Single-cell RNA sequencing allows highly detailed profiling of cellular immune responses from limited-volume samples, advancing prospects of a new era of systems immunology. The power of single-cell RNA sequencing offers various opportunities to decipher the immune response to infectious diseases and vaccines. Here, we describe the potential uses of single-cell RNA sequencing methods in prophylactic vaccine development, concentrating on infectious diseases including COVID-19. Using examples from several diseases, we review how single-cell RNA sequencing has been used to evaluate the immunological response to different vaccine platforms and regimens. By highlighting published and unpublished single-cell RNA sequencing studies relevant to vaccinology, we discuss some general considerations how the field could be enriched with the widespread adoption of this technology.
Vaccines are one of the most effective public health interventions in history and have been extremely successful in preventing illness and death from many infections. Much of this success can be attributed to the discovery of disease-causing agents and/or by the discovery of how to cultivate these pathogens to allow large-scale production of attenuated vaccines. While it is clear that effective vaccines induce protective immunological memory, the precise mechanisms by which this manifests are often poorly understood. Moreover, there are many diseases against which we have not developed successful vaccines, often a result of not fully understanding the “ideal” immune response and/or how to induce this with vaccination. Currently used techniques, such as ELISAs, ELISpots, flow cytometry, and growth inhibition assays, broadly measure responses in the T cell or humoral compartments after vaccination, but cannot agnostically measure differences in response between single immune cells [
RNA sequencing quantitatively profiles the cellular transcriptome. Polyadenylated messenger RNA (mRNA) molecules are often the target as the polyA tail is a convenient handle to selectively target the protein-coding mRNA (as opposed to other RNA types). In bulk RNA-seq studies, many thousand cells may be pooled together, obscuring heterogeneity. scRNA-seq (in contrast to bulk) allows the dissection of previously unappreciated levels of heterogeneity. This is a major motivation for embarking in scRNA-seq studies [
The relative paucity of published reports of single-cell transcriptomic responses in the context of vaccination suggests that there remains much to be learned from scRNA-seq. As with all new techniques, there are difficulties in establishing robust, scalable, and cost-effective protocols for the generation and analysis of scRNA-seq data [
This review considers the applications of scRNA-seq in prophylactic vaccine development, with a focus on infectious diseases. We use examples from several diseases to demonstrate the flexibility of the technology. We explore published and unpublished literature to highlight existing applications of this technology and provide recommendations and predictions as to how vaccinology could be enriched with its widespread adoption. To illustrate the adaptability of scRNA-seq, we present the case study of COVID-19 vaccine development and discuss the contribution unbiased transcriptional profiling could make.
Our understanding of the mechanisms underlying immune responses in health and disease has important implications for vaccine design. Previously, targeted techniques have allowed us insights into specific parts of the immunological system during development, during infection, and after infection. scRNA-seq allows the immune system to be studied in an unbiased manner. Additionally, studying single cells allows quantitation of the heterogeneity in systems and to resolve time during dynamic processes. Studying the immune response to infection can provide a window to understanding the challenges that must be overcome by vaccination. This is particularly relevant in diseases such as influenza or malaria where natural infection does not engender complete protection [
The particular innate cell types and pathways that trigger an effective adaptive immune response have been the focus of recent work by Blecher-Gonen et al. The authors used scRNA-seq to characterise the initial 48 hours of the cellular response to several fluorescently labelled inactivated pathogens [
Long-lived plasma cells are crucial to maintaining high levels of antibodies long after infection and vaccination [
Peripherally circulating CD8+ T cells have been associated with immune control of HIV [
Memory CD4+ T cells are required for long-lived immunity and are induced by vaccination strategies, including against malaria and influenza [
Rato and colleagues used scRNA-seq to investigate CD4+ T cell heterogeneity prior to HIV infection [
scRNA-seq can be deployed to simultaneously interrogate both pathogen and host transcriptomes. Transcriptional profiling at high resolution has enabled an in-depth appreciation of the cellular diversity in biological organisms and the number of transcriptional states during infection. This can allow the interpretation of immune responses to intracellular pathogens at single-cell resolution, as bulk isolates are often heterogeneous [
