Characterization of potential biomarkers of reactogenicity of licensed antiviral vaccines: randomized controlled clinical trials conducted by the BIOVACSAFE consortium

Biomarkers predictive of inflammatory events post-vaccination could accelerate vaccine development. Within the BIOVACSAFE framework, we conducted three identically designed, placebo-controlled inpatient/outpatient clinical studies (NCT01765413/NCT01771354/NCT01771367). Six antiviral vaccination strategies were evaluated to generate training data-sets of pre-/post-vaccination vital signs, blood changes and whole-blood gene transcripts, and to identify putative biomarkers of early inflammation/reactogenicity that could guide the design of subsequent focused confirmatory studies. Healthy adults (N = 123; 20–21/group) received one immunization at Day (D)0. Alum-adjuvanted hepatitis B vaccine elicited vital signs and inflammatory (CRP/innate cells) responses that were similar between primed/naive vaccinees, and low-level gene responses. MF59-adjuvanted trivalent influenza vaccine (ATIV) induced distinct physiological (temperature/heart rate/reactogenicity) response-patterns not seen with non-adjuvanted TIV or with the other vaccines. ATIV also elicited robust early (D1) activation of IFN-related genes (associated with serum IP-10 levels) and innate-cell-related genes, and changes in monocyte/neutrophil/lymphocyte counts, while TIV elicited similar but lower responses. Due to viral replication kinetics, innate gene activation by live yellow-fever or varicella-zoster virus (YFV/VZV) vaccines was more suspended, with early IFN-associated responses in naïve YFV-vaccine recipients but not in primed VZV-vaccine recipients. Inflammatory responses (physiological/serum markers, innate-signaling transcripts) are therefore a function of the vaccine type/composition and presence/absence of immune memory. The data reported here have guided the design of confirmatory Phase IV trials using ATIV to provide tools to identify inflammatory or reactogenicity biomarkers.


Results
In saline placebo-controlled studies, we evaluated live yellow fever virus (YFV) and varicella zoster virus (VZV) vaccines in YFV-naive and VZV-primed subjects respectively (Study A); a prime-boost-boost regimen with aluminum-hydroxide (alum)-adjuvanted subunit hepatitis B virus (HBV) surface antigen vaccine in initially HBV-naive subjects (Study B); and non-adjuvanted or MF59-adjuvanted subunit trivalent influenza vaccine (TIV or ATIV respectively) in subjects assumed to have been exposed previously to circulating seasonal influenza viruses (Study C); see Table 1. Vaccines or placebo were administered at Day (D)0 during the inpatient stay (D-1 ─ D5; see Fig. S1 for detailed study designs). Subjects of Study B (referred to as the HBV-1 group up to D28) also received a booster dose at D28 during the outpatient follow-up, and a third dose at D169, during their second inpatient stay (the HBV-3 group).

Reactogenicity.
Local and systemic AEs were collected during the inpatient and outpatient phases based on queries in which no symptoms of interest were specifically solicited. As expected, mostly mild reactogenicity was observed for these licensed vaccines, with the most common events being fever, injection site pain and headache. To achieve a uniform reactogenicity quantification across the groups that could serve as a basis for the correlative analyses in the subsequent studies, we calculated an integrated metric for all unsolicited local and systemic AEs per group. This was performed by summing (by subject and by day) the total recorded severity gradings across all treatment-related unsolicited local and systemic AEs, and calculating the overall group scores ( Table 1). The scores obtained for the ATIV group were considered the most informative for future use in systems biology analyses.
Vital signs, hematology, and blood biochemistry. Four-hourly recording of vital signs during the inpatient stay revealed the expected diurnal changes, with subtle decreases in the oral body temperature and heart rate (~0.2 °C and ~5 bpm) during the first two nights post-vaccination, which were seen in all groups except the ATIV group (Fig. 1). Remarkably, the ATIV group exhibited in the first night (16-24 hours post-dose) a transient rise in baseline temperature (~0.3 °C), which remained within the normal range and was also not accompanied by a decreased nocturnal heart rate. As this was not observed with TIV, the data suggest that the effect was mediated by the oil-in-water (o/w) adjuvant MF59. This effect would not have been detected in conventional vaccine Acute-phase proteins, and serum cytokines/chemokines. During the inpatient stay, a clear (~5-fold) increase in the levels of C-reactive protein (CRP; a biomarker of acute inflammation) was only seen in the ATIV group (Fig. 3a). These levels began to rise above baseline from 16 hours post immunization and peaked at 48 hours, but remained below the range associated with clinically relevant inflammation. No changes were observed in any other group.
