Pharyngeal Microbial Signatures Are Predictive of the Risk of Fungal Pneumonia in Hematologic Patients

ABSTRACT

The ability to predict invasive fungal infections (IFI) in patients with hematological malignancies is fundamental for successful therapy. Although gut dysbiosis is known to occur in hematological patients, whether airway dysbiosis also contributes to the risk of IFI has not been investigated. Nasal and oropharyngeal swabs were collected for functional microbiota characterization in 173 patients with hematological malignancies recruited in a multicenter, prospective, observational study and stratified according to the risk of developing IFI. A lower microbial richness and evenness were found in the pharyngeal microbiota of high-risk patients that were associated with a distinct taxonomic and metabolic profile. A murine model of IFI provided biologic plausibility for the finding that loss of protective anaerobes, such as Clostridiales and Bacteroidetes, along with an apparent restricted availability of tryptophan, is causally linked to the risk of IFI in hematologic patients and indicates avenues for antimicrobial stewardship and metabolic reequilibrium in IFI.

INTRODUCTION

Invasive fungal infections (IFI) still remain a major cause of nonrelapse mortality and represent one of the significant causes of expense in the management of patients with hematological malignancies (1, 2). The possibility to determine the actual risk for IFI would be of fundamental importance to drive the clinicians toward the best therapeutic option for the patients and individualize the therapy to balance between prevention of IFI and occurrence of side effects (3). Although the introduction of molecular and serological diagnostic techniques into clinical practice has significantly improved fungal diagnostics, sensitivity and specificity are still a problem.

Currently, the host-fungus interaction is being exploited to project more efficient and reliable fungal diagnostics (4), and efforts are being devoted to the implementation of clinical models predicting the infection in high-risk patients (5). Several studies have identified associations between the gut microbiome and infection risk in allogeneic hematopoietic stem cell transplant (HSCT) and graft-versus-host disease (GvHD) (6). Loss of intestinal bacterial diversity and outgrowth of opportunistic pathogens belonging to the phylum Proteobacteria and the Enterococcus genus are common during the course of HSCT and are associated with GvHD development and treatment with broad-spectrum antibiotics (6, 7). The “gut-lung axis” is commonly invoked to explain the microbiome's influence on lung immunity and inflammation, yet the lungs harbor their own microbiome and their own mucosa. As the influence of remote (gut-lung) microbe-host interactions on lung inflammation is relatively unknown (8), the assessment of lung microbial composition may better explain the role of microbial dysbiosis in lung health and disease. One recent study has shown that following HSCT, a significantly increased relative abundance of Proteobacteria in bronchoalveolar lavage (BAL) specimens is associated with histologic, immunologic, and physiologic features of pulmonary complications (9), a finding consistent with the association of Proteobacteria domination with lung dysfunction and disease (10). However, there are limited data regarding the actual predictive capacity of lung microbiome-related dysbiosis and biomarkers. In addition, although the use of BAL specimens provides valuable information to assess the microbiota of the lower respiratory tract, it represents an invasive procedure (11). Thus, considering that spatial microbial diversity in the lungs of healthy individuals seems almost absent (12) and that the lung microbiota is a community of transiently present microorganisms derived from the upper respiratory tract, the exploitation of the upper airway microbiota as a representation of the lung microbiota (11, 13) could be an attractive option that overcomes the inherent limitations associated with the sputum (11) or the bias related to the sampling method.

Based on these premises, we have designed a multicenter, prospective, observational study, termed SNIF (survey of nasal infection), in which patients with hematologic malignancies were recruited and their nasal and pharyngeal swabs collected over a 6-month period for microbiome characterization. The study was designed to identify microbial signatures that could predict the risk of IFI as a primary objective. Here, we show that the specific risk for IFI is associated with configurations of the pharyngeal microbiome along characteristic patterns.

RESULTS

The oropharyngeal microbiota of hematologic patients.

