Personalizing the Management of Pneumonia Samir Gautam, MD, PhD, Lokesh Sharma, PhD, Charles S. Dela Cruz, MD, PhD* INTRODUCTION Lower respiratory tract infections (LRTIs) are the leading cause of death in developing countries and account for more than 4 million deaths per year worldwide. 1 They result in the loss of 103,000 disability-adjusted life years annually, making pneumonia the single greatest contributor to human disease burden. 2,3 It is astonishing, therefore, that diagnosis of pneumonia in most cases (even at academic centers) still relies on decades-old and highly unreliable clinical criteria such as the chest radiograph, 4 which has a sensi- tivity less than 50% and positive predictive value less than 30%. 5 The difficulty only increases in pa- tients with underlying cardiopulmonary disease or immunosuppression; two of the populations at highest risk of death from LRTI. Microbiological culture, another pillar of pneumonia diagnosis, is similarly faulty, as it reveals a pathogen in less than half of cases. 6,7 In the absence of dependable diagnostic guide- posts, clinicians faced with any suspicion of pneumonia have traditionally resorted to treating with empiric broad-spectrum antibiotics ‘just to be safe’. However, this time-worn adage is finally being questioned, as data have accumulated to show the danger of indiscriminate antimicrobial use both to society and to individual patients. On a population level, antibiotic administration for suspected respiratory infection is now appre- ciated as a major driver of antibiotic resis- tance, 2,8,9 which in turn has been identified by the World Health Organization (WHO) as one of the biggest global threats to human health. 10 Meanwhile, the harm of inappropriate antibiotics to patients is also becoming recognized. 11 In addition to the risk of allergy and drug toxicity, Disclosure: This work was supported by grants from the NHLBI (HL126094 and HL103770 to CSD and T32- HL007778 to SG). Pulmonary Critical Care and Sleep Medicine, Center for Pulmonary Infection Research and Treatment, Yale Uni- versity, 300 Cedar Street, TACS441, New Haven, CT 06520-8057, USA * Corresponding author. E-mail address: [email protected]KEYWORDS Pneumonia Personalized Precision Individualized Immunomodulation Antibiotic resistance KEY POINTS The current approaches to diagnosing pneumonia and identifying pathogens rely on antiquated methods that have poor test characteristics. Treatment strategies are similarly crude because they rely on broad-spectrum empiric antibiotics, which promotes antimicrobial resistance, and in some cases steroids, which have numerous un- wanted side effects. Emerging genomic methods have the capability to improve microbiologic diagnosis and assess- ment of host immune responses. This information may enable the formulation of personalized treatment of patients, featuring highly selective antimicrobials and targeted immunomodulation. Clin Chest Med 39 (2018) 871–900 https://doi.org/10.1016/j.ccm.2018.08.008 0272-5231/18/Published by Elsevier Inc. chestmed.theclinics.com
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Personalizing theManagement of
Pneumonia Samir Gautam, MD, PhD, Lokesh Sharma, PhD,Charles S. Dela Cruz, MD, PhD*
� The current approaches to diagnosing pneumonia and identifying pathogens rely on antiquatedmethods that have poor test characteristics.
� Treatment strategies are similarly crude because they rely on broad-spectrum empiric antibiotics,which promotes antimicrobial resistance, and in some cases steroids, which have numerous un-wanted side effects.
� Emerging genomic methods have the capability to improve microbiologic diagnosis and assess-ment of host immune responses.
� This information may enable the formulation of personalized treatment of patients, featuring highlyselective antimicrobials and targeted immunomodulation.
INTRODUCTION
Lower respiratory tract infections (LRTIs) are theleading cause of death in developing countriesand account for more than 4 million deaths peryear worldwide.1 They result in the loss of103,000 disability-adjusted life years annually,making pneumonia the single greatest contributorto human disease burden.2,3 It is astonishing,therefore, that diagnosis of pneumonia in mostcases (even at academic centers) still relies ondecades-old and highly unreliable clinical criteriasuch as the chest radiograph,4 which has a sensi-tivity less than 50% and positive predictive valueless than 30%.5 The difficulty only increases in pa-tients with underlying cardiopulmonary disease orimmunosuppression; two of the populations athighest risk of death from LRTI. Microbiologicalculture, another pillar of pneumonia diagnosis, is
Disclosure: This work was supported by grants from thHL007778 to SG).Pulmonary Critical Care and Sleep Medicine, Center for Puversity, 300 Cedar Street, TACS441, New Haven, CT 06520* Corresponding author.E-mail address: [email protected]
Clin Chest Med 39 (2018) 871–900https://doi.org/10.1016/j.ccm.2018.08.0080272-5231/18/Published by Elsevier Inc.
similarly faulty, as it reveals a pathogen in lessthan half of cases.6,7
In the absence of dependable diagnostic guide-posts, clinicians faced with any suspicion ofpneumonia have traditionally resorted to treatingwith empiric broad-spectrum antibiotics ‘just tobe safe’. However, this time-worn adage is finallybeing questioned, as data have accumulated toshow the danger of indiscriminate antimicrobialuse both to society and to individual patients.On a population level, antibiotic administrationfor suspected respiratory infection is now appre-ciated as a major driver of antibiotic resis-tance,2,8,9 which in turn has been identified bythe World Health Organization (WHO) as one ofthe biggest global threats to human health.10
Meanwhile, the harm of inappropriate antibioticsto patients is also becoming recognized.11 Inaddition to the risk of allergy and drug toxicity,
e NHLBI (HL126094 and HL103770 to CSD and T32-
lmonary Infection Research and Treatment, Yale Uni--8057, USA
antibiotics produce profound and lasting alter-ations in the microbiome of the gut and lung(dysbiosis),12 which manifest overtly through sec-ondary infections such as Clostridium difficileinfection (CDI) but also more subtly throughalterations in host response to infection and con-tributions to diabetes mellitus, atherosclerosis, in-flammatory bowel disease, and asthma.13,14 It islikely that these direct hazards to the patientwill serve as a greater deterrent to antibiotic over-use than less tangible risks such as breedingresistance.If the current methods for diagnosing and man-
aging pneumonia are inadequate, what are the al-ternatives? In general, there are 2 strategies. Thefirst relies on guidelines, such as those put forthby the Infectious Diseases Society of America(IDSA) and American Thoracic Society (ATS)for community-acquired pneumonia (CAP) andfor hospital-acquired pneumonia (HAP) andventilator-acquired pneumonia (VAP).15,16 Theseguidelines synthesize the best available data intoan evidence-based approach for the managementof pneumonia, with goals of simplicity and the abil-ity to generalize. These goals are in part born of ne-cessity, because guidelines must be accessible tononspecialists, but it is also an inescapable conse-quence of the large, unstratified patient popula-tions in studies that inform the guidelines.Recommendations are similarly monolithic andtherefore have limited relevance for uncommon in-fections and unique hosts, such as those withcompromised immune or cardiopulmonary func-tion. Diagnosis and management become signifi-cant challenges in these patients, especiallyduring critical illness. In such cases especially,an alternative strategy is needed, one thatcombines greater diagnostic granularity with indi-vidually tailored therapy: so-called personalizedmedicine.The first step toward personalized medicine is
refinement of diagnostic categories. For pneu-monia, this requires subclassification of the syn-drome, which is currently defined broadly by (1)evidence of systemic infection (leukocytosis, fe-vers, or chills), (2) respiratory symptoms (dys-pnea, cough, sputum), and (3) new radiographicinfiltrates.17 This highly inclusive entity could,for instance, be divided into viral pneumonia,bacterial pneumonia, and noninfectious respira-tory disease using additional diagnostic tech-niques (eg, biomarkers)- a preliminary degree ofendotyping referred to as stratified medicine.18
Such subgroups remain large enough to enablewell-powered clinical trials and, thus, endotypingat this level may leverage conventional evidence-based medicine to guide management.
However, the full realization of personalizedmedicine requires a complete delineation of dis-ease mechanisms, advanced diagnostics forinterrogating these mechanisms in patients, andtargeted therapies for modulating them. Theclosest approximation to this vision is inoncology, where tumors are sequenced to iden-tify driver mutations for selective targeting (eg,with tyrosine kinase inhibitors), and the host im-mune response is assessed to determine candi-dacy for checkpoint inhibitors. Thus, the field isbeginning to adopt a “tissue-agnostic” approach,wherein therapy is guided not by histologicallydefined tissue of origin but by the molecularbiology and immunology of the tumor and host;a dramatic departure from traditional oncologicmanagement.Respiratory infection has much to learn from
this paradigm. For pneumonia, personalizationwould require a comprehensive moleculardescription of the pathogen, the host, and theimmunologic phenomena that stem from theirinteraction. This description will allow manage-ment decisions to be determined by not onlydata from empiric trials but also basic microbio-logical and immunologic principles. Examplesinclude delivery of highly selective antimicrobialsbased on pathogen taxonomy and susceptibility,and rational manipulation of dysregulated hostresponses to promote pathogen clearance andlimit immunopathology.A useful framework for understanding the com-
plex interplay of host and pathogen, based on theconcepts of resistance and tolerance, has beendefined by Ayres and Schneider19,20 (graphicallyrepresented in Fig. 1). Resistance refers to thehost’s ability to clear microbes, whereas toler-ance is a term borrowed from ecological immu-nology that describes the host’s ability toendure a microbial insult. Resistance comprisesthe host’s defenses against an invading path-ogen, including intrinsic epithelial mechanisms,innate immunity, adaptive responses, and othersas described later. Tolerance is influenced by amore varied set of factors, including the patho-logic consequences of immune effectors (eg,reactive oxygen species [ROS] released frominfiltrating neutrophils) as well as mechanismsunrelated to resistance (eg, myocardial infarctionin patients with influenza infection).21 To eliminateconfusion with the traditional concept of immu-nologic tolerance, this article refers instead toresilience, following the example of Mizgerd andcolleagues.22 Using this this framework, 4 phasesin the personalized diagnosis and managementof pneumonia can be described (summarized inFig. 1).
A
C D
B
Fig. 1. The 4 clinical phases in the management of acute pneumonia. (A) Initial clinical assessment. In this phase,the patient’s baseline level of health and immunologic competency are assessed. Simultaneously, rapid identifi-cation of pathogen is pursued. Health (dependent variable) is a conceptual term that reflects the clinical status ofthe patient: a composite of criteria including hemodynamic stability and organ function. With increasing path-ogen burden (independent variable), health declines. (B) Host response to pathogen. The slope of the curve isdetermined by tissue resilience, which depends on characteristics of the host, the pathogen, and their interaction.For instance, host resilience to Pneumocystis jiroveci pneumonia (PJP) is substantially increased in AIDS, permit-ting a high pathogen burden with little disorder but dramatically decreases in the setting of immune reconstitu-tion. Pathogen virulence also affects resilience and therefore the slope of the curve. The rapidity of decline alongthe curve (over time) is determined by the adequacy of host resistance. (C) Personalized therapy. Based on acomprehensive characterization of the host, pathogen, and their interaction, a treatment regimen consistingof antimicrobials and immunomodulators is administered. Antimicrobial therapy may have several consequences,including uncomplicated resolution of infection, pathogen killing complicated by immunologic disorder (eg, theJarisch-Herxheimer reaction), and drug toxicity. Pathogen killing and drug toxicity diminish health independentlyfrom the pathogenesis of infection, thus leading to deviation from the original curve. Immunosuppression alonemay improve health but increase pathogen burden. (D) Consequences of infection. In patients who survive infec-tion, pathogen burden returns to zero, but the effects of illness and treatment may produce a new, compromisedstate of health, which may manifest as permanent dysfunction (eg, scarring of lung parenchyma), potentiallytreatable conditions (such as dysbiosis), and/or increased susceptibility to secondary infection (compensatory anti-inflammatory response syndrome [CARS]). IRIS, immune reconstitution inflammatory syndrome. (Data from AyresJS, Schneider DS. Tolerance of infections. Annu Rev Immunol 2012:30;271–94.)
