A novel circRNA–miRNA–mRNA regulatory axis as a sex-specific biological variable in bronchopulmonary dysplasia

Abstract

Babies born prematurely with gestational age <28 weeks usually develop a severe pulmonary complication called Bronchopulmonary Dysplasia (BPD) in response to hyperoxia. BPD is a sexually dimorphic pediatric disease with no curative options. Several micro RNAs (miRNAs) are implicated in BPD and are expressed differentially in males and females. Circular (circ) RNAs serve as sponges for their dedicated miRNAs to influence transcription and translation. We tested the dimorphism of these circular RNAs in BPD to assess their biological significance and unravel a circRNA–miRNA–mRNA regulatory axis in response to hyperoxia. Our RNA-Sequencing assay identified approximately 33 000 circRNAs at the alveolar stage of development with only a handful of them being expressed differentially between males and females in the control room air (RA) and hyperoxia-treated (BPD) groups. One circRNA, i.e circNfix was found to be associated with miR204-5p that targets the downstream mRNA target Ntrk2. To prove that circNfix regulates Ntrk2, we silenced circNfix using a GapmeR and found that Ntrk2 was also suppressed, leading to an improved alveolar phenotype in BPD male pups. From the results of our study, we can propose circNfix and Ntrk2 as novel key regulators in the pathogenesis and sexual dimorphism of BPD, while simultaneously proposing the use of circNfix GapmeR towards a potential therapeutic application.

Graphical Abstract

Graphical Abstract.

Introduction

New RNA capturing, sequencing, and annotative bioinformatics tools have recently revealed novel classes of regulatory noncoding RNAs (ncRNAs) that can enrich the complexity of gene regulation. While only <2% of the human genome is transcribed into mature protein-coding mRNAs, approximately 70% of the genome is classically transcribed into noncoding (nc) RNAs, which are canonically involved in cell division, cell migration, cell death, gene transcription and protein synthesis (1). Recent publications have begun to emphasize the physiological and pathological roles of ncRNAs in various diseases, cancer being the most predominant one (2). The ability of a single ncRNA to modulate the expression of several downstream targets and their associated pathways has provided a rationale to pursue them for therapeutic drug development (2). One such group of a novel class of ncRNAs are circular (circ) RNAs, formed by the back-splicing of exons and/or introns and lacking the 5′ or 3′ ends (3). This unique class of ncRNAs is expressed in different tissue types across several species and is involved in gene expression (by regulating transcription) and translation (by sponging RNA binding proteins and miRNAs) (4). CircRNAs have increasingly been associated with different respiratory diseases and are slowly drawing attention towards the pathogenesis and progression of lung cancer, pulmonary tuberculosis, silicosis, and pulmonary hypertension (5).

These unique ncRNAs are single-stranded, covalently closed, non-polyadenylated, and highly stable due to the lack of a 5′ or 3′ end, which prevents them from exonuclease digestion (3). With widespread access to RNA sequencing, it has been possible to identify these circRNAs, which are transcribed from >20% of the expressed genes in different cells and tissues across various species (4) and are highly conserved amongst these different species (6). Owing to their abundance, stability, and diverse expression profile, circRNAs are more likely to play a vital role in regulatory pathways controlling lineage determination, cell differentiation, and function of various cell types. They are therefore emerging as important molecular modulators for several neoplastic and non-neoplastic diseases (2,7). Given the fact that circRNAs consist of a circular configuration with typical 5′ to 3′-phosphodiester bond but without a free 5′ or 3′ terminus which makes them more stable in cells, they can be considered as potential drug targets. Although the functional role of circRNAs is evolving daily, several studies have reported that circRNAs harbor specific binding sites and primarily act as miRNA sponges or gene regulators (8). Due to the broad efficacy of circRNAs on miRNA activity, there is enormous interest in understanding the impact of miRNA sponging by circRNAs, on gene regulation. This property of circRNAs has been utilized as therapeutic candidates for diseases such as cancer (9). Very recently, circFOXO3 has been reported to improve cardiac function after doxorubicin-induced cardiac injury, suggesting an integral role for circRNAs in cardiac homeostasis and dysfunction (10). Garikipati et al. have reported that overexpression of circFndc3b can reduce ischemic injury in a mouse model of myocardial infarction (11).

Bronchopulmonary dysplasia (BPD) is a neonatal respiratory condition commonly seen in extremely low birth weight preterm babies, characterized by impaired lung alveolarization and dysregulated pulmonary vascularization secondary to the exposure of the developing lung to hyperoxia and/or invasive mechanical ventilation (12–14). BPD is the most common complication of prematurity and a pivotal contributor to the significant healthcare burden associated with more extended hospital stays, higher hospital costs, frequent re-hospitalizations of affected patients through the first year of life and increased resource utilization throughout childhood (15). While many drugs have been tried to prevent or attenuate BPD, no specific and effective treatment is available; therefore, this disease is still associated with high mortality and morbidity (16). Being an orphan disease, there is limited support for concerted research efforts because there is still no specific or compelling preventive or therapeutic pharmacological agent available to mitigate the short- and long-term impact of this devastating disease (17,18). Sex-specific differences exist in the development and the severity of BPD, with premature females having an advantage over premature male babies in terms of mortality and major morbidities. In human studies, males born <28 weeks gestational age have a higher incidence of BPD with higher neonatal and infant mortality rates than females after adjusting for confounding variables (19–21). When male and female neonatal mice were exposed to hyperoxia, female mice had less inflammation, an improved pulmonary phenotype, decreased cytokines, and a significant increase in angiogenic markers, compared to males (22). We have also reported that after hyperoxia exposure, females had less morphological injury with better preservation of the pulmonary phenotype (23). The biological reasons behind these sex-specific differences are not known.

Studies on both sexes are essential for the advancement of human health, and the influences of sex on the prevalence, presentation, and progression of many disease states (including BPD) are profound (24). In the present study, we have attempted to identify different circRNAs in male and female neonatal mice pups after hyperoxia exposure, followed by deep RNA sequencing to unravel the possible involvement of these ncRNAs in determining the favorable outcome in female infants. We aim to identify how these circRNAs control gene regulation directly or indirectly to regulate the desired phenotypic expression and assess if they contribute to any sex-specific differential outcomes in BPD. Hence, we believe that exploring and understanding the molecular signaling pathway of specific circRNAs and their interactive relationships with miRNAs–mRNAs–proteins will allow us to identify potential novel drug targets.

Materials and methods

Animals and BPD model

Wildtype C57BL/6 mice (males and females) were purchased from Jackson Laboratory (Bar Harbor, ME) and set for breeding to expand the colony. All animal work was approved by the Institutional Animal Care and Use Committee of Cooper University Hospital, Camden, NJ, USA (Approval number 23-006). Newborn pups were exposed to moderate hyperoxia (60% oxygen) in an airtight Plexiglas Chamber (BioSpherix, Parish, NY, USA). Exposure to supplemental oxygen began either immediately or within 8 hours after birth, designated as postnatal day (P0) and was continued through P4 (saccular stage of lung development), after which the animals were recovered in room air (RA) until P14 (alveolar stage of lung development) (1). Lactating dams were alternated between hyperoxia and RA to prevent oxygen toxicity in the adult mice during the 4 days of hyperoxia treatment. Litter size was limited to six pups per dam for uniform nutrition and growth of the litter size. Pups were sacrificed on P14. Throughout the experiment, free access to food and water was given, and oxygen concentration was continually monitored. This model mimics the current management of human preterm babies at risk of developing BPD admitted in the neonatal intensive care units (NICU) (23).

