Discovery of dysregulated circular RNAs in whole blood transcriptomes from cystic fibrosis patients – implication of a role for cellular senescence in cystic fibrosis

Abstract

Background:

A largely unexplored area of research is the identification and characterization of circular RNA (circRNA) in cystic fibrosis (CF). This study is the first to identify and characterize alterations in circRNA expression in cells lacking CFTR function. The circRNA expression profiles in whole blood transcriptomes from CF patients homozygous for the pathogenetic variant F508delCFTR are compared to healthy controls.

Methods:

We developed a circRNA pipeline called circRNAFlow utilizing Nextflow. Whole blood transcriptomes from CF patients homozygous for the F508delCFTR-variant and healthy controls were utilized as input to circRNAFlow to discover dysregulated circRNA expression in CF samples compared to wild-type controls. Pathway enrichment analyzes were performed to investigate potential functions of dysregulated circRNAs in whole blood transcriptomes from CF samples compared to wild-type controls.

Results:

A total of 118 dysregulated circRNAs were discovered in whole blood transcriptomes from CF patients homozygous for the F508delCFTR variant compared to healthy controls. 33 circRNAs were up regulated whilst 85 circRNAs were down regulated in CF samples compared to healthy controls. The overrepresented pathways of the host genes harboring dysregulated circRNA in CF samples compared to controls include positive regulation of responses to endoplasmic reticulum stress, intracellular transport, protein serine/threonine kinase activity, phospholipid-translocating ATPase complex, ferroptosis and cellular senescence. These enriched pathways corroborate the role of dysregulated cellular senescence in CF.

Conclusion:

This study highlights the underexplored roles of circRNAs in CF with a perspective to provide a more complete molecular characterization of CF.

1. Introduction

It has become clear that circular RNAs (circRNAs) and competing endogenous RNA (ceRNA) interaction networks play important roles in biology, health and disease [1]. CircRNAs have become a research hotspot because of their close association with the development of disease [1–5], such as cardiovascular disease, metabolic syndrome disorders, inflammation, neurodegeneration and various cancers (reviewed by Misir et al. 2022 [1]).

CircRNAs are endogenous, non-coding, single-stranded, covalently-closed RNA molecules found ubiquitously expressed across the tree-of-life ranging from viruses to mammals [6]. CircRNAs are generated from the precursor mRNA back-splicing of exons when a splice donor (5′ splice site) is covalently joined to an upstream splice acceptor (3′ splice site) [7–9]. This leads to a non-canonical splicing event called back-splicing, which results in a covalently closed loop characterized by a back-splice junction (BSJ). Compared to their linear counterparts, circRNAs are unusually stable due to the lack of an open end that prevents them from being broken down through conventional RNA degradation pathways [4].

The underlying biological mechanisms of circRNA biogenesis and degradation currently remain poorly understood [8]. However, recent studies suggest that circRNAs play a crucial role in regulating gene expression by acting as a miRNA sponge, an RNA binding protein sponge, or as a translational regulator [1]. Some circRNAs are reportedly expressed in a tissue- and development stage-specific manner [1]. CircRNAs have unique functions that often rely on association with miRNAs and RNA-binding proteins. Through these interactions, circRNAs have been implicated in major cellular processes and hence, in the pathophysiology of a range of diseases. CircRNAs and their ceRNA interaction networks represent potential sources of novel biomarkers for disease diagnosis and for understanding disease [1, 2].

In the context of circRNAs, complex respiratory disorders are grossly understudied and under-explored [5]. Regarding cystic fibrosis (CF) as a respiratory disease, to date, there are no known peer-reviewed publications that examine the presence and role of circRNAs in CF disease. A recent study by Chen and colleagues in October 2021 [5] showed that out of 1013 papers that examine the role of circRNA in human diseases, only 12 papers (i.e. 1.2%) reference circRNA in relation to respiratory diseases. To date, our publication is the first to identify and discover circRNAs that are dysregulated in human CF patient cells. A portion, of these dysregulated circRNAs, is likely to be associated with CF phenotypes and could have a role as circulating biomarkers. Researching dysregulated circRNAs in CF could be a fruitful starting point to develop useful biomarkers to monitor disease progression or efficacy of treatments for CF.