Personalizing the Management of Pneumonia 873
PHASE I: CHARACTERIZATION OF HOST ANDPATHOGENCharacterization of Host
Physiologic reserveOne of the clinician’s first priorities when encoun-tering a patient with pneumonia (or any patient) isto estimate the patient’s baseline level of health,which in turn helps determine the likelihood of
their ability to survive the disease. This physio-logic reserve (depicted as the Y intercept inFig. 1A), is a composite of several parameters,including age and premorbid organ function.For instance, decreased FEV1 (forced expiratoryvolume in 1 second) diminishes the patient’sability to withstand an additional insult tolung mechanics. Likewise, impaired cardiac
Gautam et al874
performance or coronary artery patency predis-poses to heart failure and myocardial infarction,respectively. Such compromise of physiologicreserve should prompt more aggressive care,but this often manifests as broader-spectrumand less-judicious administration of antibiotics,a mistake partly based on the faulty assumptionof antibiotic safety. The authors propose thataggressive care should instead translate intomore comprehensive diagnostic characterizationof host and microbe, which in turn enables amore highly individualized therapeutic plan thatmaximizes efficacy and minimizes side effects.The initial clinical assessment also includes anappraisal of disease severity, as this guidesclinical triage and immediate management. How-ever, since this is a manifestation of the hostresponse to pathogen, it is dealt with in relationto phase II.
Box 1Innate immune responses in pneumonia
Innate immunity in the lung (reviewed in detail elsewpathology; as such, a description of its basic mechanidirected diagnostics and therapeutics, fundamental cment. In brief, alveolar macrophages and respiratorypathogens. When these defenses are overwhelmedreleased from damaged parenchyma and pathogeninnate immune signaling pathways via PRRs, which aThis process leads to the release of chemokines and ctrophils and exudative fluid into the interstitium and
Innate immune signaling: Viral nucleic acids activateepithelial cells, including the cytosolic RIG-I-like recepTLRs (eg, 3, 7, 8, and 9). Recognition of viral PAMPs bduction of type I IFNs, which promote antiviral defengenerate cytokines such as tumor necrosis factor (TNtype I and III innate lymphoid cells (ILCs) and initiaType I ILCs (similar to Th1 cells) are characterized by(similar to Th17 cells) generate IL-17 and IL-22.
In contrast, PAMPs derived from extracellular bacteriaactivate cell-surface PRRs, including TLRs (1, 2, 4, and 62, and mincle), leading to production of an overlapIL-12, and IL-23) but less type I IFN. Similar signals decytosolic recognition of bacterial PAMPs.
Sequelae of pulmonary inflammation: Ideally, recruitleads to pathogen clearance with little collateral dasponses can be deleterious because they may induceprotease release), alveolar edema, and progression tothe acute inflammatory phase must be tightly controlmechanisms mediate the resolution of lung inflammaantiinflammatory innate signaling (thereby limitingapoptotic neutrophils via efferocytosis (see the text c
It is worth noting that the pathogen may benefit fromprovides critical nutrients for further proliferation; thmune responses in certain infections, which would otnor the pathogen.
ResistanceAs defined earlier, resistance refers to the host’sability to clear a pathogen load. This term com-prises not only the innate and adaptive immunemechanisms enumerated in Box 1 but numerousother parameters, including adequacy of cough,ciliary clearance, mucus quality (influenced bypericiliary pH and mucins), release of antimicrobialpeptides and opsonins, ROS production, and thebarrier function that prevents invasive infection.25
Opportunities to therapeutically modulate thesemechanisms are touched on in relation to phaseIII and reviewed elsewhere.25 Graphically, inade-quate resistance leads to a more rapid declinedown the curve depicted in Fig. 1B.Swift recognition of an immunocompromised
state is critical, because it informs triage andempiric antibiotic therapy against unique patho-gens to which a host may be susceptible. This
here23) mediates both host defense and immuno-sms is essential to understand the targets of host-omponents of personalized pneumonia manage-epithelial cells collaborate to clear low levels of
, danger-associated molecular patterns (DAMPs)-associated molecular patterns (PAMPs) activatere expressed by both epithelial and immune cells.ytokines, which induce the extravasation of neu-airspaces.
PRRs in dendritic cells, macrophages, and alveolartors and AIM-like receptors, as well as endosomaly plasmacytoid dendritic cells (DCs) leads to pro-ses, whereas macrophages and conventional DCsF)-a, IL-1b, IL-6, IL-12, and IL-23, which stimulatete T helper (Th) 1 and Th17 adaptive responses.robust production of IFNg, whereas type III ILCs
(eg, lipopolysaccharide) and fungi (eg, b-glucan)) and C-type lectin receptors (eg, dectin-1, dectin-ping set of cytokines (including TNFa, IL-1b, IL-6,rive from the NOD-like receptors, which mediate
ment of neutrophils and other immune effectorsmage. However, overexuberant neutrophilic re-parenchymal destruction per se (eg, via ROS andacute respiratory distress syndrome.24 Therefore,led in terms of both severity and duration. Severaltion, including a switch from proinflammatory tofurther leukocyte recruitment) and clearance ofoncerning phase IV).
immune-mediated tissue destruction because itis may help explain the presence of pathologic im-herwise serve as an advantage to neither the host
Personalizing the Management of Pneumonia 875
point is most dramatically shown by the 1-hourdoor-to-needle time recommended for administra-tion of antipseudomonal antibiotics in patients withneutropenic fever, who may rapidly decompen-sate and die within hours from gram-negativerod bacterial infection if not treated promptly(discussed later).26 However, subtler examplesinclude hypogammaglobulinemic patients, whomay require adjunctive therapies such as intrave-nous immunoglobulin,27 and those with altered im-mune responsiveness caused by pathogenrecognition receptor (PRR) polymorphisms. Thissection describes some of the mechanisms thatlead to impaired host resistance, both geneticand acquired. Characterizing these defects in indi-vidual patients would facilitate personalized ther-apy for acute pneumonia in several ways:through predicting severity of disease course, indi-cating pneumonia susceptibilities that will guideempiric antimicrobial coverage, and identifyingdeficient host immune pathways that may be ther-apeutically enhanced.
Genetic determinants of reduced resistanceThe study of rare patients with primary immunode-ficiencies helps to elucidate the pathogenesis ofhuman infections, as shown by the case of a youngchild with interferon (IFN) regulatory factor 7 (IRF7)deficiency and severe influenza.28 This findingconfirmed the putative role of the IRF7 pathwayin the generation of protective type I IFN duringinfluenza infection suggested by prior animalstudies. However, to explain the interindividualvariability of pneumonia severity observed in thegeneral population, it is more valuable to identifycommon and benign genetic variants that influ-ence disease susceptibility.29 Polymorphismslinked to influenza and legionella infections thattake particularly variable clinical courses are high-lighted here.
The best-studied genetic determinant of influ-enza susceptibility is IFN-induced transmembraneprotein 3 (IFITM3), which associates with the en-dosome to block cytosolic delivery of the genomeof RNA viruses, a necessary step in their replica-tion.30 IFITM3 also plays a role in IRF3 activationand persistence of memory T cells within thelung.31 Through these mechanisms, IFITM3 poly-morphisms impair tissue resistance, leading tohigher viral burdens and worse clinical out-comes.32 For instance, the C allele produces se-vere disease when homozygous33 and is fairlyprevalent, especially in Asian people, where it isobserved in more than 50% of the Han Chineseand Japanese populations.29,34 Several smallerstudies have identified additional disease-associated single nucleotide polymorphisms
(SNPs; reviewed elsewhere35), but their clinicalsignificance is not yet clear; it is likely that influenzasusceptibility is a complex trait influenced byseveral of these loci.
Legionella susceptibility has been linked toSTING (Stimulator of IFN Genes, encoded byTMEM173/STING), an adaptor protein down-stream of cGAS and IFI16, innate immune sensorsof cytosolic DNA. Activation of this pathway elicitsa type I IFN response important for host defenseagainst viruses and certain bacteria, includingLegionella.36 Human TMEM173/STING showsconsiderable interindividual variability; forinstance, 20% of the population in the 1000 Hu-man Genome Project database express the HAQallele, which contains 3 nonsynonymous substitu-tions.37 Recently, Ruiz-Moreno and colleagues38
showed that carriage of this variant is associatedwith heightened susceptibility to legionella pneu-monia in humans. Mutations in toll-like receptors(TLRs) have also been shown to affect Legionellasusceptibility, as TLR5 truncation (affecting a sur-prising w10% of individuals) and TLR2 mutationlead to increased risk,39,40 whereas certain TLR4polymorphisms are protective.41
Genetic risk modifiers for CAP (irrespective ofetiology) have also been described. For instance,a genome-wide association study (GWAS) identi-fied several common variants in the FER gene (acytosolic tyrosine kinase that contributes toneutrophil recruitment and endothelial perme-ability) that afford marked protection from deathfrom pneumonia.42 A deleterious polymorphismin interleukin (IL)-6 and a protective SNP withinIL-10 have been recognized as well.40 Regardingnoninfluenza viral pathogens, susceptibility to se-vere rhinovirus infection in children has beenlinked to a variant of cadherin-related family mem-ber 3 (CDHR3; the receptor for rhinovirus-C),43–45
and several genetic risk factors have been identi-fied for pediatric respiratory syncytial virus (RSV)infection; because the focus is on adult diseasehere, readers are referred to recent reviews onthese topics.46,47 In addition, although not all arespecifically related to pneumonia, a plethora ofadditional polymorphisms in PRRs have beenshown to predispose to viral, mycobacterial, andfungal infections (reviewed in Refs.48,49).
The potential clinical utility of identifying suscep-tibility loci in patients with pneumonia is significant.In the acute setting, as noted earlier, such datacould improve prognostication, guide the individu-alization of empiric antibiosis, and identify therapiesthat can augment defective resistance mecha-nisms. Furthermore, identification of high-risk pa-tients could inform preventive strategies, includingmore aggressive vaccination, counseling on
Gautam et al876
exposure avoidance, and prompter administrationof antimicrobial prophylaxis after exposure (eg,oseltamivir for influenza). These measures areconsidered later in relation to phase III.
