Origin of samples

Samples used in these experiments were cryopreserved peripheral blood mononuclear cells (PBMC) or whole blood in RNA lysis buffer from a double-blind, randomised, placebo-controlled Phase 2b efficacy trial of a candidate TB vaccine, MVA85A, in BCG-vaccinated, HIV-negative South African infants (South African National Clinical Trials Register DOH-27-0109-2654, ClinicalTrials.gov NCT00953927). Infants were randomized to receive either one dose of MVA85A (1 × 10⁸ plaque forming units in 0.06 mL) or an equal volume of Candida skin test antigen (Candin, AllerMed, USA) as placebo at 4–6 months of age [4]. The trial was approved by the University of Cape Town Faculty of Health Sciences Human Research Ethics Committee, Oxford University Tropical Research Ethics Committee, and the Medicines Control Council of South Africa. Parents or legal guardians provided written, informed consent. Storage of samples for exploratory immunological analyses was fully ethically approved.

The samples used in this study were selected from a subset of 100 infants who had a small blood sample taken one day post-vaccination. The samples were selected to exclude cases (infants who went on to develop TB disease) and controls (selected to demographically match the cases), which will be used in a future study looking at correlates of risk of TB disease. Therefore none of the samples used in this study were taken from infants diagnosed with TB during the course of the trial. Cells were collected 0–7 days pre- and 28 days post-vaccination with MVA85A/placebo in cell preparation tubes with sodium heparin (CPT; Vacutainer; BD) and PBMC separated and cryopreserved. The cells were thawed and stimulated as detailed below. One day post-vaccination, 50-300 μL of whole blood was collected directly via heel prick into a tube filled with RLT buffer (RNeasy kit, Qiagen) containing 10 μL/mL β-mercaptoethanol using a BD Quikheel Lancet. The sample was immediately frozen and RNA extracted as detailed below.

Cell thawing

Cryopreserved PBMC were rapidly thawed in a 37°C waterbath and transferred to a 15 mL Falcon tube containing 10 mL R10 (RPMI, 10% FCS, 1% L-glutamine, 1% Pen-Strep and 1% sodium pyruvate). PBMC were pelleted, supernatants discarded and resuspended in 10 mL R10 with 20 μL Benzonase (Merck Chemicals Ltd.) and rested overnight at 37°C, 5% CO₂. PBMC were counted on a Casy Counter (Roche) and split into appropriate volumes for each assay. Not all assays were performed on all samples.

Ex-vivoIFN-γ ELISpot assay

The ex-vivo IFN-γ ELISpot assay was performed on thawed PBMC samples collected pre-vaccination and 28 days post-vaccination as previously described [23]. 3 × 10⁵ PBMC were stimulated in triplicate with a pool of Ag85A peptides, consisting of 66 15mers, overlapping by 10 amino acids (2 μg/ml) (Peptide Protein Research).

Gene expression assays

PBMC from pre-vaccination and 28 days post-vaccination were incubated for 12 hours with either R10 media alone (unstimulated) or pooled Ag85A peptides as described for the ELISpot (2 μg/mL). After 12 hours supernatant was removed and the PBMC resuspended in 350uL RLT buffer (Qiagen) containing 10 μL/mL β-mercaptoethanol and frozen at -20C.

Blood in RLT buffer was thawed and RNA extracted using the RNeasy kit (Qiagen) according to manufacturer’s instructions, including the optional protocol for DNA digest (RNase-free DNase kit, Qiagen). The protocol was modified in the following way for the heelprick samples due to the small volume of whole blood collected: 80% ethanol was added to precipitate the RNA (rather than the recommended 70%) and an extra wash with 350 μL RW1 buffer was performed prior to DNA digest.

Messenger RNA was amplified from the total RNA using the Illumina Totalprep kit (Ambion) according to manufacturer’s instructions. RNA quantity and quality was assessed using a Nanodrop ND-1000 Spectrophotometer and an Agilent Bioanalyser (Agilent RNA 6000 Nano Kit).

750 ng amplified cRNA was labeled and hybridized to Illumina Human HT-12 v4 beadchips as specified in the manufacturer’s instructions. Beadchips were scanned on an Illumina iScan machine and data extracted using the GenomeStudio software.

RNA-Seq was performed on pre- and post-vaccination PBMC from 6 infants (12 samples). Total RNA was sent to the Beijing Genomics Institute (BGI). Libraries were constructed using the TruSeq kit (Illumina) and sequenced on a HiSeq2000 sequencer, using paired-end reads of 90 bp and 30 M sequencing depth. Quality control was performed and the reads aligned to the genome hg19, downloaded from UCSC (http://genome.ucsc.edu/), using SOAPaligner2.21 with the following constraints: maximum number of mismatches allowed on one read is 5 bp, no gaps allowed, only repeat hits reported.

Data analysis

Accession codes

Gene expression omnibus: GSE56559 (day 1 heelprick, South African infants), GSE56561 (PBMC, South African infants) GSE40719 (UK adults).