For circNfix silencing in the animals, 6 μl (3 μl in each nostril) of 20 μM control and circNfix GapmeR was administered intranasally on P2, P4 and P6 to the control RA and hyperoxia-treated BPD pups. To rule out any toxic effect of the GapmeR, scrambled GapmeR sequence was also delivered intranasally in a similar dose as that of circNfix to only the RA control group pups.

Human tissues

Two cohorts of de-identified postmortem lung tissues were procured for this study from two different resources after necessary institutional approvals. The paraffin slides from lung tissues (Cohort I) were obtained after informed consent, and the collection and processing of these human lung samples was approved by the National Supervisory Authority for Welfare and Health in Finland, as previously reported (25). The RNA and protein from autopsy lung samples (Cohort II) were requested from the National Heart Lung Blood Institute (NHLBI) sponsored LungMAP Human Tissue Biorepository Core (BRINDL) and approved by the University of Rochester Research Subjects Review Board (RSRB00047606) after informed consent for research. Selected clinical data for both the sources are listed in Supplementary Table S1.

RNA extraction

Neonatal mouse lungs were harvested from four groups of pups: RA males (RA-M), RA females (RA-F), BPD males (BPD-M) and BPD females (BPD-F) on P14 and total RNA was isolated using the miRNeasy Kit (Qiagen, Frederick, MD, USA) following the manufacturer's instructions. Human lung RNAs were also extracted using the similar kit.

cDNA library preparation

For RNA-sequencing, total RNA was isolated from all 4 groups of neonatal mouse lungs, with three pups from each group, as described above, and a library was created using TruSeq® Stranded Total RNA Sample Preparation kit (Illumina, San Diego, CA, USA) following the manufacturer's instructions. In brief, 100 ng of total RNA was used as a starting template where the first step involved the removal of ribosomal RNA (rRNA) using biotinylated, target-specific oligos combined with Ribo-Zero rRNA removal beads to deplete cytoplasmic and mitochondrial rRNA. Following purification, the RNA was fragmented, followed by first-strand cDNA synthesis using reverse transcriptase and random primers and by second-strand cDNA synthesis using DNA Polymerase I and RNase H. The products were purified and enriched by PCR to create the final cDNA library for cluster generation. Quality assessment was performed using Agilent 2200 TapeStation (Agilent Technologies, Palo Alto, CA, USA) prior to q-PCR and RNA-Seq analysis.

RNA sequencing and circRNA analysis

Adapter contamination from the raw Fastq files was removed using cutadapt and quality control was performed on the raw Fastq files using FastQC. Clean Fastq files were next aligned to Mouse Reference Genome (mm10) using the STAR aligner, ChimSegmentMin-10 parameter and CIRCexplorer2 pipeline for circRNA annotation (26). Due to low circRNA count, differential gene expression was performed using transcripts per million (TPM) to normalize circRNA expression levels (27). The identified circRNAs and their expression levels are presented in Supplementary Table S2.

RNase R treatment, polymerase chain reaction (PCR) and quantitative real-time PCR (q-PCR)

Total RNA (2 μg) was isolated from lungs using miRNeasy Mini Kit (Qiagen, Frederick, MD, USA) following the protocol as described and treated with or without 2.5 U of RNaseR (BioVision, Milpitas, CA, USA) for 30 min at 37°C and reverse transcribed using Maxima cDNA reverse transcription kit (Invitrogen, USA) to be subsequently used for q-PCR using SYBR Green master mix (Applied Biosystems, Waltham, MA, USA) (28). To quantify the expression of circRNA transcripts, divergent primers were designed to amplify the sequence spanning the back-splice junction (BSJ). The primer sequences for circRNAs and mRNAs are presented in Supplementary Tables S3A–B and S4A–B. The primers for miR-204-5p and miR-211-5p were commercially procured from Qiagen (Frederick, MD, USA). The PCR condition for circRNA amplification was 95°C (2 min), 95°C (5 s), 60°C (20 s) for 40 cycles. The amplified circRNA PCR products were resolved on 2% agarose gel and the bands were visualized by UV irradiation. The size of the circRNA was confirmed by running 10 μl of 100 bp SYBR Safe DNA ladder (Invitrogen, Carlsbad, CA, USA) and later Sanger sequenced to confirm the BSJ. All the q-PCR for mRNA, miRNA and circRNA were followed as per MIQE guidelines.

Identification of circRNA–miRNA–mRNA interactions and circRNA silencing

Sequence information of the differentially expressed (DE) circRNAs was used to predict the associated miRNAs (29,30) and publicly available databases like miRDB, targetScan and miRTarBase were used for miRNA prediction. Mouse lung-specific miRNAs were generated from published data and compared with custom-predicted miRNAs (31,32). miRTarBase (v8.0) was used to predict circRNA-miRNA-targeted mRNAs. Based on the target site and target score, four mRNAs, three miRNAs and two circRNAs were shortlisted and later confirmed to be involved in the regulatory pathway, for predicting the sexual dimorphism of BPD. To silence the circRNAs, control GapmeR and antisense GapmeR oligos were designed against the BSJ sequence. The sequence of GapmeRs are presented in Supplementary Table S5.

circRNA pull-down assay

Antisense oligomer pulldown of circNfix was performed following the protocol of Das et al. (33), with modifications. Briefly, approximately 400 mg lung tissues were lysed in polysome extraction buffer (20 mM Tris–HCl at pH 7.5, 100 mM KCl, 5 mM MgCl₂ and 0.5% NP-40) containing a cocktail of RNAse and protease inhibitor and incubated with 100pm of biotin-labelled controlled oligomers or an oligomer complementary to the junction sequence of circNfix in 1× TENT buffer (10 mM Tris–HCl at pH 8.0, 1 mM EDTA at pH 8.0, 250 mM NaCl, 0.5% [v/v] Triton X-100) containing protease and RNase inhibitors, overnight at 4°C with constant rotation. Streptavidin-coupled Dynabeads (50 μl, Life Technologies, Carlsbad), were washed with 1× TENT buffer and incubated with the lysate (120 min, room temperature, rotation). After isolating and washing the beads three times with ice-cold 1 × TENT buffer, RNA was isolated using TRIzol following which CircNfix and associated miR204-5p miRNA in the pulldown were detected by RT-qPCR analysis. The biotinylated antisense oligonucleotide (ASO) sequences are presented in Supplementary Table S6.

mRNA differential expression analysis and pathway enrichment

The raw RNA-seq files were aligned to the Mouse Reference Genome (mm10), from which the .bam file was further used to extract the raw counts using the Feature Counts package. For differentially expressed genes (DEGs), a generalized linear model (GLM) of the EdgeR package was used. The design matrix for analyzing the DEGs was generated using four different count matrices wherein RA-M was compared to RA-F, RA-M to BPD-M, RA-F to BPD-F and BPD-M to BPD-F. Log₂ fold change of ±2 and P-adjusted value at <0.05 was considered the cut-off criterion for all the DEGs.

Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis

Targets from the circRNA–miRNA–mRNA regulatory network was used to determine the function of selected DE circRNAs. The STRING database (v11.5) was used for a broad analysis of these gene targets. Gene enrichment analysis was performed using GO, (http://www.geneontology.org/) and KEGG (https://www.kegg.jp/) databases to target specific pathways (34–36). GO enrichment analysis (false discovery rate: FDR, P-value < 0.05) of different identified target genes are involved in several significantly enriched GO terms, including biological process, molecular function and cellular component. The enrichment score for the GO terms, tissues enriched, KEGG pathways and Reactome pathways were calculated as –log₁₀ of the FDR value and were represented in bubble plots.

Construction and visualization of competing endogenous RNA network

Based on the DE circRNAs identified after q‐PCR validation, overlapping prediction of miRNAs and DE mRNAs, Cytoscape software (v3.9.0) was used to construct and visualize the circRNA-miRNA-mRNA network. R studio and Graphpad Prism were used to plot the graphs for data visualization, while bubble plots were generated using ggplot2 in R studio. Alluvial plot was used to generate the circRNA–miRNA–mRNA schematic interaction. Statistical significance was calculated by Student's t-test and considered significant with a P-value of <0.05.

Bronchoalveolar lavage

Pups were sacrificed on P14, and the trachea was cannulated with a 21G caliber needle (BD Biosciences, NJ, USA) by instilling two volumes of 300 μl of cold 1× phosphate-buffered saline (PBS), gently aspirated and pooled. Samples were centrifuged at 1000g for 10 min at 4°C. The supernatant was collected, and the total protein was quantified using the Pierce BCA Protein Assay Kit (Thermofisher Scientific, Waltham, MA, USA). The cell pellet was dissolved in 200 μl 1× PBS and cytospun for total cell count.

Histology

Both the RA control and BPD mice were anesthetized (using a cocktail of xylazine–ketamine), and tracheas were cannulated as mentioned above, and lungs harvested after perfusion followed by fixing the tissues overnight in freshly prepared 4% paraformaldehyde. The fixed tissues were then washed in fresh 1× PBS, dehydrated using a series of graded ethanol, cleared in xylene, and embedded in paraffin to be sectioned at 5 μm and later stained with hematoxylin and eosin (H&E), or for immunostaining, as previously described for morphological and cellular interpretations, respectively (23).

TUNEL assay/immunostaining/immunohistochemistry

Paraffin sections (5 μm) were processed to study cell death using the TUNEL staining kit (Roche, Indianapolis, IN, USA) as per manufacturer's instructions. Fluorescent immunostaining (IF) was performed to study angiogenesis using von Wilebrand factor (vWF) antibody (1:100, DAKO, Carpinteria, CA, USA) and immunohistochemistry (IHC) was done to study cell proliferation using Ki67 antibody (1:50, Cell Signaling, Boston, USA) and Ntrk2 antibody (1:50, ThermoFisher Scientific, Carlsbad, CA) as per our standardized methodology (15,37).

Morphometry and quantification

To study chord length (distance across alveolar air space), 5 μm thick left-lobe lung paraffin-embedded sections were stained with H/E. Multiple randomly chosen areas (at least 5–7 areas) from each section were photographed using 100× total magnification. Chord length was measured by installing the chord length plugin in the ImageJ software (National Institutes of Health, USA version 1.5). Sections with large airways or blood vessels from the lung were excluded during imaging of the slides for lung morphometry.

For quantification of cell proliferation and cell death, the entire lung section was divided into three areas, and the total number of Ki67 +ve and TUNEL +ve cells were counted manually, which was normalized with the total number of nuclei in that specific area to give a percentage of positive cells. For vWF quantification, total number of closed vessels were counted per high-power field area in one lung section. A minimum of three areas was chosen, and three to five pups were used for staining and counting.

To measure the intensity of Ntrk2 IHC, ImageJ software was used with IHC profiler plug-in to convert the images to 8-bit gray scale and the threshold adjusted to eliminate any background noise so as to measure the final staining intensity for comparison between control and BPD groups (38). Briefly, the entire slide was scanned and three cases from each group were selected for comparison.

Imaging

All images were captured on an Olympus IX70 with DP73 camera attachment. At least five to seven low-power (magnification, X20) images were acquired for quantification. CellSens software (version 7) was used to acquire the images and further modified with Adobe Photoshop 13 (Adobe Inc., San Jose, CA) to present the best quality images.

Western blot

Western blot analyses for Vinculin (1:2000) and Ntrk2 (1:500) (Santa Cruz Biotechnology, Dallas, TX) were performed on precast 4–20% SDS-PAGE gels (GenScript, Piscataway, NJ, USA), as previously described (25), by loading 30 μg of lung protein, followed by immunoblotting with the above antibodies and visualization with Odyssey infrared imaging system (LI-COR Biosciences, Lincoln, NE, USA)

Statistical analyses

Values are expressed as mean ± SEM of at least three independent experiments. Groups were compared with the Student's two-tailed unpaired t-test or one-way ANOVA (followed by Tukey's multiple comparison post hoc test) or two-way ANOVA as appropriate (followed by Bonferroni's multiple comparison post hoc test), using GraphPad Prism 10.0 (GraphPad Software, Inc., San Diego, CA). A value of P < 0.05 was considered statistically significant.

Results

Characterization and screening of circRNAs expressed in neonatal mouse lungs

Lungs harvested from the previously described four groups (RA-M, RA-F, BPD-M and BPD-F) were used for the screening of circRNAs by deep RNA sequencing, followed by identification and validation using FastQC, CircExplorer2, STAR/Feature Counts, TPM and Edge R software, as outlined in Figure 1A. Of all the total circRNAs identified in neonatal P14 mouse lungs, approximately 30 550 circRNAs were annotated from the total RNA-seq library, of which 12 319 circRNAs were expressed in RA males (RA-M), 14 064 in RA females (RA-F), 14 220 in BPD males (BPD-M) and 13 198 in BPD females (BPD-F) (Figure 1B). Upon comparing all four groups, it was observed that >3600 circRNAs were shared across all groups (Figure 1B). Notably, the analysis revealed a higher number of annotated circRNAs in the BPD groups (irrespective of being male or female) when compared to the control RA group (Figure 1B). CircExplorer2 also helped to identify the splice length of the annotated circRNA. It was interesting to note that more than 15 000 circRNAs were less than 1000 bp in the lung tissue of neonatal pups (Figure 1C). Around 2500 circRNA transcribing genes generated only one circRNA while majority of the circRNA generating genes produced <10 circRNAs in the neonatal mouse lungs (Figure 1D). As circRNAs can potentially originate from both exonic and intronic sequences, the current investigation revealed that around 95% of circRNAs were generated from exonic sequences and the remaining circRNAs were derived from intronic sequences, in the present study (Figure 1E). Most circRNAs were generated from less than ten exons (Figure 1F).

Figure 1.