Acquired defects in resistanceCertain forms of acquired immunocompromise,such as hypogammaglobulinemia, neutropenia,hematologic malignancy, steroid use, and ac-quired immunodeficiency syndrome (AIDS), arereadily recognized on history and basic laboratorystudies and cue clinicians to consider pertinentclinical syndromes, such as Pneumocystis jirovecipneumonia (PJP) in AIDS. This level of personal-ized therapy is well established in clinical practiceand needs no further elaboration here. However,more common conditions, such as diabetes,chronic kidney disease (CKD), cirrhosis, alco-holism, smoking, and advanced age (immunose-nescence), also increase risk of pneumonia but inless definable ways. Predisposition to LRTI isalso influenced by transient risk factors, such asair pollution,50 intercurrent viral infections,51 sepsis(discussed later relation to phase IV), and anti-biotic use. In addition, omission of vaccinationand/or waning immunity caused by remote vacci-nation represent, in effect, missed opportunities toimprove resistance.Is it possible for clinicians to comprehensively
catalog all of the resistance deficits present in agiven patient, quantify their individual effects, inte-grate their collective impact, and use this informa-tion tomeaningfully guide clinical management? Atpresent, the answer is clearly no, but this a prioriapproach is not the only means of assessing a pa-tient’s immunocompetence. An alternative, orcomplementary, strategy is to interrogate patientimmune responsiveness directly, using in vivo orex vivo assays. A well-known example of theformer is the tuberculin hypersensitivity test, whichreports on T-cell reactivity.52 Quantifying surfacemarkers of T-cell exhaustion, a phenomenonobserved in sepsis that predisposes to secondaryinfection (discussed further in relation to phaseIV), has also been explored as amethod for assess-ing adequacy of T-cell immunity in the clinicalsetting.53 Another conceivable approach is tran-scriptomic analysis of local immune responses atthe respiratory epithelium. This approach may beparticularly useful in the context of active infection,as defective resistance mechanisms could beidentified in situ and targeted for therapy. Func-tional assays such as these effectively integrate ge-netic and acquired defects in resistance and mayprovide clinicianswithmore concrete data than pa-tient history alone to guide empiric antibiosis,immunomodulation, and preventive strategies.
Characterization of Pathogens
First introduced in the nineteenth century, plate-based microbiological culture remains the goldstandard for identifying bacterial and fungalpathogens in the lung and for determining theirantimicrobial sensitivity. However, it has 2 majordrawbacks: long turnaround times (>36–48 hours)and poor sensitivity.54 The former requires at least2 days of empiric antibiotics, with all of the atten-dant risks enumerated later in relation to phase III,and even then antimicrobial coverage may missthe offending pathogen (eg, in the case of an unex-pected multi-drug resistant [MDR] organism). Theprolonged incubation times required for fungal cul-tures (often >2 weeks) create further risks for inad-equate antibiotic therapy, as empiric antifungalagents are rarely used outside of neutropenia.55
The inadequate sensitivity of culture was shownby a landmark study in patients with CAP, whichshowed that conventional culture failed to providea diagnosis in more than 60% of patients despiteaddition of an extensive list of infectious bio-markers.6 Similarly dismal numbers exist for pa-tients with VAP, in whom more than 50% lack anidentifiable pathogen.7 Reasons for this poorsensitivity include failure of culture to detect fastid-ious organisms and inadequate sampling methods(eg, underuse of invasive techniques). Clearly,improved microbiologic diagnostics are needed.
Invasive samplingDespite decades of debate, the question of whento obtain invasive cultures has not yet beenanswered. With regard to VAP, the European andAmerican guidelines are at odds; the former favorquantitative distal sampling, whereas the latterrecommend semiquantitative endotracheal aspi-ration.16,56 Two of the more commonly cited ran-domized control trials (RCTs) addressing thisquestion are Fagon and colleagues’57 demonstra-tion that bronchoscopy increased antibiotic-freedays, and the Canadian Critical Care TrialGroup’s58 study showing no benefit. This discrep-ancy may be attributable in part to a lack of patientendotyping, and criteria for identifying appropriatepatients for bronchoscopy should be pursued.However, methodological advances since thetime of these trials should also be considered.For instance, combining invasive sampling withnucleic acid–based or mass spectrometry–baseddiagnostics (described later) might allow moreeffective assessment of pathogen identification,burden, and antibiotic sensitivity, and therefore in-crease the efficacy of bronchosopy. More recentstudies have shown the utility of this approach(reviewed in Ref.59).
Personalizing the Management of Pneumonia 877
Nonbronchoscopic sampling (via blind catheter-ization of the lower airways) is an alternative thataddresses many drawbacks of the bronchoscopicapproach.60 These drawbacks include expense,risk to the patient, and the need for highly trainedoperators, which often produces delays thatcompromise the yield of cultures due to antibioticexposure before bronchoscopy. Although notguided specifically toward diseased portions ofthe lung, nonbronchoscopic methods still corre-late well with their bronchoscopy in a range of pa-tient populations and microbiological tests.61–66
Again, the utility of such methods is bound to in-crease when combined with rapid moleculardiagnostics.
Bronchoscopy to diagnose pneumonia in thenonintubated immunocompromised population isanother source of controversy; although it remainsstandard of care, evidence for this practiceremains sparse. A good example comes from pa-tients with hematological malignancies who pre-sent with pulmonary symptoms and/or infiltrates.A multicenter RCT showed that a noninvasivework-up in such patients (including imaging, tradi-tional culture, and biomarkers) was noninferior tobronchoscopy with respect to rates of pathogenidentification.67 The reliance on bronchoscopy inthis population is called further into question bythe impact of invasive procedures on patients’quality of life, particularly at the end stages of dis-ease; a host-specific aspect of personalized med-icine that is often neglected. However, if combinedwith molecular microbiological testing, the yield ofbronchoscopy in the immunocompromised mayimprove substantially.
A theoretic argument against invasive samplingcomes from the cystic fibrosis literature, in whichit has been shown that pathogens from differentparts of the lung may express completely differentresistance patterns, such that sampling error alonemay lead to inappropriate antibiotic selection.68
Similar differences in microanatomic bacterialcommunities have been described for patientswith advanced chronic obstructive pulmonarydisease (COPD).69 Although the results relateless to acute pneumonia, prolonged residence inan intensive care unit (ICU), and the consequentacquisition of multiple MDR strains, couldconceivably produce similar spatial heterogeneity.
Matrix-assisted laser desorption ionizationtime of flight mass spectrometryIn contrast with the conventional approach to mi-crobial recognition using colony appearanceon culture plates, matrix-assisted laser desorptionionization time of flight mass spectrometry(MALDI-TOF MS) identifies pathogens via
proteomic profiling. The technique is both(requiring only minutes) and inexpensive on aper-sample basis.70 Furthermore, it identifies notonly pathogens but also certain resistance mech-anisms by detecting products of b-lactam hydroly-sis,71 fluoroquinolone acetylation,72 and proteinsthat mediate resistance (eg, penicillin binding pro-tein 2a [PBP2a], encoded by MecA, which medi-ates methicillin resistance in Staphylococcusaureus).73 Additional techniques are being devel-oped to allow direct assessment of antibioticsensitivity via measurement of stable isotope-labeled amino acid incorporation into proteins74;a surrogate of microbial growth with much fasterkinetics than traditional growth curves.
BiomarkersPathogen-associated biomarkers are the mostwidely used complement to traditional culturetechniques. Although they remain limited in thenumber of pathogens they can detect and cannotoffer insight into antimicrobial resistance, they arewidely available, rapid (because they require nomicrobial culture), inexpensive, and in some caseshighly specific for their targets. The list includesLegionella and pneumococcal urinary antigens,Mycoplasma and Chlamydia antibodies, Histo-plasma urine antigen, galactomannan (associatedwith aspergillosis), and b-glucan (a nonspecificfungal marker). An example of recent progresscomes from Wunderink and colleagues,75 whoshowed a doubling in detection rate of pneumo-coccal CAP with the use of a second urinary anti-gen compared with a conventional assay alone(9.7% vs 5.4%). Although encouraging, it never-theless reveals that the standard urinary pneumo-coccal antigen assay (one of the best and widelybiomarkers) still misses at least 40% of diagnoses,highlighting the need for further work in this area.
Quantitative polymerase chain reactionSimple, rapid assays using quantitative polymer-ase chain reaction (qPCR) may be used to identifypathogens and resistance mechanisms usingprimers designed according to sequenced ge-nomes. Polymerase chain reaction (PCR) isextremely effective for diagnosis of viral infectionand is in common use for detecting respiratory vi-ruses in the upper respiratory tract; a surrogate forLRTI. Several caveats of this technique exist,including the potential dissociation between upperrespiratory tract infection and LRTI and the poorsensitivity for Herpesviridae (eg, herpes simplex vi-rus [HSV] and cytomegalovirus [CMV]). The latter isan important weakness given the potential patho-genic role of these viruses even in immunocompe-tent hosts (discussed later in relation to phase III).
Gautam et al878
PCR has valuable application in the diagnosis ofbacterial infections as well, for instance in identi-fying the MecA gene in S aureus. qPCR can beused to assess relative microbial burdens andpathogen dominance; an important indicator ofpotential pathogenicity, as discussed later. Thisassessment may be achieved by normalizing theamount of pathogen to the total bacterial commu-nity, which is assessed using general bacterialprimers.The rapidity of PCR is one of its greatest
strengths, and could help to remove the need forempiric antibiotics in pneumonia. The presentguidelines recommend antibiotic therapy within4 hours based on data showing increased mortal-ity with delays in therapy (although even these datahave been questioned76)77; thus, the prompt per-formance of a PCR-based diagnostic, which takesw2 hours, may allow immediate delivery of tar-geted antimicrobial therapy.