MVA85A induces an inflammatory signature one day post-vaccination

Illumina microarray gene expression analysis of 82 whole blood samples taken one day post-vaccination (37 MVA85A, 45 Candin placebo) identified 32 differentially expressed genes. These genes were largely associated with the immune response (differentially expressed genes are shown in Table 2 with genes associated with the immune response highlighted in bold). Hierarchical clustering showed that infants fall into three clusters with a mixed MVA85A/Candin cluster exhibiting a moderate level of expression of inflammatory genes (Figure 1). This analysis shows MVA85A-vaccinated infants exhibit a more inflammatory gene expression profile than those in the placebo group however the range is large in both groups and there is an overlap between the two groups. This overlap could be due to the immunomodulatory properties of Candin, which induces inflammation and may lead to functional reprogramming of monocytes associated with protection from subsequent infection [34].

PROBE_ID	SYMBOL	Fold change	AveExpr	adj.P.Val
ILMN_1791759	CXCL10	3.2	7.18	0.04
ILMN_1799848	ANKRD22	2.61	6.65	0.04
ILMN_1656310	INDO	2.53	5.68	0.04
ILMN_2114568	GBP5	2.52	8.7	0.02
ILMN_2132599	ANKRD22	2.46	7.09	0.04
ILMN_3239965	IDO1	2.35	6.17	0.04
ILMN_3247506	FCGR1C	2.13	6.59	0.04
ILMN_1782487	LOC400759	1.96	5.38	0.01
ILMN_2066849	FAM26F	1.89	5.99	0.04
ILMN_1809086	XRN1	1.74	6.83	0.04
ILMN_2053527	PARP9	1.67	6.81	0.04
ILMN_1769520	UBE2L6	1.61	11.07	0.04
ILMN_1707979	CARD17	1.57	5.07	0.04
ILMN_2326509	CASP1	1.57	8.86	0.04
ILMN_1671452	MRPL44	1.5	6.05	0.04
ILMN_1700671	ETV7	1.49	4.83	0.04
ILMN_1678454	CASP4	1.48	9.83	0.04
ILMN_1715401	MT1G	1.43	4.78	0.01
ILMN_3238525	CARD17	1.35	4.6	0.04
ILMN_1693287	POMP	1.32	9.68	0.04
ILMN_1761159	ESYT1	-1.34	8.78	0.05
ILMN_1693410	BRI3BP	-1.37	6.56	0.04
ILMN_1774828	VEZT	-1.4	5.72	0.04
ILMN_3236036	LOC283663	-1.41	5.39	0.04
ILMN_3256478	LOC100129034	-1.41	5.38	0.04
ILMN_3240997	ARAP3	-1.43	5.7	0.04
ILMN_1767612	BBS2	-1.44	5.6	0.04
ILMN_1775542	FAIM3	-1.45	9.66	0.04
ILMN_3251155	PCBP2	-1.48	5.49	0.01
ILMN_1772876	ZNF395	-1.5	6.21	0.04
ILMN_3305938	SGK1	-1.56	6.59	0.04
ILMN_1731064	CABC1	-1.59	6.1	0.02
ILMN_3229324	SGK1	-1.6	6.3	0.04

Unstimulated whole blood, genes identified using the R package limma.

The immune response to stimulation with Ag85A peptides

The antigen-specific immune response to Ag85A was assessed by IFN-γ ELISpot and Illumina microarray gene expression analysis. Infants vaccinated with MVA85A had a significantly higher post-vaccination Ag85A-specific ELISpot response than the Candin group (Figure 2a). Differentially expressed genes between the vaccine and placebo groups, and pre- and post-vaccination time points, are shown in Table 3. The genes induced following Ag85A peptide stimulation are all associated with the STAT1 pathway and exhibit a highly correlated pattern of expression. Furthermore, upregulation of this pathway occurred in infants who also had a detectable Ag85A-specific T cell response by IFN-γ ELISpot assay but not in unstimulated cells or infants who received the candin placebo (Figure 2b,c). This observation suggests that, in this population, gene expression analysis does not add substantial information to that measured by the IFN-γ ELISpot assay in capturing the response to Ag85A peptide stimulation following MVA85A vaccination.

A proportion of infants did not respond to antigenic stimulation with Ag85A peptides following MVA85A vaccination. This lack of response was observed both by IFN-γ ELISpot and gene expression analysis. It has been suggested that low or absent responses to MVA85A may be one explanation for the lack of efficacy observed in the trial [12]. Further analysis of cases and controls is under way to address this question however, in this smaller study, we next investigated some of the mechanisms underlying the magnitude of the adaptive immune response which develops following MVA85A vaccination.

