Keywords
Transcriptome, Blood, Malaria, Vaccination, IgE, Humoral, Bioinformatics, Modular repertoires
This article is included in the Sidra Medicine gateway.
Transcriptome, Blood, Malaria, Vaccination, IgE, Humoral, Bioinformatics, Modular repertoires
About 3.4 billion people, nearly half of the world’s population, live in areas at risk of malaria transmission1. Malaria infection resulted in an estimated 198 million cases in 2013 that may have caused between 367,000 and 755,000 deaths, according to the World Health Organization2. Recent concerns have been caused by rise in parasite resistance to artemisinin, which is the last effective monotherapy available for the treatment of malaria3. Cases of artemisinin resistance have been reported from much of Southeast Asia and now appear likely to reach the Indian subcontinent, with potentially dire consequences4. But significant advances over the past years have been made towards the development of an effective malaria vaccine5. Most notably this includes successful testing of a live vaccine consisting of radiation-attenuated sporozoites6, and this year licensure by regulatory authorities of the first malaria vaccine, the recombinant adjuvanted vaccine developed by global pharmaceutical GSK, called RTS,S also known by its commercial name, Mosquirix®. This is a highly significant landmark but unfortunately the efficacy of the vaccine for unknown reasons, and despite optimization attempts, remains suboptimal7,8. Thus identification of mechanisms underlying protection conferred by this vaccine, or lack thereof, may be key to the development of a broadly effective prophylactic vaccine against malaria. Unbiased “systems approaches”, consisting in profiling all the elements constitutive of a given biological system, have recently been implemented to investigate responses to vaccines9. Such an approach consisting in measuring blood transcript abundance on a genome-wide scale has been adopted for the serial profiling of responses to the influenza, pneumococcal, yellow fever or malaria vaccines10–15. In 2010, Vahey et al. reported results from a study investigating changes in transcript abundance in blood following administration of the malaria RTS,S vaccine15. In this report we share the results of a re-analysis of the data made available by Vahey et al. upon publication of their findings. We employed an innovative approach developed earlier; including by other team members part of previous publications which consists in identification of modular transcriptional repertoires – collections of co-clustered gene sets – in order to carry out modular level “fingerprinting analyses”16. This re-analysis led to original findings, with the identification of an interferon transcriptional signature at day 1 post-vaccination, correlating with protection as well as a second interferon signature at days 3 and 14 post-vaccination correlating this time with lack of protection of study subjects from subsequent challenges with the malaria parasite.
The methodology for constructing modular transcriptional repertoires has been described earlier16,17. The particular framework employed in this re-analysis has been described in an earlier study investigating responses to influenza and pneumococcal vaccines12. Briefly, nine datasets were used as input, including blood transcriptome profiles generated from patients with HIV, tuberculosis, sepsis, systemic lupus erythematosus, systemic arthritis, and liver transplant. Each dataset was clustered independently using Hartigan’s k-means clustering, using the elbow criterion to determine the optimal number of clusters for each dataset. Cluster membership information for each gene across the nine datasets was used to build a table recording the number of co-clustering events for each possible gene pair. This table was used in turn to build a weighted co-clustering network where each node is a gene and edges indicate co-clustering events with weight ranging from 1 (pair of genes belonging to the same cluster in 1 out of 9 datasets) to 9 (pair of genes belonging to the same cluster in 9 out of 9 datasets). The module selection process consisted in the identification within this large network of cliques, which are densely connected subnetworks. A principled approach was used starting in the first round with the selection of the largest subnetworks carrying the highest weight (co-clustering in 9 out of 9 datasets; corresponding to the M1 modules), followed by identification and removal from the selection pool of the next largest subnetwork and so on (with minimum clique size set at 10). When no additional modules could be identified for a given round of selection the stringency of the selection criteria was progressively relaxed (e.g. co-clustering occurring for 8 out of 9 datasets in the second round of selection, corresponding to the M2 modules; in 7 out of 9 datasets in the third round of selection, corresponding to the M3 modules, etc…). The datasets used for module construction have been deposited in NCBI’s Gene Expression Omnibus: GSE30101.