Metataxonomics and metagenomicsAn important limitation of qPCR is its inapplica-bility to microbes and resistance genes not yetfully sequenced. High-throughput techniques,including 16s (metataxonomics) and whole-genome sequencing (WGS; shotgun metagenom-ics), which offer the ability to define the respiratorymicrobiome in a comprehensive and unbiasedmanner, overcome this hurdle. They also allowthe identification of fastidious organisms such asmycobacteria that grow poorly using conventionalculture.78,79
16s sequencing relies on the use of primersagainst highly conserved sequences in the ribo-somal RNA of bacteria to amplify the variable re-gion of the gene, which in turn is used to identifyindividual taxa. Semiquantitative relative abun-dance may also be assessed. The technique israpid and inexpensive compared with WGS butprovides no insight into nonribosomal genes,including those that mediate resistance and viru-lence. WGS, which completely characterizes thegenomes of recovered microbes, holds the prom-ise of predicting antimicrobial susceptibility, but itis not yet approved for this application.80 Thereason is that the effects of subtle genetic variants(eg, SNPs in antibiotic target genes) on resistancehave not yet been characterized. To address thisissue, research groups are pursuing large GWASanalyses on thousands of clinical isolates to createa comprehensive catalog of resistance loci; thiswill serve as both a reference for patient WGSand a basis for developing models that predictresistance in novel variants.81–83
An additional challenge to overcome, whichaffects both high-throughput sequencing
techniques, is distinguishing colonizer from truepathogen. This challenge is a general caveat forany microbiological diagnostic, even with the far-less-sensitive plate-based culture; an isolatedmicrobe may represent anything from beneficialcommensal to harmful microbiota, to colonizingpathogen, to disease-causing pathogen. One fairlystraightforward method of defining a pathogen isto demonstrate its ecological dominance in therecovered bacterial population. For instance,Wunderink and colleagues84 proposed thefollowing criteria for discriminating pathogenfrom colonizer on bronchoalveolar lavage(BAL) or tracheal aspirate: total bacterial densityof greater than 104 colony-forming units (CFU)/mL, high total bacterial DNA burden, low commu-nity diversity, and a high abundance of thepathogen.85,86
However, complexities arise from interspeciesinteractions, which are known to critically affectthe virulence of a given pathogen. For example,a dominant pathogen may be detectable but notthe source of disease because it is held in checkby 1 or more cocolonizers (protective micro-biota).87 An alternative is to assess for expressionof genes, including virulence factors, which areexpressed only after a microbe makes the pheno-typic switch from colonizer to pathogen.88,89 Trig-gers for this switch include interaction withcommensals, viral infections, cigarette smoking,and air pollution.2 As an example, Molyneauxand colleagues90 showed a significant increasein overall bacterial burden as well as an outgrowthof Haemophilus influenzae specifically in patientswith COPD after rhinovirus infection.Additional complications of metaomics include
turnaround time, risks of contamination, inabilityto discriminate live from dead microbes, theextremely low abundance of microbial DNAcompared with host, and cost. For now, thisapproach may only be applicable to the ICU pa-tients, whose condition is tenuous enough andcare is costly enough to justify the additionalexpense. Patients with chronic respiratory infec-tions represent another potential target.
PHASE II: CHARACTERIZE THE HOSTRESPONSE TO PATHOGEN
As described in Box 1, the central host responseto lung infection is neutrophilic infiltration. Thisresponse explains not only the histopathologichallmark (neutrophilic alveolitis) but also theclassic clinical symptoms (dyspnea, cough, puru-lent sputum), signs (fever and hypoxemia), labora-tory abnormalities (leukocytosis and bandemia),and radiographic findings (infiltrates). However,
Personalizing the Management of Pneumonia 879
the poor specificity of each of these clinicalfeatures leads to the frequent overdiagnosis ofpneumonia and unnecessary administration of an-tibiotics. Identification of respiratory microbes bymeans of the diagnostics described in relation tophase I is helpful but only indicates the presenceof potential pathogen; it does not prove that it iscausing a clinically meaningful infection. In addi-tion, microbiologic cultures, still the gold standarddiagnostic, take days to mature. Thus, it is essen-tial to develop more sophisticated methods forinterrogating host responses that will (1) enableaccurate and rapid identification of patientswith pneumonia; and (2) discriminate betweenbacterial, viral, and other pathogen classes toguide empiric antibiosis. The first is a sine quanon of pneumonia management, whereas the sec-ond is the first step toward personalization (ie,endotyping).
The Use of Host Response to DiagnosePneumonia
Protein biomarkersBiomarkers (often present in serum, quantitative,rapidly processed, and potentially amenable topoint-of-care testing) represent a highly attractivediagnostic modality. In the context of pneumoniadiagnosis, biomarkers are used as reporters ofthe inflammatory neutrophilic response in thelung parenchyma. Balk and colleagues provide amore complete description in this issue, but the 2most commonly used biomarkers, procalcitonin(PCT) and C-reactive protein (CRP), are briefly dis-cussed here.
The biology of PCT is still incompletely under-stood, but it is known to be produced by immuneand parenchymal cells in most tissues in responseto stimulation with pathogen-associated molecu-lar patterns (PAMPs), danger-associated molecu-lar patterns (DAMPs), and inflammatory cytokines(see Box 1). PCT appears in serum at about4 hours and peaks at 6 hours, making it an effec-tive early indicator of pneumonia.91 One of theprincipal advantages of PCT is that its expressionis suppressed by type I IFN, which increases itsspecificity for bacterial rather than viral infection.Very low PCT values are helpful in ruling out bacte-rial infection and withholding antibiotics, as shownby Christ-Crain and colleagues,92 but the currentguidelines do not recommend its use in this capac-ity. A drawback to PCT is its low expression inatypical infections (ie, Legionella, Mycoplasma,and Chlamydophila)93,94 and in bacterial pneu-monia following viral infection.95
CRP is synthesized by the liver in response toIL-6, making it a less-specific marker for lung
infection. Like CRP, it appears quickly (at w6hours) but peaks much more slowly (at 36–50hours) and its clearance is delayed.96 CRPlevels correlate with pulmonary bacterial loads(measured by quantitative tracheal aspirates) inVAP97 and are more useful than PCT in the detec-tion of atypical infections.
Inclusion of additional cytokine biomarkers inthe laboratory evaluation of pneumonia, includingtumor necrosis factor (TNF)-a, IL-6, IL-8, andIL-10, mildly improves discrimination of bacterialfrom viral infections and can be used to increasesuspicion for particular bacterial pathogens (eg,Enterobacteriaceae elicit more IL-8), but these cy-tokines are not yet used in common prac-tice.94,98–100 Notably, few studies have examinedIFN-stimulated genes (ISGs), which could increasethe positive predictive value for viruses, as tran-scriptomic studies have suggested (discussedlater).
To conclude, there is ample evidence to showthat the biomarkers in current use aid in the diag-nosis of pneumonia, but they are not yet reliableenough to identify patients with nonbacterialcauses and permit withholding of antibiotics; acritical “litmus test” for pneumonia diagnostics.One of the principal limitations of biomarkers incurrent use is their poor specificity; CRP andPCT levels are increased during inflammationfrom virtually any source, acute and chronic alike,including neoplastic, rheumatologic, necrotic (eg,pancreatitis or trauma), and infectious (with littlediscrimination between pathogen classes). Thus,they are indicators of systemic inflammation, notof pneumonia, and as such largely remain a com-plement to the similarly nonspecific markers ofneutrophilic alveolitis in common use, such as fe-ver, cough, sputum, leukocytosis, and radio-graphic infiltrates.
Neutrophilia in lower airway secretionsSputum neutrophilia is the quintessential surro-gate for the alveolar purulence that characterizesbacterial pneumonia. In immunocompromisedand intubated patients, tracheal aspirates or directalveolar assessment via invasive sampling hasproved particularly useful. For instance, in a popu-lation consisting mostly of patients with hemato-logic malignancy and solid organ transplants,BAL neutrophilia was shown to have better areaunder the curve (AUC) for diagnosing pneumoniathan either PCT or CRP, using quantitative cultureas a gold standard.101 More recently, Choi andcolleagues102 showed that BAL neutrophil countgreater than 510/mL was a highly effective predic-tor of bacterial pneumonia, with an odds ratioof 13.5. BAL neutrophil count also effectively
Gautam et al880
discriminated bacterial from viral pneumonia, withan AUC of 0.855; its performance further improvedwhen combined with CRP.
TranscriptomicsIn contrast with the focused interrogation ofclinically available biomarkers, transcriptomicsprovides a global view of differential gene expres-sion in response to infection. Clustering analysis isused to identify distinct RNA expression patternsthat correlate with presence or absence of infec-tion, different classes of pathogens, diseaseseverity, and prognosis. Given the inclusion oftens or even hundreds of genes in such immuneresponse signatures, their potential sensitivityand specificity is far more robust than biomarker-based diagnostic strategies.Early transcriptomic studies established that
much of the immune response in pneumonia isconsistent across pathogen classes, but speci-ficity can be found in the activation of distinctsignaling pathways downstream of particularPRRs (eg, TLR4 activation by extracellular gram-negative bacteria vs TLR3 activation by RNA vi-ruses).103 An example of is the transcriptomicanalysis performed by Ramilo and colleagues,104
who examined blood from 36 pediatric patientsacutely infected with influenza A and 16 withStreptococcus pneumoniae (mostly pneumonia)and identified a 35-gene panel that discriminatedviral from bacterial infection with 95% accuracy inan independent cohort. Similarly, Zaas and col-leagues105 were able to establish a 30-gene viralsignature based on blood transcriptomes fromhuman volunteers subjected to viral challengewith rhinovirus, RSV, and influenza A. This findingwas validated in an independently acquired dataset, showing 100% accuracy for identifying viralinfection and 93% for bacterial infection. Tangand colleagues106 subsequently assayed wholeblood from ICU patients in respiratory failurecaused by influenza, bacterial pneumonia, andpresumed sterile systemic inflammatory responsesyndrome (SIRS). Again, they showed an ability torobustly identify viral infection throughout the5 days of follow-up, largely based on upregulationof ISGs and inhibition of innate inflammatory cyto-kines, which indicates a profound state of immu-nosuppression in influenza. However, they wereunable to establish a bacterial signature thatcould distinguish between bacterial infection andsterile SIRS. Of note, there was surprisingly littleconcordance between viral signatures in the 3studies, perhaps because of differences intraining cohorts or in bioinformatic techniques.However, the few common genes were all IFN-inducible.
Although these studies laid important ground-work for the application of transcriptomics inpneumonia endotyping, their utility was limited by2 issues. First, gene panels were too extensive(>25 transcripts) to permit analysis in standardclinical laboratories. Second, they had notaddressed the fundamental problem of how toidentify patients who need antibiotics.The first issue was addressed in a follow-up
study by Zaas and colleagues,107 who were ableto translate their viral signature into a real-timePCR-based assay using commercially availableprobes. Landry and Foxman108 studied nasopha-ryngeal swabs from patients and showed that aset of only 3 transcripts in these samples(CXCL10, IFIT2, and OASL) could predict viralinfection with 97% accuracy. More recently,Tang and colleagues109 reported a single serumbiomarker capable of discriminating influenzafrom bacterial infection with an AUC of 91% in alarge, newly enrolled cohort: IFI27, an ISG that isupregulated in plasmacytoid dendritic cells (DCs)in response to TLR7 activation. These studiesrepresent some of the best examples to date ofthe translation of transcriptomic analysis into thedevelopment of robust but technically feasibleclinical assays.Substantial progress toward solving the second
problem was made by Tsalik and colleagues,110
who assessed host expression profiles in patientswith confirmed viral infection, bacterial infection,coinfection, and sick but noninfected controls.The use of this last control was a unique andimportant feature of the study because it helpedto directly address the question of how to identifypatients with sterile respiratory illness. Althoughthe 4 signatures each required large numbers ofprobes (up to 71), they had superb test character-istics with AUCs between 90% and 99% inexternal validation analyses. A similar aim guidedRamilo and colleagues104 in their study of an anal-ogous set of patients with viral and bacterialmonoinfections, coinfection, and controls. Usingadvanced bioinformatics techniques, they identi-fied a parsimonious 10 gene classifier with a sensi-tivity of 95% for bacterial infection (compared with38% sensitivity of PCT); another step towardestablishing a rule-out test to guide withholdingof antibiotics. Note that 7 of these 10 genes over-lap with the biosignature identified independentlyby Zaas’s group using unique analyticalmethods,111 suggesting the field may beconverging on a common classifier. Furtherstudies will be necessary to confirm the utility ofthis probe set in larger cohorts and in the immuno-compromised, a population that is particularlyprone to overtreatment with antimicrobials.
