The science and medicine of human immunology

Vaccines as probes for the human immune system

Vaccination represents the most effective way to prevent infectious diseases. But to immunologists, vaccines represent excellent probes of the human immune system. First, vaccines allow a synchronized perturbation of the immune system, such that the immune response can be assessed from the earliest minutes to several decades after vaccination (40). Although the immune system is also perturbed in infections, autoimmune diseases, and cancers, in such cases the precise moment of immune perturbation cannot be known. Second, vaccines represent a diverse array of microbial stimuli (e.g., live viruses, carbohydrate vaccines, vaccines containing adjuvants) that can be used to perturb the immune system (40). Third, vaccines are widely administered to diverse human populations (e.g., neonates versus elderly; undernourished versus obese). For decades, vaccine manufacturers have relied on a single measurement—typically the magnitude of the antigen-specific antibody response (e.g., neutralization antibody titers or binding antibody titers measured by enzyme-linked immunosorbent assay)—to assess the immune response stimulated by vaccination. However, such measurements fail to capture the global architecture of the immune response to vaccination and the multiple immunological mechanisms that can mediate protection against a given pathogen (41).

Recent advances in omics technologies have enabled scientists to probe the immune response to vaccination in a comprehensive way, by analyzing the cellular and molecular networks driving immune response to vaccination. These omics technologies can measure changes in cellular subsets, transcriptome, metabolome, or epigenome, even at the single-cell level. Coupling these with computational approaches developed to analyze and interpret such data has led to the generation of the new field of “systems vaccinology.” The aim of systems vaccinology is to comprehensively analyze the immune response to vaccination with a view to defining new mechanisms and correlates of protective immunity (41, 42).

Initial proof-of-concept studies demonstrating the use of systems biology approaches to identify molecular signatures of vaccination used the yellow fever live-attenuated vaccine YF-17D (43, 44). YF-17D is one of the most successful vaccines ever developed, having been administered to more than 600 million people worldwide with an efficacy of 99% (45); a single immunization results in robust antigen-specific cytotoxic T cell responses and neutralizing antibody responses that can persist for several decades (42). Transcriptomic analysis of peripheral blood mononuclear cells (PBMCs) isolated 3 to 7 days after vaccination of healthy young adults with YF-17D revealed a gene expression profile pattern consisting of genes encoding proteins involved in antiviral sensing and viral immunity, including the type I interferon (IFN) pathway (43, 44). Computational analysis and machine learning approaches revealed signatures of early gene expression, which correlated with and predicted the magnitude of the later antigen-specific CD8⁺ T cell and neutralizing antibody responses, in an independent study (43).

Subsequently, several groups applied systems-based analysis to study immune responses to vaccination with seasonal influenza (46–50), meningococcal (51), shingles (52), malaria (53, 54), smallpox (55), Ebola (56) vaccines, as well as a candidate vaccine against HIV (57). In addition, several groups analyzed the transcriptional profile at “baseline” (i.e., prior to vaccination) to identify signatures of vaccine responsiveness prior to vaccination (49, 58–61). Furthermore, although until recently most studies have relied on transcriptional profiling to define signatures of vaccine immunity, recent studies have performed integrated analysis of orthogonal datasets to generate an integrated model of vaccine immunity. To understand the mechanisms of immunity induced by the live attenuated shingles vaccine, we constructed a multiscale, multifactorial response network (MMRN) of immunity induced by vaccination in healthy young and older adults. The MMRN revealed striking associations between orthogonal datasets, such as transcriptomic and metabolomics signatures, cell populations, and cytokine levels, and identified immune and metabolic correlates of vaccine immunity (52).