Myeloid cells and inflammation are associated with a higher ELISpot response

The following analyses were performed using only samples from infants vaccinated with MVA85A, to further investigate the mechanisms underlying the immune response to this vaccine. The R package Significance Analysis of Microarrays (SAMR) was used to identify genes whose expression correlated with the Ag85A-specific T cell frequencies measured 28 days post-vaccination by IFN-γ ELISpot assay. SAM identifies genes significantly correlating with a continuous response variable, in this case the IFN-γ ELISpot, and outputs a positive and negative set of genes based on the strength of the correlation of each gene with higher (positive) or lower (negative) values of the response phenotype [29, 35]. Since the IFN-γ ELISpot responses in this study were very low, we have subsequently defined infants as responders or non-responders. An ELISpot response was deemed positive if the average count in the positive control wells was at least twice that in the negative control wells and at least 5 spots more than the negative control wells [36]. Sets of correlating genes pre-vaccination and one day post-vaccination were generated and ranked according to their significance score (Figure 3a). This ranked list of genes was then analysed using the Broad Institute Gene Set Enrichment Analysis PreRanked function, using the Blood Transcription Modules defined by Li et al. as a reference gene set [33]. Pre-vaccination, responder infants have an over-representation of genes enriched in monocytes, activated dendritic cells and neutrophils as well chemokines and inflammatory pathways (Figure 3b). One day post-vaccination, there is a negative association between lymphoid cells and the subsequent development of an ELISpot response (Figure 3c). Moreover, there is positive enrichment of gene sets associated with myeloid cells, inflammation and an antiviral response. The expression pattern of these clusters and their relationship to the ELISpot response is shown in heatmaps in Figure 4. Higher expression of genes associated myeloid cells and inflammation pre- and 1 day post-vaccination are both associated with the development of an antigen-specific T cell response to vaccination with MVA85A, suggesting the ability of MVA to induce a strong innate response is key to its function as a vaccine vector in this population. Additionally, higher expression of genes associated with activation of lymphoid cells such as NK cells and cytotoxic T cells one day post-vaccination is associated with an absent ELISpot response 28 days later.

The ratio of myeloid to lymphoid cells has previously been associated with differences in TB disease risk in HIV-infected South African adults and susceptibility to malaria and influenza in other cohorts [37–40] and this may be another example of an outcome associated with this ratio. The gene expression profiles associated with responder infants are present pre-vaccination and show a strong overlap pre- and one day post vaccination. This suggests the baseline inflammatory profile of the infant, including the differing proportion of circulating leukocytes, is key to determining the response to vaccination. This may be influenced by genetic influences on innate immunity or environmental exposure preceding vaccination, including the response to BCG, which all infants received at birth. As MVA preferentially infects myeloid cells [41], a higher proportion of myeloid cells may lead to overall increased viral expression of Ag85A protein in infants with higher frequencies of myeloid cells. Conversely, killing of infected myeloid cells by cytotoxic T cells and NK cells may decrease antigen expression, inhibiting the development of a response to Ag85A.In a subset of samples, the RNA extracted from the PBMC collected pre- and 28 days post-MVA85A was also measured by RNA Sequencing, as part of a pilot project for future studies. Gene expression values as measured independently by these two methods are highly correlated (Figure 5), providing a technical validation for this result.

Gene expression signatures correlating with immunogenicity differ in South African infants and UK adults

Further to this, the relationships between these genes show interesting differences between the two populations (Figure 6). In the South African infants, expression of both CXCL10 and IDO1 one day post-vaccination correlates with the IFN-γ ELISpot response and with each other. In the UK adults however, expression of these genes correlates neither with each other nor with the IFN-γ ELISpot response. In the UK adults, we have previously described a negative association between expression of CTLA4 post-vaccination and the IFN-γ ELISpot response. However, this association was not found in South African infants. Furthermore, CTLA4 correlated positively with IDO1 in the UK adults but negatively in the South African infants.

Indoleamine 2,3-dioxygenase (IDO1) is an enzyme catalyzing the first and rate-limiting step in tryptophan catabolism. This enzyme has multiple physiological effects, including immunosuppression and regulation of T cells. In UK adults, expression of IDO1 in PBMC correlates with that of the inhibitory co-receptor CTLA4. A recent study showing IDO1 is a critical resistance mechanism in antitumor T cell immunotherapy targeting CTLA-4 in a mouse model of melanoma suggests the interaction between these two genes is an important component to consider when inducing an immune response for therapeutic purposes [43]. In the South African infants however, we did not observe a correlation between expression of IDO1 and CTLA4 but IDO1 correlated with CXCL10 instead, suggesting either a different role for this enzyme or that inflammatory and regulatory responses are closely coupled in this population, perhaps to protect against excessive immune responses.

These disparities suggest there may be important differences in the immune response to vaccination in these two populations and in the pathways that determine the adaptive immune response. However, it is impossible to determine whether the differences observed are attributable to age or to genetic or environmental differences between the populations. Furthermore, the post-vaccination timepoint differed in both cell type (whole blood in the infants, PBMC in adults) and time taken (one or two days post-MVA85A). PBMC were used in the pre-vaccination timepoint in both studies. Despite this, the similarities in the list of differentially expressed genes suggest these comparisons are meaningful and warrant further investigation.

UK adults have a stronger innate response to the vaccine, however the magnitude of this response does not predict the magnitude of the antigen-specific response. By contrast, in the South African infants the magnitude of inflammatory gene expression one day post-MVA85A correlated with the ELISpot response at week 4. Further investigation of these mechanisms and how they differ between the different populations in which vaccines are tested and deployed may prove important in future vaccine development strategies.