Personalizing the Management of Pneumonia 881
A final study worth mentioning compared theimmune response in patients with sepsis causedby peritonitis versus pneumonia. There was littleto distinguish between these two cohorts, sug-gesting that transcriptomic analysis is unable todelineate the anatomic source of infection, at leastwhen applied to peripheral leukocytes in late-stage sepsis (see Fig. 1A).112
Use of Host Response to Define Severity atPresentation and Guide Prognostication
As mentioned earlier, the severity of clinical pre-sentation in pneumonia primarily depends on thehost response to the pathogen and the associatedbystander immunopathology (Fig. 2). The existingmeasures of severity, including CURB-65 (confu-sion, urea, respiratory rate, blood pressure, age�65 years) and pneumonia severity index (PSI),are clinical scoring scales used to assess theend-organ consequences of this inflammatoryresponse (eg, renal dysfunction) and are princi-pally used for triage. PSI additionally accountsfor comorbidities and therefore incorporates theconcept of physiologic reserve described in rela-tion to phase I. In contrast, biomarkers reportdirectly and quantitatively on the inflammatorytone of the host in response to infection. Theyhave been used to gain insight into additional clin-ical parameters, including acute stability and
A
Fig. 2. Host response to pathogen. (A) Patterns of inflammanisms allows progression of infection (in the absence ofreaches a threshold, it may lead to an irreversible declinsuch as sepsis or ARDS. The specificity of host immune sbecause the inflammatory response degenerates to a cosite of infection. (B) Immune responses to influenza. Influeing infection, given its highly variable course. In most patantiviral medication. In some, however, secondary bacteriadisposed to excessive immune responses (poor resiliencecourse depicted by the dotted line on the left. Such patientclearance after exposure (top left arrow).
prognosis, as well as response to infection, as dis-cussed in relation to phase III. The use of bio-markers in this context is discussed briefly hereand in detail by Balk and colleagues elsewhere inthis issue (also see Torres and colleagues’113
recent review).As might be predicted, systemic levels of inflam-
matory cytokines (including IL-6, IL-10, and IFNg)are significantly higher in patients with severeCAP than in patients with nonsevere CAP and inhealthy individuals.98,114 Furthermore, IL-6 corre-lates with clinical scoring scales115,116 and predicts30-day mortality in hospitalized patients withCAP.98,114 Addition of CRP to a composite clinicalindex including both PSI and CURB-65 improves30-day mortality prediction, achieving an AUC of0.88.114 PCT on its own shows similar prognosticaccuracy to CURB-65 and scales with severity.117
VanVught andcolleagues118 provided an importantcaveat to these findings, showing that systemic cy-tokines do not correlate with PSI in the elderly.
Given that the progression to sepsis (discussedfurther in relation to phase IV) portends a worseoutcome in patients with pneumonia, it is of prog-nostic value to detect this transition. Protracted,smoldering inflammation marks the later phase ofsepsis; evidence of this in patients recoveringfrom acute pneumonia, as marked by increasedlevels of IL-6 and IL-10, was shown to correlatewith increased mortality at 1 year.119 In contrast,
B
ation in severe pneumonia. Failure of resistance mech-antibiotics). However, when the severity of infectione caused by uncontrollable inflammatory syndromesignatures decreases at these end stages of infectionmmon pattern regardless of microbe class and initialnza represents a useful example of host response dur-ients, influenza is cleared effectively, with or withoutl infection complicates the illness. Still others are pre-) and therefore follow the more precipitous clinicals would benefit from vaccination, which leads to rapid
Gautam et al882
local immune responses at the respiratory epithe-lium, as revealed by sputum cytokine profiles, areblunted in severe CAP despite exaggerated inflam-mation in the periphery.120 This discordance be-tween lung and systemic immune compartmentshighlights the importance of site selection whenassessing host responses. In influenza infection,Oshansky and colleagues121 showed the potentialfor using mucosal-specific host responses to pre-dict clinical outcomes, showing that a nasal cyto-kine profile characterized by increased monocytechemoattractant protein-3 (MCP-3) and IFN-a2could predict progression to severe disease inde-pendently of age, viral load, and neutralizing anti-body titers.
Assessment of Resilience
As mentioned in the context of host resistance,there is remarkable interindividual variability inthe severity of pneumonia caused by a given path-ogen, ranging from mild infection treated in theoutpatient setting to fulminant sepsis requiringICU admission. Physiologic reserve, pathogenburden, and resistance contribute substantially tothis variability, but host resilience, defined as thehost’s ability to tolerate a pathogen load, alsoplays a critical role. Simply put, 2 patients withsimilar baseline health and pathogen load maydevelop widely discordant disease severities, aphenomenon largely attributable to the host’s pre-disposition toward immunopathology. A uniqueexample is shown in Fig. 1B, which shows theincreased resilience to PJP observed in patientswith AIDS; although driven by a pathologic pro-cess (ie, severe immunocompromise), the patientis able to tolerate an extraordinary pathogenburden with minimal pulmonary inflammation.The data presented by Oshansky and col-
leagues121 exemplify the more common patternobserved in practice: decreased host resilienceleading to more severe disease. Despite similarphysiologic reserve (indicated by age in theseotherwise healthy children), host resistance (indi-cated by neutralizing antibodies), and pathogenburden (indicated by viral load), a subset of pa-tients progressed to severe influenza, suggestingan underlying immunologic susceptibility.Although the investigators focused on the prog-nostic value of the signature, it is notable the bio-markers (eg, IFN-a2) are known components ofthe cytokine storm that mediates immunologic dis-order, organ dysfunction, and death in extremecases.122,123 Therefore, these markers couldpotentially function as theranostics in influenza,both guiding initiation of immunosuppression andindicating response to therapy.
Substantial efforts have been made to identifythe genetic underpinnings of susceptibility to influ-enza infection and other forms of pneumonia (seeFig. 2). The topic has been reviewed else-where,46,124 and more extensively in the contextof sepsis,125 but, in these analyses, susceptibilityloci are not clearly stratified by mechanism (ie,whether they affect resistance or resilience). Onegenetic variant that seems to specifically compro-mise resilience affects CD55, which protects therespiratory epithelium from complement deposi-tion, a process implicated in the immunopathogen-esis of severe influenza.126 A second study used anintegrated genomic approach to identify suscepti-bility loci in patients with CAP that progressed tosepsis.127 First, unsupervised transcriptomic anal-ysis divided the study cohort into 2 endotypes us-ing a 7-gene classifier; one expressing sepsisresponse signature 1 (SRS1, marked by an immu-nosuppressed phenotype and increased 14-daymortality), and the other expressing SRS2. Next,genetic analysis identified a set of approximately4000 quantitative trait loci that predisposed to thehigher-risk phenotype, SRS1.At present, the clinical utility of disease-
associated SNPs is limited, but several potentialapplications can be envisioned as the list expandsand host genomics come into more routine clinicalpractice. For instance, identification of variantsthat compromise resilience may prompt moreaggressive immunosuppression. Also, from aresearch perspective, disease-associated SNPsgive mechanistic insight into human infection andrepresent future therapeutic targets.In closing, we propose that personalized anal-
ysis of host immune responses should ideally (1)confirm true infection; (2) identify bacterial pro-cesses that require antibiotics; (3) estimateseverity to guide triage and prognostication; (4)assess host resistance, as discussed in relationto phase I; and (5) characterize host resilience.The last 2 should be performed with sufficientgranularity to identify specific pathways for modu-lation as described in relation to phase III. In addi-tion, as indicated by studies, including that byOshansky and colleagues,121 test performancemay improve with integration of local respiratoryepithelial and systemic immune responses.
PHASE III: PERSONALIZED TREATMENT ANDASSESSMENT OF THERAPEUTIC RESPONSE
Armed with a clinical dataset that confirms thepresence of pneumonia, identifies the offendingpathogen and its susceptibilities, and describesthe host’s immune competence and immunopath-ologic diatheses, clinicians are prepared to devise
Personalizing the Management of Pneumonia 883
a treatment plan. This plan will have the followingaims: (1) to reduce pathogen burden, both throughdirect attack on the microbe (eg, with antibiotics)and through support of host-intrinsic resistancemechanisms; and (2) to optimize host resilience,largely through suppression of hyperactive andmaladaptive immune pathways.
A key principle that informs the following discus-sion is that clearance of bacteria in patientstreated for pneumonia is a collaboration betweenhost resistance and antimicrobials (Fig. 3A).Some patients with pneumonia may have suffi-ciently robust immunity to eradicate the infectionwithout therapy (Fig. 3B). On the other end of thespectrum are neutropenic patients dependent onantibiotics until count recovery (Fig. 3C). Theremainder of patients are somewhere in betweenthese extremes, and clinicians are responsiblefor personalizing an antibiotic regimen that bal-ances the patient’s reliance on antibiotics againstthe substantial hazards of these drugs. Antibioticchoice, dose, and duration are considered, asare so-called antibiotic-sparing interventions,including nonantimicrobial pharmaceutics (eg, re-combinant antimicrobial peptides).
Antibiotic Therapy for Bacterial and FungalPneumonia
Hazards of antibioticsAs mentioned at the outset, there is a widespreadmisconception that antibiotics are benign
A B
Fig. 3. Mechanisms of reducing pathogen burden. (A) Pathealthy hosts, brief antibiosis reduces pathogen burdennisms. (B) Pathogen clearance via host resistance alone. Ifpathogens without specific therapy. (C) Pathogen clearanccompromise affects resistance mechanisms against particuexemplified by neutropenic patients infected with pyogengression of infection. This finding contrasts with the immopportunistic infections, which affects resilience (see Fig.
medications, but the risks of antibiotic use aremyriad, with none more ominous than the growingspecter of resistance (see Fig. 1C).128
The clinical use of antimicrobials, an estimated50% of which is unnecessary,129 leads to thespread of resistance in a fairly well-describedsequence. First, antibiosis creates a selectionpressure that leads to enrichment of microfloraand pathogens with preexisting resistance, aswell as generation of de novo resistance.130 Sub-sequent transfer of resistance determinants be-tween organisms in vivo and human-to-humantransmission of resistant organisms (eg, by thefecal-oral route in the community and via clini-cians’ hands in hospital) leads to disseminationwithin a population.131 If this process continuesunchecked, the WHO warns,132 a postantibioticera will soon begin, with an estimated loss of 10million lives to antimicrobial resistance per yearby 2050.133 Even rapid deescalation of antibiotics(in cases in which a pathogen is isolated) carries asubstantial risk of selecting resistant bacteriabecause their macrobiotic effects are rapid andpersist for months after exposure.12,134
Additional hazards of antibiotics includetheir adverse drug-drug interactions and class-specific toxicities, such as the nephrotoxicityobserved with vancomycin, aminoglycosides,amphotericin, and polymyxins; a particularconcern in the ICU.135,136 Furthermore, newmech-anisms of toxicity continue to emerge, such as theability to induce mitochondrial dysfunction and
C
hogen clearance via antibiotics and host resistance. Into a level that allows eradication by immune mecha-the inoculum is small enough, healthy hosts can cleare in the setting of impaired resistance. When immuno-lar pathogens, antimicrobial therapy is essential, as isic bacteria. Impaired resistance also leads to rapid pro-unocompromise observed in patients with AIDS with1B).
