07461) and 200 mg ml^−1 of DNase (Sigma-Aldrich, D5025-150KU) or using the tumour dissociation kit (Miltenyi, 130-096-730) followed by debris removal (Miltenyi, 130-109-398). Samples were run on a m_TDK_1 cycle and centrifuged at 300g for 5 min after completion. The pellets were resuspended in cold PBS and filtered through a 70-µm cell strainer. The samples were washed twice using cold PBS and then manually counted using a haemocytometer.

Isolation of splenocytes

Whole spleens were collected from euthanized mice and placed into cold RPMI medium. The spleens were then teased through a 70-µm filter and lysed for 5 min at 37 °C using 1× BD Pharmalyse buffer (BD, 555899). Lysis buffer was quenched with medium and centrifuged at 500g for 5 min. Splenocytes were resuspended in cold PBS and filtered through a 70-µm cell strainer and washed once with PBS before being counted using the Beckman Coulter Vi-cell XR or haemocytometer.

Mouse flow cytometry analysis

In total, 1 × 10⁶ cells from either tumours and spleens were placed into 96-well V-bottom plates. Unless stated otherwise, the washing steps were performed by centrifugation at 500g for 5 min at 4 °C followed by resuspension of cells with 200 µl of buffer, with mixing done by pipetting. Cells were washed with cold PBS and stained with 100 µl of live/dead stain (Thermo Fisher Scientific, L10119) for 30 min at 4 °C. Live/dead dye was quenched with 100 µl of cold PBS and the cells were washed once with cold FACS buffer (PBS with 2% FBS). Cells were centrifuged and resuspended with 10 µl True stain FCX buffer (BioLegend, 422302) diluted to 100 µg ml^−1 with FACS buffer for 10 min. Then 90 µl of antibodies (Supplementary Tables 12 and 13) and Brilliant Stain buffer (BD, 563794) were added and incubated for 30 min at 4 °C. Next, 100 µl of FACS buffer was added to each well and the cells were washed twice with cold FACS buffer. Cells were fixed for 15 min with 100 µl of Cytofix buffer (BD, 554655) at 4 °C. Next, 100 µl of PBS was added to each sample and cells were washed twice and stored in FACS buffer at 4 °C in the dark until analysis. Initial compensation was acquired using Ultracomp eBeads (Thermo Fisher Scientific, 01-2222-42) and ArC amine reactive compensation beads (Thermo Fisher Scientific, A10346). Results were acquired using a BD Symphony A3 and analysed using FlowJo v.10.8.1 and v.10.10.0.

Tetramer production

Peptides selected to be high-affinity binders for Survivin (ATFKNWPFL), GP100A (EGSRNQDWL), GP100B (KVPRNQDWL), WT1 (RMFPNAPYL), CLDN6 or claudin 6 (KVYDSLLAL) and TRP2 (SVYDFFVWL) were purchased from GenScript. H2-Db and H2-Kb monomers were purchased as easYmers from ImmunAware (5001-01 and 5004-01, respectively), and tetramers were prepared according to the manufacturer’s instructions. In brief, peptides were reconstituted to 1 mg ml^−1 in deionized water, further diluted to 75 μM and incubated with easYmer (Eagle Biosciences, 5004-01, 5001-01) at 18 °C for 48 h. The resulting complexes were tetramerized by gradually mixing with streptavidin-APC (BD Biosciences, 554067). After incubation, a final stock concentration of 500 nM for each tetramer was achieved.

Tetramer staining

According to the manufacturer recommendations, 1–2 × 10⁶ cells were stained with tetramer diluted to 20 nM in FACS buffer. Cells were left to stain in the dark at room temperature for 1 h, washed once with FACS buffer, co-stained with surface antibodies (CD3, CD4) at room temperature for 20 min, washed twice with FACS buffer and analysed on a Cytek Aurora flow cytometer.

AIM assay

The AIM assay was performed as previously described¹². In this iteration, whole splenocytes were used rather than isolated T cells. In brief, 24 h after the last RNA-LNP vaccine, splenocytes were collected for antigen recall assay. Spleens were collected and processed as described above. 1 million splenocytes were cultured in a 96-well round-bottom plate in T cell medium containing RPMI 1640 (Gibco, 11-875-119), 10% FBS (Thermo Fisher Scientific, 35-011-CV), 1% penicillin–streptomycin (Gibco, 30-002-CI), 1% MEM non-essential amino acids (NEAA, Gibco, 11140050), 1% sodium pyruvate (Gibco, 11360070), 0.1% β-mercaptoethanol (BME, Gibco, 21985-023) and 0.2 µg tumour-associated peptides without cytokines. Peptides were chosen based on their suggested upregulation in melanoma tumours¹². Selected peptides were purchased from JPT Peptide Technologies and reconstituted according to the manufacturer’s guidelines. Co-culture was maintained for 48 h in an incubator at 37 °C under 5% CO₂. Cells were next collected and stained for the AIM assay for evaluating activation of antigen-specific T cells. Cells were centrifuged and resuspended in FACS buffer containing antibodies mix (Supplementary Table 14). Cells were washed three times and fixed using fixation medium (Thermo Fisher Scientific, GAS001S100) for 15 min at room temperature. Cells were washed once with PBS and were stored in FACS buffer at 4 °C in the dark until analysis. Results were acquired using a BD Fortessa III. Values were normalized by subtracting the percentage of AIM⁺ cells of tbe DMSO-only controls from each treatment condition. Statistical analysis was performed using Brown–Forsythe and Welch ANOVA, followed by Dunnett’s T3 multiple-comparison test.

ELISA and multiplex analysis

Plasma was analysed for cytokine concentration using ELISAs and multiplex cytokine arrays. A portion of the plasma samples were assessed for IFNα through ELISA (Invitrogen, BMS6027). Plasma was also sent for analysis by Eve Technologies for multiplex cytokine array (MD44). For quantification of spike IgG, mice that received three vaccines/vehicle doses during tumour studies were bled once the humane end point was reached. Spike-specific IgG was determined using an Anti-mouse SARS-CoV-2 IgG titre assay (AcroBiosystems, RAS-T023). Plasma was diluted 100,000-fold and run according to the company’s recommended parameters for semiquantitative analysis.

Immunofluorescence

Half of bisected tumours were fixed with 4% paraformaldehyde at 4 °C overnight. The samples were washed three times using PBS and then immersed successively into 10%, 20% and 30% sucrose for cryopreservation. Tissue was then embedded in O.C.T. (Tissue-Tek, 4583) and stored at −80 °C. Blocks of tissue were moved to −20 °C for 24 h and were sectioned at a thickness of 30 µm using the Leica cryostat, and the sections were placed onto microscope slides. Before staining, the slides were brought up to room temperature for 15 min and washed three times with PBS. Tissue was blocked at room temperature for 1 h using a blocking buffer containing 2% goat or donkey serum, 1% BSA and 0.1% Triton X-100 in PBS. The sections were then stained with primary antibodies (Supplementary Table 15) in blocking buffer overnight at 4 °C. After incubation, the sections were washed three times with PBS and stained with secondary antibodies (Supplementary Table 15) diluted in blocking buffer for 1 h at room temperature. After three 5 min PBS washes, the sections were incubated with DAPI (1:1,000 in PBS) for 10 min at room temperature. After a triple PBS wash, the sections were mounted with Prolong Glass Antifade Mountant (Thermo Fisher Scientific, P36984) and covered with a cover glass. Images were acquired using a Leica Stellaris 8 WLL Spectral Confocal Microscope. Image processing was performed using Fiji ImageJ software (NIH) and Imaris (Oxford Instruments).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.