Enabling patients to survive severe infection is a fundamental goal of critical care medicine, encompassing both infection as the primary insult leading to critical illness and secondary infections acquired after admission to a critical care unit. Aside from provision of organ support, in current clinical practice our ability to modify the pathogen-host interaction is mainly limited to the use of antimicrobials. While antimicrobial therapy represented a massive advance when first implemented at scale more than 80 years ago, and remains the cornerstone of managing most severe infections, the approach suffers from 3 main limitations.

• 1.

  Death and morbidity due to severe infection occurs despite receipt of antimicrobials with in vitro antimicrobial activity against the causative pathogen;

• 2.

  Antimicrobials have built-in redundancy: evolution of antimicrobial resistance is inevitable;

• 3.

  Antimicrobial therapy is associated with adverse ecological events for recipients, predisposing to colonization with resistant pathogens, Clostridioides difficile infection, and invasive Candida infections.

A conceptually attractive adjunctive or even alternative strategy is to therapeutically optimize the host immune response using immunotherapies. Two main strands of evidence support this approach: Firstly, host defense against pathogens is usually extremely successful. Despite frequent exposure to and colonization with pathogenic microbes such as Streptococcus pneumoniae or Staphylococcus aureus , invasive disease is a very uncommon outcome. ^,,, Identifying and augmenting the components of normally successful host defense that fail during disease therefore represents an attractive therapeutic strategy, referred to here as “host-directed” anti-infective therapy. This also extends to suppressing immunopathologic inflammatory responses. Secondly, there is substantial variability between different individuals in the outcome of infection with the same pathogen: spanning life-threatening disease to asymptomatic infection. Broadly, survival from infection requires the correct balance of resistance (ability to control the pathogen load) and tolerance (ability to limit immunopathology in response to a given pathogen load). Host-directed therapies have the potential to restore a resistant or tolerant phenotype on an otherwise susceptible or intolerant host ( Fig. 1 ).

Host-directed immunotherapies. Schematic of the role of resistance and tolerance in host defense against infection, and the role of host-directed therapies to re-calibrate these when they fail. For host health, red shading indicates clinically apparent disease, morbidity, or mortality, whereas green indicates recovery, survival, or asymptomatic infection.

( Created using PowerPoint and GraphPad Prism.)

While conceptually attractive, there are limited examples of host-directed therapies in routine clinical usage. Existing approaches mainly relate to augmenting tolerance through use of anti-inflammatory drugs, for example, immunomodulatory treatment of Coronavirus disease 2019 (Covid-19) using steroids, interleukin 6 receptor blockers, and Janus kinase (JAK) inhibitors (discussed later).

There is an unmet need to develop host-directed therapies for a broader range of indications, especially drugs to augment pathogen clearance (ie, resistance), which is especially important in the context of antimicrobial resistance. Achieving this is complicated by the difficulty in identifying therapeutic targets in severe infection. Host responses are multi-layered, incorporate redundancy, and are heterogeneous between individuals. We lack a granular understanding of which host responses deterministically contribute to successful defense against specific pathogens or cause harmful immunopathology. Innovative research strategies are required to overcome these challenges.

In this article, we will discuss how studying extreme infection phenotypes can provide a solution to this problem. In this context, an “extreme phenotype” broadly identifies the observable outcome of a pathogen:host interaction that is either at the very upper or lower end of the overall distribution, due to either host or pathogen traits ( Fig. 2 ). Extreme host phenotypes include unusual disease severity (high or low), resistance to pathogen replication, or susceptibility to low virulence organisms. While the extreme phenotype concept originates in genetics, this can be extended to any extreme observable phenotype arising as a consequence of an individual’s cumulative environmental exposures across the life course (the exposome), for example, comorbidities and age-related biological changes. Extreme pathogen phenotypes can also be informative, for example, pathogen variants with an unusually high capacity to cause severe disease in otherwise healthy people (ie, hyper-virulent).

Extreme infection phenotypes. Schematic representation of the concept of an “extreme phenotype”: clinical features of infection that are either at the very upper or lower end of the overall distribution, and can be the result of host or pathogen traits. The y -axis indicates the number of organisms with the trait, either the host or the pathogen. The x -axis indicates the impact of the host or pathogen trait on the infection phenotype, representing a spectrum between most and least harm to the host.