Functional analyses were carried out systematically for each module using commercial as well as publically available tools (primarily MetaCore™ version 5.0 and DAVID version 6.718) and results are reported on a wiki page: http://www.biir.net/public_wikis/module_annotation/V2_Trial_8_Modules. A complete list of the genes forming the modules is also available from the wiki.
The top six rounds of modules defined by this approach (M1–M6, a total of 62 modules) were used as a framework to analyze and interpret the datasets generated in the context of the Vahey et al. study: i.e. rather than carrying out analyses at the individual gene level, which assume that changes in transcript abundance for each gene occur independently from that of other genes, we performed analyses at the modular level, were changes are assessed for sets of co-clustered genes. Thus we summarize “modular response” as a single value, the percent of responsive genes for a given module. In earlier analyses the average fold change per module was also used to demonstrate that high level of concordance could be observed across microarray platforms at the modular level but not at the gene level17. For determining changes for individual subjects post-vaccination a cutoff is set against which individual genes constitutive of a module are tested. If the gene meets the set criteria it is considered “responsive”. “Module-level” data is subsequently expressed as a % value representing the proportion of responsive transcripts for a given module.
The design of the vaccine trial is described in detail by Vahey et al. and in an earlier publication15,19. Briefly, study subjects received the RTS,S vaccine, which consists of sequences of the Plasmodium falciparum Circum Sporozoite Protein (CSP) expressed in hepatitis B surface antigen and formulated with the proprietary adjuvant systems AS01/AS0220. Challenge was performed with a homologous 3D7 strain of P. falciparum delivered by 5 bites from infected mosquitoes. Samples were obtained from study participants at study entry (36 samples); on the day of the third vaccination (44 samples); at day 1 (43 samples), day 3 (43 samples), and day 14 (37 samples) thereafter; and at day 5 post-challenge (39 samples). Whole blood transcriptome profiles were generated using commercial Affymetrix HG-U133 chips. Data processing and normalization methodologies are described in the original publication. Data are available publically from the NCBI Gene Expression Omnibus (GSE18323). Only the blood transcriptional profiles generated on the day of the third vaccination and at day 1, day 3 and day 14 post-third vaccination were used in our re-analysis.
We employed a “modular repertoire approach” first described in 2008 in a research paper17, and more recently in a review16. Briefly, this approach consists in a priori identifying relationships among constituents of a given biological system, which in our case is the blood transcriptome. This makes it in turn possible to analyze transcriptional profiles as functionally interpretable gene sets rather than independent genes. Modular repertoires are established in an entirely data-driven process through the recording of co-clustering patterns of transcripts across a wide range of immune-related diseases. A collection of datasets encompassing infectious as well as autoimmune disorders and primary immune deficiency was used as input in order to capture a wide variety of immune signatures. The module construction process and modular analyses are described in detail in the Methods section.
In the original analysis of this dataset Vahey et al. report the identification: 1) of a transient signature at 24 hours post-vaccination that was not observed at subsequent time points. This signature is described as being associated with inflammatory processes elicited by the vaccine and was not associated with outcome of the infectious challenge; 2) of a signature at 5 days post-challenge that distinguish vaccinated from non-vaccinated individuals, thus directly reflecting and demonstrating the effect of vaccination; 3) of a signature at 14 days post-vaccination correlating with protection conferred by the vaccine. This 393-gene signature was identified using high resolution Gene Set Enrichment Analysis (GSEA) and consisted in transcripts belonging to the immunoproteasome pathway associated with the processing of major histocompatibility complex class I peptides.
In our re-analysis we first assessed changes in transcript abundance at the modular level. The percentage of responsive transcripts constitutive of a given module was determined for each individual at days 1, 3 and 14 following administration of the third vaccine dose in comparison to the levels obtained in samples collected just prior to that injection (see Methods for details). Hierarchical clustering was then performed at each time point to group modules (rows) and subjects (columns) based on patterns of changes in blood transcript abundance represented by the percent module response values (day 1, Figure 1A). Modules were filtered to only retain those with changes >15% in at least one subject. This analysis is unsupervised since it does not take knowledge of outcome of the infectious challenge into account. We observed nonetheless that samples tended to segregate based on whether or not the vaccine conferred protection (Figure 1A). Three modules associated with induction by interferon appeared to be the main elements driving the clustering of study subjects, with higher abundance levels being observed in subjects protected from subsequent infectious challenge. We demonstrated in our previous work that those three interferon modules represent distinct signatures that can be used for stratification of subjects with systemic lupus erythematosus21. Thus, we used in turn the same M1.2, M3.4 and M5.12 modules to stratify malaria vaccine recipients. Hierarchical clustering using only this subset of modules contributed to further separation of subjects based on the outcome of the infectious challenge (Figure 1B). The difference in % module responsiveness between protected and non-protected subjects was also statistically significant for M1.2 (p=0.0094, Mann Whitney test) (Figure 1C). M3.4 tended to be elevated compared to pre-vaccination baseline in both protected and non-protected individuals but was not different between those two groups. Abundance of M5.12 transcripts did not change following vaccination.
