These are the Genes You’re Looking For: Finding Host Resistance Genes

Abstract

Humanity’s ongoing struggle with new, reemerging, and endemic infectious diseases serves as a frequent reminder of the need to understand host-pathogen interactions. Recent advances in genomics have dramatically advanced our understanding of how genetics contributes to host resistance or susceptibility to bacterial infection. Here we discuss current trends in defining host-bacterial interactions at the genome-wide level, including screens that harness CRISPR/Cas9, natural genetic variation, proteomics, and transcriptomics. We report on the merits, limitations, and findings of these innovative screens and discuss their complementary nature. Finally, we speculate on future innovation as we continue to progress through the post-genomic era and towards deeper mechanistic insight and clinical applications.

An Expanding Toolbox for a Critical Discipline

The importance of microbiological research is constantly reaffirmed as drug resistance rejuvenates dangerous old foes, ancient plagues reemerge, and novel pathogens arise and drive world-wide pandemics. Fascination with the microscopic world stretches back to the mid-1600s when Hooke and van Leeuwenhoek first used microscopy to observe microbes. Later advances in germ theory by Pasteur, Henle, Koch, and others, solidified the importance of studying microbes, while the advent of molecular biology reshaped how microbes are studied. Now, advances in genomics are fueling significant microbiological breakthroughs. Genetic tools, like the mouse Collaborative Cross, are being applied to microbiology even as microbiology generates new genetic tools, like CRISPR-Cas9. This interplay is redefining the microbial research toolbox.

In this review, we will focus on how cutting-edge genetics and genomics screens are being applied to identify genetic determinants of host resistance to bacterial infection. We will discuss screens that harness CRISPR/Cas9, natural diversity, proteomics, and transcriptomics to understand host-bacterial interactions. Our goal is to demonstrate the power of these screening approaches and to encourage continued innovation in host-bacterial interactions research.

Harnessing CRISPR/Cas9 to Identify Host Resistance Genes

Genetic screens have fundamentally improved our understanding of biology, but until recently, these studies were limited in mammalian cell lines. Previous genetic screens in these cells were restricted to siRNA knockdown systems—which can have off-target effects, variable efficacy, batch effects, and limited throughput—or haploid cell screens, which are limited to a small number of cell lines [1]. While these studies were impactful, the sum of their limitations make it technically challenging and financially infeasible to identify genes with small effect sizes that contribute to polygenic and clinically relevant traits.

By overcoming several of these weaknesses, CRISPR knockout screens have recently become the dominant method to identify host genes and pathways involved in the host response to bacterial infection. CRISPR screens come in two forms, traditional arrayed screens or pooled selection experiments, and the vast majority of host-pathogen interaction CRISPR screens have used the latter. Pooled CRISPR selection screens first deliver a pool of single guide RNAs (sgRNA) into a population of cells expressing Cas9, then select based on a cellular phenotype, and finally compare each guide’s frequency in the selected and unselected samples using sequencing (Figure 1). By screening multiple guides within a pool and targeting multiple sgRNAs to a single gene, this pooled approach reduces noise from off-target effects, batch effects, or failed knockout compared to siRNA screens or arrayed CRISPR screens. Further, because CRISPR/Cas9 technology is rapidly being adapted to a number of primary and immortalized cell lines across a variety of species, this screening platform can be used to detect cell type- and cell line- dependent resistance or susceptibility factors. CRISPR screens can also be performed in vivo, as multiple studies have knocked out genes in murine immune progenitor cells, returned them to irradiated mice, and measured the impacts on immune cell development [2, 3]. Together, these strengths make CRISPR/Cas9 screens invaluable for understanding host immunity.

Figure 1: Interrogating the Host: Modern Screens to Identify Host Resistance Genes.

(Top) CRISPR/Cas9 Screens enable high throughput screening of genes by transfecting a pool of single-guide RNA (sgRNA) into cells and performing a selection for traits associated with host-bacterial interactions. In the provided example, different colors represent cells transfected with different sgRNA, and the green guide knocked out a gene required for pyroptosis. As such, the guide DNA is enriched following infection with pyroptosis-inducing bacteria and is identified for additional experimental follow-up work. (Bottom) Natural Genetic Diversity Screens enable researchers to identify genetic loci that confer resistance to bacterial infection. After identifying a diverse population and phenotyping individual responses to a bacterial pathogen, phenotypic diversity can be statistically associated to genotypic diversity. Following identification of genetic loci associated with phenotypic diversity (quantitative trait locus, QTL), follow-up work examines which polymorphisms and/or genes contribute to host resistance. Further, comparison and integration of individual studies into additional screens can provide a more complete picture of a QTL’s role in host biology. (Right) Dual Approach Screens involve simultaneous screening of host and bacterial genomes, transcriptomes, and proteomes to understand infection biology. These approaches essentially employ studying the joint host-bacterial interaction to learn about the pressures that the bacteria overcomes in the host and/or the systems it manipulates to promote virulence. These approaches can be critical hypothesis generating tools.

While genetic knockout systems are powerful tools, they do have limitations. For instance, genetic ablation can have unintended consequences on the genetic landscape that mask phenotypes [4, 5]. Early stop codons caused by CRISPR/Cas9 can upregulate associated genes through compensatory processes paired to nonsense-mediated mRNA decay [6]. Further, CRISPR/Cas9 screens cannot identify essential genes that moonlight in immunity. Notably some work has attempted to circumvent the latter limitation by enabling temporal control of the Cas9 machinery [7]. Despite these drawbacks, screens using and improving on CRISPR/Cas9 have probed bacterial sensing, phagocytic uptake, bacterial manipulation of the intracellular niche, and toxin susceptibility.

Several groups have harnessed CRISPR screens to examine how host immune cells respond to bacterial pathogen-associated molecular patterns (PAMPs). Parnas et al. performed a whole-genome screen (125,793 sgRNAs targeting 21,786 mouse genes) to untangle the response of myeloid cells to LPS stimulation [8]. The screen identified known factors in LPS signaling, including TLR4, MYD88, and ZFP36, as well as novel positive and negative regulators of TNF. Follow-up experiments revealed that oligosaccharyltransferase (OST) complex regulates TLR4 signaling. Interestingly, an independent CRISPR screen in a murine pro-B cell line also identified that the OST complex regulates TLR5, 7, and 9 signaling [9], suggesting that arginine glycosylation is a conserved regulator of TLR signaling. Notably, these techniques rediscovered genes with established involvement in sensing, raising confidence in the results of these screens.