( Created using PowerPoint and GraphPad Prism.)

Studying extreme infection phenotypes has the potential to amplify biological signal over noise, facilitating the identification of host responses mediating disease severity (eg, immunopathology) or protection (eg, pathogen clearance). Host responses identified in this way are rationale candidates for host-directed therapeutic targeting.

The concept of selectively studying extreme phenotypes is well established in the field of genetics, where extreme phenotype sampling is a method used to increase statistical power for identifying genetic associations compared with random sampling of all individuals with the phenotype. This approach is the foundation of the GenOMICC study ( https://genomicc.org/ ), which recruits individuals with critical illness caused by diseases of interest, rather than individuals with less severe manifestations. Infection with SARS-CoV-2 results in a wide range of outcomes, including asymptomatic infection, a mild respiratory tract infection, or viral pneumonitis. The latter, now largely restricted to unvaccinated or immunocompromised individuals, also presents as a spectrum, with a small proportion of individuals requiring critical care, thus representing an extreme phenotype of SARS-CoV-2 infection. GenOMICC performed several genome wide association studies (GWAS), ^,, comparing critical Covid-19 to controls, and identifying an unprecedented number of genotype-phenotype associations for an infectious disease. These included variants in genes encoding effectors of the host anti-viral responses (eg, IFNAR1 , IFNAR2 ) and regulators of myeloid cell inflammation (eg, TYK2 , JAK1 , DPP9 ). Orthogonal evidence from studies of bronchoalveolar lavage fluid, circulating inflammatory mediators, ^, peripheral blood mononuclear cells, and tissue immunopathology ^,,,,, concurrently identified a central role for macrophage-mediated inflammation in the pathogenesis of critical Covid-19. Specifically, expansion of pulmonary interstitial monocyte-derived macrophages was observed in critical Covid-19, occurring to a greater extent than in bacterial pneumonia or influenza virus infection. Activation signals included M-CSF, GM-CSF, IL-6, and C5a, and recruitment axes included CCR2::MCP-3/MCP-1 and C5AR1::C5a. The finding that dexamethasone reduced mortality in individuals with Covid-19 associated with hypoxia confirmed that suppressing the inflammatory response was beneficial. A humanized mouse model provided evidence this benefit was through augmenting tolerance by limiting the expansion of interstitial monocyte-derived macrophages identified in humans. In this model, dexamethasone reduced inflammatory macrophage recruitment and weight loss but increased the viral load. Signaling through Janus kinase (JAK) family tyrosine kinases is a major pathway in pro-inflammatory macrophage activation, so the identification of the association of genetic variants in the JAK kinases TYK2 and JAK1 with critical Covid-19 supported the re-purposing of the JAK inhibitor baricitinib. In particular, genetic evidence identified that increased TYK2 expression was associated with critical Covid-19. Baricitinib treatment was found to reduce 28-day mortality by 13% and remained beneficial even for individuals receiving corticosteroids or IL-6 receptor blockers. This example provides important proof of principle that insights from studying extreme infection phenotypes can directly translate into therapeutically actionable information.

The concept of HIV elite controllers provides another example of how studying extreme phenotypes may ultimately lead to therapeutic insights. Elite controllers are individuals able to suppress HIV-1 replication to undetectable or very low levels in plasma in the absence of anti-retroviral therapy, indicating unusually successful host defense. This phenotype is identified in less than 1% of people living with HIV. This phenotype can be conferred to otherwise susceptible individuals, as demonstrated by lymphocyte transfer from an elite controller to a recipient with progressive HIV infection, which was associated with a transient reduction in viral load before clearance of the donor cells.

Several facets of this extreme infection phenotype are informing development of lymphocyte-targeted immunotherapies for HIV-1 infection. Elite controllers are characterized by distinct CD8 T-cell effector phenotypes and resistance to exhaustion, ^,, identifying immune checkpoint inhibitors as a candidate approach with pre-clinical data suggesting this could enhance vaccine-induced CD8 T-cell responses. In addition, CD4 T cells from elite controllers exhibit up-regulation of the cyclin-dependent kinase inhibitor p21, which promotes resistance to viral replication, identifying CDK inhibitors as a therapeutic strategy. ^,