Defining correlates of vaccine efficacy (i.e., the capacity of vaccination to protect from infection) is more challenging. One approach to defining correlates of efficacy is the controlled human infection model (such as those that involve controlled challenge of humans with pathogenic strains of typhoid, influenza, or malaria), in which subjects can be vaccinated and subsequently challenged with the pathogen to define the mechanisms and correlates of vaccine immunity (62). Such studies have already been done in the context of the RTS,S malaria vaccine and have yielded important insights about the mechanisms of protection (53, 54). In particular, the results from these studies reveal that protection from malaria infection can be mediated by multiple immunological mechanisms involving the magnitude of serum circumsporozoite-specific antibody titers as well as the magnitude of antigen-specific CD4⁺ T cells with the capacity to produce polyfunctional cytokines (53, 54). In the case of phase 2b or phase 3 efficacy trials that involve several thousand subjects, the primary endpoint is efficacy. Typically, the trial design and sample collection schedules do not provide opportunities for retrospective analysis of immune responses. This issue has been brought to the fore recently, as the world races to develop a vaccine against COVID-19. At present, more than 160 vaccine candidates are being developed across the world, and systems vaccinology approaches may be useful to help identify early predictors of efficacy and accelerate the COVID-19 vaccine testing pipeline. Therefore, future efficacy trials should be leveraged to include sample collection schedules that facilitate systems-based immune profiling, with a view to defining mechanisms and correlates of efficacy.

An important question is the extent to which these high-throughput data can yield new insights about the immune system. Several mechanistic insights are emerging from such studies. For example, among the signatures that predicted CD8⁺ T cell responses to YF-17D was general control nonderepressible 2 kinase (GCN2; also called eukaryotic translation initiation factor 2α kinase 4), a sensor of amino acid starvation and an orchestrator of the integrated stress response (43). Studies with knockout mice showed a key role for YF-17D– induced GCN2 activation in programming DCs to initiate autophagy and enhance antigen presentation to CD4⁺ and CD8⁺ T cells (63), as well as in the control of intestinal inflammation (64). These results revealed an unappreciated link between virus-induced integrated stress response in DCs and the adaptive immune response.

Another insight that emerged from the influenza vaccine study was the impact of the gut microbiota on vaccine-induced immunity (65). Transcriptional profiling of immune responses to vaccination with the seasonal influenza vaccine revealed a striking correlation for the expression of the gene encoding TLR5—a receptor for bacterial flagellin—3 to 7 days after vaccination and the ensuing antibody response a month later (47). This prompted mechanistic investigation using mice that were genetically deficient in TLR5. Such studies revealed that TLR5-deficient mice were impaired in the vaccine-specific antibody response induced by vaccination with the seasonal influenza vaccine (66). This effect was shown to depend on sensing of bacterial flagellin by TLR5; consequently, mice administered antibiotics to deplete their microbiota or mice devoid of microbiota from birth (so-called “germ-free mice”) were impaired in their capacity to launch antibody responses to seasonal influenza vaccination (66). This experiment in mice raised the key question regarding whether the microbiota controlled immune responses after vaccination in humans. A recent clinical trial assessed the impact of administering broad-spectrum antibiotics to healthy adults before and after seasonal influenza vaccination; despite a reduction in gut bacterial load by four orders of magnitude and a long-lasting restriction in gut bacterial diversity, antibody responses were unaffected (67). However, in a second trial of subjects with low preexisting antibody titers, there was significant impairment in H1N1-specific neutralization and binding immunoglobulin G1 (IgG1) and IgA responses (67). In addition to these effects on the adaptive immune response, antibiotic administration also enhanced inflammatory signatures (including AP-1/NR4A expression), observed previously in the elderly (49), as well as perturbation of the serum metabolic profile, with a reduction in serum secondary bile acids by three orders of magnitude. The change in serum bile acids was highly correlated with AP-1/NR4A signaling and inflammasome activation. Integrative analysis of multi-omics datasets revealed significant associations between bacterial species and metabolic phenotypes, highlighting a key role for the microbiome in modulating human immunity (67). These studies show that systems approaches can help to identify early “signatures” that could predict the later immunogenicity or efficacy of a vaccine and yield new mechanistic insights.

Probing the human immune system in infections

Despite the invention of vaccines and antibiotics, infectious diseases continue to exert a major toll on human health, as evidenced by the recent COVID-19 pandemic (68). In 2016, lower respiratory infections, diarrheal diseases, tuberculosis (TB), and HIV/AIDS accounted for nearly 7 million deaths per year (69), and these diseases alone were the leading cause of disability-adjusted life years (DALY), accounting for 12.1% of all DALYs (69). Animal models have been developed for several infectious diseases and have yielded many valuable insights about mechanisms of immunity, yet differences in pathogenesis between these models and the human model have posed challenges.