Once a PAMP is sensed, the immune response must be regulated to avoid excessive inflammation and/or sepsis. Several groups sought to understand this regulation using CRISPR screening, including one that identified the complement peptidase Cpb1 as a positive regulator of Caspase-11-mediated pyroptosis in response to LPS, Salmonella, and Shigella in murine macrophages [10]. Their experimental validation revealed that CPB1 enables the C3/C3aR signaling pathway to amplify TLR4 and IFNAR signaling cascades, enhancing Caspase-11 activity and sepsis severity in mice and humans. An additional series of CRISPR screens revealed that autophagy genes modulate TNF signaling to counteract IFN-γ-mediated cell death and sepsis in murine models [11]. These studies demonstrate the power of CRISPR screens to identify genes involved in harmful immune responses in vitro, which could also serve in vivo as biomarkers or druggable targets for sepsis.

Beyond sensing and signaling, many groups have used CRISPR/Cas9 screens to examine how host genes facilitate intracellular bacterial lifestyles, particularly in phagocytic host cells. One recent screen identified host genes required for Salmonella enterica uptake into PMA differentiated human THP-1 macrophages [12]. This study complemented a phagocytosis screen that used magnetic sorting to identify genes required for uptake of beads or other substrates (red blood cells, myelin, zymosan) conjugated with ferrous nanoparticles by differentiated human U937 macrophages [13]. Both screens identified NHLRC2 as a driver of phagocytosis, with Yeung et al. demonstrating that the protein likely plays a role in both macrophage phagocytosis and differentiation. Finally, a third screen examined the later stages of phagosome development and found that the bicarbonate transporter SLC4A7 is required by phagocytes to acidify their vacuoles and kill bacterial pathogens [14]. Together these screens uncovered dozens of genes involved in phagocytosis; however, they also highlight an old screening maxim, “You get what you screen for.” Each group performed a screen for phagocytosis, but what they identified varied based on cell type, phagocytic substrate, timing, and other experimental parameters. Thus, considering these factors is critical when designing screens and interpreting results.

While the studies above focused on host genes required to resist bacterial infection, several screens also identified host proteins that are hijacked by bacteria to aid in pathogenesis. For instance, a herculean series of screens identified hundreds of host factors required for Legionella pneumophila to enter phagocytic cells, replicate, and eventually kill the host cell [15]. This study provides a comprehensive view of how previously identified host factors and newly identified genes work together to facilitate pathogenesis. Additional work has focused on host factors that enable bacteria-mediated cytotoxicity. For instance, one CRISPR screen that focused on Vibrio parahaemolyticus-induced cell death found that host cell surface sulfation allows the bacteria to adhere and kill cells using its type-III secretion system 1, while host cell surface fucosylation enables the type-III secretion system 2 translocon to enter the host membrane and facilitate cytotoxicity [16]. Similar work with enterohemorrhagic E. coli demonstrated that host sphingolipid biosynthesis is critical for both type-III secretion- and Shiga toxin-mediated cell death [17]. Finally, a CRISPR screen by Tao et al. identified certain family members of the Wnt receptor Frizzled (FZD1/2/7) as the predominant colonic receptors for the Clostridium difficile toxin TcdB [18]. These studies highlight CRISPR screen identification of disease risk loci, which are potential targets for host-directed therapeutics to reduce bacteria-induced pathology.

Together this work demonstrates the impact that CRISPR screens have already had on understanding host resistance to bacterial infection. However, this likely represents the tip of the iceberg of what can be learned by harnessing this approach. Researchers across disciplines are constantly working to advance the technology and expand the number of questions that can be asked (Box 1). As these techniques develop, so will our understanding of host-bacterial interactions.

Box 1: Moving Beyond Single-Gene Knockouts in CRISPR Screens.

The many advantages of CRISPR screens have inspired researchers to expand the technology to address additional questions. Two major advances in CRISPR screens have included using Cas9 constructs designed to modulate gene expression rather than ablate it and knocking out multiple genes to understand genetic networks. While these approaches have had limited use in host-bacterial screening, we will spend some time discussing these platforms as they will likely become more prevalent in the immediate future.

Through engineering of Cas9 machinery, researchers have developed systems to increase (CRISPRa) or decrease (CRISPRi) host gene expression without directly disrupting the gene [83]. In particular, CRISPRa technology represents a major innovation as its one of few techniques that can be used for gain-of-function screening. Further, unlike traditional cDNA overexpression systems, CRISPRa modulates expression at the native locus and thus keeps cis-regulatory effects intact. While no bacterial host resistance screens have been performed using CRISPRa, a recent publication used it to identify norovirus resistance genes [84]. Interestingly, while one weakness of CRISPRa and CRISPRi is that different sgRNA can yield different intensities of overexpression or knockdown, some groups have actually engineered sgRNA that intentionally differ in efficiencies in order to study the impacts of titrated gene dosage on phenotypes [85].

In addition to advances in modulating expression, some studies seek to examine how perturbing multiple genes in a single cell impacts host phenotypes (Figure i). Approaches using standard CRISPR/Cas9, CRISPRa, and CRISPRi technologies have all attempted to co-transfect sgRNA into cells, determine which sgRNA were expressed, and measure the impacts of those co-expressed sgRNA on the host. There are several different approaches to examine combinatorial effects, primarily differing in how to determine which sgRNA were transfected into cell. One technique, called CRISP-Seq, pairs CRISPR screening, fluorescent sorting, and single-cell RNA-sequencing [2]. By placing genes for fluorescent markers and unique guide indexes on the sgRNA plasmids, the authors were able to use cell sorting and single-cell sequencing to connect the sgRNA(s) expressed in each cell to the resulting changes in the transcriptome. While CRISP-Seq has great promise, the number of sgRNA combinations tested is limited by single cell RNA-sequencing throughput as well as the number of available fluorescent markers. Further, techniques that depend on guide indexes are subject to spontaneous recombination during lentiviral preparation than can decouple the unique guide indexes and sgRNA [86]. An alternative approach, called direct-capture Perturb-seq, leverages single-cell RNA-sequencing paired to a modified library preparation to directly identify expressed sgRNA during sequencing [87]. By directly sequencing guides this method lacks some of the limitations associated with CRISP-Seq, though is still bound by single-cell RNA-sequencing throughput.

