Predicting Hyperoxia-Induced Lung Injury from Associated Intestinal and Lung Dysbiosis in Neonatal Mice

Background: Preclinical studies have demonstrated that hyperoxia disrupts the intestinal barrier, changes the intestinal bacterial composition, and injures the lungs of newborn animals. Objectives: The aim of the study was to investigate the effects of hyperoxia on the lung and intestinal microbiota and the communication between intestinal and lung microbiota and to develop a predictive model for the identification of hyperoxia-induced lung injury from intestinal and lung microbiota based on machine learning algorithms in neonatal mice. Methods: Neonatal C57BL/6N mice were reared in either room air or hyperoxia (85% O2) from postnatal days 1–7. On postnatal day 7, lung and intestinal microbiota were sampled from the left lung and lower gastrointestinal tract for 16S ribosomal RNA gene sequencing. Tissue from the right lung and terminal ileum were harvested for Western blot and histology analysis. Results: Hyperoxia induced intestinal injury, decreased intestinal tight junction expression, and impaired lung alveolarization and angiogenesis in neonatal mice. Hyperoxia also altered intestinal and lung microbiota and promoted bacterial translocation from the intestine to the lung as evidenced by the presence of intestinal bacteria in the lungs of hyperoxia-exposed neonatal mice. The relative abundance of these bacterial taxa was significantly positively correlated with the increased lung cytokines. Conclusions: Neonatal hyperoxia induced intestinal and lung dysbiosis and promoted bacterial translocation from the intestine to the lung. Further studies are needed to clarify the pathophysiology of bacterial translocation to the lung.

DOI: 10.1159/000513553 oxidant stress and result in lung injury. Neonatal rodents exposed to prolonged hyperoxia displayed impaired alveolarization comparable to human bronchopulmonary dysplasia [1,2]. High concentrations of inhaled oxygen were found to increase the partial pressure of oxygen in the intestinal tissues of mice [3]. Preclinical studies have demonstrated that hyperoxia injures the distal small intestine and disrupts the intestinal function in neonatal animals [4,5]. Disruption of the gut barrier function induces translocation of pathogens through the epithelial layer to the lamina propria and then to the mesenteric lymph nodes, where they invade the bloodstream and disseminate to other sterile organs [6].
Hyperoxia exposure has been observed to change the bacterial composition of neonatal mice and rats' intestines [7,8]. We found that hyperoxia increased intestinal permeability and the translocation of bacteria from the gastrointestinal tract to the liver and spleen in newborn rats [9]. The aims were to investigate the effects of hyperoxia on the lung and intestinal microbiota and the communication between intestinal and lung microbiota and to develop a predictive model for the identification of hyperoxia-induced lung injury from intestinal and lung microbiota based on machine learning algorithms in neonatal mice.

Experimental Groups
We conducted experiments in accordance with regulations of the Institutional Animal Care and Use Committee of Taipei Medical University. Time-dated pregnant C57BL/6N mice were housed in individual cages with free access to laboratory food and water. Within 12 h of birth, the litters were pooled and randomly redistributed among the newly delivered mothers, and the pups were randomly assigned to be reared in room air (RA) or O 2 -enriched air. The pups in the hyperoxia (O 2 , normobaric) group were reared in an atmosphere containing 85% O 2 during postnatal days 1-7. The pups in the RA group were reared in RA during postnatal days 1-7. On postnatal day 7, the mouse pups were euthanized, lung and intestinal microbiota were sampled from the left lung and lower gastrointestinal tract, and right lung and terminal ileum were harvested.

Lung and Intestinal Histology
To standardize the analysis, sections were obtained from the right middle lobe, stained with hematoxylin and eosin, and assessed for lung and intestinal morphometry. Mean linear intercept, an indicator of mean alveolar diameter, was assessed [10]. Radial alveolar count is the average number of alveoli transected by a perpendicular line drawn from the center of a respiratory bronchiole to the nearest septal division. Vascular density was determined with the von Willebrand factor (vWF) immunohistochemistry reaction.
Microvessel density was determined by counting the vessels with positive vWF staining in an unbiased manner and in a minimum of 4 random lung fields at ×400 magnification [11]. The intestinal mucosal injury was scored using a scale from 0 to 5 [12].