Characterization of circRNAs in neonatal mice lungs. (A) Workflow of circRNA analysis and interpretation using different publicly available databases. (B) Venn diagram showing the overlapped circRNAs detected in different groups. (C) Number of circRNAs generated depending on splicing event. Approximately >14 000 circRNAs were spliced around 500 nucleotides while very few were >7000 nucleotides. (D) Number of circRNAs derived from the genes in lung tissues. (E) Number of exonic circRNAs and intronic circRNAs detected in different samples. (F) The exon number distribution for circRNAs detected with CircExplorer2. RA, room air; BPD, bronchopulmonary dysplasia; M, male; F, female.

To identify functional circRNAs, we analyzed the DE circRNAs between the four groups of this study using the transcripts per million (TPM) method (Figures 2A, B). The expression of the screened circRNAs was normalized in a unit of TPM and validated by qPCR. Interestingly, all the experimental comparisons identified more than 200 DE circRNAs in different comparison groups. For example, RA-M versus RA-F had 250 DE circRNAs, BPD-M versus BPD-F had 233, RA-F versus BPD-F had 259, and RA-M versus BPD-M had 247 DE circRNAs. It was interesting to observe that 23 circRNAs were differentially expressed between all the 4 groups and validated by q-PCR after RNA seq screening (Supplementary Figure S1), and only three showed a significant difference between RA-M versus BPD-M or BPD-M and BPD-F groups, as confirmed by RNA-Seq and q-PCR (Figure 2C). There were still others which showed differences between the four groups by qPCR but did not give a significant score from RNA-Seq data (not shown).

Figure 2.

Differential expression analysis of circRNAs. (A) Table showing the differentially expressed (DE) circRNA selected based on transcripts per million (TPM). The blocks highlighted in blue shows a significant difference between the groups with a P value <0.05. (B) Heatmap of selected DE circRNAs in the four groups; blue bar shows upregulated circRNAs while red bar shows downregulated circRNAs. (C) Q-PCR showing expression of the 3 DE circRNAs (circNfix, circRad52 and circBoc) in the four groups. CircNfix and circRad52 are significantly upregulated while circBoc is downregulated in BPD-M as compared to BPD-F. (D) Representative agarose gel showing the clean expression of the three circRNAs with the correct band size (∼125 bp for all of them) shown in Figure 2C. Lanes with ‘+’ show the sample while lanes with ‘–’ show ‘no RT’ samples without any cross contamination. (E) Representative Sanger sequencing chromatograms results showing the validation of the DE circRNA. The red line on the sequence denotes the splice junction from where the divergent primer sequences were designed. (F) RNAseR treatment to the three DE expressed circRNAs shows that these RNAs were not denatured by RNAseR thus confirming the three DE expressed circRNAS as real, valid and stable. Two housekeeping genes (Vinculin and Gapdh) were included as linear mRNA controls to show fold change levels after RNAseR treatment. The mRNA was degraded for Vinculin and Gapdh, while it remained unaffected for circNfix, circRad52 and circBoc. N = 4; one-way ANOVA with Dunn's post hoc correction shows *P< 0.05.

circRNAs validation by q-PCR and RNAseR treatment in lung tissue

From the RNA-seq data, we extracted 23 circRNAs that were abundant and differentially expressed between the four groups (Figures 2A, B). Only those DE circRNAs with a log₂ fold change of ±2 and P-value <0.05 were validated by q-PCR using divergent primers for precise amplification (Supplementary Figure S1) and subsequently confirmed by RNAseR digestion for only those circRNAs that showed a dramatic altered expression between RA and BPD groups, irrespective of being males or females. The efficiency of digestion was compared with a few housekeeping mRNAs like Vinculin, Gapdh, β-actin and Hprt (Supplementary Figure S2). Based on the scoring criteria from the annotated circRNAs, there were 3 circRNAs (circNfix, circRad52, circBoc) which had a score of 2 and showed a significant change altered by q-PCR between either between RA and BPD in one sex (male) or the two BPD groups in the two sexes (BPD-M versus BPD-F) (Figure 2C). The PCR amplified products of these 3 circRNAs, were run on agarose gel to depict that the targets amplified were clean products without any contamination in the ‘no-RT negative’ control (Figure 2D). Subsequently, PCR products were Sanger sequenced to check for the BSJ (Figure 2E). Furthermore, total RNA was digested with RNase R to conclude them as RNA circles in lung tissue. RNase R exonuclease treatment confirmed the circular nature of the DE targets as they were resistant to RNAse R digestion, whereas the linear Vinculin and Gapdh mRNA were degraded after RNase R digestion, thus confirming the circular nature of the tested circRNAs (Figure 2F).

Identification of circRNA–miRNA–mRNA regulatory axis in BPD

To identify the miRNAs thought to be regulated by the DE circRNAs, the validated circRNA sequences were used to predict the target miRNAs. The seed sequence and prospective target mRNAs were obtained from miRDB and miRTarBase (v8.0) (29,30). miRDB identified 51 circRNA-associated miRNAs (Figure 3A). All the current listed circRNA-associated miRNAs from miRDB and the other miRNAs from miRTarBase were compared with previously published 801 miRNAs expressed in mouse lungs (31,32). Only 4 miRNAs (miR-211-5p, miR-204-5p, miR-299a-3p and miR-432) which were common in all the databases and showed a significant difference only between 2 groups (RA-M versus BPD-M) were analyzed to identify the target site and target score to confirm if our screened circRNAs were indeed regulating any of the selected target miRNAs. We were able to pick only two circRNAs (circNfix and circRad52) that were differentially expressed between RA-M and BPD-M or BPD-M and BPD-F by q-PCR validation (Figures 2C and 3B; Supplementary Figure S1). Based on a search of all the accessible databases, testing by qPCR and computational analyses, we did not find any of the other differentially expressed circRNAs (shown in Supplementary Figure S1) to have a suggested linear relationship with their potential target miRNAs and mRNAs. Since circNfix showed a significant difference over circRad52 and showed a linear relationship with its prospective miRNAs and mRNA, we selected circNfix as the preferred circRNA for subsequent study. Further, our computational analysis revealed that circNfix was targeted by two miRNAs (miR-211-5p and miR-204-5p) (Figure 3C). While there was no significant difference in the expression of miR-211-5p between RA-M and BPD-M, miR-204-5p was significantly upregulated in the BPD-M group as compared to RA-M group (Figure 3C). Similarly, we tried to validate miR299a-3p and miR-432 (targets of circRad52) by q-pCR and found that while, miR299a-3p was significantly upregulated in the BPD-M group as compared to BPD-F group (Supplementary Figure S3), miR432 was not expressed in P14 lungs (data not shown). In the above-mentioned miRNA comparison, circBoc did not exhibit an expression pattern with any specific miRNA. Hence, we chose to proceed with only circNfix as a confirmed, validated circRNA differentially expressed in the neonatal male lungs between the RA and BPD groups. The 2 miRNAs targeted only by circNfix in the web of all functional mRNAs were plotted using Cytoscape (Figure 3D).

Figure 3.