As mentioned above, significant advances have been made performing CRISPR screens in vivo [2, 3], however, loss-of-function screens in whole model organisms are still predominantly performed using random mutagenesis (e.g. N-ethyl-N-nitrosourea (ENU) mutagenesis). These tried-and-true techniques have been thoroughly reviewed elsewhere [19, 20], but we want to draw special attention to the role these systems have played in understanding host-bacterial interactions in zebrafish. This is perhaps best illustrated by a pair of landmark papers by Tobin et al. who used an ENU mutagenesis screen to identify a critical role for lta4h in modulating zebrafish susceptibility to Mycobacterium marinum infection [21]. Interestingly, they found that lta4h activity must be carefully balanced, as excessive activity results in accumulation of the proinflammatory LTB₄ eicosanoid, while reduced activity leads to accumulation of the anti-inflammatory Lipoxin A₄ [22]. Tobin et al. confirmed this optimal immune balance in humans, as human SNPs associated with LTA4H gene expression confer heterozygous advantage [21] and LTA4H genotype appears to serve as a predictive biomarker for the efficacy of the anti-inflammatory drug dexamethasone [22]. This marriage of zebrafish and human genetics leading to clinical translation demonstrates the power of random mutagenesis screens. However, the future of loss-of-function screening in zebrafish is unclear as impressive work has been done to develop zebrafish CRISPR screens [23–26], though these techniques have not yet been applied to host-pathogen biology.

Leveraging Natural Genetic Diversity to Identify Host Resistance Genes

An additional limitation of genetic knockout systems is that they cannot reveal which genes contribute to natural inter-individual differences. This is because the majority of inter-individual differences are not large genetic ablations but rather single nucleotide polymorphisms (SNPs) [27]. Therefore, complimentary techniques that harness natural diversity are necessary to identify naturally occurring host resistance and susceptibility genes (Figure 1).

The technical and statistical frameworks that leverage natural variation to identify host resistance genes differ but are broadly categorized into linkage analysis and association analysis [28]. For many years, linkage analysis has been the predominant method for studying Mendelian traits, where highly penetrant variants of large effects are mapped using genotyped family pedigrees. This kind of analysis is still very powerful, especially in finding rare variant “inborn errors of immunity” [29], such as mutations in IL-12 and IFN-γ signaling that predispose to mycobacterial and Salmonella infection [30]. However, for identifying more common resistance alleles for polygenic traits, genome-wide associations studies (GWAS) are the predominant discovery method. In GWAS, members of a genetically diverse population are assessed for a phenotype (e.g. infection status, disease severity, cytokine response during infection, leukocyte counts, etc.). Then, markers across the genome are tested for association with the trait, asking how well each marker explains the observed phenotypic variation when accounting for genetic relatedness and other covariates (usually through principal components analysis) [31]. These studies depend on linkage disequilibrium maps [32, 33], which are used to select tag SNPs (to reduce the number of markers that need to be genotyped) and to understand individual association signals using the patterns of genetic inheritance along chromosomes. These techniques have identified loci associated with genetic resistance, susceptibility and quantitative host-bacterial interactions (quantitative trait loci (QTL)). We will discuss three applications of natural variation in host-bacterial studies: (1) leveraging natural variation in model organisms through association mapping, (2) leveraging natural genetic variation in humans using GWAS, and (3) examining natural genetic variation in vitro using cellular GWAS.

Identifying Host Bacterial Resistance Genes by Using Natural Diversity in Mouse Models

Many studies have paired the strengths of invertebrate (Box 2) and vertebrate models of infections with the natural diversity present within these species to study host-pathogen interactions. Here we will focus on host genetic screens of bacterial systems that use the Collaborative Cross (CC) mouse model, which consists of ~100 recombinant inbred lines derived from eight diverse founder strains (C57BL6/J, 129S1/SvlmJ, A/J, NOD/ShILtJ, NZO/HiLtJ, CAST/EiJ, PWK/PhJ, and WSB/EiJ) [34]. These diverse yet inbred strains allow nearly genetically identical replicates and mechanistic follow-up. While QTLs revealed by the CC lines are quite large (the average recombination distance is ~10Mb), individual CC strains that are outliers for a trait of interest can be crossed to produce F₁ and F₂ progeny for more sensitive QTL mapping. This section will complement recent reviews on the impact of CC mice on host-pathogen biology [35, 36], as we unpack a number of recent host-bacterial papers with exciting implications.

Box 2: Natural Diversity Screen using Animal Models.

Caenorhabditis elegans and Drosophila melanogaster have frequently been used in natural diversity screens aimed at understanding host-pathogen interactions, as researchers have collected, lab-adapted, phenotyped, and sequenced wild-isolates of these species. A classic example used natural diversity to identify that a single amino acid substitution in the C. elegans neuropeptide Y receptor homolog (NPR-1) drives a differential response to bacterial prey in wild-isolates [88]. More recent work has sought to understand how the microbivorous worms avoid pathogen exposure in the wild. One group leveraged a diversity panel to identify two separate single amino acid substitutions in a neuronal ubiquitin E3 ligase (HECW-1) that affect Pseudomonas aeruginosa avoidance [89]. Later work has also leveraged diversity to identify QTLs for P. aeruginosa [90] and Serratia marcescens [91] avoidance, though no causal genes were identified.

