Humoral and cell-mediated responses to ARCT-021 are weakly correlated

Total RNA was extracted from the whole blood of 106 healthy volunteers who provided written informed consent for participation in a phase I/II randomised, placebo-controlled clinical trial that assessed the safety and immunogenicity of ARCT-021 (clinicaltrial.gov NCT04480957). Gene transcript levels were determined at baseline (pre-dose 1), day 2, 3, and 8 for all vaccinees. In addition, gene expression was profiled at day 29 (pre-dose 2), 30 and 36 among vaccinees in the two-injection cohort. The magnitude of antibody responses to ARCT-021 was measured by the Luminex immuno-assay against the full-length recombinant S protein, and the S-reactive T-cell responses were assessed using IFNγ ELISPOT assay following stimulation with overlapping S protein peptide pools. The effects of vaccine dose and subject demographics on antibody and T-cell responses has been previously reported15. Among the ARCT-021 vaccinees, there was variation in S-binding IgG titers (Fig. 1a) and T-cell IFNγ responses (Fig. 1b) of >100 fold. We found low or weak correlation between S-binding IgG titers and T-cell IFNγ responses (Fig. 1c), although both measurements approximate to a Gaussian distribution (Supplementary Fig. 1a). These observations provided us with an opportunity to dissect the molecular signatures driving humoral and cell-mediated immunity to sa-mRNA vaccination.

Fig. 1: ARCT-021 triggers an early induction of transcripts related to innate immune responses and antigen presentation.

a Day 29 IgG antibody titers and b Day 15 S-specific T cell responses in vaccinated individuals (n = 78), ranked from smallest to largest values. c. Correlation matrix showing the relationship of IgG titers (day 15, 29, 36, 43, 57) and day 15 S-specific T cell responses after ARCT-21 vaccination. The colour intensity is proportional to the correlation coefficient. d. Differentially expressed genes (DEGs) detected in all subjects receiving ARCT-021 (n = 78), as compared to the placebo group (n = 28) for day 2, 3, 8, 30 and 36. DEGs were identified based on fold-change >1.3 and false discovery rate-adjusted p value < 0.05 (Benjamini-Hochberg step-up procedure). Red bars indicate number of upregulated DEGs and blue bars indicate number of downregulated DEGs. e Volcano plot displaying genes that were most differentially expressed at day 2, 3 and 8 after vaccination. The most differentially regulated genes are annotated on the volcano plot. Volcano plot for day 30 and 36 are shown in Supplementary Fig. 1c. f Top 20 Blood Transcription Modules (BTMs) that are positively enriched (Benjamini-Hochberg adjusted p value < 0.05) at day 2, 3, 8, 30 and 36 in vaccinated subjects compared to placebo controls. Colour intensity and size of the dots is proportional to the −log10 transformed Benjamini-Hochberg adjusted p values. g Clustergram showing the log2-transformed fold-change of DEGs present in the top two BTM modules highlighted in f, “Enriched in monocytes (II) (M11.0)” and “Regulation of antigen presentation and immune response (M5.0)” at day 2, 3, 8, 30 and 36. The colour-gradient from blue to red indicates log2-transformed fold change (day 2/day 1) values from −2 to 2 respectively.

Early innate immune responses to ARCT-021

To examine the early innate immune gene signatures induced by ARCT-021, we conducted NanoString profiling using the nCounter Human Immunology v2 Panel (NanoString Technologies), which allows quantification of 579 immune-related transcripts with high precision16. The greatest number of differentially expressed genes (DEGs) (fold change >1.3; false discovery rate [FDR]-adjusted p-value < 0.05, Benjamini-Hochberg step-up procedure) was detected at day 2, and the number of DEGs decreased three-fold at day 3 (Fig. 1d). Secondary vaccination generated similar changes in transcriptional responses at day 30 relative to day 2 (Fig. 1d), with strong correlation observed for expression of transcripts between these 2 time points (Supplementary Fig. 1b). Notably, the most upregulated genes comprised cytokines, chemokines and interferon (IFN)-stimulated genes, including CXCL10, CCL2, CCL8, LAMP3, IFIT2 and MX1 (Fig. 1e, Supplementary Fig. 1c). Visualisation of gene networks by Ingenuity Pathway Analysis (IPA) revealed a tight interacting network of transcription factors responsible for the induction of these cytokines and chemokines, including IRF3, IRF7, STAT-1 and STAT-2 (Supplementary Fig. 2).