In the past decade or so, there has been a growing effort to study immunity to infections in humans. An obvious limitation here, unlike the case with vaccination, is that the precise moment of natural infection cannot be known, thus posing an obstacle to analyzing immune responses during the earliest stages of infection. Nonetheless, there are some notable examples of addressing this issue, such as the FRESH study (Females Rising through Education, Support, and Health), which is a longitudinal study to identify and analyze participants immediately after they are infected with HIV (70). The FRESH program is located in the Umlazi township of South Africa, where the HIV prevalence rate in young women is as high as 66% at age 23. Investigators are collecting blood samples from 300 noninfected women aged 18 to 23, with a view to studying the earliest events that occur after natural infection with HIV (70). This FRESH study is providing key mechanistic insights into the biological factors that predispose to HIV infection. For example, the composition and diversity of the cervicovaginal microbiome were found to have a noticeable effect on the local host inflammatory response and the risk of HIV acquisition (71). In addition, there was profound CD8⁺ T cell activation and proliferation (>70% of CD8⁺ T cells in some individuals, the majority being HIV-specific by tetramer staining) in untreated acute infection. These cells rapidly became proapoptotic and were defective at producing antiviral cytokines such as IFN-γ, which suggested that HIV infection caused abnormal activation of CD8⁺ T cells (72), different from that observed with live viral vaccines such as YF-17D or smallpox (73).

Another area that has emerged in recent years is the profiling of transcriptional gene expression signatures in the blood to enhance our understanding of the host immune response to infection (74–84). Mycobacterium tuberculosis has been a scourge of humanity since ancient times, having survived more than 70,000 years, and currently one-third of the world’s population are carriers of the latent pathogen and are at risk for developing active disease (85). One of its mysteries is that an estimated 10% of infected adults succumb to full-blown TB, but most are able to control the disease throughout their lives (85). Blood transcriptional analysis of tuberculosis has revealed IFN-inducible genes whose expression diminished after successful treatment (76). Coupled with sensitive radiography techniques, blood transcriptomics have revealed heterogeneity in patients with active tuberculosis and asymptomatic people with latent tuberculosis, suggesting a continuum of infection and immune states (76). A recent study using high-dimensional mass cytometry demonstrated that latent TB is associated with enhanced cytotoxic responses and natural killer (NK) cells (86). Finally, a meta-analysis of 24 datasets containing 3083 transcriptome profiles from whole blood or PBMC samples of healthy controls or patients with active or latent TB from 14 countries yielded a three-gene set in whole blood that was robustly diagnostic for active tuberculosis, suggesting the clinical utility of such a signature (87, 88).

An especially fruitful area of research has been the analysis of immune responses in HIV-infected individuals, which has provided new insights about antigen design strategies for vaccination. A small proportion of HIV-infected individuals develop broadly cross-reactive neutralizing antibody responses that can neutralize a broad array of HIV viruses. Monoclonal antibodies have been isolated from such subjects and have been used to define critical antigenic epitopes that represent the targets of the broadly cross-reactive neutralizing antibodies (89–91).

Probing the human immune system in autoimmunity

Autoimmune diseases represent a diverse family of some 80 distinct diseases, such as rheumatoid arthritis, type I diabetes, and systemic lupus erythematosus (SLE), which afflict 5 to 9% of the global population (92). In healthy individuals, the immune system is considered to be “tolerant” of proteins present within the host’s own body, so-called “self” proteins. However, in humans who develop autoimmune diseases, there is a complex and dynamic interplay between host genes and the environment that results in the loss of immune tolerance to “self antigens,” leading to the development of autoreactive T and B cells that attack the body’s own tissues (93). Such autoreactivity can be organ-specific (e.g., rheumatoid arthritis) or systemic (e.g., SLE).