Early work harnessing Drosophila diversity to study bacterial susceptibility performed candidate diversity screens targeting immune genes on the second [92, 93] and X [94] D. melanogaster chromosomes. Following these studies, a novel resource containing hundreds of genotyped strains was devised for genome-wide D. melanogaster association mapping: the Drosophila melanogaster Genetic Reference Panel (DGRP) [95]. One study using the panel found that genetic variation in the kri, S6k, mad2 locus and the BomBc1 antimicrobial peptide locus associated with resistance to E. faecalis [96]. Another expanded our understanding of fly gut immune defense by pairing the DGRP with Pseudomonas entomophila [97]. Researchers found that animals that showed higher survival cleared the infection faster, launched a more controlled reactive oxygen species response, maintained protein synthesis, and preserved gut stem cell activity. Additionally, twenty-seven survival QTLs were mapped, including several SNPs in the Nrk and Gyc76C genes, which the authors confirmed play key roles in immunity.

By using the standardized DGRP, it becomes easier to compare results across studies. For instance, one impressive screen infected the DGRP with Metarhizium anisopliae or P. aeruginosa and identified a number of sex-dependent QTLs associated with survival, including in the kri, S6k, mad2 locus that associated with E. faecalis survival above [98]. Further, by mining DGRP studies, they found that geotaxis, oxidative stress, blood glucose, starvation, and sleep patterns all correlate with post-infection survival.

Curiously, despite significant amounts of genetic diversity in laboratory zebrafish strains [99], we are unaware of any work that has leveraged this diversity to study infectious disease. This may represent an exciting future direction for the field.

Several studies have identified natural variation in responses and susceptibilities of CC mice to Klebsiella pneumoniae [37], Salmonella enterica serovar Typhimurium (S. Typhimurium) [38], P. aeruginosa [39], and the periodontitis-associated bacteria Porphyromonas gingivalis and Fusobacterium nucleatum [40]. Even with relatively small cohorts (35 – 48 CC lines), the S. Typhimurium and K. pneumoniae studies identified putative QTLs underlying susceptibility [37, 38]. Additional work by one of the authors of this review demonstrated that the CC mice encompass a broad phenotypic spectrum after Mycobacterium tuberculosis infection and further showed that host genetic background modified the BCG vaccine’s protective efficacy against tuberculosis [41]. Intrinsic host resistance to M. tuberculosis and the response to the BCG vaccine were not correlated, demonstrating the genetic complexity of M. tuberculosis protection.

As noted above, an advantage of the CC is the ability to follow-up phenotypes with mechanistic experiments. For instance, in the original M. tuberculosis study [41], the mouse line CC042 had unusual susceptibility to infection. Further work using F₂ progeny to map the QTL revealed that CC042 mice have a 15-base deletion in the Itgal gene that resulted in impaired lymphocyte trafficking and M. tuberculosis susceptibility [42]. Interestingly, work by a separate group on CC042 identified that the same deletion drives susceptibility to S. Typhimurium [43], highlighting a conserved role for the protein in pathogenesis. Unexpectedly, the CC042 Itgal deletion is absent in the original CC parental strains and instead arose de novo during inbreeding. The high frequency of these de novo mutations during the inbreeding process may have ramifications on reproducibility across CC generations [44]. Importantly, these studies demonstrate that experimental follow-up can surmount the large QTLs obtained from CC studies and reveal very specific genetic mechanisms for differences in susceptibility.

There have been two notable attempts to connect CC bacterial pathogenesis studies to human GWAS cohorts. The first studied periodontal models, using P. gingivalis and F. nucleatum infection[45]. Unfortunately, even with F₂ progeny, the identified QTLs were massive—the two significant QTLs contained 80 genes, while the eight suggestive QTLs contained 1,309 genes. Thus, while the authors note some overlap of their QTLs to loci identified in aggressive periodontitis and chronic periodontitis human GWAS cohorts, the implications are unclear. Additionally, work from Lorè et al. examined P. aeruginosa susceptibility using the CC system and identified a QTL containing 31-protein coding genes [46]. By examining the associations between their QTL and SNPs in the human syntenic region from a cystic fibrosis GWAS, the authors identified two SNPs in DPYD, rs10875080 and rs11582736, which associated with a modest delay in the age of first P. aeruginosa infection in the human cohort. Follow-up work has yet to connect Dpyd to resistance in the CC lines. Together, these data demonstrate both the promise of the CC mice to identify possible human resistance genes and the challenge of going from QTL intervals to causal variants.

The CC mapping power problem has been highlighted throughout this section and is due in part to the number of strains available. Recent power analyses in the CC indicate that for studies of polygenic traits explained by many loci of small effect, more CC strains are required than may be currently available [47]. In lieu of more CC strains, another approach may be to leverage more refined phenotypes (“endophenotypes”) that explain a larger proportion of the variance. While the broader genetics field has taken an eQTL approach, the infectious disease field has the opportunity to move away from polygenic traits such as survival or burden and instead leverage bacterial expression as novel refined intermediate phenotype that is likely to be caused by very specific host variants. The advent of diverse but reproducible model populations such as the CC, in combination with next generation bacterial technologies, make these dual host-pathogen mapping approaches possible.

Identifying Host Resistance Genes Using Natural Human Diversity

GWAS of human infectious diseases present several unique challenges. First, even with covariates (i.e. age, sex, comorbidities, and unknown factors with principal components analysis), a successful GWAS requires large population sizes, generally in the thousands. Second, obtaining exposed controls is challenging, so many infectious disease GWAS simply use population controls (e.g. blood donor banks [48–50]). This approach has a significantly lower logistical and financial cost compared to obtaining carefully matched controls who were exposed to the pathogen and resisted infection without a protective memory response, but results in significantly lower power. Third, integrating individual GWAS into the broader scientific literature is difficult, as differences in populations, sample sizes, disease definition, and study design can severely impair reproducibility across cohorts. Despite these challenges, GWAS can reshape medicine by informing drug design and providing possible clinical biomarkers. For example, an early GWAS of hepatitis C virus infection [51] led to the development of a genetic biomarker that helped direct recombinant interferon treatment [52].