In a first analysis of the SNIF database, we focused on the oropharyngeal microbiome, owing to its higher richness, that may increase the likelihood of uncovering associations with the risk of IFI for being the main source of the lung microbiota in adult tracts (14) and for the potential protective role in pulmonary infections (15). The taxonomic composition of the oropharyngeal microbiome is shown in Fig. 1. Consistent with previous reports (16, 17), Firmicutes, Bacteroidetes, Proteobacteria, Actinobacteria, and Fusobacteria represented the major phyla, accounting for nearly the entire spectrum of bacteria detected in the oropharynx, followed by low-abundance phyla with a cumulative abundance of less than 1% (Fig. 1A). At the genus level, Streptococcus was found to be the most abundant, followed by Veillonella, Prevotella, Neisseria, Actinomyces, Haemophilus, Fusobacterium, and Rothia, common genera of the pharynx (14, 16, 17) (Fig. 1B).

FIG 1.

Oropharyngeal microbiome of hematologic patients is dominated by bacteria commonly associated with the oropharynx. (A) Bar plot showing bacterial composition (abundance percentage) of each sample at the phylum level. Taxa are differentiated by colors. Samples are ranked based on the abundance of Firmicutes. (B) Behavior of the cumulative proportions of taxa at genus level in the complete cohort versus number of observed taxa. The color of each taxon represents the prevalence of the taxon within the community. Taxa are ordered by increasing percentage value, and the first 10 most abundant genera are explicitly indicated by labels. These 10 taxa present a prevalence >50%.

Hematologic patients at different risks for IFI harbor distinct oropharyngeal microbiomes.

To define the potential predictive value of the oropharyngeal microbiome, each sample was associated with the overall risk of IFI based on the following criteria, i.e., the duration of neutropenia, antibiotic usage, and the dynamic risk proposed by SEIFEM (epidemiology survey of invasive fungal infections in hematological malignancies) (3). The details for risk stratification in each hematological malignancy, based on diagnosis, phase, and type of treatment, as well as the algorithm at the basis of the dynamic risk definition, are described in reference 3. Upon application of these criteria, samples were assigned to high risk (HR) for IFI if collected during a condition of prolonged neutropenia or administration of broad-spectrum antibiotics or if satisfying the conditions for dynamic high risk (dHR) according to the SEIFEM algorithm. Conversely, samples were assigned to low risk (LR) for IFI if collected outside periods of prolonged neutropenia or broad-spectrum antibiotics administration or if satisfying the conditions for dynamic low risk (dLR) according to the SEIFEM algorithm. The distribution for age, gender, and hematological disease for the HR and LR populations considered separately are shown in Fig. S1 in the supplemental material. The overlap between the different HR and LR populations is shown in Fig. S2. Specifically, 66.4% and 94% of the LR samples are identified by all three criteria or at least two of them, respectively, indicating a substantial degree of overlapping in the classification of patients as low risk for IFI (Fig. S2A). Conversely, only 14.3% and 29.6% of HR samples are identified by all three criteria or at least two of them, respectively (Fig. S2B to E). The difference is driven by a group of samples that are recognized as HR by the dynamic risk proposed by SEIFEM but not by prolonged neutropenia and administration of broad-spectrum antibiotics. This is in line with the dynamic risk factors that go beyond the occurrence of neutropenia and the use of broad-spectrum antibiotics (3). The analysis of alpha and beta diversities revealed that LR and HR samples were associated with distinct microbiota for all the criteria. Indeed, a higher richness and evenness, as measured by observed operational taxonomic units and Chao1 and Shannon indexes, were observed in the LR groups, i.e., no prolonged neutropenia or broad-spectrum antibiotics or assigned to dLR, than the HR groups (Fig. 2A). This result was expected, as the HR group is associated with a long period of neutropenia and, therefore, more intense pharmacological treatment that severely affects the microbial composition and diversity. The oropharyngeal microbiota of the two groups also showed significant differences in compositional structure, as measured by Jaccard and Bray-Curtis indexes (Fig. 2B). Indeed, principal coordinate analysis derived from Jaccard and Bray-Curtis distances revealed significant differences between the LR and HR groups (Fig. S3).

FIG 2.

Microbiome composition differs between LR and HR oropharyngeal samples. (A) Boxplots of observed OTUs, Chao1, and Shannon alpha diversity indexes grouped by risk of IFI according to three distinct criteria (presence or not of prolonged neutropenia, dynamic LR, or HR according to SEIFEM algorithm, and use or not of broad-spectrum antibiotics). Significance was evaluated by applying a Kruskal-Wallis test (the P value is indicated). (B) Boxplots of Jaccard and Bray-Curtis beta diversity indexes evaluating distances within (LR, green; HR, red) or between (gray) LR and HR samples according to the three distinct criteria described for panel A. Significance was evaluated by applying a Kruskal-Wallis test (the P value is indicated).