To dissect the molecular pathways induced by ARCT-021, we utilised the blood transcription modules (BTMs), a curated database comprising of an integrated large-scale network of publicly available human blood transcriptomes17. We found that BTMs related to activation of myeloid cells, antigen presentation, cell cycle and interferon signalling were significantly enriched on days 2, 3 and 30. The majority of transcripts returned to baseline levels at day 8 and day 36 (Fig. 1d, f-g). The upregulated transcripts in the enriched in monocytes (II) (M11.0) module included Toll-like receptors (TLR1-8), leukocyte immunoglobulin-like receptors and inflammatory mediators (Fig. 1g, Supplementary Fig. 3a), indicating that ARCT-021 vaccination leads to an early activation of innate immune responses. Furthermore, the majority of upregulated transcripts related to antigen presentation constitute the MHC-I signalling pathway, including induction of transcripts involved in peptide degradation and processing (LMP2, LMP7), peptide loading (TAP1, TAP2, TPN, CLIP) and MHC-I molecules, which could facilitate antigen presentation to CD8 T-cells (Supplementary Fig. 3b). Taken together, our data showed that vaccination with ARCT-021 triggers an early and strong induction of the innate immune response, leading to upregulation of genes involved in pattern recognition receptor signalling, cytokine and chemokine signalling and MHC-I antigen presentation.

Expression of BSIG transcripts 1 day post-vaccination correlates with humoral responses to ARCT-021

An unsupervised hierarchical clustering of the 231 upregulated DEGs at day 1 post-ARCT-021 vaccination indicated 3 distinct clusters: placebo group, C1 (purple) and C2 (yellow) (Fig. 2a). The top 10 differentially regulated genes between C1 and C2 were related to complement activation (C1QB, C2, SERPING1) and inflammatory response (CCL2, IL1RN, IDO1, TNFAIP6) (Fig. 2b, Supplementary Fig. 4a-b). The majority of participants in C1 were younger adults who received at least 3 µg of ARCT-021 (Fig. 2c). In contrast, participants in C2 consisted of either older adults or those who received lower doses (1µg-3µg) of ARCT-021 (Fig. 2c), indicating that the transcriptional responses to ARCT-021 in older adults are more blunted.

Fig. 2: Magnitude of induction of BSIG transcripts is positively correlated with humoral responses.

a Gene and sample hierarchical clustering based on expression profiles of upregulated DEGs on day 2. Three distinct sample clusters, named placebo, C1 (purple) and C2 (yellow) are detected. The colour-gradient from blue to red indicates log2-transformed fold change (day 2/day 1) values from −3.5 to 3.5 respectively. b. Volcano plot displaying genes that were most differentially expressed between C1 and C2 on day 2. The top 10 most differentially regulated genes are annotated on the volcano plot. Fold-change cut-off of 1.3 and Benjamini-Hochberg adjusted p-value 0.05 are indicated as dotted lines on the volcano plot. c. Donut plots showing the demographics of C1 and C2. Young and elderly subjects are highlighted in blue and yellow-orange respectively, with higher doses of vaccines indicated in darker colours. Overall percentages in the young and elderly subjects in C1 and C2 are also displayed. d. S-specific T-cell responses at day 15 post-vaccination in C1 (n = 55) and C2 (n = 23) clusters. e. Anti-S IgG titers (log2-transformed) of subjects in clusters C1 and C2 at day 15, 29, 36, 43 and 57 post-vaccination. For time-points from day 36 onwards, analysis is based on subjects who received the second dose of the vaccine. Box plots in d-e represent 25–75% intervals, with lines indicating medians. The whiskers represent 10–90% intervals. Unpaired, two-sided, Student’s t-tests were used for comparisons for d-e. f Pearson correlation of log2-transformed IgG titers at day 29 with BSIG score. BSIG score is calculated as the arithmetic mean of the top 10 genes that most distinctly separate C1 and C2, as indicated in b. *P < 0.05, **P < 0.01, ***P < 0.001.

Next, we examined if the C1-C2 dichotomy affected anti-S IgG titers and S-specific T-cell IFNγ responses. While no significant differences were seen in T-cell responses (Fig. 2d, Supplementary Fig. 4c), anti-S IgG titers at day 15, 29, 43 and 57 were significantly lower in C2 compared to C1 (Fig. 2e, Supplementary Fig. 4d). To verify that the gene expression differences between C1 and C2 were indeed correlated with anti-S IgG titers, we calculated the arithmetic mean of the log2 fold change values from the top 10 transcripts that most distinctly separate C1 and C2 (termed as BSIG score) for each sample, and performed a correlation analysis with anti-S IgG titers. Indeed, the BSIG score was significantly correlated with anti-S IgG titers (Fig. 2f), thus supporting an association between the magnitude of induction of the BSIG transcripts at 1 day post-vaccination with humoral responses to ARCT-021.