Expression analysis of mRNAs and miRNAs regulated by circRNAs. (A) Venn diagram depicting the overlapping miRNA selection from different databases. There were four miRNAs (arrow) that was common to the 3 databases used in this study. Q-PCR showing significant upregulation of (B) circNfix in BPD-M as compared to RA-M with subsequent upregulation of (C) miR204-5p in BPD-M over RA-M and no significant change in the expression of miR211-5p in both the RA-M and the BPD-M groups. miR204-5p and miR211-5p are considered potential targets of circNfix *P< 0.05; Vinculin and U6 are the normalizing reference genes. (D) The circRNA regulatory network constructed and visualized on Cystoscape software showing the central circNfix (pink circular node) with distribution of target miRNAs (yellow diamonds) and mRNAs (multiple green circular nodes). (E) GO enrichment analysis, and (F) KEGG enrichment analysis was performed to study the possible roles of all the enriched genes involved in different cellular processes involved in BPD and the level of gene enrichment on various pathways. Based on the most significant adjusted p-value, the top 4 processes were neurogenesis, regulation of development, cell differentiation and multipotency for GO pathway while the top pathways of KEGG analysis that was detected in the neonatal lung circRNA network were involved in many cancers. The ordinates (left) represent the top pathways and the enrichment score (right). The size of the dots represents the number of genes in the pathway which are proportional to the amounts of RNAs with significant differential expression. Dots with different colors represent distinct Q values (Q values are corrected P values). At least n = 3–6 samples were used for Q-PCR in (B) and (C). Student's two-tailed t-test was used for comparison; *P< 0.05

Gene enrichment and pathways analysis

We performed GO and KEGG functional enrichment analysis using the gene ontology database to explore the functional significance of the target genes involved in the molecular interaction and mechanisms underlying the circRNA–miRNA–mRNA regulatory network. GO enrichment analysis depicted the hits for various pathways in biological process, cellular components and molecular function (Figure 3E). As shown in this figure, several GO terms like neurogenesis and regulation of cell differentiation are enriched for the target genes of circRNAs. The top GO biological processes were the regulation of molecular function, regulation of cell differentiation, positive regulation of developmental process and negative regulation of transcription and metabolic processes. Interestingly, transcription regulator complex, organelle, nucleus and nucleoplasm were among the top hits in the GO cellular component. Similarly, transcription regulator activity, transcription factor binding and sequence-specific DNA binding were among the top GO terms involved in various molecular functions. KEGG pathway enrichment analysis was also applied to examine the level of gene enrichment in various pathways of circRNA involvement. Based on the most significant adjusted p-value, we identified several developmental regulating signaling pathways involved in BPD, while simultaneously being involved in pluripotency, cancer, Hepatitis B and Hippo signaling pathways that are also involved in cancer and other developmental processes (Figure 3F).

circRNAs are involved in the outcome of sexual bias in BPD

To analyze the functions of the validated circRNAs further, we performed differential gene expression analysis of the aligned mRNA data generated from the neonatal lung tissue and identified >3000 differentially expressed mRNAs between different sub-groups in the complete dataset. As shown in the Figure 4A, comparative volcano plots were created illustrating the highly differentially expressed genes in blue and low expressed DEGs in orange using a log₂ fold change along with P value <0.05 (Figure 4A). To analyze the role of circRNA in BPD, we compared the circRNA-miRNA targets with the DE mRNAs (Figure 4B). The analysis of the DE mRNA and circRNA targets is depicted through heatmap in Figure 4B. In the DE mRNA data, there were 14 circRNA–miRNA–mRNA targets, of which there were 3 mRNA targets of circNfix alone: Itpripl2, Nuak1and Ntrk2; the latter two showed significant changes between RA-M and BPD-M, with only Ntrk2 mRNA in BPD-M showing higher expression over RA-M, which was validated by q-PCR (Supplementary Figure S4; Figure 4C). Nuak1 was higher in the RA-M as compared to BPD-M, while there was no change in the expression of Itpripl2 between these 2 groups (Supplementary Figure S4). The expression of Neurotrophic tyrosine receptor kinase 2 (Ntrk2) was further confirmed by western blotting in RA-M and BPD-M groups (Figure 4D). Since Ntrk2 showed a robust validation both at the mRNA and the protein level, we proceeded with only circNfix, for which we could confirm a linear circRNA–miRNA–mRNA regulating pathway.

Figure 4.

Differentially expressed genes for mRNA analysis in BPD lung tissue. (A) Volcano plots showing the differentially expressed mRNA between RA-M and RA-F groups. Blue dots represent upregulated while orange dots represent downregulated genes. (B) Venn diagram showing the DEGs compared between RA-M and BPD-M. There were three mRNA targets (arrow) that aligned with the linear axis of RA-M and BPD-M (Itiprpl2, Nuak1, Ntrk2) and were differentially expressed and significantly upregulated (blue bar) in BPD-M when compared to RA-M in the depicted Heatmap. Of the 3 targets, only Ntrk2 showed similar upregulation between the two groups by (C) Q-PCR, and (D) Western blotting (representative image with its corresponding densitometric quantification on the side). (E) Biotinylated antisense oligonucleotide (ASO) circNfix pull down assay showing enrichment of circNfix (left panel) and miR204-5p (right panel) in the lungs by q-PCR. At least n = 3–4 samples were used for Q-PCR in (C and E). Student's two-tailed t-test shows *P< 0.05 or **P< 0.005; U6 and Vinculin are the reference genes for Q-PCR.

Validating the circRNA–miRNA–mRNA axis for circNfix in experimental BPD

Prior to using circNfix GapmeR for confirming the regulation of Ntrk2 by circNfix, we used mouse scrambled GapmeR sequence as a control to rule out any off-target effects of the GapmeR (Supplementary Figure S5). The BAL total cell count (Supplementary Figure S5A), total protein content in the BAL (Supplementary Figure S5B) and the chord length (Supplementary Figure S5C) (parameters used to assess any structural or functional changes in the lungs after any treatment) showed no difference between the wildtype RA group and the control GapmeR treated RA group) which made us confident in using circNfix GapmeR to assess the function of circNfix. Similarly, there was no obvious difference in the expression of circNfix target, Ntrk2 either at the RNA level by Q-PCR (Supplementary Figure S5D) or the protein level by immunoblotting (Supplementary Figure S5E) or immunostaining (Supplementary Figure S5F). From our screening analysis using publicly available databases as described earlier in the text, we found that circNfix was increased in BPD-M; its two miRNA targets- miR204-5p was also increased in BPD-M while there was no change in miR211-5p in BPD-M when compared with RA-M. To confirm our hypothesis that circNfix directly regulates its mRNA target Ntrk2, we performed q-PCR as well as western blotting and showed that Ntrk2 was significantly increased in BPD-M over RA-M both at the RNA level as well as the protein level (Figures 4C, D).

Since we had hypothesized that circNfix can regulate Ntrk2 without the intervention of miR204-5p and to prove that both the RNAs coexist via sponging, we did a biotinylated circNfix-biotinylated oligonucleotide pulldown assay in lung fractions to check for their association. By Q-PCR we found that there was approximately a 3-fold enrichment of both circNfix (Figure 4E, left panel) as well as miR204-5p (Figure 4E, right panel) in the biotinylated ASO sample as compared to biotinylated scrambled control oligo sample suggesting the possibility of an interaction and association with each other. There was minimal amplification for Vinculin and U6, the two reference genes for circNfix and miR204-5p, respectively with the control and circNfix ASOs.