Upon applying the Adonis test to Jaccard and Bray-Curtis diversities when a linear simple multifactorial model of the dynamic risk, prolonged neutropenia, broad-spectrum antibiotics, and occurrence of mucositis was used, all the variables, except mucositis, had significantly different profiles for both metrics. Of note, the dynamic risk accounted for the largest percentage of variation (Table 1).

TABLE 1.

Multivariate analysis

Parameter	Df	SumsOfSqs	MeanSqs	F.Model	R2	Pr(>F)
Jaccard
Dynamic risk	1	2.361346588	2.361346588	7.203580256	0.024817729	0.001
Prolonged neutropenia	1	0.721783724	0.721783724	2.201890654	0.00758594	0.001
Broad-spectrum antibiotics	1	0.594637061	0.594637061	1.81401401	0.00624963	0.001
Mucositis	1	0.340901612	0.340901612	1.039962593	0.003582873	0.33
Residuals	278	91.12890092	0.327801802	NA	0.95776383	NA
Total	282	95.1475699	NAa	NA	1	NA
Bray-Curtis
Dynamic risk	1	1.579643048	1.579643048	5.350979996	0.018505499	0.001
Prolonged neutropenia	1	0.839875287	0.839875287	2.845045193	0.009839129	0.001
Broad-spectrum antibiotics	1	0.608823486	0.608823486	2.062366115	0.00713236	0.003
Mucositis	1	0.265042627	0.265042627	0.897821693	0.003104971	0.614
Residuals	278	82.06735358	0.295206308	NA	0.961418042	NA
Total	282	85.36073802	NA	NA	1	NA

^aNA, not applicable. Df: degrees of freedom; SumsOfSqs: Sums of Squares; MeanSqs: mean squares; F.Model: F-statistics; R2: R-squared; Pr(>F): P value from F-distribution.

Collectively, these results indicate that the different criteria provide a consistent framework to identify potential signatures of the oropharyngeal microbiome that could predict the risk of IFI.

The risk of IFI is associated with distinct microbial genera.

To identify these signatures, we performed high-dimensional class comparisons using linear discriminant analysis of effect size (LEfSe) (18), by which we could detect significant differences in the bacterial communities between LR and HR groups. Despite differences at the level of some genera, a common signature could be identified that was associated with the risk of infection among the differently stratified groups. Specifically, we observed the relative predominance of members of Bacteroidetes (i.e., the Prevotella genus), Firmicutes (i.e., Clostridia with the Lachnospiranaceae family), and Actinobacteria phyla, in addition to a higher abundance of most oral taxa (Veillonella, Neisseria, Leptotrichia, and Gemella) in the LR groups. Conversely, the relative predominance of members of Firmicutes (i.e., the Staphylococcus and Enterococcus genera) variably associated with Lactobacillaceae and Gram-negative Proteobacteria (i.e., Acinetobacter and Stenotrophomonas genera) was observed in the different HR groups (Fig. 3 to 5 and Fig. S4 and S5). Similar results were obtained when samples were assigned to HR or LR for IFI if collected or not during the occurrence of mucositis (Fig. S6). The relative abundances of these genera are shown in Fig. S4.

FIG 3.

Oropharyngeal microbiomes have a different genus composition in function of prolonged neutropenia. Taxonomic visualization of statistically and biologically consistent differences between samples collected during the presence or absence of prolonged neutropenia. The cladogram simultaneously highlights high-level taxa and specific genera. Taxa (circles) are colored red when significantly (LEfSe, P < 0.05) associated with the presence of prolonged neutropenia, green when significantly associated with the absence of prolonged neutropenia, and yellow when not significantly associated with either group. The size of each circle is proportional to the abundance of the corresponding taxon in all samples. The histograms of the linear discriminant analysis (LDA) scores are computed for genera significantly associated with either the presence (red) or the absence (green) of prolonged neutropenia. LEfSe has been applied with default alpha values for the analysis of variance (ANOVA) and Wilcoxon tests (0.05), and the LDA effect size has been evaluated by setting the absolute value of the logarithmic LDA threshold equal to 3.5. Other LEfSe parameters have been set to the default.