Researchers have boosted sample size by assembling massive databases of genotyped individuals with clinical information from electronic medical records (e.g. UK biobank [53], eMERGE network [54]) and self-reporting (23andMe [55]) and performing simultaneous GWAS across many phenotypes. For example, a study by 23andMe recruited a cohort of >200,000 European-descent participants to fill out a survey about their experiences with common infectious diseases [56]. The group identified 59 genome-wide significant SNPs associated with 17 different infectious diseases. They also identified pleiotropic SNPs that affect multiple diseases. For example, rs1978060, in TBX1, is associated with tonsillectomies, childhood ear infections, and myringotomy, which implicates the gene broadly in bacterial susceptibility. Further, they found SNPs in HLA class I genes tended to correlate with viral disease susceptibility, while bacterial diseases were more likely to have significant SNPs in HLA class II genes, in line with each proteins’ canonical role in immunity. Notably, they also found associations in autoimmunity-associated genes, highlighting potential tradeoffs between infection protective alleles and inflammation homeostasis alleles.

Even without massive databases, dozens of bacterial susceptibility GWAS papers have been published over the last decade (Table 1). The specific loci identified for each disease have been reviewed elsewhere [57, 58], so we will focus on how GWAS results are validated and used to enhance our understanding host-pathogen biology and/or genomics.

Table 1:

Human Genome-Wide Association Studies for Bacterial Infection

Pathogen(s) and/or Disease(s)	References
Mycobacterium tuberculosis and/or Tuberculosis	[48–50, 60, 106–113]
Mycobacterium leprae/Leprosy	[114–118]
Staphylococcus aureus, S. aureus Bacteremia, and/or S. aureus Infective Endocarditis	[119–123]
Streptococcus pneumoniae, Neisseria meningitidis, Pneumococcal or Meningococcal Bacteremia/Sepsis/Meningitis, and/or Pneumonia Associated Sepsis	[102, 124–127]
General Bacteremia and/or Non-Typhoidal Salmonella (NTS) Bacteremia	[77, 128]
Oral Bacterial Pathogen Burden (in Periodontal Disease Patients)	[129–131]
Enteric Fever/Typhoid Fever	[132]
Helicobacter pylori	[133]
Pseudomonas aeruginosa (in Cystic Fibrosis Patients)	[134]
Common Infectious Disease (23 Pathologies Tested)	[56]

Validation.

One of the most critical questions following GWAS is whether the identified SNPs are true hits or a consequence of sampling bias. The gold standard for validating SNPs is to replicate the association in additional, unrelated cohorts. Unlike the original GWAS which requires genome-wide significance (traditionally p < 5×10⁻⁸) to overcome the large number of multiple comparisons, targeted replication of the association can have weaker statistical significance as the correction burden is lower. Examples of successful replication are discussed throughout this section.

Importantly, failure to replicate GWAS can occur for a number of reasons. For example, the 23andMe study replicated known associations with shingles and tonsillectomy, but failed to replicate several other reported genome-wide associations [56]. This may be due to differences in populations or criteria for patient inclusion, as well as false positives. Population differences are particularly difficult to overcome, as differences in allele frequencies and in linkage disequilibrium among populations can result in insufficient power for SNPs that are rare in the replication population or even make replication impossible if the SNP is not present. Further, complex gene x gene and gene x environment interactions may prevent certain alleles from having an impact in certain populations. Alternatively, differences in disease definition across studies can have dramatic impacts on patient inclusion, leading to profound differences in the actual phenotype being measured. Determining whether a replication failure is due to the heterogeneity of infectious disease studies or false positives can be difficult and is most adequately addressed by carefully considering the differences between the specific cohorts in question.

Finally, the “Winner’s Curse,” where the effect size of reported GWAS hits are often larger than their true biological effect, can substantially contribute to lack of replication. Analysis of the results of 100 quantitative trait GWAS papers that attempted replication by Cameron and Pe’er [59] revealed that 31% replicated at Bonferroni-corrected p-value thresholds, compared to 75% predicted by the effect sizes in the discovery cohorts. Winner’s Curse accounted for most of this difference, resulting in a predicted replication rate of 37%. When analysis was restricted only to the 39 papers that attempted replication within the same continental ancestry and reported numbers of the replication cohort for each locus being tested, the replication rate increased to 43%, consistent with the predicted replication rate considering Winner’s Curse (45%). Thus, as with all science, interpretation of GWAS findings are a continual process, and as replication, functional, and clinical follow-up studies are conducted, the most robust findings will stand the test of time.

Biological Insight.

Understanding the mechanistic underpinnings of GWAS findings can take different forms. Some researchers focus on how implicated gene(s) affect infection. Such experiments involve knocking down or knocking out the gene and assessing how cells or mice change their response to the pathogen. Other studies focus on uncovering (a) which genetic variant out of the many in linkage disequilibrium causes a locus to associate with pathogen resistance and (b) how that variant regulates the associated gene(s) to induce resistance (e.g. coding mutation, enhancer mutation, splicing variant, etc.).

We will discuss Curtis et al. as an example for how GWAS can generate insight into infection biology. In their work, they identified ASAP1 as a novel human resistance factor against M. tuberculosis [50]. After identifying several SNPs in ASAP1 as putative hits in their original GWAS (~11,000 Russian individuals), the group validated these SNPs in an additional Russian dataset and in a previously published cohort from Ghana and The Gambia. To connect the variation to ASAP1 function, the group examined how host genotype and Mycobacteria infection affect expression in immune cells. They found the rs4733781 risk allele associates with reduced baseline and post-infection expression of ASAP1 in dendritic cells. Finally, they demonstrated that knockdown of ASAP1 in patient-derived dendritic cells reduced dendritic cell matrix digestion and motility. These data suggest that ASAP1 helps activate the adaptive immune system, as suppression hinders migration of dendritic cells to lymphatic tissues.

In contrast, work by Zheng et al. exemplifies how GWAS can provide insights into genetic regulation as well as host-pathogen interactions [60]. In their work, they performed an initial GWAS (~2,100 Han Chinese individuals), followed by two smaller replication studies to identify two SNPs associated with tuberculosis. Following up on rs6114027, located in an TGM6 intron, they revealed that the risk allele associates with reduced TGM6 expression in patient immune cells. Based on luciferase assays, they demonstrated that the intronic risk allele suppressed expression relative to the protective allele. Further work with patients and mouse models confirmed the importance of TGM6 in M. tuberculosis resistance. Together, these data both identified an allele-specific and causal role for the rs6114027 containing intron in regulating TGM6 expression, as well as demonstrated the importance of the gene during M. tuberculosis infection.