Preclinical animal models of human autoimmune diseases have provided important clues to these diseases in the form of possible mechanisms and etiology. Studies using mouse models of spontaneous autoimmune diseases [e.g., the experimental autoimmune encephalomyelitis (EAE) model for multiple sclerosis (MS); NOD mice for diabetes; the MRL/lpr model for lupus] have provided powerful tools to assess mechanisms and test novel therapeutic concepts (94, 95). Furthermore, mice genetically deficient in particular genes have established critical regulators of autoimmune processes, and the use of engineered models of autoimmune disease, such as mice transgenic for T and B cell receptors specific for autoantigens, has yielded a wealth of mechanistic insight (94, 95). However, the failure to translate promising therapies from these animal models to humans has led to a reassessment of the extent to which such models reflect the complexity and heterogeneity of the clinical presentation of the human disease. For example, of the ~200 treatments that prevent or delay the development of type 1 diabetes in the NOD mouse model, only a few have been shown to have clinical promise, and in only a fraction of patients (94–96). Some of the failures in translation were likely due to differences in the experimental design in the preclinical versus clinical studies, but species differences in the immune systems and human variations have likely been a major impediment.

The discovery of TNF-α as a central regulator in rheumatoid arthritis and the development of therapeutic anti-TNF-α antibodies have established a new paradigm for the treatment of autoimmune diseases (97). Ironically, it was an experiment with human tissues—the evaluation of cytokine expression in human rheumatoid synovium—that represented the key experiment in defining the cytokines that were overexpressed in the disease site (98). Many cytokines were overexpressed, so which of those represented the best therapeutic target was unclear. This was analyzed by blocking various cytokines in rheumatoid synovial cultures, which revealed that blockade of TNF-α reduced interleukin-1 (IL-1) synthesis as well as all other pro-inflammatory cytokines found in joints. This was the first demonstration that TNF-α could be an important therapeutic target (98). Concurrent experiments in mice supported the concept of a TNF-α-dependent cytokine cascade (99). Subsequent experiments in animal models demonstrated that administration of anti-TNF-α monoclonal antibody ameliorated both inflammation and joint damage in the collagen type II model of arthritis (100). This provided the rationale for anti-TNF therapy in rheumatoid arthritis, which has heralded a revolution in the introduction of protein therapeutics, termed “biologics.”

Genome-wide association studies (GWASs) have been performed over the past decade, with the goal of analyzing single-nucleotide polymorphisms (SNPs) in genes associated with a disease in large patient populations, and have revealed the polygenetic basis of many autoimmune diseases (101). Originally it was hoped that genetic studies of autoimmune diseases would discover one or a few defective genes that could explain the pathology and be useful diagnostically. This launched a number of GWASs, but unfortunately these efforts have shown that many loci contribute very small effects. However, some alleles. such as the human leukocyte antigen (HLA) alleles, do show statistically significant associations. Because different HLA alleles bind to distinct antigenic peptides, and because these HLA-peptide complexes are specific targets for T cell recognition, this points to a potential role for T cells in initiating and maintaining the disease (102).

Transcriptomics of patient PBMCs and whole blood have revealed alterations in the type I IFN network in SLE, Sjögren’s syndrome, inflammatory myopathies, and a subset of systemic sclerosis patients. SLE is an autoimmune disease characterized by loss of tolerance to nucleic acids, including double-stranded DNA and ribonucleoproteins, with diverse clinical manifestation. Recently, Pascual and colleagues analyzed transcriptional changes in the blood of 158 pediatric lupus patients (103). Using computational approaches, they identified transcriptional signatures of the type I IFN and plasmablast responses as the most robust biomarker of disease activity. They detected increased enrichment of neutrophil transcripts during progression to active nephritis. Strikingly, they were able to stratify the patients into seven groups according to distinct molecular correlates of disease activity. This study reveals the molecular heterogeneity underlying disease manifestation in SLE and suggests that future trials should implement “precision medicine” therapies that target the specific molecular anomalies in genetically and clinically complex autoimmune diseases.

