Enhancers of Host Immune Tolerance to Bacterial Infection Discovered Using Linked Computational and Experimental Approaches

We anticipate that host tolerance will be a valuable strategy to reduce the morbidity and even mortality associated with acute infection, such as bacterial septicemia, when it is not possible to achieve sterile immunity. For example, recent work has shown that immunity to severe life‐threatening infection with malaria is underpinned by acquired mechanisms of disease tolerance,[ 45 , 46 ] and this new understanding suggests that a future focus on tolerance‐inducing treatments is warranted. The use of a broad‐spectrum host tolerance‐inducing drug may be especially useful before a vaccine is available for a novel infectious pathogen or when bacteria develop resistance to existing antibiotics or antivirals. Although the present study focused on a single treatment of tolerance‐inducing compounds, we anticipate that these drugs might synergize with antibiotic and antiviral therapies when used in combination by both inducing tolerance and reducing the burden of pathogens, which would decrease the likelihood of spreading to others while also further reducing tissue damage. However, infection tolerance could result in organisms never fully clearing the pathogen and persistence of low‐grade infection. From a public health perspective, a tolerant individual could also continue to spread the infection and endanger nontolerant individuals around them.[ 47 ] Therefore, careful consideration of public health outcomes must be considered before translation of tolerance drugs into human patients.

A unique aspect of Xenopus embryos is the lack of adaptive immunity development, enabling the study of innate immune responses in the presence of a complex organ system interaction without the overlapping adaptive response. This experimental platform enables natural isolation of the innate immune system without having to engineer an organism that could result in unpredictable effects. Although immature Xenopus elicit strong immune responses,[ 17 ] their responses are different and therefore any insights from the Xenopus embryo model will need to be validated in more adult‐like infection models. Nevertheless, the success of this integrated experimental‐bioinformatics screening approach used here demonstrates the value of the Xenopus infection model for gaining insight into mechanisms of host tolerance to infection and for discovery of broad‐spectrum tolerance‐inducing drugs. Using transcriptomic signatures and the Xenopus embryo infection screen, we identified that the hypoxia pathway involving prolyl‐4 hydroxylase and HIF‐1α play key roles in modulating infection tolerance and showed that metal ion chelators and a prolyl‐4 hydroxylase inhibitor can act as potential inducers of an active tolerance response to infection. Thus, this combined experimental and computational platform may help to accelerate the development of a new class of tolerance‐inducing therapeutics that could complement antibiotic therapies in the fight against life‐threatening infectious diseases, such as bacterial septicemia.

All experimental procedures involving Xenopus embryos were approved by the Institutional Animal Care and Use Committees (IACUC) and Tufts University Department of Laboratory Animal Medicine under protocol M2014‐79 and by the Office of the IACUC at Harvard Medical School under protocol IS00000658‐3. Xenopus laevis embryos were fertilized in vitro according to standard protocols in 0.1X Marc's Modified Ringer's solution (MMR; 10 × 10−3 m Na+, 0.2 × 10−3 m K+, 10.5 × 10−3 m Cl–, 0.2 × 10−3 m Ca2+, pH 7.8) and housed at 18 °C. Since exclusively sterile filtered solutions, including MMR culture medium, were used to avoid confounding effects of seasonally varying environmental bacteria, we dosed embryos overnight with 2.64 μL KoiZyme probiotic solution (Koi Care Kennel, Las Vegas, NV) per 1 L of 0.1% MMR to avoid dysbiosis (Figure S7, Supporting Information). Sterile solutions were used for all studies. Embryos were staged according to Nieuwkoop and Faber.[ 48 ]

Bacteria (Table 2 ) were streaked out from frozen glycerol stock of target bacterial strain on sheep blood agar media plates overnight at 37 °C. A single colony was selected from the streak and grown in LB medium at 37 °C. Exponentially growing bacteria were pelleted and resuspended in sterile saline/dextrose with 15% glycerol at a concentration of 1X109 CFU mL−1. Faber‐Nieuwkoop stage 13/14 embryos were injected with fresh bacterial suspensions using borosilicate glass needles calibrated for a bubble pressure of 25–30 kDa and 0.4 sec pulses to deliver 103–104 CFU of bacteria to each embryo (Figure S1, Supporting Information). Bacterial injections were performed with embryos submerged in 3% Ficoll prepared in 0.1X MMR. Even though bacteria were grown and concentrated following a constant methodology and embryos were collected and grown under the same conditions, variation in survival rates was observed between experiments. This was likely due to differences in bacterial growth in situ, which could be affected by differences in the genetic background of each individual from different egg clutches. For these reasons, every individual comparison within an experiment was conducted using eggs from a single fertilization and a single suspension of bacteria and replicate studies were conducted with completely independent batches of embryos.