Identifying Host Resistance Genes Using Cellular GWAS

While the previous section demonstrates the wealth of knowledge obtained through traditional GWAS, challenges of heterogeneity still hinder these studies. Further, there are inherent limitations to what phenotypes can be measured in humans, making it difficult to understand how the SNPs contribute to specific molecular and cellular changes that impact pathophysiology. Several groups have utilized complementary cellular GWAS methods to identify SNPs that alter host cell-pathogen interactions following stimulation or infection of primary immune cells [61–66], differentiated induced pluripotent stem cells (iPSC) [67], or lymphoblastoid cell lines (human B cells immortalized ex vivo through Epstein Barr Virus infection) [68–71]. Shared advantages of these systems are (1) consistent dose and timing across cell lines, (2) controlled bacterial genetics and/or PAMP sources, (3) same donors across technical and biological replicates, and (4) a greater number of testable phenotypes. Collectively, these systems have identified a plethora of novel SNPs and genes that contribute to host resistance to infection.

One form of cellular GWAS attempts to uncover the genetic factors that contribute to variable transcriptional responses to pathogenic stimuli. Similar to studies that identify QTL associated with gene expression (eQTL), these studies seek to understand response eQTL (reQTL), or eQTL that emerge in response to stimulation. For instance, work on patient-derived monocytes stimulated with IFN-γ or LPS [61] or patient-derived dendritic cells stimulated with LPS, influenza virus, or IFN-β [62] revealed a number of reQTLs. Similar work found reQTLs and response chromatin accessibility QTLs (response cQTLs) in iPSC-derived macrophages following stimulation with INF-γ and/or S. Typhimurium [67]. Interestingly, these studies found overlap between their reQTLs and SNPs identified in bacterial, inflammatory, and/or autoimmune disease GWAS. More recent work has identified QTL associated with microRNA regulation during monocyte stimulation [72]. These studies demonstrate that understanding response QTLs provides mechanistic information that complements traditional GWAS.

Other work has examined how variation affects the host immune response. Specifically, the Human Functional Genomics Project has identified genetic variation that affects pathogen sensing and cytokine production by immune cells ex vivo. Li et al. exposed peripheral blood mononuclear cells from over 100 European donors to LPS, heat-killed M. tuberculosis, or heat-killed Candida albicans [64]. Following stimulation, the authors measured six cytokines and identified six novel QTLs, including two QTLs involved in M. tuberculosis-induced IL-8 and one QTL involved in LPS-induced IL-10. They found high levels of pleiotropy, with all genome-wide hits also associating with additional pathogen-induced cytokine responses. Following integration of their data in the National Human Genome Research Institute (NHGRI) GWAS catalog, their SNPs also associated with infectious disease susceptibility, autoimmunity, and curiously, heart disease. Later work by the same group expanded the concept to measure cytokines from macrophages, peripheral blood mononuclear cells, and whole blood from over 400 individuals in response to 18 different killed-pathogens, TLR-ligands, or metabolic stimuli [65]. The group found 18 genome-wide associated SNPs in 17 independent loci with a high degree of replication using their previous study [64] as a replication cohort. One of the top hits from this study was located in the TLR1, TLR6, and TLR10 locus, and affected IL-6 production following PBMC stimulation with polyinosinic:polycytidylic acid (poly I:C). This locus is under positive selection, suggesting its importance in host resistance. Interestingly, an earlier study also found a number of SNPs that correlate with IL-6 production at this locus following stimulation of whole blood with Pam₃CSK₄ [66]. This convergence highlights the importance of TLR 1/6/10 signaling in cytokine production and demonstrates the ability for even highly divergent cellular GWAS to identify common loci.

One of our labs carries out cellular GWAS screening with a system called Hi-HOST (High-throughput Human in vitrO Susceptibility Testing) using live pathogens. The original Hi-HOST screen was performed using 352 lymphoblastoid cell lines (from donors of western European or Nigerian descent) infected with S. Typhimurium or S. Typhi [68]. We screened for QTLs involved in host cell invasion, pyroptosis, bacterial replication, and intracellular bacterial survival. We found that rs514182, an eQTL for the methionine salvage enzyme APIP, associated with pyroptosis during S. Typhimurium invasion [69]. Follow-up work revealed that APIP negatively regulates pyroptosis by reducing the abundance of the methionine-derived metabolite methylthioadenosine. The same locus is associated with systemic inflammatory response syndrome, as well as non-typhoidal Salmonella (NTS) bacteremia and sepsis [73]. Finally, metabolites in the salvage pathway regulated by APIP, in particular methylthioadenosine and S-adenosyl homocysteine, are reliable sepsis biomarkers [73] that also directly regulate S. Typhimurium virulence [74, 75]. Another hit from the Hi-HOST screen was rs8060947, an eQTL for VAC14, which correlated with S. Typhi invasion [70]. VAC14 reduces cholesterol levels in the plasma membrane, leading to reduced S. Typhi docking and invasion. This SNP also associated with Typhoid fever susceptibility, malarial risk [76], and modestly with NTS, S. pneumoniae, E. coli, and Acinetobacter bacteremia [77].

Hi-HOST recently evolved into the Hi-HOST Phenome Project (H2P2), which screened an additional 528 lymphoblastoid cell lines from four populations, with eight different stimuli (seven live bacterial and fungal pathogens and one bacterial toxin), for 79 phenotypes [71]. Several of the 17 genome-wide significant hits demonstrated pleiotropy, including rs953897, located in the transcriptional repressor ZBTB20. rs953897 associated with Chlamydia trachomatis burden and 20 additional H2P2 phenotypes at p<0.05. ZBTB20 knockdown validated three of the strongest associations: C. trachomatis burden and induced IL-6 production, and S. Typhimurium-induced pyroptosis. Further, integration of the H2P2 data into the eMERGE PheWAS and other clinical datasets revealed that a number of the top H2P2 SNPs associate with human disease, including the ZBTB20 SNP (rs953897) with HCV susceptibility and a SNP in CXCL10 (rs2869462) with inflammatory bowel disease. Future work is exploring the biology behind the H2P2-identified SNPs and further adapting the Hi-HOST screening platform.