Such approaches have also revealed a key role for IL-1β in systemic-onset juvenile idiopathic arthritis and have led to the development of anti-IL-1-based therapeutics (104). In the case of MS, IFN-β therapy has been used as a drug therapy, but despite its clinical use, the treatment modality involved a protracted regimen of uncertain clinical benefit (105). Transcriptomics has revealed transcriptional responses to IFN-β1α and has enabled the stratification of patients according to therapeutic response (106). Furthermore, systems approaches have been used to probe the immune response in patients with celiac disease, an intestinal autoimmune disease. Administration of gluten to patients with celiac disease not only induced the expansion and mobilization into the circulatory system of gliadin-specific CD4 T cells 6 days after challenge, but also induced CD8 T cells and γδ T cells with oligoclonal T cell receptors with similar kinetics (107). This same pattern of co-mobilized CD4⁺, CD8⁺, and γδ T cells was seen in the blood and central nervous system tissue of mice induced to develop EAE, which suggests that this may be a general feature of autoimmune diseases. Surprisingly, upon analysis of the specificities and functional properties of these different T cells, the CD4⁺ and γδ T cells were found to be pathogenic, whereas the CD8⁺ cells were found to be suppressive and able to kill the CD4⁺ cells in vitro (108). This indicates a complex dynamic in autoimmunity that will be fascinating to investigate further.

Clearly, analysis of the human immune system in the context of autoimmune diseases is yielding novel insights about the mechanisms and potential targets for therapy. Future studies could be aimed at using systems-based approaches to define the immune mechanisms and gene-environment interactions that lead to autoimmunity. However, unlike with vaccination, it has been challenging to assess the earliest stages of the response, and thus the etiology of the disease has been difficult to ascertain. One solution to this problem may be to establish human cohort studies such as the TEDDY study (The Environmental Determinants of Diabetes in the Young), a multi-center prospective cohort study that was designed to follow children with and without a family history of type 1 diabetes to understand the environmental factors that contribute to the disease (109, 110). In this study, 424,790 newborn children younger than 4 months were screened for high-risk HLA alleles, of whom 21,589 had the qualifying haplotypes and 8676 were enrolled. Of this group, 672 developed persistent antibodies to insulin, glutamic acid decarboxylase, or islet antigen 2, and these subjects are being followed for 15 years. It is predicted that 390 children will develop diabetes. Blood samples are being collected and banked from the day of birth and every 3 months, and stool samples collected (for microbiome analysis) quarterly, throughout the 15-year period (110). Clinical metadata (infections, medication, immunizations), exposure to dietary and other environmental factors, negative life events, family history, tap water, and measurements of psychological stress will also be collected. The major goals of the TEDDY study are to discover genes associated with the development of autoantibodies and type 1 diabetes, to identify environmental exposures that modify the risk of autoimmunity, and to use omics technologies to define signatures of the transcriptome, metabolome, epigenome, or microbiome that correlate with and predict autoantibody formation and disease progression (110). Such system analysis will not only yield novel biomarkers but also reveal mechanistic insights about the sequalae of immunological events, starting at the very earliest stages, that culminate in disease. Studies such as TEDDY also provide unique opportunities for discovering novel autoantigens—for example, by analyzing the B cell repertoire and isolating monoclonal antibodies from plasmablasts, or analyzing the T cell receptor repertoire using the recently discovered GLIPH algorithm (grouping of lymphocyte interactions by paratope hotspots) (34).