Western Blot Analysis of Lung VEGF
Lung tissues were homogenized and incubated with antibodies against VEGF (Santa Cruz Biotechnology) and anti-β-actin (Sigma-Aldrich, St. Louis, MO, USA) and subsequently with horseradish peroxidase-conjugated goat anti-mouse IgG (Pierce Biotechnology, Rockford, IL, USA).

16S rRNA Gene Sequencing
The methods of 16S rDNA analysis were described in Yang et al. [13]. The library preparation follows the protocol of 16S Ribosomal RNA Gene Amplicons for the Illumina MiSeq System. The alpha diversity indices were calculated using the estimate_richness function from the phyloseq package. UniFrac distances were calculated using the GUniFrac package (v1.1) to assess the community dissimilarity between groups [14]. The enrichment analysis between groups was analyzed using the linear discriminant analysis effect size method with the Wilcoxon-Mann-Whitney test (at α = 0.05) and logarithmic linear discriminant analysis score more than 2 and visualized as cladogram using GraPhlAn [15,16].

Predictive Modeling to Classify RA-and O 2 -Treated Groups Using Intestinal and Lung Microbiota
Machine learning algorithms have been shown to be effective for comparing microbiota and predicting diseases [17,18] further depict the discriminative power of the lung and intestinal microbiota, the genus-level relative abundances of microbiota and lung cytokines were further incorporated as input features for distinguishing between RA-and hyperoxia-reared mice. Two machine algorithms (decision tree and Bayes network) were applied to predict the classes of the mice based on lung cytokines and microbiota. Due to the small sample size, leave-one-out cross-validation was incorporated to evaluate the predictive performance. One mouse was selected as the test set, and the remaining mice were used as the training set to create a machine learning model to predict the test sets to which mice belonged. This process was repeated for all mice in the data set. Subsequently, several evaluation measures including accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve were used to evaluate the predictive performance.

Survival Rate and Body Weight
Four pregnant C57BL/6N mice gave birth to a total of 22 pups; 9 and 13 pups were distributed to the RA and hyperoxia groups, respectively. All 9 mice reared in RA Hyperoxia-induced intestinal injury and decreased intestinal tight junction expression on postnatal day 7. a Representative intestinal sections from mice exposed to postnatal RA or hyperoxia that were stained with hematoxylin and eosin. The mice reared in RA exhibited a normal intestinal mucosal framework, and the mice reared in hyperoxia had malalignment and distended basolateral intercellular spaces in the epithelium; additionally, they exhibited a significantly higher intestinal injury score than did RA-reared mice. b, c Representative immunohistochemistry and Western blotting for occludin and ZO-1. Both occludin and ZO-1 (arrow) were observed on the side adjacent to the cell membranes of the enterocytes. The hyperoxia-reared mice exhibited disrupted occludin and ZO-1 immunohistochemistry between adjacent enterocytes. The hyperoxia-reared mice exhibited significantly lower occludin and ZO-1 protein levels than did RA-reared mice. The insert shows the positively stained cells and nuclei in more detail. Data are presented as mean ± SD p value was calculated by Student's t test. *p < 0.05, ***p < 0.001. RA, room air; ZO-1, zonula occludens.
survived. However, 2 mice reared in O 2 -enriched air died on postnatal day 5. Mice reared in hyperoxia exhibited significantly lower body weights on postnatal day 7 than did those reared in RA (3.54 ± 0.47 g vs. 5.72 ± 0.53 g, p < 0.001).

Hyperoxia-Induced Intestinal Injury and Decreased Intestinal Tight Junction Expression
The mice reared in RA exhibited a normal intestinal mucosal framework and well-defined intercellular space at the basal portion of the enterocytes (Fig. 1a). The mice reared in hyperoxia had malalignment and distended basolateral intercellular spaces in the epithelium and exhib-ited a significantly higher intestinal injury score than did RA-reared mice. Both occludin and ZO-1 were observed on the side adjacent to the cell membranes of the enterocytes (Fig. 2b, c). The hyperoxia-reared mice exhibited disrupted occludin and ZO-1 immunohistochemistry between adjacent enterocytes. The hyperoxia-reared mice exhibited significantly lower occludin and ZO-1 protein levels than did RA-reared mice.