Exposure to non-tuberculous mycobacteria (including the attenuated Mycobacterium bovis used in BCG vaccination), Candida spp., and non-typhoidal Salmonella spp. generally does not result in clinically significant disease among immunocompetent individuals. However, invasive, recurrent, or chronic infections with these organisms is an identifier of Mendelian susceptibility to mycobacterial disease (MSMD). MSMD represents an extreme infection susceptibility phenotype, caused by mutations in one of greater than 20 components of the interferon gamma response (including IFNG , IFNGR1 , IFNGR2 , IL12 B , IL12RB1 , IL12RB2 , IL23R ), predominantly acquired through autosomal recessive inheritance. As a result of impaired interferon gamma production or signaling (or both), affected individuals have a defect in macrophage intracellular killing of specific pathogens. In addition to the weakly virulent organisms already mentioned, this susceptibility appears to extend to additional intra-macrophagic organisms such as Histoplasma and Leishmania . Importantly, identifying that defective interferon gamma signaling is causal in the MSMD phenotype means that treatment with exogenous interferon gamma can be offered to patients with mutations resulting in impaired production. Although this strategy has not been evaluated in a clinical trial, in part due to the rarity of MSMD, cumulative evidence suggests there is clinical benefit. ^,,, Furthermore, by identifying interferon gamma as a key mediator of host resistance to mycobacterial disease, adjunctive interferon gamma has been evaluated as an adjunct to antimycobacterial treatment in patients without MSMD, with evidence from a small clinical trial indicating there may be clinical benefit. This example highlights that defining the mechanistic basis of extreme susceptibility to specific pathogens can identify non-redundant elements of the host response against that pathogen that could be targeted using host-directed anti-infective therapies.

The immune system exerts demonstrable selective pressure on pathogen evolution, resulting in the emergence of immune-adaptive pathogen variants. For example, HIV-1 escape from CD8 T-cell recognition, host-adaptation of Mycobacterium abscessus spp. and S. aureus during chronic endobronchial infection, ^, and differential S aureus toxin specificity in response to inter-species diversifying immune selective pressure. Experimental passage of Legionella pneumophila or S aureus through macrophages in cell culture results in emergence of adaptive mutants with enhanced intra-macrophagic survival. Therefore, within the genetic heterogeneity naturally present in all bacterial species, certain adaptations will represent a footprint of immune selective pressure exerted by the host. Such variants will be resistant to specific host defense mechanisms (eg, cytotoxic T-cell recognition, macrophage intracellular killing) and be identifiable by their association with distinct clinical phenotypes during infection, for example, hyper-virulence ( Fig. 3 A ), thus representing extreme pathogen phenotypes. While more challenging to identify than extreme host phenotypes, studying extreme pathogen phenotypes also has the potential to deliver therapeutic insights. Specifically, by asking the question “ which host responses has a hypervirulent pathogen variant evolved to escape ?” Immune-adaptive pathogen variants could be used as tools to identify elements of the host response as therapeutic targets.

Extreme pathogen phenotypes. ( A ) Identification of immune-adaptive pathogen variants. ( B ) Invasive disease potential of pneumococcal capsular serotypes. Data from reference. The graph shows the OR of isolates recovered from invasive disease versus carriage, stratified by capsular serotype. An OR greater than 1 ( dotted line ) indicates a serotype is more likely to be associated with invasive disease than carriage, identifying the capacity of serotype 1 pneumococci to cause invasive disease as an extreme phenotype. OR, odds ratio.

(Created in BioRender. Russell, C. (2026) https://BioRender.com/9xxegiv .)