Infection tolerance state was determined by a combination of the survival rate and pathogen burden in a group of embryos. At selected time points after infection, embryo survival was assessed by microscopic evaluation. At the same time points, embryo lysate was spiral plated onto selective media plates by the Eddy Jet 2 Spiral Plater, cultured, and counted using the Flash & Go Automated Colony Counter to determine the concentration of viable pathogen. The survival and amount of viable pathogen could then be combined into a single score that reflected the host infection tolerance state, the Host Pathogen Response Index (1) HPRI=%embryosurvivallogCFU+1+1

Embryos were exposed to ion scavenging and hypoxia‐inducing compounds from Faber‐Nieuwkoop stage 13/14, including DFOA (Sigma‐Aldrich; St. Louis, MO), hydralazine (Sigma‐Aldrich), L‐mimosine (Sigma‐Aldrich), and 1,4‐DPCA (Santa Cruz Biotechnology; Dallas, TX) (Figure S1, Supporting Information). Compounds were dissolved in DMSO and further diluted in 0.1X MMR to a range of micromolar concentrations (Table 1). Embryos infected with A. hydrophila were dosed with drugs in 12‐well plates (n = 10 embryos per well) to assess therapeutic potential. Embryo survival, levels of viable pathogen, and HPRI were assessed for each treatment at 24, 48, and 120 h postinfection. To test the primary mechanism of action for 1,4‐DPCA, embryos were co‐treated with 1,4‐DPCA and a HIF‐1α inhibitor (Santa Cruz Biotechnology). 1,4‐DPCA treatment was also tested across the circadian cycle by flipping light cycles from the normal 12/12 light/dark cycle to a 12/12 dark/light cycle using isolated light boxes within the culture incubators. Following initial drug screening and mechanism studies, prophylactic and postinfection treatment was compared for 40 × 10−6 m 1,4‐DPCA treatment at 24 h postinfection.

To determine the effects of bacterial infection and treatments on expression profiles in embryos, microarray analysis was performed at the conclusion of two experiments (Figure S1, Supporting Information). Separate microarray analyses were used to assess: 1) The effect of clinically obtained pathogen strains (Table 2) on embryos at 4 h and 28 h postinfection and 2) The therapeutic effect of prophylactic and postinfection treatment with 1,4‐DPCA on embryos infected with A. hydrophila at 24 h postinfection. For each experiment, RNA from Xenopus embryos (n = 2–3 replicates per condition) was separately extracted and purified using the RNeasy Micro Kit (Qiagen; Venlo, Netherlands) and microarray measurements were performed using the GeneChip Xenopus laevis Genome 2.0 Array (Affymetrix; Santa Clara, CA) at the Advanced Biomedical Laboratories (Cinnaminson, NJ). Microarray data were extracted from CEL files, Robust Multi‐array Average (RMA)[ 49 , 50 ] normalized in Matlab (Mathworks; Natick, MA), and expression data was log2‐transformed. The limma package[ 51 ] in R was used to determine differentially expressed genes after infection and treatment relative to uninfected controls using the Benjamini–Hochberg false discovery rate (FDR).[ 52 ] Heatmaps displaying genes that undergo the greatest differential expression were produced using the gplots package.[ 53 ] To identify tolerant‐specific genes, we filtered for genes that were differentially expressed (FDR<0.05) in the active gram negative tolerance state (K. pneumoniae and A. baumannii infections) and removed any genes that were differentially expressed in the gram negative sensitive state (P. aeruginosa and A. hydrophila infections).

The infection tolerance expression signature was assessed using gene ontology (GO) enrichment tools to highlight pathways relevant to infection tolerance. Tolerance‐specific Xenopus gene names were first converted to human orthologs based on the HGNC Comparison of Orthology Predictions (HUGO Gene Nomenclature Committee at the European Bioinformatics Institute; https://www.genenames.org/). The PANTHER Classification System[ 26 ] (http://pantherdb.org/) was used to perform a functional classification analysis based on GO[ 27 , 28 ] biological processes, molecular function, and Reactome[ 29 ] (https://reactome.org/) pathways amongst genes of the tolerance signature for Homo sapiens. Pathways were identified using the PANTHER Overrepresentation Test and sorted based on the number of genes identified along each pathway.

Gene networks for each bacterial infection were analyzed and visualized in Matlab (Mathworks; Natick, MA) to identify subnetworks and motifs of interacting genes that were activated in infection tolerance and sensitive states. The overall gene network was built using known molecular interactions from the KEGG (https://www.genome.jp/kegg/) and TRRUST databases (https://www.grnpedia.org/trrust/).[ 18 , 19 , 20 ] The active subnetworks and motifs for each infection and treatment condition were identified by searching the overall network for genes found to be significantly differentially expressed by limma analyses. Activated subnetworks and motifs include genes that were interconnected with at least one other gene and disconnected gene nodes were removed from each network. For each network, the importance of each node was determined by calculating its degree centrality, which counts the number of edges connecting to each node. The same methods were applied to previously published and processed gene expression data from S. pneumoniae‐infected mice[ 6 ] and LPS‐exposed primates.[ 7 ]

Statistical analyses of survival and pathogen load were performed using one‐way ANOVAs and Tukey's or Šídák's multiple comparisons test. These experiments used n = 3–6 replicates per group per time point with n = 10 embryos/replicate and data are presented as mean ± standard deviation. Statistics were calculated using Prism Version 9.1.2 (GraphPad; San Diego, CA). For microarray experiments, expression data were normalized by RMA in Matlab, log2‐transformed, and groups (n = 2–3 replicates per group) were compared with the R package limma using a linear model and the Benjamini–Hochberg false discovery rate (FDR) adjustment method. Across experiments, differences were considered significant for p < 0.05.