Hyperoxia-Impaired Alveolarization and Angiogenesis in Neonatal Mice
The mice reared in hyperoxia exhibited large thinwalled air spaces and significantly higher mean linear in- Hyperoxia-impaired alveolarization and angiogenesis and increased lung cytokines on postnatal day 7. a Representative lung sections from mice exposed to postnatal RA or hyperoxia that were stained with hematoxylin and eosin. Those from mice reared in hyperoxia exhibited large thin-walled air spaces and significantly higher MLI and lower RAC than did those from mice reared in RA. b, c Representative immunohistochemistry and Western blotting for VEGF and vWF, and (d) lung cytokines. The mice reared in hyperoxia exhibited decreased VEGF (arrow) and vWF (arrow) immunoreactivity. Western blot analysis and semiquantitative analysis revealed that the mice reared in hyperoxia exhibited sig-nificantly decreased VEGF protein expression and vascular density than did those reared in RA. The mice reared in hyperoxia exhibited significantly higher IL-1β, TNF-α, and MIP-2 levels than did those reared in RA. The insert shows the positively stained cells and nuclei in more detail. Data are presented as mean ± SD p value was calculated by Student's t test. *p < 0.05, ***p < 0.001. RA, room air; MLI, mean linear intercept; RAC, radial alveolar count; vWF, von Willebrand factor; IL-1β, interleukin-1β; TNF-α, tumor necrosis factor-α; MIP-2, macrophage inflammatory protein-2; VEGF, vascular endothelial factor.  tercept and lower radial alveolar count than did those from mice reared in RA (Fig. 2a). The mice reared in hyperoxia exhibited significantly decreased VEGF and vWF immunoreactivity (Fig. 2b, c). Western blot analysis and semiquantitative analysis revealed that the mice reared in hyperoxia exhibited significantly decreased VEGF protein expression and vascular density than did those reared in RA.

Hyperoxia Increased Lung Cytokines Levels
The mice reared in hyperoxia exhibited significantly higher IL-1β, tumor necrosis factor-α, and MIP-2 levels than did those reared in RA (Fig. 2d).

Hyperoxia Altered Intestinal and Lung Microbiota
Hyperoxia treatment was revealed to influence the community richness and reduce alpha diversity in the intestine microbiota (Fig. 3a). Nonmetric multidimensional scaling also revealed that hyperoxia-reared mice displayed an intestine microbiota profile different from that of the RA-reared mice (Fig. 3b). In hyperoxia-reared mice, several significant microbial changes occurred (as shown in Figure 3c and d. Firmictes, Epsilonbacteraeota, and Actinobacteria were significantly decreased in the microbial composition at the phylum level compared with in RA-reared mice (Fig. 3c, d).
Evaluation of the lung microbiota revealed that hyperoxia-reared mice exhibited significantly increased alpha diversity compared with RA-reared mice (Fig. 3e). Nonmetric multidimensional scaling analysis showed that hyperoxia-reared mice displayed different lung microbiota profiles compared with RA-reared mice (Fig. 3f). Significant changes occurred as shown in Figure 3g and h.

Hyperoxia Promoted Bacterial Translocation from the Intestine to the Lung
In this study, to explore the relevance of the gut-lung axis, we analyzed the types of strains at the genus level exist in the intestines of both types of mice but only in the lungs of hyperoxia-reared mice (Table 1). In addition, Spearman correlation coefficients revealed a significant positive correlation between lung microbiota translocated from the intestine and pulmonary inflammatory cytokines ( Table 2).

Use of Machine Learning Algorithms for Accurately Classifying RA-and Hyperoxia-Treated Groups Based on Lung Cytokines and Microbiota
The lung and intestinal microbiota (10 from lungs and 10 from intestines) analyzed in Table 1 and 3 lung cytokines were incorporated to develop prediction models. The use of the microbiota combined with cytokines yielded accurate predictive results evaluated through leaveone-out cross-validation. Two algorithms, including a Bayes network and decision tree, yielded satisfactory predictive performance. The Bayes network performed better than did the decision tree for accuracy, sensitivity, specificity, and area under the curve, attaining values of 94.4, 88.9, 100%, and 0.963, respectively. The decision tree attains the values of accuracy, sensitivity, specificity, and area under the curve 88.9, 88.9, 88.9%, and 0.940, respectively. Moreover, to further identify the important variables for the classification of RA-and hyperoxiareared mice, the random forest algorithm was incorporated to select discriminative features, as shown in Figure 4.