Several groups have recently used dual scRNA-seq to profile virally infected cells and draw insights from transcriptome information [
Responses to infection can also be interrogated in a more regulated setting using controlled human infection models. These involve the direct inoculation of an infectious agent in order to evaluate the subsequent immune response and/or potential protective efficacy of interventions. Barton et al. have already discussed the use of transcriptomics in controlled human infection models [
In the context of malaria, Tran et al. set out to profile the differences in the bulk blood transcriptome of challenge-protected and challenge-nonprotected volunteers during and after malaria immunisation [
Mpina and colleagues assessed variations in NK, NKT, and MAIT cell populations using samples from a CHMI study of Tanzanian adults challenged with
It is clear that scRNA-seq could provide valuable insights for vaccine redesign and targeting in the context of controlled human infection models. Indeed, owing to the relatively common use of these models in malaria vaccine development, most of the published analyses relate to this pathogen. Investigation of other pathogens used in challenge models, such as influenza,
Generating, curating, and characterising single-cell datasets of well-known pathogen isolates at various stages of infection will allow exploration of pathogen diversity and plasticity, which will ultimately aid in the identification of vaccine targets. The transcriptional variation of several malaria parasite species has been interrogated using scRNA-seq [ International collaborations including the Human Cell Atlas [ Equally, new scRNA-seq reference maps following vaccination could provide vaccine-induced signatures that are known to correlate with long-lived protection from disease and/or infection. For example, one could envisage a viral vector response reference map, where the immunological responses to several virally vectored vaccines are profiled. More importantly, vaccination is almost exclusively performed in young children and early adolescence [
The evaluation of the immune response to vaccination in both the preclinical and clinical phases is central to the prediction of success in disease protection. Antibody titres are correlates of protection for many, if not most, vaccines and vaccine candidates [
Immunogenicity of vaccines is modulated by a number of factors including vaccine antigen, vaccine platform, and adjuvant. scRNA-seq allows the impacts of these to be investigated with a high degree of specificity. scRNA-seq can also be used to characterise the heterogeneity in response to different vaccine regimens. Sheerin et al. performed comparative transcriptomics of the response to the capsular group B meningococcal vaccine (4CMenB), administered concomitantly with other vaccines and on its own, and the response to its constituent antigens or several comparator antigens in mice ([
Summary of prophylactic vaccinology publications using scRNA-seq.
Reference | Cell type | Species | Vaccine pathogen | scRNA-seq method |
---|---|---|---|---|
Afik et al. 2017. | CD8+ T cells | Human | Yellow fever | Smart-seq |
Upadhyay et al. 2018. | Plasmablasts | Human; Rhesus | Influenza; SIV | Smart-seq |
Neu et al. 2019. | Plasmablasts | Human | Influenza | Smart-seq/Spec-Seq |
Cirelli et al. 2019. | B cells | Rhesus | HIV | Smart-seq |
Waickman et al. 2019. | CD8+ T cells | Human | Dengue | 10X |
Sheerin et al. 2019. | Neutrophils | Mouse | Neisseria meningitidis serogroup B | 10X |
Darrah et al. 2020. | T cells | Rhesus | Tuberculosis | Seq-Well |
Vaccine human challenge studies (VHCS), a subset of controlled human infection models discussed above, involve the direct evaluation of vaccine efficacy by administration of an infectious agent to human volunteers after vaccination. Since the inception of VHCS such as those by Theodore Woodward in the 1940s [
In addition to effects modulated by vaccine regimen, platform, and immunisation route, there is ample evidence to suggest that gene expression following immunisation can be affected by adjuvant selection. Transcriptomic evaluation of nonhuman primates (NHP) and human responses to vaccine adjuvants, with and without vaccination, has largely been restricted to bulk and/or microarray analyses [
For many pathogens, our understanding of protective epitopes is incomplete. In the study of B cells, peptide arrays and phage displays are methods that have been used for the discovery of linear epitopes [
scRNA-seq methods have the potential to improve antigen screening and selection, by providing a more accurate picture of the immune response generated by vaccines with different antigenic make-ups. The development of algorithms that reconstruct T cell receptor sequences from single-cell data allows parallel analysis of the T cell transcriptome and TCR clonotype in multimer sorted antigen specific cells [
Methods using the transcriptome to identify vaccine-responsive T cells without knowing their epitope specificity have also been developed, which allows a broad understanding of the T cell repertoire in response to vaccination. Fuchs et al. used scRNA-seq to identify a genetic signature of virus-responsive cells by performing scRNA-seq on dye-labelled antigen-specific cells [