In Study A, levels of pentraxin 3 (PTX3; Fig. 3b), a protein related to CRP which also plays a role in the innate response 18 , showed that there was more background variability in the assay. Still, a relatively modest (~30%) increase could be seen 24 hours after immunization with ATIV. No clear response was seen with the other vaccines.
Interferons (IFNs) are indispensable for control of viral infections and immune regulation. During the inpatient phase, early serum responses of IFN-γ-induced protein 10 (IP-10/CXCL10) and the interferon-inducible monocyte chemotactic protein 1 (MCP1/CCL2) were only observed for the ATIV and YFV groups. For IP-10, the ATIV elicited robust responses (4 FC from baseline) with high inter-subject variability at D1, while for YFV the levels gradually increased from D3, peaking (at 2 FC) at D7 (Fig. 4a). For MCP-1, low (<1.5 FC) detections were observed in the ATIV group at D1 (Fig. 4b), also with a high degree of variability. After the inpatient phase, low MCP-1 responses (≤1.4 times baseline values) were seen for all vaccine or control groups, suggesting an impact of the changes in trial setting rather than a vaccine-related effect.
Whole blood transcriptomics. Peripheral blood transcriptomics data were grouped by previously described functional clusters of blood transcriptional modules (BTMs 19,20 ) containing genes regulating IFN-γ, CD4 stimulation or innate cell (monocytes, neutrophils, DCs)-associated responses (Fig. 5). As before, there were significant differences between the inpatient and outpatient time points (at false discovery rate [FDR]-adjusted P-value < 0.05), especially in the expression of globin genes (indicative of different blood cell frequencies). As www.nature.com/scientificreports www.nature.com/scientificreports/ this was not consistently seen across all BTMs and vaccine groups, these differences were most likely associated with the specific kinetics of the different phases of the innate and early adaptive response. The data also revealed distinctly different kinetic patterns between live and subunit vaccines, and allowed discrimination based on the participants' priming statuses (for both HBV groups) and on the vaccine adjuvantation (for the ATIV and TIV groups).
In Study A, the (previously immunologically naïve) YFV group exhibited a gradual increase in innate cell-and IFN-associated transcripts in the first week. This was followed, between D2 and D28, by gene responses involved in CD4 stimulation, which peaked at D14. The early IFN-associated response was the main discriminant between the two live vaccine groups. Indeed, in the VZV vaccine recipients, who were all seropositive for the virus, activation of CD4-stimulating genes between D2-D5 was followed by responses of innate cell-associated transcripts at D7 and D14, and then by a second activation of adaptive cell-related transcripts at D21. These patterns were associated with differences in the vaccine composition and in the participants' immune statuses.
In contrast to the live vaccines, the HBV vaccine in Study B elicited in both naive and primed subjects only modest innate cell-related responses, which were seen in the first week and had contracted to baseline at D14. In addition, a low-level activation of IFN-related responses was detected, but only in the naive subjects of the HBV-1 group. No activation of CD4 stimulation-related genes was observed in either group of subjects.
In Study C, both groups of influenza vaccine recipients exhibited similar kinetic patterns, showing early peaks of IFN-related and innate cell stimulation-associated genes at D1. This was followed by CD4-stimulating www.nature.com/scientificreports www.nature.com/scientificreports/ responses peaking at D5, with earlier detections and higher-level responses for the ATIV group. The latter observation was most likely associated with the presence of the MF59 adjuvant. A second peak of innate cell-associated genes was detected at D5, and these responses were in both groups sustained up to D21.    www.nature.com/scientificreports www.nature.com/scientificreports/ More detailed inspection of the activated BTMs in Study C revealed similar functional patterns between both groups, but with marked, statistically significant inter-group differences in response levels ( Fig. 6). Indeed, while only low-level signals were detected after TIV administration, with significant responses for only a few modules, higher-level responses were seen in the ATIV group. Responses included upregulation of modules relating to innate immunity and IFN-related responses, and downregulated T-cell activation-associated modules, both observed at D1. This was followed by upregulation of late-phase response modules governing CD4, C-MYC and PLK1 signaling, seen between D4 and D6 (at FDR-adjusted P-value < 10 −4 ). In both groups, very low responses for genes associated with monocytes and T cells were already detected at D0 (at FDR-adjusted P-value < 0.05), however the effect size was small (area under the curve [AUC] < 0.65), and there were no associated genes with significant differences in expression).