Probing the human immune system in cancer

The development of checkpoint blockade inhibitors has revolutionized cancer immunotherapy (28) and has led to durable survival outcomes in some patients with metastatic disease (111). Combination therapy of previously untreated advanced melanoma patients with nivolumab and ipilimumab (monoclonal antibodies that respectively target the PD-1 and CTLA-4 receptors on T cells) resulted in significantly greater objective-response rates and progression-free survival relative to ipilimumab monotherapy alone (112, 113). The development of these immune therapy regimens occurred as a result of fundamental discoveries using the mouse model (114, 115). However, a major issue with checkpoint inhibitors is that only a fraction of cancer patients experience durable clinical benefit (28). Thus, the identification of molecular signatures that can be used to predict nonresponsiveness to checkpoint inhibitors would be valuable in identifying patients who would be viable candidates for treatment. In this context, tissue-based immune monitoring is providing important insights into the cellular and molecular mechanisms of immune checkpoint therapy. For example, in a trial of ipilimumab, analysis of the RNA transcriptome in pre- and post-treatment tumor tissues revealed major transcriptional changes related to T cell signaling and activation after CTLA-4 blockade. In particular, ICOS⁺ T cells were increased in tumor tissues from patients with ipilimumab, and the increase in ICOS⁺ T cells was accompanied by similar increases in the blood (116). To test the hypothesis that ICOS⁺ CD4 T cells may play a role in the therapeutic effect of CTLA-4 blockade, similar studies were conducted in mice; although anti-CTLA-4 treatment was shown to induce tumor rejection in 80 to 90% of mice, the efficacy was less than 50% in ICOS-deficient mice (116).

In addition to tissue-based immune monitoring, emerging studies demonstrate that analysis of peripheral blood can yield new insights into mechanisms and biomarkers of tumor immune therapy. A recent study used immune profiling of blood from patients with stage IV melanoma before and after treatment with anti-PD-1 antibody (pembro-lizumab); the clinical failure in many patients was not due to a lack of invigoration of exhausted T cells, but rather due to the high tumor burden prior to anti-PD-1 therapy (117). Thus, the magnitude of reinvigoration of circulating exhausted T cells determined in relation to pretreatment tumor burden correlated with clinical response.

These studies highlight the power of human immune profiling technologies to identify signatures and mechanisms of antitumor immunity. In this context, emerging evidence suggests that patients receiving immunotherapy can be stratified into responders versus nonresponders based on the composition of their gut microbiomes (118–120). Future work should thus be aimed at using multi-omics technologies to delineate the transcriptional, metabolomic, epigenetic, microbiome, and cellular networks that regulate antitumor immunity, with a view to obtaining molecular signatures of responsiveness to cancer immunotherapy, similar to what has been accomplished with analysis of vaccination-induced responses (52, 67).

Scientific challenges: The obstinately diverse species

One major scientific challenge is dealing with diversity in the human model. Peter Medawar once called us the obstinately diverse species (125), a reference to our diverse genetic makeup. However, profound demographic, socioeconomic, and environmental changes have resulted in even further diversification with respect to health and age distributions. For example, The State of Food Security and Nutrition in the World 2019 states that maternal and child undernutrition contributes to 45% of deaths in children under 5, yet overweight and obesity are increasing in almost all countries and contribute to 4 million deaths globally (126). Indeed, paradoxically many populations are afflicted by the dual burdens of undernutrition and overweight. In 2018, Africa and Asia bore the greatest share of all forms of malnutrition, accounting for 90% of all stunted and wasted children but also nearly 75% of all overweight children worldwide (126). Disparity in life expectancy is also wide; as of 2018, Monaco had an average life expectancy of 89.52 years, whereas Chad was lower at 49.81 years (127), only moderately better than the 35 years in 17th-century Britain. These wide disparities no doubt impinge on the physiological states of humans and on variations in their immunological states in particular. This in turn is likely to affect responses to vaccination or susceptibility to inflammatory diseases.

The immunological state of an individual can vary longitudinally with time as a result of environmental changes, aging, or developmental changes in early life. In this context, Olin et al. used mass cytometry to analyze the changes in blood leukocyte composition in 100 newborn children during the first 3 months of life. The blood leukocyte composition of preterm and term children differed at birth but converged on a shared trajectory (128). Alternatively, there can be variations in the immunological states in individuals within a given population (because of genetic, microbiome, and environmental differences) (129, 130), or there can be variations in the immunological states between different populations (Fig. 2). This diversity is, however, an excellent source of natural variation or perturbation that enables effective interrogation of the human immune system in a population to draw predictive insights, and is thus a major advantage of the human model (131). To address this issue, Carr et al. used systems approaches to establish the cellular profiles of 670 healthy individuals and demonstrated high interindividual variation but low longitudinal variation at the level of cellular subset composition. Influenza vaccination induced expansion of antigen-specific B cell clones, but the cellular subset structure was elastic and returned to the individual’s unique baseline state. The largest influence on immune variation between individuals was cohabitation, with 50% less variation between individuals who share an environment than between people in the wider population (130). Such variations can lead to drastic differences in the responsiveness of the immune systems, such as wide differences in the immunogenicity and efficacy of vaccines. Therefore, a major challenge for human immunology is to embrace this diversity by probing the immune system in diverse human populations, with a view to obtaining mechanistic insights about the factors that lead to this variation. To understand this diversity, it must be accurately captured and recorded. This means aiming to record and standardize as much clinical metadata as possible during human studies. This can be facilitated through close collaboration between immunologists/basic scientists and clinicians. As precision medicine and wearable health technology advance, more and more of these data will become available (132, 133).