To prove further that circNfix regulates Ntrk2, we silenced circNfix using a BSJ-specific GapmeR against circNfix and analyzed the expression of miR204-5p and Ntrk2 (respective miRNA and mRNA targets of circNfix) by q-PCR, and protein expression of Ntrk2 by western Blotting. Following circNfix silencing, it was observed that circNfix expression was decreased, while miR204-5p expression increased further in the treated BPD group as compared to wildtype BPD-M (Figures 5A, B). Although there was no significant change in Ntrk2 between the treated GapmeR BPD group and wildtype BPD group at the RNA level (Figure 5C), there was a significant decrease of Ntrk2 protein level in the BPD-M group after treatment with circNfix GapmeR (Figures 5D, E). In summary, the administration of circNfix GapmeR resulted in a reduction of the expression of basal circNfix with a simultaneous increase in the expression of its miRNA target–miR204-5p, and a decrease in the expression of its mRNA/protein target, Ntrk2, specifically in BPD males. This outcome confirms the direct linear regulation of circNfix; however, it does not completely exclude any intermediary mediation through miR204-5p.

Figure 5.

CircNfix GapmeR regulates Ntrk2 and miR204-5p. Q-PCR showing (A) expression of circNfix after circNfix GapmeR was administered intranasally to newborn pups. There is a significant decrease in the expression of circNfix in both the RA-M and BPD-M groups after treatment with circNfix GapmeR, confirming the binding potential of the GapmeR. (B) Expression of miR204-5p in both RA-M and BPD-M groups after GapmeR treatment. miR204-5p (a target of circNfix) is increased considerably in the BPD-M group treated with GapmeR as compared to the untreated BPD-M group. (C) Expression of Ntrk2 (mRNA target of circNfix) after circNfix GapmeR treatment. There is a significant increase of Ntrk2 in the RA-M group treated with GapmeR as compared to the untreated RA-M group, while there is no change in the expression in BPD-M group with or without treatment. 18S and U6 are the reference genes for normalization. (D) Representative WB showing expression of Ntrk2 with and without treatment with GapmeR, along with densitometric quantification. There is a significant decline of Ntrk2 in the RA-M and BPD-M treated groups as compared with the untreated wild type groups. Dashed lines represent merged images from uncut raw gel data and aligned for comparison. Vinculin is the loading control. (E) Representative IHC staining showing decreased expression of Ntrk2 after GapmeR treatment (panel IV) as compared to wild type BPD (Panel III). It is interesting to note high expression of Ntrk2 in the bronchioles (dashed circles) as compared to alveoli in the RA group (Panel I), but after Hyx treatment, Ntrk2 is also seen to be expressed in the alveoli (Panel III), which is markedly diminished after circNfix GapmeR treatment. The bronchioles in panel IV show complete absence of Ntrk2. Scale bar 200 μm. The bottom graph panel shows quantification of staining intensity in the three groups. N = 3–11 samples for (A–C); one-way ANOVA with Dunn's post hoc correction shows *P< 0.05; **P< 0.005; ***P< 0.001.

Suppression of circNfix improves the alveolar phenotype

The intranasal administration of circNfix GapmeR worked successfully in our hands, evident from the reduced expression of circNfix and its miRNA (miR204-5p) and increased mRNA (Ntrk2) and protein (Ntrk2) targets by q-PCR and Western Blotting in the RA model (Figure 5). In the experimental BPD model, circNfix GapmeR also decreased circNfix, but increased miR204-5p expression and decreased Ntrk2 protein levels (Figure 5). Since decreased inflammation is accompanied by improved alveolar morphology resulting in normal histological phenotype, we measured the chord length (a morphometric assessment for alveolar space, which is usually enlarged in response to hyperoxia) to conclude that there was no difference in chord length between the RA group and the RA group treated with GapmeR (Figure 6A).

Figure 6.

circNfix GapmeR improves alveolar phenotype in BPD lungs. (A) H/E sections of the lung showing improved chord length in the BPD treated group as compared to the untreated group. Right panel shows measurement of chord length in the four different groups. (B) TUNEL staining showing decreased cell death (white arrowheads). (C) IHC showing increased cell proliferation, as evident from Ki67 staining (black arrows). (D) Increased angiogenesis/vasculogenesis in the treated BPD group as compared to untreated BPD, as evident from merged vWF/DAPI IF staining. vWF is a marker for blood vessels while DAPI stains the nucleus. Bottom panels (of B–D) show quantification of the respective stainings. Scale bar 200 μm; N = 3–10 for (A); n = 3 for (B–D); one-way ANOVA with Brown-Forsythe post hoc correction shows *P< 0.05; ** P< 0.005; ***P< 0.001, ****P< 0.0001.

Classically, increased cell death, decreased cell proliferation, and decreased and dysregulated angiogenesis/vasculogenesis are considered canonical hallmarks of BPD. We, therefore, tested these three physiological processes in the lungs after GapmeR treatment and showed by TUNEL staining that there was decreased (but not statistically significant) cell death in the BPD + GapmeR treated group as compared to wild type untreated BPD group; interestingly, the RA + GapmeR treated group showed significantly more positive TUNEL cells as compared to wild type RA or wildtype BPD or BPD + GapmeR treated groups (Figure 6B). Cell proliferation studied by Ki67 immunohistochemistry revealed increased proliferation with RA + GapmeR. Importantly, BPD + GapmeR treated group had a significantly higher number of proliferating cells as compared to wild type BPD alone group (Figure 6C). While there were fewer blood vessels in the RA + GapmeR group, there was a significantly increased number of blood vessels in the BPD + GapmeR treated group as compared to wild type BPD alone group (Figure 6D), by vWF (marker for large blood vessels) staining. All three staining parameters were quantified to show the recovery in cell death, cell proliferation, and angiogenesis after GapmeR treatment.

Translational significance of the novel circRNA–miRNA–mRNA axis in clinical BPD

We wanted to test if the above axis holds true for human babies or not. For that purpose, the expression of circNFIX, miR204-5p and NTRK2 was verified by q-PCR, Western blotting and IHC in only male infants. There was a significant increase of circNFIX (Figure 7A) as well as its miRNA/mRNA targets, miR204-5p (Figure 7B) and NTRK2 (Figure 7C–E) in the human BPD cases as compared to control babies, consistent with our experimental moderate BPD model. This suggests that circNFIX might have a regulatory role in hyperoxia and/or ventilation induced BPD in human infants via its interaction with miR204-5p and NTRK2.

Figure 7.