While this review has focused on methods that interrogate the host to identify genes that contribute to resistance, there is a major caveat in studying host-pathogen interactions in this way: these approaches generally require treating pathogens as static, homogenous stimuli, rather than dynamic and diverse components of the host-pathogen interaction. There is substantial value in incorporating pathogen biology into approaches to identify host resistance genes, including through bacterial diversity informed GWAS as described in Box 3.

Box 3: Using Bacterial Diversity Informed GWAS to Identify Host Resistance Genes.

In our discussion of animal models of natural diversity, single genetically identical strains of pathogen were used to identify loci associated with pathogen outcome. However, in human cohorts, participants can have dramatically different bacterial strains, serovars, or even species pooled into the same experimental group. This can make it difficult to identify associated loci, as pathogen diversity can confound the host response. An alternative approach to traditional GWAS incorporates pathogen diversity by genotyping both host and pathogen DNA from participants to stratify results by bacterial lineage (Figure ii). This approach has been applied to M. tuberculosis, where one study ran GWAS on their entire patient cohort, as well as separate GWAS on patients infected with the Beijing lineage or those infected with a non-Beijing lineage [49]. While, no significant associations were noted in the pooled GWAS, the researchers were able to find a statistically significant association with susceptibility to the non-Beijing lineages (rs1418425, intergenic between CD53 and LRIF1, odds ratio 1.62). This association was found in their primary and replication cohorts, however, it is interesting to note that the association was only significant among older individuals.

Beyond this approach of stratifying GWAS by bacterial taxonomy, a second approach involves using hypothesis free genome-to-genome analyses to identify interactions between host and pathogen genetics by measuring associations between human and pathogen SNPs (Figure ii). This information can then be used to stratify GWAS results using interacting genes. The genome-to-genome approach has proved fruitful for viral human GWAS with HIV [100] and HCV [101]. In these viral studies multiple associations were found between HLA genes and viral proteins, as well as an association between rs12979860 (an intronic SNP in IFNL4) and the HCV protein NS5A. However, the increased multiple comparison burden from the larger size of bacterial genomes have thus far made this approach recalcitrant for bacterial human GWAS, though it has been attempted in a small pneumococcal disease cohort [102]. Future work with large bacterial GWAS cohorts should consider sequencing host and bacterial DNA, as identifying host-bacterial interacting genes could provide important insight into both host resistance alleles and host-pathogen interactions.

Interrogating the Host-Pathogen Interface through Dual RNA-seq

In this last section, we will discuss techniques that use the bacteria as a biological sensor to understand the pressures inflicted by the host (Figure 1). This can take many forms, including examining the bacterial-host protein interactome or tracking bacterial transcriptional and genomic adaptation to the host (Box 4), or by simultaneously measuring how the bacteria and host respond to one another during infection. We will discuss the latter by examining the impact of dual-RNA sequencing on understanding host-pathogen interactions. Notably, while this method does not directly implicate host genes in being important for resistance, these screens do identify host processes that appear to place significant selective pressures on the bacteria and therefore represent an important hypothesis generating tool.

Box 4: Identifying Important Host Processes by Studying the Bacteria.

Many bacteria manipulate host processes to thrive within an intracellular space. Understanding which host proteins bacteria have evolved to interact with can shed light on which host genes are necessary for virulence. Two techniques can be used to uncover interactions between bacterial and host proteins: co-immunoprecipitation approaches or BioID-based screens. A recent preprint used the former by generating 15 S. Typhimurium mutants containing chromosomally-tagged effector proteins and infecting HeLa (Human epithelial cell line) and RAW264.7 (Murine macrophage cell line) cells [103]. Following infection, secreted effectors were immunoprecipitated and host interacting proteins were identified using mass spectrometry. A similar BioID screen ectopically overexpressed five BirA-tagged S. Typhimurium effectors in human epithelial cells to identify effector-interacting proteins [104]. Notably, there were considerable differences in the results from these studies, with some of the interactions from the infection experiment failing to replicate in the BioID paper. The lack of replication may be the result of effector overexpression failing to model infection conditions. Thus, the strengths and weaknesses of these systems must be considered when designing protein interactome screens.

The above examples rely on the fact that bacterial proteins have evolved over countless generations to manipulate host proteins. However, genomic approaches can also be used to study adaptation in real time. For example, recent work by Crofts et al. used a human infection model of Campylobacter jejuni to understand how the pathogen’s biology changes during infection [105]. The authors infected volunteers with a genetically diverse population of C. jejuni and collected feces from participants in order to (a) compare the C. jejuni transcriptome within humans to the transcriptome in other conditions, and (b) track how selective pressures in humans influenced C. jejuni evolution. The authors noted that genes involved in hydrogen peroxide detoxification, iron acquisition, and antimicrobial peptide resistance were upregulated, suggesting that the bacteria were responding to host innate and nutritional immunity. By studying the pathogen’s genomic adaptation, the authors found that a number of mutants showed consistent selection patterns across patients. Notably, mutations in seven out of eight flagellin modifier genes in the C. jejuni genome were under selection in humans, suggesting an important role for flagellin modifications and/or host flagellar sensing during C. jejuni infection. Together, these experiments uncovered clues from the bacterial response to humans that reveal host selective pressures.

Improvements in RNA-sequencing technology has enabled capturing and sequencing of host and pathogen RNA during in vitro and in vivo infection. This technology, called dual RNA-seq or dual-seq, directly examines correlations between host and bacterial transcripts, providing preliminary evidence of how the two entities respond to each other. Dual-seq has been reviewed elsewhere [78], so here we briefly discuss how the technology can identify host processes that restrict bacterial virulence. In contrast to traditional dual-seq studies that measure the host & pathogen transcriptional profiles during infection, measuring restriction requires comparing the transcriptional profiles of resistant and susceptible hosts. Work by Thänert et al. used dual-seq to understand differential S. aureus susceptibility in C57BL/6 and A/J mice [79]. Infected tissues from susceptible A/J mice showed evidence of elevated inflammation and hypoxia, correlating with bacterial upregulation of genes involved in anaerobic metabolism and acid resistance. This likely drives the sepsis-related mortality in these mice. In contrast, S. aureus in resistant C57BL/6 mice increased expression of de novo amino acid synthesis pathways, antimicrobial peptide resistance genes, and stress response genes, implying that nutritional and innate immunity restrict bacterial virulence.