The confounding effects of genes can be addressed to some degree by performing studies in monozygotic twins. Immune studies along these lines have shown some strong genetic influences in young children receiving vaccines, but these genetic dependencies dwindle with age. One study reported that 70% of variation had little or no genetic dependency, reinforcing the idea that this is an adaptive system, adapting to microbial exposure similar to the way the nervous system responds to sensory input (61). An additional way to analyze the effects of genes is to study immune responses in subjects with monogenic lesions that consist of highly penetrant genetic mutations that exhibit large phenotypic effects. Such human knockouts, such as individuals with a common stop codon polymorphism in the ligand-binding domain of TLR5 (TLR5^392STOP), are unable to mediate flagellin signaling and are susceptible to infection with a flagellated organism, Legionella pneumophila (134). Given the aforementioned studies demonstrating a role for suboptimal antibody response of TLR5 knockout mice to seasonal influenza vaccination (66), it would be of interest to use high-throughput systems approaches to profile immune responses in such human mutants. Indeed, associations of SNPs in genes involved in innate and adaptive immunity and with measles and rubella vaccine responses have been described (135).

In the case of microbial imprinting, one particularly striking example of microbial influence is the case of cytomegalovirus (CMV), a type of herpes virus that has representatives throughout the animal world. Monozygotic twins who were discordant for CMV had many alterations in their immune system; almost 60% of 200+ variables indicated that this virus has an unusual ability to change many aspects of the immune system (61). Other studies have shown that CMV⁺ young adults have better antibody responses to a flu vaccine than age-matched CMV comparators, and that even more profound effects can be seen in mice infected with mCMV prior to influenza infection, with substantially reduced flu titers, indicating a profound protective effect (136). This result suggests that CMV is largely beneficial to the immune defense, which may explain its ubiquity around the world and in many organisms.

Scientific challenges: From data to knowledge to understanding

Sydney Brenner has said that systems biology is “low input, high throughput, no output” biology (137). A major challenge in the field of systems immunology in humans, and indeed for the field of systems biology in general, is learning how to extract knowledge and understanding from the vast sea of data that will be generated using profiling studies. Identification of omics-based signatures that predict susceptibility to autoimmune disease or vaccine response does not provide any mechanistic insight per se. The concept of knowledge-based (“gnostic”) predictors uses groups of biologically related genes that are coordinately regulated in response to infection, autoimmune disease, or vaccination (138). Although such tools are of great value, it is experimental validation of the signatures that will ultimately yield new biological insight. Data generated in the human model will enable the formulation of novel hypotheses about mechanisms of immune regulation. Mechanistic validation of this hypothesis will require the use of animal models such as gene knockout mice. An example of this approach involves the TLR5-microbiome-influenza vaccination study discussed above (47, 66, 67). The first clue came from systems analysis of immune responses to influenza vaccination in humans, which identified TLR5 as an early correlate of the later antibody titers to vaccination (47), raising the hypothesis that TLR5 could be functionally relevant in modulating antibody responses to influenza vaccination. This was tested in mice, where flagellin in the gut microbiota was shown to signal through TLR5 and provide adjuvant signals to influenza vaccination (66). The relevance of the gut microbiota as an adjuvant to antibody responses to influenza vaccination was subsequently shown to be important in humans in a clinical trial (67). Therefore, this represents an iterative cycle of inquiry that started with correlative observations in a human study, which led to a mechanistic study in mice, which was then validated in a perturbation study in humans.