Gautam et al884
ROS damage,137 indicating the field’s incompleteknowledge on the subject. Antibiotics may alsoinduce potentially catastrophic hypersensitivity re-actions, such as anaphylaxis, toxic epidermal nec-rolysis, and drug rash with eosinophilia andsystemic symptoms (DRESS) in susceptible hosts.In addition, antimicrobial agents destabilize the
microbiome, producing a state known as dysbio-sis. As mentioned earlier, this may affect the devel-opment and course of various disease, includingdiabetes, atherosclerosis, and asthma.13,14 How-ever, more immediate for patients is the risk ofCDI, which accounts for roughly 29,000 deathsper year in the United States.138 All antibiotics,even low-risk classes, predispose to CDI, and theireffects are cumulative with respect to number ofagents, dose, and duration.Based on abundant mouse data, it is likely that
gut dysbiosis predisposes to pneumonia aswell139–144; consistent with this is clinical evidencethat oral probiotics protect ICU patients from VAP,as described in relation to phase IV. Furthermore,antibiotic use may lead to secondary pneumoniacourtesy of pathobionts: normally benign florathat may overgrow and cause infection in thesetting of dysbiosis, analogously to C difficile.145
In addition, it has been shown in humans that car-riage of certain taxa within the nasal microbiomecorrelates with improved adaptive immunityto respiratory pathogens, namely responses toinfluenza A vaccination.146 Antimicrobials maycompromise this component of the microbiota aswell.Considering this litany of potential hazards, it is
not surprising that the unnecessary use of antibi-otics has been shown to increase mortality incertain patient populations, including those withsepsis in the ICU.147
Antibiotic selectionSelection of a particular antibiotic is guided largelyby susceptibilities, but, except in cases of MDRs, afair breadth of choice usually remains. Personal-izing this decision should first take into accountpotential adverse reactions, perhaps in the futureusing pharmacogenomic techniques to predicttoxicity,148 although this approach may find moreuse in the context of chronic lung infections, whichrequire more prolonged courses of therapy.149
Second, clinicians must decide between bacterio-static versus bactericidal agents, although, asargued by Spellberg and colleagues,150 thedistinction is arbitrary and in general populationsthere seems to be no advantage to bactericidaldrugs despite the folklore belief. Nevertheless,there are specific circumstances in which eachmay be desirable. Drawing on the example of
endocarditis and meningitis, for which bactericidaldrugs are recommended based on the relativepaucity of immune effectors at these sites of infec-tion, heavily immunosuppressed patients maybenefit more from bactericidal drugs. In contrast,bacteriostatic drugs that inhibit protein synthesismay improve immune resilience; for instance inpostviral pneumonia.151,152 In addition, an argu-ment has been made for the use of bacteriostaticdrugs in pneumococcal pneumonia because lyticagents increase the generation of pneumolysin, aproinflammatory toxin with numerous harmful ef-fects, including myocardial toxicity.153
Antibiotic dosingIn recent years, some of the basic assumptions onwhich current dosing protocols are based havebeen called into question.154–156 For instance,the practice of administering antibiotics at theirmaximum tolerable dose is based on the prevailingnotion that low doses create selection pressure forthe emergence of resistance, whereas high dosesof antibiotics kill microbes before resistancecan develop.157 Among others, Read and col-leagues155 have challenged this belief, arguingthat, when MDRs are present at the start of infec-tion, they are likely held in check by other micro-biota not affected by resistance mechanisms thatcompromise microbial fitness; in the presence ofantibiotics, these protective microbiota are killed,allowing the resistant organisms to flourish un-abated, a phenomenon called competitive release.In contrast, when MDRs are absent at the outset,they recommend a high-dose regimen for prevent-ing development of resistance according to theconventional argument.The implications of this model for patients with
pneumonia are potentially practice-changing. Pa-tients with severe infection or immunocompromisewould still be given the conventional high-doseprotocol, but, for those with robust physiologicreserve and fairly mild disease, outpatient therapywith the lowest clinically effective dose may be theoptimal regimen. However, close follow-up tomonitor for underdosing and treatment failure (aparticular concern in drug hypermetabolizers)would be essential. Given this risk, and that sub-therapeutic antibiosis may promote resistance,158
navigating this lower bound of the therapeutic win-dow would require great vigilance if adopted intoclinical practice.
Duration of therapyDetermining the optimal duration of therapy is acrucial feature of personalizing pneumonia man-agement. Antibiotic courses have shortened toas little as 5 days, and effort has been made to
Personalizing the Management of Pneumonia 885
identify biomarkers that may guide even earliercessation. For instance, protocols such as stop-ping therapy when PCT decreases to 20% of itspeak have been shown to reduce antibioticdays.159 Taking abbreviated courses to theextreme, it has been shown that even a singleday of antibiotics can have significant clinical ef-fect, as a dose of ceftriaxone given before acourse of linezolid substantially improved curerates.160 Besides minimizing antibiotic exposure,an additional theoretic benefit of short courses issuggested by models showing that brief therapeu-tic pulses may reduce the risk of inducing resis-tance without compromising pathogen killing.161
Conceptually, the optimal duration is a functionof pathogen burden, adequacy of host resis-tance, and efficiency of chemotherapeutic killing.Rather than attempting to predict this a priori, itmay be preferable to use a theranostic strategythat follows an indicator of microbial persistence,either indirectly using host response (eg, PCT) ordirectly using a microbial marker (eg, serum gal-actomannan and b-glucan in aspergillosis).162
CAP guidelines do incorporate a fair degreeof personalization, as the recommended lengthof therapy varies depending on host response in-dicators. Given the success of the current guide-line-based approach, as shown in a large RCT byUranga and colleagues,163 the bar would be highfor any potential alternatives.
Antimicrobial Therapy for Viral Pneumonia
The administration of neuraminidase inhibitorssuch as oseltamivir for influenza is well establishedin clinical practice, but management of other formsof viral pneumonia is less clear despite their sub-stantial clinical burden. In one series of patientsin the ICU with severe CAP, 36% had a viral causewithout bacterial coinfection on BAL, and, withinthis group, rhinovirus, parainfluenza, and humanmetapneumovirus were all more frequently recov-ered than influenza.164 HSV may be an additionalcontributor to severe respiratory disease, even inimmunocompetent hosts, as it has been shownthat 21% of nonimmunocompromised patientson prolonged mechanical ventilationhave evidence of HSV bronchopneumonitis byhigh viral titer on BAL-specific and HSV-specificnuclear inclusions in cells recovered on BAL or bi-opsy.165 Likewise, CMV may have pathogenic ef-fects in previously immunocompetent critically illpatients.166
In immunocompromised populations, thesepathogens are routinely treated,167 but it maybe advantageous to treat in select immunocom-petent patents as well. For instance, ribavirin is
highly effective therapy for upper and lower res-piratory tract infection from RSV in hematologicalmalignancy and carries few side effects, particu-larly in the oral formulation.168–170 Given thesefeatures, as well at its additional activity againstparainfluenza and human metapneumovirus,ribavirin may prove useful in immunocompetentpatients with severe viral pneumonia, althoughdata to this end are currently lacking.
An alternative, or complement, to chemotherapy-based regimens for pneumonia is a diverse collec-tion of therapeutics that includes synthetic antimi-crobial peptides,171 engineered bacteriophagelysins,172 neutralizing antibodies (eg, againstinfluenza),173 and antibodies targeting pathogen-associated toxins (eg, pneumolysin).174 Thesetherapeutics are reviewed by Czaplewski and col-leagues175 but are also mentioned here for theirutility in personalized therapy.
Lytic bacteriophages epitomize this class of‘antibiotic alternatives’.176 Reemerging after theirinitial description in the preantibiotic era, theseviruses have potent bactericidal effects onactively replicating cells and are highly specificfor particular bacterial species, so their dysbioticeffects are minimal. In addition, they have lowpotential for generating antimicrobial resistanceor host toxicity. Although still largely the purviewof basic research, this approach may eventuallytranslate to the clinic, perhaps as a last resortfor respiratory pathogens with extended drugresistance.
Host-Directed Therapies for Improving HostResistance
Most of the measures discussed earlier promotepathogen clearance predominantly through directtoxic effects on the microbe. However, some func-tion by blunting virulence (eg, antibodies thattarget bacterial toxins or neutralize viruses), leav-ing host resistance mechanisms to clear the atten-uated pathogen. A third strategy, not mutuallyexclusive with the others, is to bolster host resis-tance directly using immunotherapeutics.177 Tothis point, the clinical application of such therapyhas largely been restricted to chronic infectionswith mycobacteria and aspergillus unresponsiveto antimicrobials.178,179 The use of such strategiesas chimeric antigen receptor-T therapy and sup-plemental cytokine therapy in this context pro-vides an instructive model for acute pneumonia.A notable example from this literature is the admin-istration of recombinant IL-2 to a patient with
Gautam et al886
idiopathic CD41 lymphopenia and antibiotic-refractory Mycobacterium avium-intracellulairelung disease, with resultant resolution ofinfection.180
Another concept worth exploring is the use ofsupportive therapies that promote nonimmuno-logic aspects of host resistance, such assecretion clearance, including routine chest phys-iotherapy, which has been shown to decrease theincidence of VAP.181 Along similar lines, coughaugmentation may be useful in a select group ofpatients to prevent or manage VAP, althoughmeta-analyses show that it does not seem toimprove time to extubation in the general ICUpopulation.182 An as-yet unexplored directionwould be to counteract the known defects inmucociliary clearance in critically ill183 and intu-bated184 patients by improving mucus rheology.One approach to doing so is the use of cysticfibrosis (CF) transmembrane regulator (CFTR)modulators such as ivacaftor, which has beenshown to potentiate the function of CFTR in pa-tients without CF.185,186
Host-Directed Therapies for ImprovingResilience
As stated earlier, the principal determinant ofseverity in most cases of pneumonia is the immu-nopathology associated with the host response,not the virulence of the pathogen. A portion ofthis immunopathology is attributable to collateraldamage from essential immunological defensemechanisms, while another is simply due toexcessive inflammation. Ideally, immunosuppres-sive agents should selectively target the latter,but in practice, medications like glucocorticoidspotently inhibit both. However, when simulta-neously treating with antibiotics, resistance mech-anisms play a less pivotal role in eradication ofmicrobes, and therefore the impaired resistanceinduced by immunosuppressive therapy may bean acceptable sacrifice for the reduction of patho-logic inflammation. Macrolides represent a uniqueexample among antibiotics in that they simulta-neously clear pathogen and dampen inflamma-tion. The latter effect was strikingly revealed by ameta-analysis that showed a mortality benefit inCAP even in patients with macrolide-resistant bac-teria.187 Similar dissociation of clinical efficacyfrom microbicidal activity was shown in CF.188
Antimicrobial therapy also has the potential toexacerbate immunologic disorders. This exacer-bation occurs via release of PAMPs from lysedpathogens; the so-called Jarisch-Herxheimer re-action (see Fig. 1C). Often observed in the earlystages of treatment of cellulitis and spirochetal
disease, this phenomenon is best known for itsrole in PJP therapy in patients with AIDS. In suchpatients, the insufficiency of host defenses permitsthe proliferation of fungi to high levels within thelungs. On initiation of antimicrobial therapy, fungallysis leads to a massive bloom of cell wall compo-nents, including b-glucan, which elicits an intenseinflammatory response through dectin-1 that mayresult in ARDS.189 It is therefore common practiceto treat these patients simultaneously with steroidsto avert the potential immunopathologic response.More targeted approaches have also beenexplored, such as cotreatment with b-glucan syn-thesis inhibitors (echinocandins), which has shownefficacy inmousemodels.189What role the Jarisch-Herxheimer reaction might play in other causes ofpneumonia has not been explored in detail.A more heated debate surrounds the use of
immunosuppression in non-PJP pneumonia.Torres and colleagues190 were able to solve thisproblem using a fairly simple endotyping strategyas they limited administration of steroids to pa-tients with a hyperinflammatory phenotype, asindicated by CRP level greater than 150 mg/dL.A contemporaneous study similarly showed abenefit to steroid use in severe pneumonia; unsur-prisingly, the mean CRP in the study cohort wasalso greater than 150 mg/dL.191 These successeshighlight the value of personalizing therapy, evenif to a rudimentary degree. As the sophisticationof host diagnostics increases, it should bepossible to endotype in much finer detail,enabling more effective prediction of responseto immunosuppression.Another possible explanation for the failure of
steroids in early trials relates to the immunologicnonspecificity of these agents. In this sense,steroids might be considered antipersonalizedtherapy because they indiscriminately inhibitimmune pathways across the spectrum fromprotective to pathologic. Instead, patient stratifica-tion according to immune pathway dysregulationshould be used to target immunotherapy and mini-mize side effects. One such targeted therapeuticstrategy is the use of PRR antagonists,192 whichcould in principle halt the inflammatory paroxysmat its source. However, the TLR4 antagonist, eri-toran, failed to improve outcomes in sepsis (evenin a subgroup analysis of the 50% with pneu-monia)193 despite its demonstrated protectionagainst endotoxemia in healthy volunteers.194 Itmay be that PRR antagonism is most effectiveearly in the disease process (as suggested by an-imal models as well195), and that advanced dis-ease requires a very different approach (includingimmunostimulation, for instance), as explored inrelation to phase IV.