Discussion
Our in vivo model demonstrated that hyperoxia exposure during the first 7 days after birth induced intestinal and lung injury in neonatal mice, as evidenced by increased intestinal injury scores, decreased intestinal barrier integrity, increased lung inflammation, and reduced alveolarization and angiogenesis. Hyperoxia-induced intestinal and lung injuries were associated with intestinal and lung dysbiosis. Several pathogenic bacteria present in the gastrointestinal tract were found in the lungs of hyperoxia-exposed newborn mice, and this suggested a significant positive correlation between the translocation of lung microbiota from the intestines and the presence of pulmonary cytokines. These findings suggest that translocation of intestinal bacteria to the lungs contributes to hyperoxia-induced lung injury and that depletion of the intestinal microbiome might reduce hyperoxia-induced lung injury. These results support the concept that the lung microbiome represents a novel therapeutic target for the prevention and treatment of lung injury [19].
Alterations in the intestinal microbiota composition contribute to the pathogenesis of many pulmonary diseases including chronic obstructive pulmonary disease and asthma [20,21]. In this study, we demonstrated that hyperoxia-induced intestinal and lung injury and altered intestinal and lung microbiota in neonatal mice. The lung microbiota was enriched with intestinal bacteria, and intestinal abundance was significantly positively correlated with lung cytokines. These results indicate that the intestinal microbiota contributes to the pathophysiology of hyperoxia-induced lung injury in neonatal mice. These findings are compatible with the finding of Dolma et al. [22], who found that in neonatal mice, the lung structure and mechanics are protected and inflammation is decreased for germ-free specimens.
In this study, we found that the lung microbiome was primarily dominated by the phyla Bacteroidetes, Firmicutes, Proteobacteria, and Actinobacteria on postnatal day 7, which is consistent with the findings of Singh and colleagues [23]. We also observed that several gut-associated bacteria in the lungs of hyperoxia-reared mice and their relative abundance were significantly positively correlated with lung inflammation. Bacterial translocation is defined as the passage of viable microorganisms from the gastrointestinal tract through the epithelial barrier to the mesenteric lymph nodes and other extraintestinal organs and sites [24]. Neonatal hyperoxia-induced intestinal injury fulfills 2 of the 3 primary mechanisms promoting   [25]. These results suggest that translocation of intestinal bacteria to the lungs contributes to hyperoxia-induced lung injury in neonatal mice.
To investigate the effects of biomedical features on hyperoxia, partial dependence plots were incorporated to evaluate the marginal effects of variables on class probability for classification. Figure 4b-e illustrates partial dependence plots depicting the effects of important variables on the hyperoxia-reared group. As suggested in Figure 4b, when the value for the most influential variable, IL-1β, is more than 331, a high propensity exists for the hyperoxia-reared group. In addition, Figure 4c demonstrates that when MIP-2 is higher than 1,011, the probability that the specimen belongs to the hyperoxia-reared group is higher. The results from the proposed computational approaches correspond well with medical insights. Therefore, we believe that our predictive models can be incorporated into clinical practice and may provide interpretable biomedical features for future research.

Conclusions
This study demonstrated that hyperoxia induced intestinal and lung injury and dysbiosis and that the lung microbiota was enriched with gut-associated bacteria in neonatal mice. These findings suggest that the translocation of intestinal bacteria to the lungs contributes to hyperoxia-induced lung injury in neonatal mice. These results recommend the potential for manipulation of the gut microbiota in the prevention and treatment of hyperoxia-induced lung injury. Further investigation is required to clarify how intestinal bacteria translocation to the lung occurs after hyperoxia exposure.

Statement of Ethics
The animal study was reviewed and approved by Animal care and Use Committee at Taipei Medical University (LAC-2019-029).