Inter-subject variability in gene expression was observed in all vaccine groups, as illustrated by the responses in the individual ATIV recipients (Fig. 7). We also sought to determine whether the high degree of variability in IFN-related protein responses in serum (see Fig. 4) was correlated to inter-subject variability in gene responses in this group. The data revealed substantial inter-individual variability in gene response profiles. Focusing on the IFN-associated modules, we observed that 5, 6 and 4 of the 20 subjects did not respond at D1, D2 and D5, respectively. Most of these non-responders did respond at one or two of these time-points, and one subject did not respond at any of them. Such between-subject variability should be considered in future confirmatory studies. Ideally, this should be compensated for by increasing the sample size, and be accounted for in the data analysis.
Further individual analysis of the relationships between the D1 IFN signatures and serum IP-10 responses was done by stratification of the IP-10 levels by the subject-matched 'early transcriptomic responder' status. The data revealed a clear link between these two parameters (Fig. 8a), which was confirmed when we plotted the IP-10 levels at D0 (baseline) and D1 against enrichment levels for one specific IFN-related module (DC.M5.12 20 ; Fig. 8b).

Discussion
Using licensed safe antiviral vaccines as benchmarks, we undertook linked, residential clinical studies to generate a data-set of vaccine-elicited responses up to one month post-vaccination. The data set contained clinical events as well as responses at the level of gene transcripts. Previously, we described some highly specific aspects of the safety and adaptive immune response evaluations for some of these trials [15][16][17] . Here, we report the reactogenicity and early vaccine response data. The data suggest that the inflammatory response, with respect to physiological parameters, serum markers and innate signaling transcripts, is a function of both the vaccine composition (live vs. inactivated; nonadjuvanted vs. adjuvanted) and the presence of immune (CD4 T-cell or B-cell) memory in the recipient. www.nature.com/scientificreports www.nature.com/scientificreports/ As a first approach, we assessed the impact of immune memory after the low-level stimulation by the alum-adjuvanted recombinant subunit HBV vaccine, for which 3 doses are required to obtain protective responses 21 . The vaccine elicited low-level gene responses, which lacked CD4 activation-related transcripts in both naive and primed subjects, and IFN-related transcripts in primed subjects. In contrast to the gene responses, no differences between the first or third dose were detected with respect to inflammatory responses (CRP, innate cells), vital signs or adverse event reactogenicity (or lack thereof). This suggests that aluminum salt is a poor activator of immune transcripts, which is consistent with other studies 4,22,23 . Furthermore, this suggests that the expression of transcripts depends on the individual's pre-existing status of immune priming. Unfortunately, this www.nature.com/scientificreports www.nature.com/scientificreports/ study could not discriminate mechanistically between the immune system components that were responsible for the differences between naïve and primed individuals, as no studies on blood cell populations were performed.
In parallel, we assessed the impact of the o/w adjuvant MF59, using a subunit vaccine type administered to subjects with presumably comparable immune statuses at baseline. Aligned with the potent activation of innate immunity reported for MF59 [24][25][26] , the ATIV induced the most prominent innate response. This response was characterized by a sequential increase in mean body temperature and lack of expected decrease in mean heart rate in the first night. This was followed at D1 by activation of IFN-related and innate cell-related genes, by increased monocyte and neutrophil counts (coinciding with decreased lymphocyte counts), and by increased PTX3, serum IP-10 and MCP-1 protein responses, and then by a CRP response at D2. The observed serum PTX3 response is consistent with the increased PTX3 expression seen in the muscles of MF59-injected mice, and maybe also with the fact that PTX3 can act as an endogenous adjuvant of certain antimicrobial responses 26,27 . These data aligned with other evaluations of both MF59-containing influenza vaccines and conventional subunit TIVs 26,[28][29][30] . They are also consistent with clinical data for vaccines containing AS03, an o/w adjuvant that contains α-tocopherol 22,31 . Notwithstanding the evident quantitative differences in transcriptomic responses (Fig. 6b), the signatures of these responses were qualitatively similar for both vaccines. These signatures were characterized by a sharp peak around D1, when monocyte and neutrophil counts increased, and a prominent role for genes associated with early IFN signalling as well as for antigen-processing and presentation-related genes. Interestingly, the concurrent increase in peripheral myeloid cells and blood depletion of lymphoid cells, as seen at D1-2 (a phenomenon also seen with AS03-adjuvanted HBV vaccine 22 ) may reflect local recruitment of lymphoid cells. However, this would need to be confirmed by direct investigation of the events in tissue at the sites of the immune response.