To explore whether certain circRNAs are dysregulated in the context of CF, we searched for circRNAs in the sequenced whole blood transcriptomes of people with CF who are homozygous for the F508delCFTR pathogenetic variant [10] and examined their expression relative to wild-type controls. F508delCFTR accounts for two-thirds of CF alleles worldwide and 90% in the United States. F508delCFTR represents a Class II pathogenic variant that leads to CFTR protein misfolding, premature degradation by the endoplasmic reticulum (ER) quality control system, and impaired protein biogenesis, severely reducing the number of CFTR molecules that reach the cell surface [11].

In this study, a Nextflow-based bioinformatics pipeline (named circRNAFlow) was developed to conduct medium-scale circRNA analyzes. Next, previously sequenced whole blood transcriptomes of 20 CF cases (all homozygous for F508delCFTR) and 20 healthy non-CF controls [10] were downloaded from GEO. Subsequently, circRNAFlow was utilized to identify and characterize dysregulated circRNAs in CF patients compared to the healthy controls. The new results generated from analyzing whole blood transcriptomes of CF samples were assessed to answer the following questions: Are circRNAs dysregulated with CF? What are the identities and characteristics of dysregulated circRNAs in CF? How can such new information be utilized to better understand CF and possibly develop new theragnostic options, such as biomarkers, for CF patients?

2. Methods

2.1. Data

In a previous study, whole blood transcriptome sequencing was performed with 20 CF and 20 non-CF samples [10]. The fastq files for the whole blood transcriptomes were downloaded from NCBI Gene Expression Omnibus (GEO) (accession: GSE124548) [10].

2.2. Development of circRNAFlow nextflow pipeline

To perform the analysis, a circRNA analysis pipeline called circRNAFlow was developed utilizing Nextflow [12]. CircRNAFlow provides support for docker and singularity containers [13, 14]. CircRNAFlow development is, in part, based on a DCC-centric protocol [15, 16].

2.2.1. An overview of circRNAFlow

Briefly, we describe circRNAFlow (Fig. 1). The data pre-processing steps are grouped into two parts (a) quality and adapter-trimming and (b) molecular-reference filtering (rRNA removal). Quality assessment was done by FastQC (v0.11.9) [17]. Adapter-trimming with user supplied sequences was done by FlexBar (v.3.5.0) [18]. Once this step was completed, rRNA was removed by aligning to an rRNA database using bowtie2 (v2–2.3.0) [19]. The next steps focused on the circRNA analyzes. STAR (v2.7.9a) [20] was first invoked on the rRNA filtered, adapter-removed data to map reads. STAR was run thrice; first-of-pair, second-of-pair, and finally in a paired end mode. The ENSEMBL human GRCh38 reference genome and gtf (GRCh38.107.gtf) were used for purpose of mapping. The DCC (v0.5.0) [21] program ran next to perform back-splice junction detection with paired end read orientation, annotation, and other means to help identify circRNA. CircRNAFlow final steps include a script invoking CircTest (GitHub commit 0f5a86c) [21] for comparing circRNA expression relative to corresponding linear host genes between groups. These results are augmented with plotting routines (Fig. 2) and clusterProfiler [22] functional enrichment analysis calls (Fig. 3).

Fig. 1.

The flowchart shows the conceptual steps encapsulated in circRNAFlow. The ovals are the primary processing steps e.g., FastQC [17], FlexBar [18].The yellow ovals represent the data pre-processing steps which include the pre-alignment quality control assessment. The blue ovals represent alignment and circRNA identification steps. The green ovals represent the generation of plots (Fig. 2) and the clusterProfiler plots (Figs. 3A–D). The gray ovals represent circTest [21] which compute differential expression between case and control using the default parameters or a plethora of parameters. The arrows represent the flow of the data. The small filled-in closed dots represent input files introduced to the pipeline from external sources e.g., the addition of circAtlas identifier [23] to annotate circRNA. The small open dots represent intermediate processing steps, carried out by circRNAFlow e.g., filtering, splitting, or merging of data. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 2.