Serotype 1 S pneumoniae are associated with disease that is clinically and epidemiologically distinct from pneumococcal disease due to other serotypes, representing putative adaptive variants. They exhibit a strong association with invasive pneumococcal disease (including extra-pulmonary) in young healthy hosts including empyema, peritonitis and salpingitis ( Fig. 3 B). ^,, A further characteristic is their ability to cause disease outbreaks, often in conditions of close living such as prisons and schools, but also epidemics of meningitis in sub-Saharan Africa. ^,,,, Overall, serotype 1 pneumococci represent an extreme pathogen phenotype, behaving as hyper-virulent primary pathogens whereas most other pneumococci are opportunists. In a study providing proof of principle of this approach, clinical isolates of serotype 1 S pneumoniae were identified with an unusual resistance to macrophage intracellular killing. This led to the identification of macrophage microbicidal responses these variants escaped from, which were confirmed to be key mediators of normally successful bacterial killing against S pneumoniae and related gram-positive organisms ( S agalactiae , S pyogenes , and Enterococcus faecium ). Specifically, P2RX7 signaling was identified as a host defense factor that could be augmented in mouse and zebrafish larvae models by re-purposing of the anti-histamine clemastine. Treatment with clemastine led to pathogen clearance without the requirement for conventional antimicrobials. This example shows that extreme pathogen phenotypes, characterized by hyper-virulence, can be used to pinpoint elements of host defense that normally control the pathogen and can be therapeutically targeted. While the genetic basis for serotype 1 S pneumoniae hypervirulence has not been established, distinct gene expression profiles have been described compared with less virulent isolates. Several other examples of extreme pathogen phenotypes exist and could be leveraged to define host defense mechanisms, for example,.

• •

  S aureus USA300 North American epidemic clone, associated with necrotizing pneumonia, severe skin and soft tissue infections, and metastatic complications during bacteraemia. ^,,

• •

  Hypermucoviscous Klebsiella pneumoniae (hvKp), associated with hematogenous liver abscesses and endogenous endophthalmitis.

While the genetic basis for these extreme pathogen phenotypes remains to be confirmed (and may be context specific), candidate genes exist for both examples: the Panton-Valentine leucocidin (PVL) toxin, enterotoxin variants, and the arginine catabolic mobile element for USA300, and the regulator of mucoid phenotype genes (eg, rmpA ) for hvKp. More definitively, a bacterial GWAS has established the role of PVL in tropical pyomyositis caused by S aureus .

The examples considered so far rely on clinical, epidemiological, and microbiological observation to identify extreme phenotypes to study. Approaches for patient stratification also offer opportunities to identify extreme infection phenotypes and may be particularly well suited to infections or syndromes with a high degree of clinical heterogeneity. For example, critical illness in COVID-19 is a clearly identifiable, homogenous sub-group. In contrast, identifying an extreme phenotype within a more heterogeneous syndrome is more challenging and well exemplified by the difficulties identifying therapeutically-relevant sub-groups within sepsis (comprehensively reviewed in ). Important proof of principle comes from the similarly heterogeneous syndrome of acute respiratory distress syndrome, where two disease subphenotypes (hyper-inflammatory and hypo-inflammatory) can be reproducibly identified in different cohorts, using routinely available clinical variables, associated with differences in outcome and also response to ventilation strategy and simvastatin. ^, An inevitable tension exists in efforts to classify patients into subphenotypes: clinical applicability of the approach (ie, can the relevant classifying measurements be made in routine clinical practice) versus the depth of biological insight (using deep phenotyping methods not routinely available). While decades of sepsis immunobiology have established that some patients suffer organ dysfunction due to excessive inflammation (loss of tolerance), whereas others exhibit immune exhaustion and susceptibility to secondary infections (loss of resistance), until recently this had not led to successful drug discovery. However, a recently reported approach for patient stratification in sepsis has succeeded in identifying a clinically applicable and therapeutically relevant approach, using measurement of circulating ferritin and HLA-DR expression on CD45/CD14 monocytes. Patients with ferritin greater than 4420 ng/mL were classified as having a hyper-inflammatory macrophage-activation like syndrome subphenotype, and patients with ferritin less than or equal to 4420 ng/mL and monocyte HLA-DR expression less than 5000/cell were classified as having sepsis-induced immunoparalysis subphenotype. ^, These represent extreme phenotypes within sepsis, and patients with ferritin less than or equal to 4420 ng/mL but HLA-DR expression greater than or equal to 5000/cell were considered unclassifiable and not belonging to one of these extremes. The macrophage-activation like phenotype was treated with the interleukin-1 receptor antagonist anakinra and the sepsis-induced immunoparalysis phenotype was treated with recombinant interferon-gamma. In a randomized, placebo-controlled trial of this approach, personalized host-directed immunotherapy was associated with an improvement in day 9 Sequential Organ Failure Assessment score. This is a particularly noteworthy finding considering the large number of negative trials investigating immunotherapy in sepsis. This example highlights the role of biomarker-based patient stratification in defining extreme phenotypes that would otherwise not be observable.