Though the majority of the ATIV recipients exhibited robust transcriptomic responses, the high level of inter-subject variability in transcriptomic responses was remarkable, as observed for example in the interferon modules. This heterogeneity may be explained by the subjects' variable degrees of immune memory derived from previous infection and/or vaccinations with different matching or non-matching strains (which we did not investigate), in combination with the small sample size (since heterogeneity was also observed for the other vaccines). It is also possible that the gene-set enrichment analysis based on absolute values of gene expression (rather than on fold-changes) was not a sufficiently sensitive method. While variability in individual measurements of gene expression may also be due to technical variability (which could have prevented detection of enrichment in some cases), this is unlikely to have played a significant role, given the remarkable uniformity of the analyzed control RNA samples. In any case, the variability in gene expression observed in some ATIV recipients over time did not appear to have affected their ability to mount strong influenza-specific antibody and cellular responses, as previously reported 15 .
Preclinical data suggest a link between o/w-adjuvant-mediated activation of innate immunity and reactogenicity, which is corroborated by clinical data (reviewed in ref. 32 ). Indeed, for an AS03-adjuvanted pandemic www.nature.com/scientificreports www.nature.com/scientificreports/ influenza vaccine, there was a trend for an association between the occurrence of medium/high-level reactogenicity and increased IP-10 levels, but no such putative link was seen for CRP or other inflammatory markers in serum 3 . In the current study, individual analysis detected a relationship between the D1 IFN signatures and serum IP-10 responses, however it remains to be determined to which extent these immune phenotypes correspond with reactogenicity for this vaccine, and whether such links are translatable to other vaccines and/or adjuvants.
Pre-vaccination signals for monocytes-related and T-cell-related modules were weak for both influenza vaccines, and were shown to be enhanced after vaccination. In the abovementioned study with the AS03-adjuvanted pandemic A/H1N1 vaccine 3 , the subjects experiencing medium/high-level reactogenicity shared a baseline gene expression associated with certain B-cell phenotypes, suggesting that transcriptomic changes before vaccination might be predictive of clinical events post-vaccination. An impact of the baseline transcriptome on the innate vaccine responses has also been reported for the YFV and HBV vaccines 33,34 , and the links with reactogenicity warrant further investigation. Pre-vaccination signals will be evaluated in-depth in the forthcoming larger clinical trials using the adjuvanted influenza vaccine.
Finally, we evaluated two different live-attenuated vaccines (inducing variable levels of pathogen-associated molecular patterns-mediated innate immunity), in the absence (YFV) or presence (VZV) of immune memory. Although based on very different viruses, both vaccines exhibited a gradual and (as compared with subunit vaccines) suspended increase in gene responses. Due to the kinetics of viral replication, peaks were detected around D5-D7, around the same time as the monocyte response peaked in blood. The YFV vaccine, a potent inducer of innate immunity, can elicit long-term activation of 'trained' monocytes and NK cells, which is for some activation markers detectable for up to 2 months post-vaccination [34][35][36][37] . This aligns with the readily detectable innate gene expression and with the early serum IP-10 and CRP responses observed here, in the absence of immune memory. By contrast, the lack of significant activation of IFN-associated genes by the VZV vaccine may be linked to the pre-existing antibodies in these subjects (which may have reduced viral replication), and to the higher level of vaccine attenuation as compared to the YFV vaccine. Indeed, previous data demonstrated that even with a higher-dosed VZV vaccine as the one used here, the induced transcriptomic responses remained relatively low 38 .