CircNfix may regulate NTRK2 in human BPD patients. Q-PCR showing a trend towards increased expression of (A) circNFIX, (B) significant increase of miR204-5p and (C) NTRK2 in human BPD male babies as compared to control babies. VINCULIN and U6 are the reference genes. (D) Representative WB showing a similar increase in expression of NTRK2 in BPD babies as compared to full term controls. Dashed lines represent merged images from uncut raw gel data aligned for comparison. Graph on the right-side panel shows densitometric quantification of NTRK2 between the two groups. (E) Representative IHC showing increased expression of NTRK2 in the lungs of babies with clinical diagnosis of RDS and BPD as compared to controls. Babies with RDS also had significantly higher expression of NTRK2 similar to the confirmed cases of BPD. Note the enlarged simplified alveolar epithelium with large alveoli, in the BPD group, a characteristic feature of the disease. Scale bar 200 μm N = 3–5; C: control; RDS: respiratory distress syndrome. The bottom graph panel shows quantification of staining intensity in the three groups. N = 3–5; Student's two-tailed t-test was used for comparison *P< 0.05; ***P< 0.0005.

The results from our experimental BPD model potentially confirms that circNfix might sponge miR204-5p by minimally inhibiting its activity and/or function; suppressing circNfix via a specific GapmeR improved the lung injury caused by hyperoxia exposure in the neonatal mouse lungs, while a similar linear axis could be playing a role in clinical BPD thus highlighting the possible translational impact of our data, as depicted in the cartoon (Figure 8).

Figure 8.

Cartoon summarizing the regulation of circNfix. Upon hyperoxia exposure, circNfix is upregulated with subsequent upregulation of miR204-5p and increased expression of Ntrk2. CircNfix may sponge miR204-5p which may result in the reduced availability of this miRNA leading to an increased expression of Ntrk2. Due to a linear increase of all the interacting RNAs, the lungs present with a BPD phenotype.

Discussion

CircRNAs are emerging as a new field of targeted drug therapy and have been the focus of biological research in the last few years. With the advent of new sequencing technologies, it has been possible to identify most of the existing circRNAs within a specific cell/tissue type; yet only the biological function of a handful of circRNAs has been studied in detail. As the field of neonatal care is improving, efforts are being made to improve treatment options. In that context, circRNAs are emerging as a new class of therapeutic targets. So far, only a handful of published work have reported the involvement of some novel circRNAs in BPD lungs (39,40). However, because different laboratories have different hyperoxia exposure protocols ranging from 4 days to 14 days (41), the expression of key genes during lung development is also altered. This results in varied circRNA expression in different hyperoxia models; circABPD1 was elevated in BPD lungs (42), circFANCL (43) and circABCC-4 (44) were elevated in the peripheral blood of human preterm infants diagnosed with BPD; several other non-annotated circRNAs with unknown functions (43) were also reported from BPD patients in a generalized context without mentioning their role in males and females. Since BPD is a sex-biased condition, in the present work, we have identified circNfix as a novel circRNA regulating its target mRNA, Ntrk2 via a modulating miRNA, miR204-5p highlighting the sexual dimorphism, specifically in experimental BPD murine males.

We used a developmentally appropriate (saccular stage of lung development) mouse model administered with moderate hyperoxia (60% O₂) to mimic human BPD for maximal translational relevance to the current standards of the NICU. BPD is a sexually biased neonatal disease where female babies do better than their male counterparts. The better prognosis of females compared to males is not yet fully understood. Therefore, we wanted to test if circRNAs contribute towards explaining this dimorphism in BPD so that this new class of ncRNAs can be exploited to be used as targeted and personalized drug candidates for this devastating neonatal disease for which there are no specific or effective drugs available, to date.

We started with circRNA screening from lung tissue obtained from our mouse experimental BPD model using ribosomal RNA depleted RNA sequencing and observed several circRNAs that were differentially expressed in the BPD lungs. There were 3 noteworthy circRNAs (circNfix, circRad52 and circBoc) that looked the most promising in terms of DE either in the RA-M versus BPD-M or BPD-F versus BPD-M groups, because our goal was to identify those circRNAs that were different in these two groups to rationalize the fact that BPD females have a better prognosis over BPD males. Concurrently, since we also wanted a direct linear relationship between these three circRNAs and their respective downstream miRNAs and mRNAs, we selected only circNfix over circRad52 and circBoc because the latter circRNAs did not fit our criteria of the regulatory loop, although circRad52 and circBoc showed a significant difference in the 4 studied groups. Hence, we proceeded with circNfix only, for subsequent analysis. Interestingly, circNfix has been involved in a number of adult diseases such as ovarian cancer (45), cardiac hypertrophy (46), pituitary adenoma (47), hepatocellular carcinoma (48) and lung cancer (49). For the first time, we report the expression and involvement of circNfix in a specific neonatal disease (BPD).

Currently, we are flooded with information highlighting circRNAs acting as sponges for miRNAs because the sponging of circRNAs to different miRNAs suppresses the downstream translation by competing with endogenous RNAs. Several investigators have proved this regulation with luciferase reporter activity. Our approach is unique because we have tried to address the functionality of circRNAs, miRNAs and mRNAs in a linear fashion rather than focusing only on their sponging effect on miRNAs, as reported in multiple publications. A recent review on circRNA–miRNA regulatory axis discussed that circRNAs can regulate miRNA expression or activity through various mechanisms including miRNA sponging by circRNAs (50). They suggest that circRNAs can regulate miRNA activity, directly through ‘sponging’ or indirectly through ‘miRNA biogenesis’. From the existing miRNA databases, we found 2 miRNAs (miR204-5p and miR211-5p) that regulated several mRNAs, including Ntrk2, and were under the direct influence of circNfix. Ntrk2 showed significantly higher expression in BPD-M as compared to RA-M, both at the level of transcription as well as translation. Ntrk2 is one of the Hypoxia-inducible factor (HIF)-1 target genes that is increased in newborn mouse lungs during hypoxia following subsequent reoxygenation with 100% hyperoxia, both at the mRNA and protein levels (51). Contrary to this, Dong et al observed a significant decrease of Ntrk2 in neonatal mouse lungs exposed to hyperoxia for 14 consecutive days and emphasized this decrease due to a notable increase of miR29 which targets Ntrk2 (52). Similar to this observation, Shrestha et al. also reported a decrease in Ntrk2 in neonatal mouse lungs exposed to 100% hyperoxia for 7 consecutive days (53). However, in the present study, we observed an opposite effect to what has been reported by others, as discussed above. Ntrk2 was upregulated in the pups exposed to moderate BPD (60% O₂) as evident from RNA-seq data, Q-PCR data, Western blotting and immunostaining at P14. We also saw a similar upregulation of Ntrk2 in P7 pups exposed to either 60% or 100% O₂ for 7 consecutive days or 60% O₂ in pups exposed to 14 consecutive days (data not shown). Whatever the condition, hyperoxia induced stress upregulated Ntrk2 in the lungs of neonatal wildtype C57Bl6/J mice, while Dong et al. (52) reported a decrease in Ntrk2 in ICR mice (outbred strain) by Q-PCR, Western blotting and immunostaining. This is an important finding because we have earlier reported that different strains of mice may exhibit distinct levels of gene expression, although all mouse strains contain the same set of genes (23). Intriguingly enough, Shreshta et al have reported decreased Ntrk2 expression in wild type C57Bl6/J pups only at the RNA level. It is to be noted that Ntrk2 is involved in cell proliferation (54,55) and different strains of mice with varying concentrations and duration of O₂ exposure at different stages of lung development may account for variable results in response to injury as is reflected in the expression of Ntrk2 from different research laboratories, as mentioned above. Furthermore, a similar expression pattern of NTRK2 mRNA and NTRK2 protein, in the males of both experimental and clinical BPD suggests a strong involvement of this gene in hyperoxia-induced stress. Surprisingly, NTRK2 has been identified as a fusion gene in several pediatric tumors (56–58) and is often mutated in various cancers (cancergenome.org). This places NTRK2 as a top candidate for targeted therapy for which there are currently three clinical trials in progress (www.cancer.gov/research/participate/clinical-trials).