Upon antigen stimulation, B and T cells proliferate and undergo clonal expansion; the BCR or TCR sequences are effectively a “clonal barcode.” This can provide information on antigen specificity and cell ancestry. A great strength of scRNA-seq is the ability to obtain unbiased transcriptome and V (D) J gene transcript usage information from the same cell. With the advent of new scRNA-seq workflows, a proliferation of bioinformatics tools to analyse these data has necessarily occurred. BALDR, an example of such a bioinformatic pipeline, is able to reconstruct the paired heavy and light chain immunoglobulin gene sequences from scRNA-seq data derived from Illumina short reads ([
In a similar fashion, TraCeR [
In the context of dengue virus (DENV) vaccination, approaches centred solely on B cell-mediated protection have limitations [
Using scRNA-seq data, it is possible to predict cell trajectories, that is, to computationally order cells along putative trajectories, by inferring how much progress an individual cell has made through a given process (such as cell differentiation). The above analyses by Waickman et al. could be extended by using pseudotime tools such as Monocle [
Pairing TCR sequence and transcriptome information allows the discovery and exploration of new cell populations. Afik and colleagues performed scRNA-seq on Yellow Fever Virus (YFV) vaccine-reactive and other CD8+ T cells ([
Recently, efforts to produce and characterise monoclonal antibodies (mAb) have made impressive progress. mAbs with broadly neutralising activity against specific antigens largely act through their Fab fragment specificities. Our comprehension of the way antibody specificities interact with B cell function has remained limited due to the intricacies of polyclonal antibody responses. Neu et al. developed the Spec-seq protocol to tackle this challenge ([ With improvements in droplet-based scRNA-seq methods, the scale (thousands of single cells) and order of events (BCR sequence and transcriptome information first, mAb generation thereafter) could be changed. Rather than characterising every mAb “blind” (i.e., without any prior information on its cell of origin), mAbs could be selected on the basis of transcriptome information and clonal family position in a hope to only generate and characterise high affinity/avidity candidates. Combining this with fluorescence-assisted cell sorting (FACS) to isolate vaccine antigen-specific B cells could provide a powerful new workflow to produce monoclonal antibodies against specific pathogens. An example of this type of workflow is put forward by Goldstein and colleagues [
Immunocompromised individuals are at risk of higher acquisition and complication rates of many vaccine-preventable infectious diseases such as seasonal influenza, respiratory syncytial virus (RSV), and bacterial pneumonia. In parallel, immunological responses to vaccination are often less efficient compared to healthy adults [
Optimising vaccine immunogenicity in immunosuppressed populations is paramount, but heterogeneous underlying mechanisms of immunosuppression make this challenging. Causes of altered immune states range from pathological conditions (including primary immune deficiencies and/or acquisition of chronic viral infection with HIV or cytomegalovirus (CMV)) to physiological states (including neonates, pregnant, and older persons) and iatrogenic immunosuppression following organ transplant or treatment for autoimmune conditions. With respect to pathological conditions, coinfections may further complicate vaccination, for example, HIV with hepatitis B or hepatitis C virus.
Study of the immune response at the cellular level in conditions of immunosuppression has demonstrated the nuances of vaccination responses. For example, the contribution of humoral and cellular responses to both influenza and RSV vaccination is altered in older compared to younger adults [
Cancer vaccines are different from those protecting against infectious diseases in many ways; most notably that they can be used in therapeutic and personalised capacities [
Single-cell transcriptomic profiling can be incorporated into neoepitope selection and vaccine manufacture workflows. Petti and colleagues performed matched whole-genome sequencing and droplet-based scRNA-seq on samples from patients with acute myeloid leukaemia [
Recognising that therapeutic responses are varied in pathologically identical tumours and even in genetically homogeneous cancer cells [
Single-cell sequencing must overcome a number of challenges prior to its wholesale adoption in the field of cancer vaccines. These hurdles are present in many scRNA-seq experiments but have specific consequences in cancer vaccine discovery. A major difficulty is “drop-out.” This happens when a transcript or an allele in a heterozygous mutation is not captured or amplified and can occur at 10–50% of mutation sites [
In December 2019, a cluster of patients with pneumonia of unknown cause was linked to a seafood wholesale market in Wuhan, China, later confirmed to be associated with infection with a new beta-coronavirus, now known as SARS-CoV-2 [ Understanding how the virus interacts with the host has been aided by the COVID-19 Cell Atlas ( A logical next step would be to generate a high-resolution multiomic cell atlas of the host immune response to COVID-19 infection in the periphery and in lung tissue [ Using scRNA-seq to understand humoral immunity and guide vaccine-mediated antibodies against SARS-CoV-2 has shown particular promise. Wen and colleagues profiled PBMC of convalescing COVID-19 patients [ Research on COVID-19 is evolving quickly; many studies have not yet been peer reviewed and there have been concerns about the robustness of the peer review process of studies that have [