Some strengths and intrinsic limitations of the present study should be highlighted. On the one hand, the controlled inpatient setting employed in this study enabled detection of subtle vaccine-specific changes in vital signs in the first night post-vaccination, that would have gone undetected in conventional trials. On the other hand, this setting also entailed the risk of an 'incarceration' effect, with the consequent deviations of physiological/laboratory safety parameters from their individual 'normal' values. The placebo-controlled study design was therefore vital to discriminate between such an effect and any vaccination-related impacts on inflammatory parameters. Interestingly, the incarceration effect seen in the blood chemistry was not mirrored by the CRP, chemokine or transcriptomic responses. This might be explained by the transient nature of the latter responses, peaking in the first few days, or by the assumption that these changes only manifest locally, due to low-level spillover from the draining lymph node into the serum. A limitation was that these small-scale observational studies with only exploratory endpoints did not allow comparative analyses to be done between groups, or definitive conclusions on biomarkers to be drawn. They also did not allow analyses on the effects of any factors potentially affecting innate immunity and reactogenicity, such as ethnicity 16 , age (even within this population of healthy adults 3 ), or gender 39 . These factors would also need to be considered for development of predictive personalized biomarkers of vaccine safety.
In conclusion, the current data informed the design of subsequent research phases. Based on its distinct physiological and immunological response patterns, the ATIV was selected (together with other vaccines not evaluated here) for further studies in confirmatory Phase IV clinical trials, and in human muscle biopsy studies. This research aimed to associate the transcriptomic changes in peripheral blood with those in sorted cell populations, and to dissect out the local responses underlying differential reactogenicity to different adjuvants. These results are supported by parallel preclinical BIOVACSAFE studies on adjuvants, as done for alum and MF59 in mice 7,10 . The collected data will be subjected to integrated systems biology studies, to identify the biomarkers that could accelerate development of safer and more effective new vaccines.

Materials and Methods
Study design. Studies A, B and C (ClinicalTrials.gov: NCT01765413, NCT01771354 and NCT01771367, respectively) were subject/laboratory-blinded, randomized, placebo-controlled studies, conducted from February 2013 through September 2014 at the University of Surrey Clinical Research Centre (Surrey, Guildford, UK) in accordance with the Helsinki Declaration and Good Clinical Practices. Protocols (nos. CRC305A-C) and CONSORT checklists were published previously [15][16][17] . Ethical approvals were obtained from the NRES Committee London -Surrey Borders (nos.: 12/LO/1871, 12/LO/1899, 13/LO/0044 respectively). All participants provided written informed consent prior to enrolment. Blood collection for biochemistry, hematology, whole blood transcriptomics and immunology parameters, and monitoring of reactogenicity and vital signs, was performed at scheduled time-points before and after immunization (Fig. S1).
Objectives. The study objective was to generate an exploratory 'training' set of data characterising reactogenicity as well as physiological and innate immune responses following immunisation with licensed vaccines. This data set would then be used to identify putative biomarkers of vaccine reactogenicity in subsequent research phases, that could guide the design of follow-on large-scale non-residential confirmatory studies, using selected vaccines and time points.
Participants and demographics. Healthy male and female volunteers 18-45 years of age were enrolled and randomized into two (Study B) or three (Studies A and C) study groups (Fig. S3). Each subject acted as their own control for kinetics with comparison from the baseline preimmunization levels of the measured biomarkers.
A total of 49, 25, and 49 healthy adults enrolled in Studies A, B and C, respectively, were vaccinated, and all but two subjects completed the respective studies (Table 1 and Fig. S3). These two subjects (one each in the VZV or HBV-3 groups) withdrew consent not due to an AE, either during or after the inpatient phase, and were excluded from the analyses. Demographic profiles were broadly balanced between the groups in terms of mean age and ethnicity, but there was a male preponderance in the YFV, VZV and HBV groups. No YFV viremia was detected in the recipients of live YFV vaccine, at any time point.

Vaccinations.