FIG 4.

Oropharyngeal microbiomes have a different genera composition in function of dynamic risk stratification. Taxonomic visualization of statistically and biologically consistent differences between samples allocated to low risk (dLR) or high risk (dHR) for IFI by dynamic risk stratification. The cladogram simultaneously highlights high-level taxa and specific genera. Taxa (circles) are colored red when significantly (LEfSe, P < 0.05) associated with dHR, green when significantly associated with dLR, and yellow when not significantly associated with either group. The size of each circle is proportional to the abundance of the corresponding taxon in all samples. The histograms of the LDA scores are computed for genera significantly associated with either dHR (red) or dLR (green) groups. LEfSe has been applied with default alpha values for the ANOVA and Wilcoxon test (0.05) and the LDA effect size has been evaluated by setting the absolute value of the logarithmic LDA threshold equal to 3.5. Other LEfSe parameters have been set to the default.

FIG 5.

Oropharyngeal microbiomes have a different genera composition in function of broad-spectrum antibiotics use. Taxonomic visualization of statistically and biologically consistent differences between samples collected or not during treatments with broad-spectrum antibiotics. The cladogram simultaneously highlights high-level taxa and specific genera. Taxa (circles) are colored red when significantly (LEfSe, P < 0.05) associated with the use of broad-spectrum antibiotics, green when significantly associated with the absence of broad-spectrum antibiotics use, and yellow when not significantly associated with either group. The size of each circle is proportional to the abundance of the corresponding taxon in all samples. The histograms of the LDA scores are computed for genera significantly associated with either the use (red) or lack of use (green) of broad-spectrum antibiotics. LEfSe has been applied with default alpha values for the ANOVA and Wilcoxon test (0.05), and the LDA effect size has been evaluated by setting the absolute value of the logarithmic LDA threshold equal to 3.5. Other LEfSe parameters have been set to the default.

The risk of IFI is associated with distinct microbial metabolic activities.

To improve the predictive value of microbial signatures (19), we have assessed microbial functional activity in dHR and dLR groups. By using PICRUSt2 and the annotated MetaCyc database, abundances of functional pathways were inferred from 16S data, and LEfSe was applied to evaluate significant association with either HR or LR samples. In particular, genes involved in the biosynthesis of l-tryptophan (trp), glycolysis, and homolactic fermentation were more abundant in HR than LR patients, while genes involved in the fatty acid elongation and the starch degradation pathways (abundantly present in oral bacteria [20]) were more abundant in LR than HR patients (Fig. 6A and Fig. S7). The KEGG database confirmed the association of trp biosynthesis with the HR group (Fig. 6B and Fig. S8). The increased trp auxotrophy would also predict different availability of trp in the HR and LR groups. We measured these levels in a subgroup of pharyngeal samples from 14 consecutive patients undergoing HSCT during the study period and found that trp levels were indeed lower in HR than LR patients (Fig. 6C). Moreover, and importantly, while the levels of l-kynurenine (kyn), resulting from host catabolism of trp, were not different between the two groups, the production of indole-3-aldehyde (3-IAld), an indole that reflects the microbial consumption of trp, was significantly lower in HR than LR patients (Fig. 6C). These findings suggest an apparent restricted availability of trp for microbial consumption in HR patients. Consistent with this, amino acid biosynthetic pathways were abundantly present in HR samples (Fig. S7).

FIG 6.

LR and HR oropharyngeal samples differ in tryptophan metabolism. (A and B) Box plots of trp biosynthesis pathway (A) and module (B) inferred by PICRUSt2 analysis according to MetaCyc and KEGG databases, respectively. These predicted metagenome functions were indicated by LEfSe as significantly differentially represented in the high- and low-risk groups. (C) Tryptophan (trp), kynurenines (kyn), and indole-3-aldehyde (3-IAld) levels (nmol/liter) were measured in oropharyngeal samples (n = 63; LR, 27; HR, 36) and expressed as means ± standard deviations (SD). *, P < 0.05 LR versus HR, unpaired t test.

