Altered microbiota-host metabolic cross talk preceding neutropenic fever in patients with acute leukemia

Abstract

Despite antibiotic prophylaxis, most patients with acute leukemia receiving mucotoxic chemotherapy develop neutropenic fever (NF), many cases of which remain without a documented etiology. Antibiotics disrupt the gut microbiota, with adverse clinical consequences, such as Clostridioides difficile infection. A better understanding of NF pathogenesis could inform the development of novel therapeutics without deleterious effects on the microbiota. We hypothesized that metabolites absorbed from the gut to the bloodstream modulate pyrogenic and inflammatory pathways. Longitudinal profiling of the gut microbiota in 2 cohorts of patients with acute leukemia showed that Akkermansia expansion in the gut was associated with an increased risk for NF. As a prototype mucolytic genus, Akkermansia may influence the absorption of luminal metabolites; thus, its association with NF supported our metabolomics hypothesis. Longitudinal profiling of the serum metabolome identified a signature associated with gut Akkermansia and 1 with NF. Importantly, these 2 signatures overlapped in metabolites in the γ-glutamyl cycle, suggesting oxidative stress as a mediator involved in Akkermansia-related NF. In addition, the level of gut microbial-derived indole compounds increased after Akkermansia expansion and decreased before NF, suggesting their role in mediating the anti-inflammatory effects of Akkermansia, as seen predominantly in healthy individuals. These results suggest that Akkermansia regulates microbiota-host metabolic cross talk by modulating the mucosal interface. The clinical context, including factors influencing microbiota composition, determines the type of metabolites absorbed through the gut barrier and their net effect on the host. Our findings identify novel aspects of NF pathogenesis that could be targets for precision therapeutics. This trial was registered at www.clinicaltrials.gov as #NCT03316456.

© 2021 by The American Society of Hematology. Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0), permitting only noncommercial, nonderivative use with attribution. All other rights reserved.

Figures

Graphical abstract

Figure 1.

Consort diagram of patients, samples, and major analytic steps. All serum samples from cohort 1 and some from cohort 2 had been exhausted in prior projects; thus, the remaining samples from cohort 2 were used for analysis.

Figure 2.

Gut microbiome and NF in patients with AML. (A) Relative abundances of genera in the 2 cohorts. Mean relative abundances across all samples are shown. Genera with a relative abundance < 1% are shown in aggregate. (B) LDA coupled with effect size measurements (LEfSe) in the 2 cohorts. Differentially abundant OTUs (LDA score > 3.0; P < .05) in high-risk samples are shown as red bars to the right; those more abundant in low-risk samples are shown as blue bars to the left. “f” and “g” in the plot for cohort 2 indicate family and genus, respectively. All significant taxa in cohort 1 were classifiable to the genus level, thus “f” and “g” notations are not used. High- and low-risk samples are defined by whether were followed or not by a new episode of NF within 7 days, respectively. (C) Relative abundance of Akkermansia in stool samples followed vs not followed by a new episode of NF within 7 days, with the 2 cohorts combined. A horizontal jitter was added for better visualization. The P value was derived from Welch’s t test. (D) Receiver operating characteristic curve analysis for the relative abundance of Akkermansia as a predictor of a new episode of NF within the next 7 days. AUC, area under the curve; CI, confidence interval.

Figure 3.

Serum metabolomics associated with NF. (A) Distribution of serum metabolites into pathways. (B) Volcano plot showing the magnitude and significance of metabolomics changes within 24 hours before NF. High-risk (preceding NF within 24 hours) and low-risk samples (not preceding NF within 24 hours) were compared. Points to the right (left) of the vertical line (no change) represent metabolites that were increased (decreased) before NF. Select metabolites of interest in this work are labeled (the orange circle representing p-cresol sulfate is nearly completely hidden behind 2 green circles). The horizontal lines represent q = .05 and q = .1, above which the metabolites are statistically significant according to the corresponding threshold. p values from Welch’s t-test were corrected for multiple testing to derive the q values. (C) Distribution of metabolites with q < .1 in (B) into pathways. (D) Overrepresentation analysis using subpathways. Metabolites with q < .05 in (B) were considered for overrepresentation of their subpathways using a hypergeometric test with a corrected P value threshold of .05. Each central node represents an overrepresented subpathway, and the peripheral nodes connected to the central node represent the metabolites within that subpathway that are significant in (B). The number of these metabolites in each overrepresented subpathway is proportional to the size of the central node. The complete list of metabolites is provided in supplemental Data 1.

Figure 4.

Serum metabolomics associated with gut Akkermansia. (A) Distribution of serum metabolites into pathways, provided to facilitate comparison with (C). (B) Volcano plot showing the strength and significance of the association between serum metabolite levels (outcome variable) and abundance of gut Akkermansia (predictor) in the nearest prior stool sample. For each metabolite, a separate mixed effect regression was created, adjusting for sex, use of parenteral nutrition, serum sample day, and patient number (random effect). The regression coefficient for Akkermansia was considered its effect size (x-axis), and the corresponding P value was corrected for multiple testing to derive the q value (y-axis). Select metabolites of interest in this work are labeled. The horizontal lines represent q = .05 and q = .1, above which the metabolites are statistically significant according to the corresponding threshold. (C) Distribution of metabolites with q < .1 in (B) into pathways. (D) Overrepresentation analysis using subpathways. Metabolites with q < .05 in (B) were considered for overrepresentation of their subpathways using a hypergeometric test with a corrected P value threshold of .05. Each central node represents an overrepresented subpathway, and the peripheral nodes connected to the central node represent the metabolites within that subpathway that are significant in (B). The number of these metabolites in each overrepresented subpathway is proportional to the size of the central node. The complete list of metabolites is provided in supplemental Data 2.

Figure 5.

Changes in the level of select metabolites before NF. (A) A total of 310 serum metabolites were associated with NF, 51 metabolites were associated with abundance of gut Akkermansia, and 9 metabolites were associated with both. (B-I) Serum samples preceding vs not preceding NF within 24 hours were compared for select metabolites using Welch’s t test. (B-E) show colonic bacteria–derived metabolites from dietary tyrosine and tryptophan degradation. (F-H) show select metabolites in the γ-glutamyl cycle. Metabolites in (B-H) were measured by UPLC-MS/MS. GGT in (I) was measured fluorometrically. The boxplots show the median and interquartile range.