Participants were admitted to the research center on D-1 to acclimatize and for pre-immunization blood sampling. All blood samples were taken within 15 minutes of the designated time. All treatments were administered into the deltoid muscle or overlying subcutaneous tissue at 08:00 h (±15 minutes). All participants received only a single injection when immunized with either a single vaccine or a placebo, according to the group allocation. In the vaccine groups in Study A, participants seropositive for anti-VZV antibodies (as confirmed by serology and immunization history) received live-attenuated VZV (Oka strain) vaccine (Varilrix supplied as 10 3.3 plaque-forming units; GSK), while participants seronegative for anti-YFV antibodies and without a prior history of YFV vaccination received live-attenuated YFV vaccine (Stamaril containing ≥ 1000 LD 50 units; 17 D-204 strain; Sanofi Pasteur). Treatments were administered subcutaneously (deltoid region) as a single 0.5 mL dose, delivered at D0. In Study B, participants seronegative for anti-HBV antibodies and without a prior history of HBV vaccination received intramuscular alum-adjuvanted (rDNA) HBV vaccine (Engerix-B1 containing 20 μg adsorbed HBsAg, GSK). The HBV-1 group received a priming dose at D0. A second HBV dose was administered as an outpatient vaccination at D28, and then the same participants were re-admitted to the research center on D168 and given a third dose at D169 (designated group HBV-3). In Study C, participants assumed to be primed with hemagglutinin via vaccination or natural infection, received a single 0.5 mL IM dose of either non-adjuvanted or MF59-adjuvanted seasonal trivalent influenza surface antigen (inactivated) vaccine (TIV -Agrippal or ATIV -Fluad, respectively; Novartis Vaccines) at D0. Both vaccines contained the recommended composition for the 2012/2013 northern hemisphere influenza season 15 . The control groups included in each study received saline placebo intramuscularly (groups B and C) or subcutaneously (group A) at the same time points as the vaccinees. Placebo subjects were later pooled to one group (N = 20), excluding the data from D167 onwards for the control group for HBV-3 (N = 4), which were analyzed separately (see Fig. S3).
In the inpatient phase, participants were admitted to the chronobiology ward of the Surrey Clinical Research Center at D-1 where they remained for 7 days and 6 nights, and allowed home after the 20:00 h blood draw on D5. While inpatients, they were held to controlled diet, exercise and sleep routines (lights on/off at 07: Reactogenicity. Reactogenicity (local and systemic unsolicited AEs, classified using the medDRA-preferred terms) were recorded throughout the study. Participants were queried for any AEs on a daily basis during the inpatient stay and at each outpatient visit, but there were no symptoms of interest specifically solicited. Local complications from indwelling cannulas such as phlebitis were also checked for daily and recorded, but were excluded from reactogenicity calculations. Laboratory AEs were also excluded from reactogenicity calculations. To obtain a uniform reactogenicity quantification across the groups, that could serve as a basis for the correlative analyses in the subsequent studies, AEs were first classified by the clinicians as either related or unrelated to immunization, and only the AEs that were considered as related were included in the calculation. Each AE was then described by the participant, or was measured by the investigators (in the case of local inflammatory reactions). AEs were quantified by referencing to the appropriate FDA tables (Guidance for Industry -Toxicity Grading Scale for Healthy Adult and Adolescent Volunteers Enrolled in Preventive Vaccine Clinical Trials, FDA, Sept. 2007) adapted to the protocol, and were graded as mild (no interference with activities), moderate (some interference) or severe (significant interference), or they were quantified using the measurement scale provided for local injection site reactions. The standardized grading was then translated to a numerical value on a 1, 2, 3 scale, and multiplied by the AE duration (expressed in days), to generate a reactogenicity score. For each vaccine group, the number of treatment-related AEs with an onset between D0 and D7 post-immunization, sum of the associated reactogenicity scores (the vaccine "reactosum"), proportion of participants recording any treatment-related AE with an onset between D0 and D7 post-immunization, and the mean participant reactosum, were calculated.
Physiological and laboratory assessments. Vital signs (heart rate, oral temperature, and diastolic/systolic blood pressure measured following five minutes in a supine position were recorded four-hourly (±15 minutes) on D0-3, and twelve-hourly on D4-5 of the inpatient stay and at outpatient visits. Standard laboratory blood chemistry (liver, renal and bone panels), CRP, and hematology (automated Full Blood Count with white cell, erythrocyte sedimentation rate) assays were performed as described 16 . They were performed once daily at D-1 ─ D5, and at each outpatient visit (Fig. 1). CRP concentrations were assayed in a two-step process: initially a standard sensitivity assay was used, and samples with a value < 10 mg/L were then re-assayed using a highly sensitive CRP assay, and this value alone was reported. In a subset of participants (ATIV, TIV and placebo groups in Study A; N = 20, 21 and 8, respectively), serum CRP and PTX3 concentrations were measured in stored serum samples at more frequent time points, using for PTX3 an in-house ELISA assay, as described previously 40 .