Collectively, these results reveal the existence of a transcriptionally active oropharyngeal microbiota that may impact lung immune status and suggest that not only microbial composition but also active functional activity characterizes the pharyngeal microbiota.

Validation experiments in mice.

We have resorted to a murine model of Aspergillus pneumonia for experimental validation and biological interpretation of the possible causal contribution of the discriminating taxa to the risk of IFI. For this purpose, C57BL/6 mice were intranasally infected with live A. fumigatus conidia and assessed for microbial composition and function in the lung. Using a customized 16S rRNA sequencing protocol amplifying the V3-V4 region, the most represented phyla in the lung of uninfected mice were Proteobacteria (55%), Firmicutes (31%), Actinobacteria (6%), and Bacteroidetes (4%) (Fig. S9A), a phylum distribution similar to that previously reported (21). Except for Bacteriodetes, which expanded, no major alterations were observed at the phylum level in infection (Fig. S9A). However, a significant shift in bacterial richness and diversity as well as community structure was observed. The Lactobacillaceae family (Lactobacillus genus), mostly represented in uninfected mice, slightly decreased, while strictly anaerobic bacteria belonging to both Firmicutes (Clostridiales order, Lachnospiranaceae and Clostridiaceae Family_XI and Blautia) and Bacteroidetes (Prevotellaceae, S24-7, and Bacteroides) phyla expanded (Fig. S9B). The expansion of strictly anaerobic Clostridiales occurred in the early phase of the infection, followed by a more sustained expansion of Bacteroidetes (Fig. S9C), consistent with the observation that Bacteroidetes are pH sensitive and fail to prosper in an acidic inflammatory environment (22). Within Proteobacteria, Enterobacteriaceae (Morganella and Escherichia-Shigella) expanded in infection (Fig. S9B). Strictly anaerobic bacteria are protected from fungal pneumonia, as indicated by the increased susceptibility to the infection upon treatment with metronidazole or the broad-spectrum antibiotics, meropenem and piperacillin-tazobactam, known to cause severe declines among obligate anaerobes (23) and less with ciprofloxacin (experimental protocol depicted in Fig. 7A). The decreased abundance of strict anaerobes (Clostridiales and Bacteroidetes) was associated with the expansion of facultative anaerobes of Enterobacteriaceae (Fig. 7B) and concomitantly increased fungal growth and pathology (Fig. 7C) in the lung. In agreement with these findings, Prevotella oris, a representative member of Bacteroidetes, and Peptostreptococcus anaerobius, a representative member of Clostridiales, inhibited conidial germination upon coculturing with the fungus (Fig. S10). This finding suggests that anaerobic bacteria serve a protective role in infection by attenuating fungal virulence and limiting the expansion of facultative anaerobes known to promote lung inflammation (24). Of interest, 3-IAld, more abundantly present in LR than HR patients, also protected from infection and inflammation upon administration to Aspergillus-infected mice, as indicated by reduced fungal burden and improved histopathology (Fig. 7D). 3-IAld treatment was associated with an increased expression of the aryl hydrocarbon receptor (AhR) and its target gene Cyp1a1 (Fig. 7E), in line with 3-IAld working as an agonist of AhR (25). 3-IAld treatment also induced the expression of Rorc and the production of IL-22 (Fig. 7E), indicating the expansion of protective type 3 innate lymphoid cells (ILC3) (25). Thus, murine Aspergillus pneumonia recapitulates the human microbiological findings while offering plausible explanations for the possible causal relationship between the microbial landscape, metabolic activity, and risk of IFI.

FIG 7.

Murine model of aspergillosis validate the human findings. (A) Schematic representation of the protocol with antibiotics and A. fumigatus infection. i.n., intranasal. (B and C) Microbial composition (by qPCR) (B) and fungal growth (means ± SD) and histology (periodic acid-Schiff staining) (C) of lung from C57BL/6 mice intranasally infected with viable resting A. fumigatus conidia and treated with antibiotics 2 weeks before and continuing for a week after the infection. Pip/Taz, piperacillin-tazobactam. (D and E) Fungal growth, histology (D), and transcription factor gene expression (RT-PCR) and cytokine production (ELISA) (E) in infected C57BL/6 mice treated with an oral formulation of indole-3-aldehyde (3-IAld) 4 times the week before the infection. Assays were done at the end of treatments. **, P < 0.01; ***, P < 0.001, treated versus untreated (none) mice. Naïve, uninfected mice.