A random forest regression model (Supplementary Fig. 5a) was used to identify the immune transcripts that could most accurately predict anti-S IgG titers. 75% of the samples were used as training data and the remaining 25% as test data to evaluate the accuracy of our machine learning model. Our first model generated a decision tree that was able to predict anti-S IgG titers with 84.8% accuracy, with root mean square error (RMSE) of 1.43. The relative feature importance of the individual genes involved in generation of the decision tree is as depicted in Supplementary Fig. 5b. After hyperparameter tuning, we ascertained that the random forest regression model based on the log2 fold-change values of the top 6 genes (KLRB1, CCL2, FCER1G, ITGA6, MSR1 and GPR183) further improved the accuracy to 96.0% (RMSE = 1.70) (Supplementary Fig. 5c). Notably, many of these genes are involved in leukocyte migration (GO:0060900, FDR = 0.0179), highlighting the pivotal role of chemotaxis in generation of humoral immune responses to ARCT-021.

Since the magnitude of BSIG upregulation is associated with higher antibody response, we further examined upregulated immune transcripts that best differentiate responders from non-responders. Responders were defined by subjects that seroconverted with at least a 4-fold rise in antibody titers15. In agreement with the random forest regression model (Supplementary Fig. 5b), we identified MSR1 and FCER1G to be most significantly increased in responders (Supplementary Fig. 6a-c). Both MSR1 and FCER1G were also significantly increased in responders compared to non-responders among the older adults (Supplementary Fig. 6d). However, the magnitude of upregulation for MSR1 and FCER1G was not associated with severity of side effects (Supplementary Fig. 6e-f). Likewise, BSIG score was not found to be associated with severity of side effects (Supplementary Fig. 6g), demonstrating that the transcriptomic signatures for ARCT-021 immunogenicity and reactogenicity are distinct.

Immune signatures at day 8 correlated with spike-specific T-cell response to ARCT-021

Next, we performed pathway enrichment for downregulated genes following ARCT-021 vaccination. Negatively enriched BTMs were mostly related to T-cells and NK cells (Fig. 3a), with genes that encode for surface receptors on T-cells and NK cells being significantly downregulated at day 2 and day 30 (Fig. 3b). However, expression of downstream T-cell signalling transcripts remained unchanged (Supplementary Fig. 7), suggesting that ARCT-021 vaccination did not lead to global suppression of T-cell activity. Instead, the downregulation of these T-cell transcripts could indicate increased migration of T-cells out of the peripheral blood, and likely to lymphoid organs. In addition, the majority of downregulated NK cell-related genes encoded for inhibitory receptors (KLRB1, KLRC1, KLRD1), suggesting an early activation of NK cells post-vaccination (Fig. 3b). To ascertain if the differences in downregulated DEGs post-vaccination could predict differences in T-cell responses, we performed an unsupervised hierarchical clustering, which generated 3 distinct clusters: placebo group, T1 (blue) and T2 (pink) (Supplementary Fig. 8a). However, the T1-T2 dichotomy was unable to discriminate subjects based on vaccine-induced T-cell response (Supplementary Fig. 8b). Instead, gene expression signatures at day 8 identified transcripts that were associated with T-cell responses. Interestingly, a total of 18 transcripts (6 positively correlated, 12 negatively correlated) were found to be significantly correlated with CD8 T-cell responses (Fig. 3c). We calculated a TSIG score for each sample by subtracting the arithmetic log2 fold change values of the 12 negatively correlated genes from that of the 6 positively correlated genes. The TSIG scores were significantly correlated with T-cell response, with correlation coefficient of 0.566 (p < 0.0001) (Fig. 3d), indicating that the expression levels of these 18 transcripts at day 8 could discriminate subjects with strong from weak T-cell response. Of note, many of the positively correlated transcripts were associated with T-cell maturation and expansion (CD27, LEF1, CD44, XBP1), as well as proteasome degradation (PSMB5)18,19,20,21,22. Conversely, negatively correlated transcripts were mostly associated with complement activation (MASP2, C1QA, CFB), T-cell proliferation (PDCD1LG2), dendritic cell activity (TNFSF11, CD209, IFNA1/13) and type I interferon response (IFNA1/13).