Personalizing the Management of Pneumonia 887
PHASE IV: SECONDARY THERAPIES TOADDRESS THE CONSEQUENCES OFINFECTION AND TREATMENTAddressing the Immunopathology ofPneumonia-Associated Sepsis
As alluded to inBox1, lung infectionmay run an un-complicated course with an appropriate immuneresponse that results in pathogen clearance fol-lowed by prompt resolution of inflammation. How-ever, severe infection in susceptible hosts (ie, thosewith poor resilience) may result in a complex syn-drome of immune dysregulation known as sepsis.The pathophysiologic details are beyond the scopeof this article and not yet fully established,196 buttwo of the key features are uncontrolled, persistentinflammation and a profound state of immunosup-pression that affects both innate and adaptive im-munity, called immunoparalysis. Therapeuticmeasures for modulating both aspects have beenexplored and are discussed here.
Resolution of proinflammatory responseThe initiating phase of sepsis involves a hyperin-flammatory reaction to microbial PAMPs andDAMPs produced by damaged tissue, followedby activation of complement, endothelial cells(which leads to tissue edema and leukocyteextravasation), neutrophils (which induce damagecaused by ROS and proteases), and the coagula-tion cascade (causing microthrombosis and coa-gulopathy), all of which interact in potentiallyamplifying loops that may degenerate into a se-vere systemic state of inflammation. However,numerous immune mechanisms are in place tocontrol the magnitude and promote the resolutionof this potentially devastating process. Proresolu-tion mechanisms include elimination of proinflam-matory cytokines, neutrophil apoptosis andefferocytosis, and a switch in macrophage pheno-type from inflammatory to reparative (or replace-ment via monocyte influx).197 Steroids werediscussed earlier, the antiinflammatory propertiesof which may help to limit the magnitude of inflam-matory response in sepsis, but therapeuticsdesigned to stimulate resolution have also beenproposed.197
Much attention in inflammatory resolution hasbeen focused on the use proresolving mediators,including lipids known as resolvins, lipoxins, andmaresins, but most studies to date have been pre-clinical.198 However, some intriguing observationaldata indicate a protective role in CAP foraspirin,199,200 which is known to generate potentlipoxins201; prospective studies are now underwayto evaluate for an ameliorative effect in sepsis.202
Similarly, statins lead to the production of lipoxins,
and established use before presentation is associ-ated with a reduced incidence of CAP (in aretrospective analysis of the JUPITER [Justificationfor the Use of Statins in Prevention: an InterventionTrial Evaluating Rosuvastatin] trial),203 and possiblyan improvement in mortality. However, conflictingstudies and potential confounders such as theso-called healthy user effect must be addressedbefore drawing definitive conclusions.204
Personalized modulation of inflammatory reso-lution is likely to require metabolomic analysis, firstin research studies to establish the differences inlipid milieu between normally resolving pneumoniaand protracted disease and then in patients todetect specific molecular deficiencies. Supple-menting these patients with synthetic analoguesto steer the immune response toward homeostasismay prove a valuable complement to immunosup-pressive agents that are intended to dampen itsseverity.198
Reversal of immunosuppressionWithin the lung, local immune responses are blunt-ed in the wake of viral and bacterial infectionthrough several mechanisms, including generationof a reparative, antiinflammatory milieu dominatedby transforming growth factor beta.205 However,as pneumonia progresses to sepsis, a profoundstate of immunosuppression seems to develop af-ter about 3 days, placing patients at high risk ofsecondary infection, about half of which is respira-tory.196,206 It is during this late stage of sepsis,termed compensatory antiinflammatory responsesyndrome (CARS), that most deaths occur.207,208
No clinical trials have yet examined lung-specificinterventions to support patients through thisvulnerable stage, but there is a substantial bodyof work on reversing the systemic state of immu-nosuppression. This article focuses on the use ofimmunostimulatory cytokines and checkpoint inhi-bition, but see van der Poll and colleagues209 formore on the topic.
Granulocyte-macrophage colony–stimulatingfactor (GM-CSF) promotes granulocyte produc-tion, survival, phagocytic function, and extravasa-tion into tissue. It also reverses the downregulationof (human leukocyte antigen, antigen Drelated (HLA-DR), an important contributor toand biomarker of immunoparalysis in advancedsepsis. The potential efficacy of GM-CSF wasshown in a double-blind multicenter trial in which38 patients with low HLA-DR expression (most ofwhom presented with pneumonia) were random-ized to receive GM-CSF or placebo. The treatmentarm showed complete normalization of HLA-DRexpression, restored responses to TLR stimula-tion, improved APACHE (Acute Physiology And
Gautam et al888
Chronic Health Evaluation) scores, and decreasedduration of mechanical ventilation and ICU stay,without significant side effects.53 This biomarker-guided (ie, theranostic) immunomodulatoryapproach represents an important example ofpersonalized treatment of pneumonia and a modelfor future studies. Of note, a related cytokine thatsimilarly stimulates granulocyte production, gran-ulocyte colony–stimulating factor (G-CSF), hasbeen studied in the context of neutropenic pneu-monia, but evidence is accumulating to showthat the resultant neutrophil reconstitution can pre-cipitate ARDS and therefore G-CSF should beavoided in these patients.210
IFNg, the quintessential T helper 1 (Th1) cyto-kine, exerts potent stimulatory effects on granulo-cytes to promote clearance of bacterial and fungalpathogens. Human studies have mostly beenlimited to case reports and results have beenmixed,211 but administration is generally well toler-ated and there is some evidence for efficacy.212
For instance, Dignani and colleagues213 describedcomplete resolution of antimicrobial-refractorypulmonary aspergillosis in 3 patients after admin-istration of IFNg; similar success was seen in 2cases of invasive aspergillosis and 1 of candidi-asis, all involving the lung.214 In select patients,this may prove a valuable adjunctive therapy forpneumonia; further insights are sure to be gener-ated by an RCT examining its role in the treatmentof patients with septic shock (https://clinicaltrials.gov/ct2/show/NCT01649921).IL-7 predominantly affects adaptive immunity,
promoting T cell proliferation, activation, survival,and trafficking to infected tissue. It has shownpromise in preclinical models of pneumonia,215
and is currently the focus of a multicenter clinicaltrial in septic patients (https://clinicaltrials.gov/ct2/show/NCT02960854).Checkpoint inhibitors, as applied to sepsis, have
been studied mostly in mice, but they may find usein severe pneumonia given the evidence of T-cellexhaustion in a postmortem examination of septicpatients (more than half of whom had evidence oflung infection)216 and evidence for improved path-ogen clearance following checkpoint blockade inpreclinical models of acute pneumonia.217,218
Although some of the trials discussed earlierused a biomarker-based determination of candi-dates for immunostimulation, selection of patientsfor therapy may be improved by a more compre-hensive immunophenotyping, such as throughgene expression signatures or multimarker proteinassays, which may improve not only prediction ofresponse but also tailoring of therapy to individ-uals’ specific immune defects. As shown by only11% of postsepsis deaths being attributable to
secondary infection,206 not all patients require im-mune stimulation. More sophisticated diagnosticsshould at the least distinguish patients needingimmunosuppression (as discussed in relation tophase III), from those who need stimulation.
Protection and Restoration of Microbiome
The iatrogenic toll of antimicrobials continues tobe underestimated, as described in relation tophase III, but nowhere more so than in the gut. Inaddition to predisposing to CDI, antibiotics selectfor resistant bacteria and create a state of dysbio-sis, which has several harmful consequences.These consequences derive in part from the erad-ication of commensals, which normally function toprotect against outgrowth of pathobionts, aphenomenon termed colonization resistance.219
Also, through metabolism of dietary fiber, healthygut microbiota synthesize short-chain fatty acids,which positively influence systemic immune func-tion and maintenance of gut epithelial integrity.Compromise of these mechanisms caused bydysbiosis promotes gut translocation of bacteriaand PAMPs, which exacerbates the prolonged,smoldering inflammation of sepsis and in somecases produces frank infection.220,221
As explained in relation to phase III, dysbiosis islikely to increase risk of pneumonia. Severalmicrobiome-protective strategies, besides mini-mizing unnecessary antimicrobial exposure, havebeen proposed. One creative solution involvescoadministration of activated charcoal with antibi-otics, which decrease intestinal but not plasmaantibiotic levels, thus protecting the gut micro-biota.222 More attention has been given to the liter-ature on oral probiotics, which shows both a trendtoward lowering incidence of VAP223–225 and a sig-nificant delay in acquisition of Pseudomonas aeru-ginosa respiratory colonization.223 Meta-analyseshave differed in their conclusions regarding thesedata,226,227 but there does seem to be a substantialclinical effect, amounting to an approximate 20%reduction in VAP as estimated by Siempos andNtaidou.228 This effect was confirmed as signifi-cant by the most recent meta-analysis on the sub-ject.229 Even stronger data support the use ofprobiotics in mitigating the risk of CDI in patientsreceiving antibiotics: a Cochrane analysis showeda number needed to treat of only 12 in patientswith a CDI risk greater than 5%.230 Thus, especiallywhen treating pneumonia in a patient with high riskof CDI, probiotics should be strongly considered.