DISCUSSION

Here, we have described the potential role of oropharyngeal microbiota in predicting the risk of IFI in hematologic patients. We have found that loss of alpha diversity, associated with decreased abundance of Clostridiales, Bacteroidetes, and common oral taxa, the relative expansion of Staphylococcus and Enterococcus, and a distinct metabolic profile characterize the HR patients. Interestingly, depletion of common taxa typically associated with oral health, including Actinomyces, Gemella, and Veillonella, has been reported in oral microbiota of patients undergoing 5-fluorouracil- or doxorubicin-based chemotherapy and negatively correlated with mucositis severity (26). Conversely, the role of Staphylococcus species in chronic inflammatory airway diseases (27) and infections in various hematological malignancies (28) is well documented. Interestingly, the expansion of Staphylococcus in oral and stool microbiome has been reported in a leukemic patient prior to invasive mucormycosis (29). The increased prevalence of the Enterococcus genus in HR patients is also of particular interest, as it is present not only in chronic obstructive pulmonary disease (30) but also in the pharyngeal microbiome of patients with hematological malignancies (31) and is causally associated with GvHD in acute myeloid leukemia (AML) patients (32). In addition, in a mouse intravenous chemotherapy model, upon 5-fluorouracil administration, C. albicans infection resulted in increased abundance of the Enterococcus genus in the oral mucosa, potentially responsible for the mucosal barrier impairment and fungal invasion (33).

The metabolic profiling helped to elucidate the possible pathogenic role of the HR-associated microbes to dysbiosis. While the glycolytic pathways are generally associated with inflammation, the increased biosynthesis of trp is of particular interest, as bacterial trp prototrophy has been associated with immune evasion and adaptation (34, 35). Indeed, on a metabolic level, the increased trp biosynthesis was associated with low levels of trp and indole observed in HR patients. Considering the influence of host and microbial trp metabolism on human health status (36, 37) as well as the occurrence of metabolic immune programming during pulmonary infections (38), these results point to a dysregulated metabolic activity linked to bacterial virulence and host inflammation (37). Fecal indole, as a potential surrogate marker for microbial diversity and specific taxa, has recently been described in HSCT recipients (39). As antibiotics alter the metabolic state of bacteria (40), it is likely that the more intense prophylactic regimen in HR patients is responsible for the observed metabolic profile in addition to the loss of bacterial diversity. In this regard, our murine results suggest that anaerobic bacteria serve a protective role in infection by attenuating fungal virulence and limiting the expansion of facultative anaerobes known to promote lung inflammation (24). In addition, they confirm the deleterious effects of anaerobe-targeting broad-spectrum antibiotics on the outcome of fungal pneumonia (23, 41, 42) and support the human findings. Indeed, the failure of strict anaerobes Clostridiales and Bacteroidetes to expand was associated with increased fungal virulence and the expansion of pathogenic Proteobacteria. Thus, although specific Prevotella species may exhibit different properties (43), the general beneficial role of Bacteriodetes in lung homeostasis (43), immune tolerance (44), and prevention of Th17 cell activation (45) is consistent with the abundance of Prevotella spp. in LR patients and loss in HR patients. Similar to what was observed in patients with AML following induction therapy (46), in our HR cohort, we did not observe the expansion of pathogenic Proteobacteria, probably because of first-line therapy with broad-spectrum piperacillin-tazobactam or carbapenems. Interestingly, however, Gammaproteobacteria that usually benefit from airway inflammation (47) and whose domination in the feces was a strong independent predictor of pulmonary complications in HSCT (7) were more abundantly present in HR than LR patients. Both Acinetobacter baumannii and Stenotrophomonas maltophilia are fatal infectious complications in HSCT (48, 49).