We next used a similar approach to classify subjects at days 3 and 14 post-third vaccination. Subjects once again segregated based on whether or not protection is conferred by the vaccine (Figures 2A & 2B). Notably, however, at these time points the signature showed a decrease in levels of transcript abundance in comparison to baseline pre-vaccine samples in subjects that were not protected. Thus conversely with the signature described at day 1, signatures at days 3 and 14 correlated with lack of protection by the vaccine. Differences between protected and non-protected groups where highly significant for M1.2 (Figure 2C , day 3 p<0.0001, day 14 p<0.0001, Mann Whitney test). M3.4 and M5.12 did not show significant differences between those groups. Notably, we found that the genes constitutive of M1.2 do not overlap with the day 14 immunoproteasome signature described by Vahey et al. Taken together results of our reanalysis of the Vahey dataset using a modular repertoire framework led to an original finding, by demonstrating the association between diverging day 1 and days 3 and 14 interferon signatures and protection conferred by the RTS,S vaccine.
We have shown in an earlier work that the three interferon modules that were described above tend to become elevated sequentially in patients with systemic lupus and may be associated with differential induction of type I and type II interferon in this disease21. Furthermore lupus disease severity was found to correlate significantly with M5.12 levels. We have also shown that an interferon response dominated by M1.2 and M3.4 was transiently increased 1 day following vaccination with the trivalent influenza virus12. O’Gorman et al. recently demonstrated that this transient interferon response is mediated by flu antigen-specific IgG immune complexes rather than engagement of pathogen-associated molecular pattern receptors22. The day 1 interferon response observed in the context of malaria vaccination could similarly be the result of engagement of CSP-specific IgG immune complexes since it occurs following administration of the third dose of RTS,S, at a time when a pre-existing humoral response would have been elicited by the first two doses.
But most peculiar is the fact that this increased modular interferon response in protected individuals at day 1 is followed by a persistent decrease in abundance of M1.2 transcripts below the pre-vaccination baseline in individuals that were not protected by the RTS,S vaccine. Indeed, in over 10 years of investigating blood transcriptome responses in a wide range of clinical and experimental settings the authors have not encountered a single instance of such a sustained and uniform decrease in abundance of interferon-inducible transcript. What is especially striking is the clear cut association between lack of protection conferred by RTS,S with the decrease in abundance of M1.2 transcripts seen in Figure 2C. This implies that the immunological mechanism underlying this suppressed interferon signature may be key to overcoming current limitations of sub-unit malaria vaccination.