Prevention of Future Infection
Vaccines have been called the most effectivemedical intervention ever devised because of their
low cost, ability to prevent disease, and continuedefficacy in the presence of drug resistance.231
They remain the mainstay in the prevention ofpneumonia, as exemplified by the highly effectiveantipneumococcal and antiinfluenza agents.Although in some ways the antithesis of personal-ized medicine, because they are given to hugepopulations with minimal stratification, vaccinedevelopment and delivery must be improved todecrease the burden of preventable illness andreduce antibiotic use.232
With regard to personalized prevention of pneu-monia, Evans and colleagues233–235 have devel-oped a provocative pharmacologic approachwherein inhaled TLR agonists (specifically TLR2/6 and TLR9 ligands) are used to induce a state oftissue resistance; this has been shown to protectmice from both influenza and bacterial pneumonia.Numerous potential applications can be envis-aged for such technology, including prophylaxisin patients with hematologic malignancies aftermyelosuppressive therapy that induces prolongedneutropenia. This prophylactic strategy is currentlyunder investigation as part of a phase I clinical trial(https://clinicaltrials.gov/ct2/show/NCT03097796)and warrants further study.
Sequelae of Pneumonia
Although primarily a lung infection, pneumoniashould be considered a systemic illness,236 withmanifestations in numerous extrapulmonary or-gans, including heart, kidneys, and brain (reviewedby Restrepo and colleagues237). As mentionedearlier, premorbid compromise in these systemsdecreases the patient’s physiologic reserve andacute ability to survive infection.
However, there is also an increasing apprecia-tion of the longer-term consequences ofpneumonia. In addition to the well-describedarchitectural distortion that may complicate necro-tizing pneumonia, as well as the bronchiectasisthat may result from repeated infection (exempli-fied by patients with cystic fibrosis), there is anincreased risk of developing obstructive diseasein patients who have an episode of pneumonia inearly life.238
Outside the lung, there is a strong associationwith cardiovascular events, including an increased30-day incidence of heart failure (w15%),arrhythmia (w5%), and acute coronary syndrome(w5%).239 Up to 20% of deaths from CAP areattributable to these complications.240 Further-more, although cardiovascular risk is highestimmediately after pneumonia, it remains increasedfor 10 years.241 The mechanisms underlyingincreased cardiovascular risk in pneumonia include
inflammation-associated endothelial dysfunctionand thrombophilia, aswell asmicrobe-specific pro-cesses such as the pneumolysin-inducedmyocyteinjury andmicroabscesses observed in S pneumo-niae infection.242,243
A strong body of literature suggests that pneu-monia can precipitate cognitive decline as well.For instance, one study showed that one year afterhospitalization for CAP, one-third of patients over65 had moderate to severe impairment, and anadditional third showed mild impairment.244 Therelationship was shown to be bidirectional, inthat premorbid cognitive dysfunction predisposesto pneumonia (likely because of increased risk ofaspiration), and pneumonia in turn leads to cogni-tive impairment.245 Functional status, quality oflife, and mood also decline substantially after anepisode of pneumonia.246,247
Renal dysfunction frequently complicatessepsis associated with pneumonia by mecha-nisms relating to systemic inflammation and he-modynamic compromise that are only nowbecoming clear248; however, to our knowledge,the long-term risk of CKD postpneumonia hasnot been studied. Thirty-day readmission ratesare greatly increased after pneumonia (7%–12%),249,250 as is long-term mortality (40% vs25% for those hospitalized for other condi-tions).251,252 Thus, long-term sequelae both withinthe lung and without can be severe and representimportant opportunities for personalization (eg,treating with aspirin or high-dose statin to preventmajor cardiovascular events in patients withvascular risk factors).
SUMMARY
The practical implementation of personalizedpneumonia management depends heavily on theclinical setting, which spans from the ambulatoryclinic to the academic ICU, where there are vastlydifferent levels of patient acuity and availableresources (Fig. 4). For instance, ambulatory pro-viders do not have access to advanced diagnos-tics such as next-generation sequencing on BALbut should also not need them for the manage-ment of mild CAP. The focus in that context shouldbe on developing tools that quickly and reliablydiscriminate between bacterial pneumonia, viralpneumonia, and noninfectious disease, perhapsusing qPCR-based host response profiling.Because of the impracticality of waiting for culturedata to guide antibiosis in this setting, pathogencharacterization will be limited to rapid assayssuch as viral PCR on upper airway specimens,mass spectrometry on sputum, and/or pathogen-associated biomarkers such as the pneumococcal
Fig. 4. Personalized pneumonia diagnosis. The diagnostic modalities enumerated in phase I and II and shownhere in relative order of cost and availability. Determining the extent of diagnostic work-up (where to set theslider in the upper-right frame) depends on patient factors and test attributes (shown in the left frame). Sickerand higher-risk patients may warrant more comprehensive and expensive testing in order to ensure appropriateantimicrobial coverage and guide immunomodulation. Simpler diagnostics may be appropriate for milder pneu-monia, although they put the patient at risk of unnecessary empiric antibiosis, which promotes the spread ofresistance and carries numerous potential side effects. As the cost of advanced diagnostics decreases and theiravailability broadens, the slider should shift upward, bringing the goal of personalized pneumonia managementcloser to realization. Next-Gen, next-generation; Rx, treatment; WBC, white blood cell.
Gautam et al890
urine antigen. The principal goal is to identify andtreat patients with antimicrobial-sensitive infec-tions and spare those without, thus reducing themassive overuse of antibiotics in the clinic andspread of resistance.In contrast, for sicker patients in the ICU, more
elaborate testing should be considered. Forinstance, it may be justifiable to perform aseveral-thousand-dollar host transcriptomic ana-lyses to identify candidates for targeted immuno-modulation, because even steroids (a fairly crudeform of such therapy) are known to reduce thelength of stay in the ICU, the daily costs of whichare commensurate with such studies. Further-more, the use of bacteriologic NGS may beconsidered in such patients to facilitate institutionof highly selective antimicrobials, especially assequencing costs decrease and antimicrobial sus-ceptibility prediction improves.Conceptually, a well-designed personalized
treatment plan consisting of both antimicrobialsand immunomodulation would reduce the hyster-esis usually observed during the course of severepneumonia; that is, the deviation from the resil-ience curve depicted in Fig. 5A, B. This hysteresis
often derives from the immunopathologic conse-quences of infection, which include ARDS, renalfailure, and CARS (the downward curve inFig. 5A). Thus, even when the offending pathogenis cleared, the patient may be left with significantdebility and increased risk for secondary infec-tion. Meanwhile, excessive immunosuppressionmay seem to improve a patient’s clinical statusbut also impairs pathogen clearance and in-creases susceptibility to infection (see Fig. 5B).The ideal therapeutic regimen would thereforeinvolve selective antimicrobials with minimal tox-icities to the host, plus tailored immunomodula-tion that offsets the downward deviation fromthe curve; the combination should be designedto return patients directly to their premorbidstates (Fig. 5C).Ultimately, thismay requireamultiomicdiagnostic
platform that deeply characterizes the host, path-ogen, and their interaction alongside a comprehen-sive suite of antimicrobial therapeutics (comprisingnot only antibiotics but also inhibitors of virulencefactors and promoters of host resistance mecha-nisms) aswell as immunomodulators that offsetmal-adaptive host responses to infection and promote
A B C
Fig. 5. Hysteresis in treatment and recovery. (A) Hysteresis caused by complications of infection or drug toxicity.Administration of antibiotics adds to the potential complications of the infection. Furthermore, failure to offsetthe immunopathologic consequences of infection can predispose to secondary infection. (B) Hysteresis caused bycomplications of immunosuppression. Immunomodulatory drugs such as steroids offset immunologic disorderbut also carry the risk of secondary infection. (C) Absence of hysteresis caused by balanced antibiosis and immu-nomodulation. The ideal combination of antibiotics and adjunctive therapies results in diminished hysteresis,which may be achieved through the use of highly selective antimicrobials, targeted immunosuppression, andminimizing the risks associated with both.
Personalizing the Management of Pneumonia 891
resolution of inflammatory responses. Biomarkersshouldbedeveloped toguide subsequent cessationof antibiotics. Complementary strategies for pre-venting secondary infections and restoring micro-biomic homeostasis should also be developed. Inshort, the goal is to not only improve survival frompneumonia but also limit the possible systemic andlong-term consequences of infection.
Although a lofty vision, the obstacles to the real-ization of this goal are less technical than practical.Much of the necessary technology, including hosttranscriptomics, pathogen NGS, and multiplexprotein analyses, already exists. What is needednow is a recognition both within the field andbeyond of the clinical burden of pneumonia andthe hazards of overuse of antibiotics; this shoulddrive further research into the mechanisms ofpneumonia, development of diagnostics and ther-apeutics, streamlining of technology to reducecosts, and methods for effective clinicalimplementation.
REFERENCES
1. Ferkol T, Schraufnagel D. The global burden of res-
piratory disease. Ann Am Thorac Soc 2014;11(3):
404–6.
2. Cookson W, Cox MJ, Moffatt MF. New opportunities
for managing acute and chronic lung infections.
Nat Rev Microbiol 2018;16(2):111–20.
3. World Health Organization. Disease and injury
regional estimates, 2000–2011. 2018. Available at:
http://www.who.int/healthinfo/global_burden_dise
ase/estimates_regional_2000_2011/en/. Accessed
April 20, 2018.
4. Albaum MN, Hill LC, Murphy M, et al. Interobserver
reliability of the chest radiograph in community-
acquired pneumonia. PORT Investigators. Chest
1996;110(2):343–50.
5. Self WH, Courtney DM, McNaughton CD, et al.
High discordance of chest x-ray and computed
tomography for detection of pulmonary
opacities in ED patients: implications for diag-
nosing pneumonia. Am J Emerg Med 2013;
31(2):401–5.
6. Jain S, Self WH, Wunderink RG, et al. Community-
acquired pneumonia requiring hospitalization
among U.S. Adults. N Engl J Med 2015;373(5):
415–27.
7. Waters B, Muscedere J. A 2015 update on
ventilator-associated pneumonia: new insights on
its prevention, diagnosis, and treatment. Curr
Infect Dis Rep 2015;17(8):496.
8. Murphy TF. Vaccines for nontypeable Haemophilus
influenzae: the future is now. Clin Vaccin Immunol
2015;22(5):459–66.
9. Guest JF, Morris A. Community-acquired pneu-
monia: the annual cost to the National Health
Service in the UK. Eur Respir J 1997;10(7):
1530–4.
10. Bush K, Courvalin P, Dantas G, et al. Tackling anti-
biotic resistance. Nat Rev Microbiol 2011;9(12):
894–6.
11. Blaser MJ. Antibiotic use and its consequences for