Fig. 3: Transcriptional correlates of spike-specific T cell responses are observed at day 7.

a Top 10 blood transcription modules (BTMs) that are negatively enriched (Benjamini-Hochberg adjusted p-value < 0.05) at day 2, 3, 8, 30 and 36 in vaccinated subjects compared to placebo controls. Colour intensity and size of the dots are proportional to the −log10 transformed Benjamini-Hochberg adjusted p-values. b Clustergram showing the log2-transformed fold-change of DEGs present in the top BTM modules highlighted in, “T-cell activation (I) (M7.1)”, “Enriched in T cells (I) (M7.0)” and “Enriched in NK cells (I) (M7.2)” at day 2, 3, 8, 30 and 36. The colour-gradient from blue to red indicates log2-transformed fold change (day 2/day 1) values from −1 to 1 respectively. c Heatmap showing day 8 transcripts that are significantly correlated (p-value < 0.05) with log2-transformed S-specific T cell counts at day 15 in ARCT-021 vaccinated individuals. Transcripts are ranked by Pearson correlation coefficient values. d. Pearson correlation of log2-transformed S-specific T cell counts at day 15 with TSIG score. TSIG score for each sample was calculated by subtracting the arithmetic log2 fold change values of the 12 negatively correlated genes from the 6 positively correlated genes shown in panel c.

Given that the top 3 positively correlated genes (CD27, LEF1 and PSMB5) were previously demonstrated to be critical for CD8 T-cell expansion, maturation and MHC-I presentation18,19,20,22, we examined if the random forest regression model based on log2 fold-change values of these genes could accurately predict spike-specific T-cell response. The accuracy of the model based on CD27, LEF1 and PSMB5 was found to be 70.43% (RMSE = 2.11) (Supplementary Fig. 8c), suggesting that the induction of these transcripts is important in inducing spike-specific T-cell responses.

Comparison of ARCT-021 with other vaccines

To understand how the host response to sa-mRNA vaccination fares against those to other vaccines, we performed a comparative analysis of our data with other published vaccine trials (Supplementary Table 1). Immune-related genes induced by different vaccines were visualized using a correlation matrix showing pairwise correlations of the mean gene log2-transformed fold changes between vaccines. When we compared gene expression at 1 day post-vaccination, ARCT-021 exhibited a strong correlation with live viral vectors (MRKAd5-HIV and rVSV-ZEBOV vaccines), adjuvanted vaccines (HepB + AS01B/AS01E and H5N1 + AS03) and mRNA vaccines (BNT162b2). However, little overlap was seen with the live-attenuated vaccines, unadjuvanted (H5N1) and polysaccharide vaccines (Pneumovax23) (Supplementary Fig. 8d). Closer examination of the host response kinetics for each vaccine showed that the peak transcriptional responses for the live-attenuated vaccines occurred at later time-points, at day 7 post-vaccination as compared to other vaccines10,23,24,25, which occurred at day 1 post-vaccination.

To investigate if the weak correlation with live-attenuated vaccines was attributed to differences in kinetics of vaccine-induced host response, we visualized the correlation matrix at time-points where peak transcriptional responses were observed for each vaccine. Indeed, the correlation with YF17D was much stronger (Fig. 4a), suggesting that the transcriptional alterations in YF17D were more prolonged and delayed as compared to ARCT-021. Among all the vaccines analysed, rVSV-ZEBOV displayed the strongest correlation with ARCT-021, and this finding was consistent across 2 different studies26,27 (Fig. 4b). Collectively, these observations suggest that a sa-mRNA vaccine that is able to induce the innate immune responses observed albeit in a more delayed fashion could potentially engender immunogenicity that approaches those of LAVs.

Fig. 4: ARCT-021 gene signatures are correlated with live viral vectors, adjuvanted and mRNA vaccines.

a Correlation matrix showing pairwise correlations of the mean gene log2-transformed fold changes between the different vaccines at time-points with peak transcriptional responses. Transcriptional responses peaked on day 7 for live-attenuated vaccines and Pneumovax, day 3 for H5N1 and day 1 for all other vaccines. Size and intensity of dots are proportional to the magnitude of correlation coefficient. b Pearson correlation of log2-transformed fold-changes of transcripts in subjects receiving rVSV-ZEBOV vs ARCT-21 at day 1 post-vaccination26,27. r = correlation coefficient, p = significance of correlation.