Overall, these results, while consistent with previous findings in gut (7), provide evidence for significant differences in oral microbial composition of patients at different risks for IFI. Supported by the high number of patients enrolled, the results clearly highlight the major contribution of neutropenia, the associated antibiotic use, and the occurrence of mucositis in the differences observed between HR and LR patients. As such, these results may pave the way for further studies to uncover associations between the many different risk factors for IFI and the changes in the microbiome. In addition, considering the tolerability and rapidity of the oropharyngeal swabbing, including the advantage of self-collection (50), this sampling method appears highly feasible and relatively low cost. Thus, if corroborated by further studies, the loss of alpha diversity associated with the loss of beneficial Clostridiales and Bacteroidetes could help delineate patients at risk of IFI, thereby providing information for antimicrobial therapy optimization. Indeed, HR patients would benefit from an antifungal prophylactic-based approach, as opposed to LR patients, for whom a fungal diagnostic-based approach is recommended to reduce overtreatment and unintended collateral damage to beneficial commensals. Discriminating patients who will benefit or not from antimicrobial prophylaxis will also help reduce the antimicrobial resistance crisis. In this regard, this study is a proof-of-concept demonstration that microbial metabolites, such as 3-IAld, known to protect from GvHD in murine HSCT (51), also reduce the risk of IFI in these patients. Indeed, 3-IAld could protect mice against aspergillosis, likely by activating AhR in ILC3 and inducing the production of interleukin-22 (IL-22). This mechanism recently has been shown to protect mice against mucosal candidiasis and dextran sodium sulfate-induced colitis (25, 52). Our results would expand the potential application of 3-IAld in the respiratory tract, in line with its suggested efficacy in cystic fibrosis (53). These wide-ranging activities are associated with a favorable safety profile (54), as expected for a molecule endogenously produced under physiological conditions, a fundamental prerequisite for clinical translation. More generally, this type of study reinforces the notion that mining the microbiota for microbial- and metabolite-based therapies could inform future interventional strategies in patients at risk for IFI.

MATERIALS AND METHODS

Study design and data collection.

The study was approved by the institutional review board of the University of Perugia, Italy, protocol number 2014-026. All subjects gave written informed consent in accordance with the Declaration of Helsinki. A total of 173 patients diagnosed with hematological malignancies were enrolled between November 2015 and November 2017. Clinical records from each patient were reviewed, and demographic and clinical data, including age, gender, disease, chemotherapeutic regimen, neutropenia, antimicrobial treatments (Table 2), transplantation type, and conditioning regimens (see Table S1 in the supplemental material), were recorded. The risk of IFI rather than the actual incidence of IFI was chosen as a surrogate primary endpoint to overcome the inherent diagnostics variability and inconsistencies of a multicenter study. In addition, as the epidemiology of IFI in hematological patients has shifted in recent years from a yeast (mainly Candida spp.) to a mold (mainly Aspergillus spp.) predominant etiology (55), the term IFI points to Aspergillus spp. as the most common causative agent in this study.

TABLE 2.

Characteristics of patients

Parameter	Value(s)
Patients
Total, n	173
Age in yr, median (range)	54.5 (25–65)
Female, n (%)	72 (41.6)
Male, n (%)	101 (58.4)
Pharyngeal swabs (no.)	945
Underlying disease, n (%)
Acute myeloid leukemia	51 (29.5)
Acute lymphoid leukemia	9 (5.2)
Lymphoma	58 (33.5)
Myeloma	44 (25.5)
Chronic lymphatic leukemia	3 (1.7)
Myelofibrosis	2 (1.1)
Chronic myeloid leukemia	1 (0.6)
Aplastic anemia	1 (0.6)
Others	4 (2.3)
Disease status, n (%)
At diagnosis	59 (34.1)
Complete remission	63 (36.4)
Partial remission	19 (11)
Refractory/relapse	14 (8.1)
Progression	18 (10.4)
Treatment, n (%)
No. of cycles	424
Myeloablative chemotherapy	201 (47.4)
Reduced intensity chemotherapy	67 (15.8)
Others (i.e., monoclonal antibodies, protease inhibitors, and immunomodulatory drugs)	156 (36.8)
Cycles of steroid therapy
No. of cycles	295 (69.6)
Duration, >7 days	94 (31.9)
Neutropenia (therapy related), n (%)
Severe (<500 neutrophils/μl)	213 (50.2)
Duration, >10 days	135 (63.4)
Antibiotics, no. (%)
Total administered	458
Prophylaxisa	105 (22.9)
Therapy	353 (77.1)
Broad-spectrum antibiotics (i.e., piperacillin-tazobactam or ceftazidime ± aminoglycoside or tigecycline; carbapenem ± aminoglycoside or tigecycline)	258 (73)
Narrow-spectrum antibiotics (i.e., fluorochinolones or oral penicillin but not antimicrobial combinations)	95 (27)
Antifungals, no. (%)
Total administered	159
Prophylaxis (fluconazole [n = 42], micafungin [n = 17], l-AmB [n = 14], itraconazole [n = 17], voriconazole [n = 3], posaconazole [n = 11], anidulafungin [n = 1])	105 (66)
Therapy (voriconazole [n = 9], l-AmB [n = 18], caspofungin [n = 10], anidulafungin [n = 3], micafungin [n = 1], posaconazole [n = 1], itraconazole [n = 2], fluconazole [n = 6], isavuconazole [n = 4])	54 (34)