Such examples illustrate that animal models are key for validating hypotheses generated in human studies. However, this approach of functionally validating genes, one gene knockout mouse at a time, does not have the throughput to rapidly validate the large number of genes that are typically contained in signatures. Therefore, additional approaches for validating hypotheses generated from human studies such as human organoid cultures offer promise in this regard. This, coupled with recent development of CRISPR-based high-throughput screens in primary cells, offers a potential way to rapidly screen and validate immune signatures (139). In addition, recent advances in genetics and microbiome research are facilitating the design of better mouse models that may more closely mimic humans. For example, mice might better mimic humans if their genomes have been engineered to include modified single bases in coding and noncoding regions or even entire networks (140), or if they are infection-experienced (25, 26), transplanted with human microbiota (141, 142), or genetically outbred and collaboratively crossed (143). Such models would represent an invaluable tool for the mechanistic validation of results observed in humans.

Practical and cultural challenges

A particular challenge to human studies is access to human samples. Although most human studies to date have focused on analyzing blood samples, emerging studies are also analyzing immune responses in other tissues, including the draining lymph nodes (using fine needle aspirates) (144), bone marrow (to assess long-lived bone marrow plasma cells) (145), and even liver (using fine needle aspirate) (146). Thus, establishment of systems for sample banking and management could facilitate human immunology studies tremendously. Another practical challenge is standardization of methods and techniques to profile immune responses in well-characterized human cohorts. An excellent example of this is the NIH-funded Human Immunology Project Consortium (HIPC), a collaborative network of researchers established to characterize the human immune system before and after vaccination or infection (www.immuneprofiling.org/hipc/page/show). Through HIPC, well-characterized human cohorts are studied using omics technologies and multiple computational methods. Cross-center and cross-assay analyses enable rigorous standardization of assays and analytical methods.

In addition, there are numerous cultural challenges. The academic ideal has been that of the lone hero/heroine going into a cave and slaying some ferocious dragon of a problem. But science and medicine are actually quite collaborative and team-dependent, and there has been a cultural inertia in academia to reward (or even acknowledge) this. Yet the medical and science worlds are just too complex and not getting any simpler, and to this must be added the computational, statistical, and technical expertise of the team. Ironically, in biotech and pharmaceutical companies, it is a given that tackling important disease areas requires a team approach. Here, the academic approach seems hopelessly outdated and is especially so when faced with the range of skill sets needed to advance human immunology; no one group can do this alone. This also means ensuring that group efforts are rewarded with a fair distribution of the credit in terms of promotions and funding. However, the emphasis on team effort and technology should not detract from the enormous value of individual creativity. To a large extent, scientific discovery has always been, and will continue to be, driven by the creativity and conceptual ideas of individual scientists. Technology, while having a transformative impact on biology, will never be a substitute for biological insight. Therefore, academic institutions and grant-awarding bodies should help to build a culture that stimulates and rewards the creative impulses and aspirations of individual scientists while nurturing a spirit of team science and robust collaboration.

An important cultural challenge is the notion that systems-based approaches to profile human immune response are not hypothesis-driven. This criticism reflects a long-standing philosophical debate on the nature of the scientific method itself. It is unquestionable that hypothesis-driven deductive research has represented the bedrock of the modern scientific method, and Popper has even argued that the so called “hypothetico-deductive” method alone is sufficient for the progress of science (147). Even so, it is clear that observational science or exploratory research that is not necessarily aimed at testing a specific hypothesis has led to major scientific advances. An example of this concerns Darwin’s observations in his field notebooks, without any apparent focus, during his voyage on the HMS Beagle to the Galapagos, which subsequently led to the theory of evolution by natural selection. Another example involves Galileo’s discovery of the four large moons of Jupiter and the rings of Saturn when he peered through the telescope. A more recent example is the collection of data by the Hubble Telescope about the origins of the universe. In the biological sciences, the omics revolution has stimulated a whole new era of science in which the starting point is data collection. Systems-based approaches represent hypothesis-enabling research rather than hypothesis-driven research.