Gene expression profiles. Microarray analysis. Peripheral blood was drawn into PAXgene tubes (PreAnalytiX) and RNA was extracted on the automated QIAcube system (Qiagen) using the PAXgene Blood RNA kit (Qiagen) according to the manufacturer's instructions. Quality control and quantification of isolated RNA was analyzed with an Agilent 2100 Bioanalyzer (Agilent Technologies) and a NanoDrop 1000 UV-Vis spectrophotometer (Thermo Fisher Scientific). Microarray experiments were performed as single-color hybridization, and RNA was labeled with the Low Input Quick-Amp Labeling Kit (Agilent Technologies). In brief, mRNA was reverse transcribed and amplified using an oligo-dT-T7 promoter primer, and labeled with cyanine 3-CTP by T7 in vitro transcription. After precipitation, purification and quantification, 0.75 μg labeled cRNA was fragmented and hybridized to custom whole genome human 8 × 60 K multipack microarrays (Agilent-048908) according to the supplier's protocol (Agilent Technologies). Scanning of microarrays was performed with 3 μm resolution and 20-bit image depth, using a G2565CA high-resolution laser microarray scanner (Agilent Technologies). Microarray image data were processed with the Image Analysis/Feature Extraction software G2567AA v. A.11.5.1.1 (Agilent Technologies), using default settings and the GE1_1105_Oct12 extraction protocol.
Microarray normalization and quality control. Blinded primary readouts of the microarrays were read, background corrected, normalized and controlled for quality using the R package limma 41 (version 3.30). For background correction, the gProcessedSignal from the primary readouts was used. Between-array normalization was done using the quantile method in limma. Quality control relied on density plots, testing for outliers, visualization with principal component analysis and visual inspection of individual array images. The normalized data was locked and submitted to project management. Next, the data was unblinded for further analysis. All primary readouts and the background corrected and normalized data are available from the Gene Expression Omnibus (GEO) database under the BioProject identifier PRJNA515032 (https://www.ncbi.nlm.nih.gov/ bioproject/?term=PRJNA515032).
Differential gene expression analysis. Prior to the analysis, the hybridization control samples were removed from the data set, and the gene expression values were averaged for each probe over all replicates of that probe on the microarray, using the 'avereps' function from limma. Differences in gene expression for each vaccine at each time point tested were assessed using a three-factor linear model in limma. The expression was fit to time point, group (vaccine vs placebo) and subject. The contrast tested for a given vaccine and a given time point was the interaction between the difference in expression between this time point and the D0 time point, and the difference between the given vaccine and placebo, as follows: (V Dn − V D0 ) − (P Dn − P D0 ), where V stands for the given vaccine, P stands for placebo, Dn stands for the given time point, and D0 stands for the sample collected at vaccination. The p-values were corrected for false discovery rate using the Benjamini and Hochberg (BH) procedure 42 .
Gene set enrichment analysis. Gene set enrichment was tested with the CERNO algorithm implemented in the R package tmod 43 , version 0.40, with the MSD metric for ordering the genes 44 . For testing, the gene sets (BTMs) defined by Li et al. 19 and Chaussabel et al. 20 were used. P-values were corrected using the BH procedure; gene set enrichments with q < 0.05 were considered significant. Enrichment was visualized with the tmodPanelPlot function from tmod.

Analysis of individual variability.
For each individual at each timepoint, the genes were ordered by the measured expression, and the gene set enrichment was performed on the ordered list of genes, using CERNO statistic as implemented in tmod.
Correlation analysis. To test for enrichment in genes correlated with a given cytokine response, the following procedure was applied. First, for the given parameter, the Spearman correlation coefficient between this parameter and the expression of each gene was calculated. Next, the genes were ordered by the decreasing absolute correlation coefficient, and the CERNO enrichment test was applied to the ordered list of genes.
Statistical methods. Safety and immunology parameters were reported descriptively and tabulated as SEM bars/ ribbons. Computations were performed in R software (https://www.r-project.org/).