To unravel the various genes involved in the physiological changes after hyperoxia exposure, specifically in the development phase when the insult has occurred, up to the time when the lungs have been allowed to recover, GO and KEGG analyses were run to examine the roles of the DE circRNAs, miRNAs and mRNAs, to find that the most enriched RNA elements were associated with overall developmental pathways like cell differentiation and pluripotency. This is logical because at P14, different developmental changes are still in process with several overlapping genes involved in organogenesis, in toto. To our surprise, many cancer regulating pathways were enriched by KEGG analysis, which we believe is the first report in neonatal lungs exposed to moderate hyperoxia. It is known that survivors of BPD are more likely to contribute to deranged alveolar injury in adulthood (59) but it is not known whether survivors of BPD have an increased likelihood of falling prey to cancer, later in life. Hence, it would be interesting to dissect out these pathways from the KEGG analysis for a more organized deep dive study in the future because most of these genes are predominantly overexpressed in lung cancer by regulating cell migration and proliferation and some of them if downregulated or mutated are indicators of bad prognosis. This leaves the question that babies who are predisposed to BPD during infancy may be susceptible to lung cancer at some stage in adulthood if other comorbid conditions exist due to their pulmonary complications. At the same time, most of these pathways are attributed to normal lung development too, and therefore remain quiescent in adult life unless they become aberrant at some stage of growth and development where the odds of contributing to an increased susceptibility for gastric or colorectal cancer or Hippo signaling later in life may become apparent in adulthood, due to an early insult in the developing stage. The same hypothesis may hold true for NTRK2 fusion in pediatric cancers which might evolve due to hyperoxia-induced stress in the fragile BPD lungs, which needs to be explored further.

To prove that circRNAs do not necessarily have to regulate the target mRNAs only via miRNA sponging, we administered GapmeR specific for circNfix to silence the same. This is the first time that we are reporting the in vivo administration of GapmeR, intranasally to neonatal pups with successful delivery. To check for efficient suppression of circNfix using circNfix GapmeR, a q-PCR validating the result showed a convincing knockdown. In the RA groups treated with GapmeR, while there was no difference in chord length, there was an increase in cell death and cell proliferation with decreased vascularization, when compared to RA group alone (Figure 6B–D). This suggests that while GapmeR administration in the non-injured state does impact cell signaling pathways, the healing/normal developmental process after injury, is sufficiently robust to restore the normal pulmonary phenotype after hyperoxia treatment (Figure 6A). Additionally, there was improvement in chord length, decrease in cell death, increase in cell proliferation, and initiation of neovascularization and angiogenesis after treatment, all of which were disrupted in BPD. This clearly shows that circNfix is involved in lung morphogenesis and development and repairs the damaged lung towards a normal pulmonary phenotype. As a result of hyperoxia initiated BPD, circNfix is upregulated, causing an increase of downstream target Ntrk2 with a simultaneous intermediary increase of miR204-5p. Upon suppression by circNfix GapmeR, Ntrk2 is decreased while miR204-5p is increased, thus recovering the BPD phenotype. Since Ntrk2 is involved in cell proliferation, its upregulation in BPD caused increased cell death, while its downregulation after circNfix GapmeR treatment-initiated cell proliferation. Whether circNfix was able to sponge miR204-5p was beyond the scope of the present study. Nevertheless, from our findings, we can conclude that the regulation of BPD by circNfix is not through miRNA sponging, but through alternate mechanism of miRNA regulation that needs further investigation (50).

Based on our results, we can conclude that hyperoxia causes an increase in circNfix and miR204-5p in the developing lungs, activating Ntrk2 to be overexpressed. This upregulation results in BPD characterized by simplified alveolar epithelium, decreased cell proliferation and angiogenesis, and increased alveolar cell death. During the process, circNfix may sponge overexpressed miR204-5p. Whether the sponging effect decreases the activity of miR204-5p is unclear, and more studies have to be undertaken to prove this. However, on suppressing circNfix via a targeted GapmeR, the activity and/or expression of miR204-5p is also decreased with a simultaneous decrease of Ntrk2. This linear suppression of all the interacting RNA elements reverses BPD and helps restore the diseased phenotype as has been summarized in Figure 8. Since Ntrk2 is upregulated in both mice and humans upon hyperoxia exposure, and is downregulated after administration of LNA circNfix GapmeR, this GapmeR has the potential to be considered as a putative therapeutic candidate for Ntrk2 suppression because two NTRK inhibitors (Larotrectinib and Entrectinib) have been FDA approved with promising results (60,61).

Our study has some limitations. A lot of the genes expressed in RNA seq data did not match the q-PCR results probably because in RNA-Seq analysis, multiple transcript variants are encoded while in Q-PCR, only a single variant is amplified from the region that spans the primers (62). Although our study focused on screening DE circRNAs in the four groups, the stringency of different databases prevented a cumulative comparison between multiple groups. Also, most of the accessible databases have an ensemble of genes derived from adult diseases with limited information on pediatric conditions, with even more limited information on neonatal diseases. Further, there is moderate information available on the role of circRNAs in BPD with varied miRNA targets, and none of these miRNA targets have been previously reported to be involved in this disease, even though there is abundant literature on miRNA microarray for BPD in humans as well as murine models. This raises the question - whether the data is unavailable in all the open databases or whether the expression level is different for different animal hyperoxia models/experimental BPD models. All the data reported so far, have utilized the ‘sponging’ theory of circRNAs with miRNAs, and there is not a single study that explains the direct linear relationship between the different interacting RNA molecules, as has been presented by us in the current work. Hence, it was essential to conduct this work to highlight the modulatory role of circRNAs in protein translation, independent of miRNA interference. We, therefore, chose to focus on circRNAs with statistical significance between RA-M vs BPD-M that also had a suggested linear relationship with potential target miRNAs and mRNAs. Using miR204-5p antagomir for suppressing the target miRNA and studying the consequential effect of Ntrk2 due to miR204-5p inhibition, was beyond the scope in the present context. Moreover, we believe our work has laid the foundation for other researchers to revisit the action of circRNAs in regulating miRNAs and mRNAs. Our ongoing work will confirm the mechanism of action of these novel ncRNAs in more detail and prove their potency as future therapeutic candidates.

Data availability

The RNA-seq data is available at ENA with accession number PRJEB66464. Raw data of the gel images are presented in Supplementary Figure S6A–C. The annotated comparisons between different groups are presented in: Supplemental Tables S7 and S8.

Supplementary data

Supplementary Data are available at NARMME Online.

Funding

No external funding.

Conflict of interest statement. None declared.