Beyond inter-animal variation, other work has probed heterogeneity across cells within a single animal. One study examined M. tuberculosis permissibility in alveolar (permissive) and interstitial (restrictive) lung macrophages and revealed a link between host metabolism and bacterial fitness [80]. The transcriptional differences followed traditional M1/M2 polarization schema, with permissive alveolar macrophages displaying high fatty acid metabolism and M2-like gene expression, which triggers fatty acid metabolism, growth, and iron storage in the invading M. tuberculosis. In contrast, the restrictive interstitial macrophages are more M1-like, sequester iron, reduce potassium and chloride transport, and drive expression of stress-response, iron savaging, and dormancy genes in their associated bacteria. Similar results were found in a dual-seq study examining S. Typhimurium growth dynamics, as they found that Salmonella use the effector SarA to drive macrophages into a permissive M2-like state and enable rapid intracellular bacterial growth [81]. This suggests that clinical manipulation of host metabolism or macrophage polarization could impact bacterial virulence and disease progression.

Improving single-cell RNA-sequencing technology will have dramatic impacts on the ability for these studies to uncover cellular heterogeneity in bacterial susceptibility. Previous work has validated this tool for dual-seq [82], though it remains underutilized. Future work could apply single-cell dual-seq to dissect why neighboring cells within a single tissue have drastically different outcomes to bacterial infection.

Concluding thoughts

While this review focused on host-bacterial interactions, these screening approaches have also been adopted and advanced by virologists, mycologists, and parasitologists to study host-pathogen biology. The work presented here demonstrates that screens can dramatically advance the field. However, it is important to note that simply compiling lists of associations do not, by themselves, illuminate how genes confer resistance to infection. Rather, following-up screens with additional studies provides key mechanistic insight into host-pathogen biology (See Outstanding Questions Box).

Outstanding Questions Box.

• Do CRISPR screen results replicate under different conditions? What are the implications of the concordance or discordance?

• Do results from Collaborative Cross mice translate to human cohorts and how can such results be efficiently integrated?

• How do loci from Collaborative Cross or human GWAS studies influence host-pathogen interaction outcomes on the molecular and cellular levels?

• How can targeted genetic engineering be paired with natural host diversity to identify genetic modifiers of genes?

• How do loci that impact host susceptibility to bacterial infection impact other aspects of host biology? Are there consequences to chronic disease in humans?

• What impact do microbial genetics have on the results from screens?

• How can microbial diversity be screened simultaneously with standard host screens to expand the clinical relevance of the results?

We discussed how innovative technology has advanced the field; however, new approaches are constantly on the horizon. For example, as discussed in Box 1, advances in CRISPR/Cas9 technology will likely expand the breadth of questions answered by CRISPR screens. Additionally, while many screens above treat the invading pathogen as a biological constant and/or a signaling PAMP, significant insight can be obtained by simultaneously screening host and pathogen diversity. As discussed in Box 3, this concept has already been acknowledged in human GWAS studies, however, future in vitro and animal model screens should also consider inclusion of pathogen diversity (whether natural diversity or bacterial knockout pools) to significantly improve modeling of complex host-pathogen interactions. In particular, we think that including bacterial diversity in approaches such as the Collaborative Cross model, Hi-HOST, or dual-seq studies could yield fascinating results. Similarly, expanding the unique phenotypes measured in natural diversity screens will improve our understanding of host susceptibility, particularly enabling pleiotropic loci identification via screen integration. Finally, improvements on methods such as single-cell RNA-sequencing, epigenetic modification detection, iPSC differentiation, high-throughput microscopy, flow cytometry, and computational toolkits will revolutionize the screens listed in this review as they expand the number of quantifiable phenotypes, provide single cell resolution, and increase throughput & statistical power. Thus, over 350 years after the birth of the field, it remains an exciting time to be a microbiologist.

Figure i: Approaches for Multi-Gene CRISPR Knockout Screens.

Two approaches to assay the impacts of multi-gene deletions include CRISP-seq [2] and Direct Capture Perturb-seq [87]. In CRISP-seq (Top), sgRNA plasmids include a unique fluorescent marker and unique guide index (UGI). This allows two unique ways of determining which sgRNA(s) were transfected into each cell assayed by single cell RNA-seq. First, during the library generation, flow cytometry can be used in order to determine which fluorescent markers are expressed in the cell. Second, the UGI can be detected directly during RNA-sequencing and mapped back to the sgRNA plasmid. Together these features enable indirect identification the sgRNA expressed in the cell as well as the resulting transcriptome changes in each. In contrast, Direct Capture Pertub-seq (Bottom) simply uses a modified single cell RNA-seq library preparation that enables sgRNA to directly sequence each expressed sgRNA alongside the transcriptome.

Figure ii: Pathogen Informed GWAS.

Infections occur in the context of both host and pathogen diversity. Simultaneous sequencing of host and pathogen DNA enables pathogen diversity to be captured during GWAS studies. This can take two forms. First, GWAS can be stratified by bacterial strain, so that the heterogeneity of the infected state is significantly reduced and biological differences between individuals are more likely driven by host genetic diversity. Second, hypothesis-free genome-to-genome analysis looks for statistical associations between the bacterial and host genomic diversity. These statistical associations may mark sites of host-pathogen evolutionary conflict and implicate host genes in the host-bacterial interaction.

Highlights.

• Understanding host-microbe interactions remains critical as growing antibiotic resistance, new outbreaks, and reemerging pathogens put lives at risk.

• Microbiologists and infectious disease researchers have leveraged and advanced the genomic technological breakthroughs of the last decade.

• Advances in CRISPR/Cas9-mediated genome editing and reduced sequencing costs have made CRISPR screens the dominant loss-of-function and gain-of-function screening platform.

• Advances in sequencing, high-throughput technologies, and resources for model organisms and humans have dramatically improved natural diversity screens. Such resources include consortia with repositories for electronic medical records and human cell lines, and model organism diversity panels.

• Integrating pathogen diversity into host resistance screens help further define the genetic landscape of host-pathogen interactions.