Here we putatively attribute this decrease in abundance of interferon-inducible transcripts and subsequent lack of protection to the elicitation by the vaccine of an antigen-specific IgE response. This assertion is based on an array of converging evidence, as outlined below:
Engagement of the high affinity IgE receptor, FCER1, mediates decreased responsiveness to interferon-inducing stimuli. Gill et al. have shown that constitutively plasmacytoid dendritic cells (pDCs) isolated from patients with allergic asthma produce reduced levels of interferon alpha in response to the influenza virus in vitro when compared with pDCs isolated from non-asthmatic controls23. They also demonstrated that production of interferon alpha by pDC stimulated in vitro with the virus is significantly decreased upon cross-linking of the FCER1 receptor23. Similar findings have been reported more recently in PBMCs exposed to Human Rhinovirus (HRV)24. This is to our knowledge the only immune-mediated mechanisms of suppression of interferon responses that may explain the decrease in M1.2 observed following RTS,S vaccination. Thus we hypothesize that the suppression by RTS,S of levels of interferon inducible transcripts results from formation of IgE-CSP immune complexes, with anti-CSP IgE being elicited in earlier rounds of vaccination (Figure 3). IgE-antigen immune complexes would cause cross-linking and downstream signaling through the FCER1 that is expressed at the surface of leukocytes of the myeloid lineage. While IgG levels have recently been correlated with protection conferred by RTS,S25, to our knowledge the elicitation of IgE responses by this vaccine has thus far not been reported. Furthermore, our hypothesis is supported by evidence independently linking IgE responses, and specifically engagement of the FCER1, to susceptibility to malaria. Perlmann et al. identified IgE as a pathogenic factor in malaria, with immune complexes contributing to excess TNF induction in peripheral blood mononuclear cells in vitro26. Furthermore, mice deficient for the high affinity IgE receptor showed increased resistance to malaria infection, specifically implicating FCER1 expressing neutrophils as pathogenic mediators27. A more recent study has established a link between asthma and atopic dermatitis and delayed development of clinical immunity to P. falciparum28. Notably, in addition to shifting cytokine balance by promoting IL10 and TNF production, engagement of high affinity IgE receptors has been reported to critically impair phagocytic function of monocytes, a mechanism that is essential for the control of malaria infection29.
Gaining an understanding of immunological mechanisms that confer protection via immunization with the RTS,S malaria vaccine, or conversely that prevent it, can help address decisively the global health challenges caused by malaria infection. In this report we identify a candidate blood transcriptional signature correlating with protection following subsequent infectious challenge. Furthermore we establish a potential link between the peculiar decrease in abundance of interferon-inducible transcripts observed at days 3 and 14 following administration of the third dose of the vaccine and the possible elicitation of an IgE response in a subset of individuals that subsequently fail to be protected by vaccination. The validity of the model that we are proposing here can easily be tested by groups having ready access to samples obtained from subjects enrolled in the RTS,S vaccine trials. If this model holds true it would also open the possibility through the choice of appropriate antigens or adjuvants, or other immune modulating agents, to design strategies aiming at preventing, suppressing or skewing the development of IgE responses and thus confer high rates of protection against malaria infection through prophylactic immunization.
DR: data analysis & interpretation, manuscript preparation; SP: software & database development; DC: data analysis & interpretation; manuscript preparation.
This work was made possible through funding support from NIH (U01AI082110, U19AI089987, U19AI08998 and U19AI057234) to DC and SP, and the Qatar Foundation to DR and DC.
I confirm that the funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Views | Downloads | |
---|---|---|
F1000Research | - | - |
PubMed Central
Data from PMC are received and updated monthly.
|
- | - |
References
1. Kester KE, Cummings JF, Ofori-Anyinam O, Ockenhouse CF, et al.: Randomized, double-blind, phase 2a trial of falciparum malaria vaccines RTS,S/AS01B and RTS,S/AS02A in malaria-naive adults: safety, efficacy, and immunologic associates of protection.J Infect Dis. 2009; 200 (3): 337-46 PubMed Abstract | Publisher Full TextCompeting Interests: No competing interests were disclosed.
Competing Interests: No competing interests were disclosed.
Alongside their report, reviewers assign a status to the article:
Invited Reviewers | ||
---|---|---|
1 | 2 | |
Version 2 (revision) 18 Jul 17 |
read | |
Version 1 29 Sep 15 |
read | read |
Provide sufficient details of any financial or non-financial competing interests to enable users to assess whether your comments might lead a reasonable person to question your impartiality. Consider the following examples, but note that this is not an exhaustive list:
Sign up for content alerts and receive a weekly or monthly email with all newly published articles
Already registered? Sign in
The email address should be the one you originally registered with F1000.
You registered with F1000 via Google, so we cannot reset your password.
To sign in, please click here.
If you still need help with your Google account password, please click here.
You registered with F1000 via Facebook, so we cannot reset your password.
To sign in, please click here.
If you still need help with your Facebook account password, please click here.
If your email address is registered with us, we will email you instructions to reset your password.
If you think you should have received this email but it has not arrived, please check your spam filters and/or contact for further assistance.
Comments on this article Comments (0)