The wholesale adoption of scRNA-seq in vaccinology is limited by hurdles relating to analysis, technical issues, and the experimental questions that can be asked with the technology (see Appendix
While there is increasing development of computational tools and reference databases [
There are several challenges related to the scRNA-seq technology itself. Every scRNA-seq protocol begins with the preparation of a single-cell suspension of the tissue of interest. When making inferences from scRNA-seq experiments, there are two inherent assumptions related to this step. These are that (i) the cellular composition of the suspension is a faithful representation of the original tissue and (ii) sample preparation results in insignificant (or no) transcriptional changes. Enzymatic, as well as mechanical, dissociation can result in biases of cellular representation as certain cells may be more sensitive to enzymes or dissociation [
By beginning reverse transcription using a poly (T)-oligonucleotide, the majority of scRNA-seq technologies use the mRNA polyA tail to synthesise the first strand of cDNA. Challenges with this approach include an inability to capture nonpolyadenylated microRNAs and regulatory RNAs [
Ultimately, most cell-cell and extracellular cell-pathogen interaction is protein-mediated. For scRNA-seq, inferences about cell-cell interactions occurring between receptor-ligand pairs can be made using repositories of ligands, receptors, and their interactions, such as CellPhoneDB v2.0 [
The spatial position of cells in tissues strongly influences function, yet there remains no truly single-cell, unbiased, spatial transcriptomics approach. Several approaches, however, are reaching cellular resolution (e.g., Slide-seq with 10
Profiling the immune response to both natural and artificial pathogen exposure by scRNA-seq has advanced our ability to identify favourable immunological profiles. The capability of scRNA-seq to concurrently examine the global gene expression, antigen-specificity, clonality, and individual copy number variants (CNVs) and infer the developmental trajectory of immune cells offers a powerful toolbox to appraise host responses to vaccine candidates. Certain areas have not yet been tackled by scRNA-seq, including critical confounders of immunogenicity such as coinfections and age, vaccine platforms, and adjuvants. In addition, we were unable to find any studies primarily using scRNA-seq to better understand adverse events following vaccination, or any other vaccine safety metrics, with the exception of one study discussed above [
scRNA-seq is now a widespread research tool; the number of diseases and areas of research in which it is being applied is growing. Improvements in the scale of adoption, robustness and ease of use of reagents, instrumentation, and computational tools mean that scRNA-seq will continue to be used more. The utility of this tool in vaccine design and development is contingent on the particular questions which are being asked (see Appendix
Beyond the anticipated improvements in system efficiencies and the increased availability of reagents, what is the best way that scRNA-seq can be applied systematically in vaccine design, development, and evaluation? This will depend on specific hypothesis-driven, experimental, and analytical considerations. In Box
(i) Cellular responses to various adjuvants and vaccines (ii) Transcriptional and antigen-specific responses to adjuvants and vaccines (iii) Site-specific immunity induced by vaccinations (for example, at mucosal surfaces following HIV vaccination, in the liver following liver-stage malaria vaccination, or in the lungs following tuberculosis vaccination [ (a) Spatial transcriptomics, soon to provide resolution at the single-cell level, could also be used (iv) Single-cell transcriptomic signatures associated with neutralising antibody responses (v) The accordance between protective transcriptional signatures in vaccine human challenge studies in nonendemic and endemic countries (i) Assess the need for scRNA-seq as the primary experimental technique and contemplate whether the question can be answered by established techniques (e.g., ELISAs, ELISpots, and flow cytometry) (ii) Consider using bulk RNA-seq as an adjunct to scRNA-seq (e.g., bulk RNA-seq on all samples, with scRNA-seq on a subset to allow computational deconvolution) (iii) Tailor the particular type of scRNA-seq to the experimental question (e.g., will alternative splicing be of interest? Use full-length transcript profiling if so) (iv) Longitudinal gene expression and TCR/BCR profiling to track antigen-specific clones (v) Define heterogeneity of cellular response in protected individuals (vi) Generate monoclonal antibodies from TCR/BCR sequences for rational vaccine redesign (vii) Use scRNA-seq as “backbone” to other technologies (e.g., G&T seq/CITE-seq) (viii) Combine scRNA-seq with flow cytometry/FACS and/or magnetically assisted