^aSeventy-three of 105 (69.5) are PCP prophylaxis, and 32 of 105 (30.5) are antibacterial prophylaxis.

Sample collection, processing, and sequencing for microbial composition.

A total of 945 pharyngeal swabs were collected, on average, monthly from the diagnosis and up to 6 months (Table 2). Swabbing was performed by the treating physicians at the time of outpatient visits according to standard procedures. Swabs were stored in liquid Amies medium (Copan Diagnostics Inc.) in each participating center before delivery to the University of Perugia for sample extraction. On the day of extraction, pharyngeal samples were thawed on ice, transferred into 2-ml Eppendorf tubes, and centrifuged at 4°C for 15 min. The supernatant was collected and stored, while the pellet further processed for DNA extraction and sequencing as described in the supplemental material.

Mice, infection, and treatment.

C57BL/6 female mice, 6 to 8 weeks old (Charles River Laboratories, Calco, Italy), were anesthetized by inhalation of 3% isoflurane in oxygen before intranasal (i.n.) instillation of 2 × 10⁷/20 μl saline viable resting conidia from the A. fumigatus AF293 strain (56), treated, and analyzed as described in the supplemental material. Murine experiments were performed according to the Italian Approved Animal Welfare Authorization 360/2015-PR and legislative degree 26/2014 regarding the animal license obtained by the Italian Ministry of Health, lasting for 5 years (2015 to 2020).

Murine 16S RNA sequencing and qPCR.

DNA was isolated from murine lung using glass beads (Sigma) followed by a QIAamp DNA minikit (Qiagen). The bacterial microbiota was evaluated by 16S rRNA. The V3-V4 region of the bacterial 16S genes was sequenced using the MiSeq platform (Illumina). Sequencing libraries were prepared using a NEXTERA XT DNA sample preparation kit (Illumina). Statistical analysis was performed as described in the supplemental material. Quantification of bacteria was done from standard curves established by quantitative PCR (qPCR) as previously described (25, 57). qPCR was performed with Power SYBR green PCR master mix (Applied Biosystems) using a StepOne Plus (Applied Biosystems) thermocycler. A short segment of the 16S rRNA gene was amplified using a conserved 16S rRNA-specific primer pair to determine the total amount of commensal bacteria. Using the same genomic DNA from each sample, group-specific primers were used to determine the amount of bacteria in the groups Enterobacteriaceae, Clostridia, and Bacteroidetes. Bacterial numbers were determined using standard curves constructed with reference bacteria specific for each bacterial group analyzed.

Real-time PCR.

Real-time PCR was performed using the iCycler iQ detection system (Bio-Rad) and iTaq universal SYBR green supermix (Bio-Rad). Total RNA was extracted using the RNeasy minikit (Qiagen, Milan, Italy) and reverse transcribed with Sensiscript reverse transcriptase (Qiagen). Amplification efficiencies were normalized against β-actin.

Cytokine and metabolite quantification.

The levels of IL-22 in lung homogenates were determined by using a specific enzyme-linked immunosorbent assay (ELISA) kit according to the manufacturer’s instructions (BioLegend). l-Tryptophan (trp), l-kynurenine (kyn), and 3-IAld were determined in pharyngeal samples by ultrahigh-performance liquid chromatography (UHPLC) and tandem mass spectrometry (MS/MS) as described in the supplemental material.