cell separation to isolate rare or specific cell types [ (ix) Multiplex samples according to genotype to reduce sample preparation time, reagent and sequencing costs, and batch effects [ (x) Preserve leftover cells either by sorting into a plate or by preserving in fixative for later use (for example, in the reanalysis of an interesting sample) (i) Plan with, budget, and include bioinformaticians who are capable of working with scRNA-seq data from study conception onwards (ii) Use freely available packages that are regularly maintained. See Table 1 in the review by Zeng and Dai [ (iii) Upload data files to Gene Expression Omnibus [ (iv) Consider submission of data to other relevant databases such as Human Cell Atlas and/or TCGA (v) Consider analysing a similar published dataset with the experimental dataset to increase statistical power and/or to ensure a novel pipeline reproduces results in a previously published study (vi) Make freely available the code which relates to bespoke analyses
Research question considerations
Ample consideration of the particular research question to be answered is required for any experimental design, but in particular scRNA-seq experiments Potential questions that can be asked in scRNA-seq experiments
Does vaccination result in previously uncharacterised single-cell states? What is the temporal sequence of cellular processes taking place after vaccination? How does T cell and/or B cell clonal diversity change in response to vaccination? What are the transcriptional differences among vaccine reactive/antigen-specific cells? To what extent are vaccine-induced transcriptional changes reflected at the cellular (bulk vs. single-cell sequencing) and protein level (transcriptomics vs. proteomics)? What are the single-cell transcriptional differences in vaccine response between vaccines A and B? What is the single-cell transcriptional profile in peripheral lymphocytes (or organ) given a particular vaccine platform (e.g., virus-like particles), regardless of the antigen that is delivered? Other specific experimental considerations will influence experimental design
Breadth vs. depth
Are lowly expressed genes of particular interest? For transcripts that are lowly expressed (e.g., transcription factors), full length sequencing approaches may be better than 3 Are rare cell types to be profiled? Increasing cell number and maintaining read depth relatively low allows more power to detect rare cell populations (that may exist at <1% in frequency) [ Will comparisons be made between different conditions (e.g., prime alone vs. prime-boost)? What are the qualities and expression levels of the marker genes of cell types of interest? Transcriptional bursting can result in substantially different transcript quantities and apparent gene expression levels [ What is the overall budget of the project? Cost/cell profiled is an often-used metric for budgeting scRNA-seq experiments What facilities and expertise are available? Has technical and experimental advice been sought by nonconventional means (consider the active scRNA-seq community on Twitter (particularly #scRNAseq and #scQA), ResearchGate, medRxiv, bioRxiv, EMBI-EBI training ( Sample preparation and processing considerations
Processing samples up to the point where scRNA-seq can be performed is a process that can greatly affect the outcome of the experiment
Are there unchangeable constraints on sample collection/processing times? What effects, if any, will these have on the results? Do samples have to be processed immediately or is there a window where gene expression will not be affected, without specific preservation? Will the samples be preserved (e.g., flash frozen, fresh frozen, or formalin-fixation and paraffin-embedding) [ scRNA-seq usually requires mechanical or enzymatic dissociation of samples to produce a single-cell suspension. Certain factors will affect this process
How fragile/robust are the samples? How well is the tissue dissociated? Which single-cell isolation procedure will be used (e.g., microdissection, reverse emulsion droplets, FACS into plates, and/or nanowell isolation)? Are there preexisting protocols in the published literature that describe isolation techniques from the sample of interest (consider published, preprint, and commercial (e.g., 10x Genomics technical notes) literature)? Will the cells be preserved (e.g., cryopreservation using DMSO, methanol fixation, or storage in commercially available formulations to preserve cells and their RNA) [ Can a method be used that does not dissociate tissue (e.g., Slide-seq) as a baseline to check for dissociation effects? This will likely add extra cost and require additional technical expertise Once cells (or nuclei) are in a suspension, the starting point for RNA capture is achieved
Which is more apt and feasible to answer your specific research question, nuclei or cells isolated from the sample [ What are the characteristics of the cells of interest (e.g., size and adherability)? Volume can vary widely cell-to-cell, this affects the absolute number of transcripts and can be reflected in the detected number of genes per cell [ What quantity of input material (cells or nuclei) is there? Assessing RNA quality. Before embarking on costly library preparation for every sample, it is important to ensure that the RNA has not significantly degraded:
RNA quality is typically measured using the RNA Integrity Number (RIN) algorithm [ Is there enough of the sample to produce a RIN score test on each sample? If the site, conditions, or timing of sample collections is variable, it may lead to differences in RNA degradation between samples (and batch effects, discussed below). If a RIN score step can be built into each experiment before embarking on library preparation, then it can prevent spending money on low-quality samples Replicates, scale, sequencing, and batch considerations
Batch effects
Batch effects are random technical artefacts which occur during handling/processing. If batches correspond to different biological conditions, then it is largely impossible to determine what differences are biological vs. artefacts Avoid batch effects by
sorting cells from different biological conditions into different wells of the same plate using genetic variants to post hoc assign sequenced cells back to their genetically unique donor using the expression of an inserted genetic construct (not recommended) using barcoded antibodies (Cell Hashing) to label samples after dissociation, but before cell-capture step, to multiplex samples [ Batch effects may be corrected by a number of bioinformatics tools and/or packages [ Experiment scale
How many cells will be tested largely depends on the level of heterogeneity of the sample and on the number of available cells Plate-sorted single cells are limited in the amount that can be handled compared to microfluidic platforms, which enable studies with several thousands of cells (Figure Method of amplification
Either exponential PCR-based amplification or linear in vitro transcription (IVT) amplification is usually used. IVT incorporates less PCR bias and erroneous bias as it is based on an unamplified RNA template [ Transcript position
Some protocols provide full-length transcript data, whereas others amplify only the 3 Non-full-length transcript methods allow increased throughput of cells, while full-length transcripts are advantageous if splice variants are important, looking to detect genetic variants or when studying species that have poorly annotated genomes Sensitivity
This is the ability of an assay to capture an mRNA molecule from a single cell within the final library. Low sensitivity protocols have a disproportionate effect on weakly expressed genes (e.g., genes encoding cytokines) If weakly expressed genes are to be evaluated, consider higher sensitivity methods or consider “clean-up” procedures such as rRNA removal [ Ultimately, the particular protocol must be decided on an individual experiment basis. It is also important to note that while scRNA-seq methods have greater sensitivity than bulk RNA-seq methods, bulk methods have higher accuracy [
scRNA-seq technologies that have been critical to allowing increments in experiment scale. Achievements over the past three years have more or less continued this pace; for example, combinatorial fluidic preindexing has increased the throughput of droplet-based single-cell RNA sequencing up to 15-fold. Figure adapted from references [
We performed an initial scoping review of the literature using MEDLINE/PubMed to identify major themes present in the vaccinology literature. Thereafter, we performed predefined searches in the context of the vaccinology themes we identified: controlled human infection studies, correlates of protection, profiling the immune response to infection, cancer vaccines, understanding host-pathogen interactions, comparing vaccine regimens and responses, adjuvants, animals as models for human diseases and natural infections of livestock, antigen screening/selection, and effects of coinfection on vaccination. Predefined searches included using MeSH terms: “Sequence Analysis, RNA”, “Vaccinology”, “Vaccines.” In January and February 2020, we searched for published literature in MEDLINE/PubMed and Embase and for preprint/not yet peer-reviewed literature in bioRxiv, medRxiv, Wellcome Open Research and arXiv. Further searches were conducted using free text to expand our capture of scRNA-seq (e.g., (“scRNA-seq” OR “single cell sequencing” OR “single cell RNA sequencing”)), vaccine (e.g., (vaccin
After performing the searches, we refined the scope of our review to include only scRNA-seq studies—as opposed to those exclusively considering bulk RNA-seq—performed in humans or animal models of human disease published after 2015, unless they were considered sufficiently relevant to the narrative of our review.
The views expressed in this article are those of the authors and not necessarily those of the NHS, the NIHR, or the Department of Health.
EB is an NIHR Senior Investigator. The authors declare that there are no conflicts of interest regarding the publication of this paper.
The authors would like to thank Alexandra Spencer and Adrian Hill of the Jenner Institute for critical support and revision in the preparation of the manuscript. AN is supported by the Rhodes Trust, Green Templeton College, University of Oxford, and the Avant Foundation. TC received funding from the NIHR and a Wellcome Trust Training Fellowship for Clinicians (211042/Z/18/Z). CMN is a Wellcome Trust Sir Henry Wellcome Postdoctoral Fellow (209200/Z/17/Z). EB is supported by the NIHR Oxford Biomedical Research Centre.