Dysregulated circRNAs in CF samples compared with healthy controls. Plots of three up regulated circRNAs in CF (blue bars) compared to non-CF (red bars) are: (A) hsa-TTTY10_0001; (B) hsa-ATP8B4_0010; (C) hsa-OLA1_0013. Plots of three down regulated circRNAs in CF compared to non-CF healthy controls are: (D) ENSG00000285557 a novel circRNA; (E) hsa-MARK3_0091; (F) hsa-tfrc_0049. The Y-axis is the ratio of circRNA reads against the total reads (linear mRNAs and circRNAs) [21]. The blue and red dashed lines represent the mean of the ratios for the CF patients and the healthy controls respectively. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 3.

ClusterProfiler [22] over-represented: (A) GO biological process terms; (B) GO molecular function terms; (C) GO cellular component terms; and (D) KEGG pathways in dysregulated circRNA host genes in whole blood transcriptomes of CF patients. The x-axis of the clusterProfiler plot shows the GeneRatio, defined as the genes of interest in the GO term divided by total genes of interest.

2.2.2. Additional details of circRNAFlow

Data analysis was supplemented by differential expression analysis with CircTest [21], coupled with annotation of known circRNAs via circAtlas [23] and a plotting step (Fig. 2). The custom plotting scripts produced plots incorporating circAtlas [23] IDs, which accompanied CircTest output and plots as well as perform some minor data wrangling.

During the differential expression analysis, the Circ.filter function was used with various parameter combinations to help researchers understand their effects. Parameter settings included filter.count, filter.sample, and percentage. These parameters respectively set thresholds for read counts, samples meeting that read count, and circRNA reads as percentages of total reads. The CircTest R package utilizes the BSJ counts as input to perform differential expression tests generating the p-values and adjusted p-values for each circRNA. CircTest [21] employs the beta binomial distribution to model the ratios of the circular BSJ counts to the total read counts. The null hypothesis states there are no differences between the groups. The alternative hypothesis states there are differences between them. The null hypothesis is compared with the alternative hypothesis using the likelihood ratio test with ANOVA to identify circRNAs that differ in their relative expression (circular versus host gene) between the groups (Table 1). The p-value is obtained from ANOVA and the adjusted p-value is computed by the Benjamini-Hochberg method. The circRNAs are independently regulated compared to the host gene (Table S1).

Table 1.

Most highly differentially expressed circRNAs in whole blood of individuals with CF disease (absolute log₂ FC > 1; adjusted p- value < 0.1). The log₂ FC is the natural logarithm of the ratio of the average BSJ count for the CF patients and the average BSJ count for the healthy controls; FC is fold change. None of the dysregulated circRNAs passed the significance level of adjusted p-value < 0.05. However, 19 dysregulated circRNAs passed the following filters (absolute log₂ FC > 1; adjusted p- value < 0.1). The CRAFT annotation key: L is circRNA length; M is the number of miRNA binding sites; R is the number of RNA binding protein binding sites; O is the number of ORFs with a stop codon; na indicates the CRAFT software did not generate results probably due to the gene being recently identified.

Chr	Start	End	Gene/CircAtlas identifier	Strand	Adjusted p-value	log2 FC	L	M	R	O
Y	20,507,352	20,521,300	TTTY10/hsa-TTT Y10_0001	−	0.094	3.44	1237	44	32	1
19	54,898,113	54,910,065	NCR1/ hsa-NCR1_0003	+	0.099	3.32	9022	323	311	1
2	105,049,171	105,049,350	MRPS9/ hsa-MRPS9_0010	+	0.057	2.37	180	0	3	0
2	86,688,232	86,701,949	RNF103-CHMP3/ hsa-RNF103-CHMP3_0001	−	0.069	2.00	13,718	256	302	1
12	94,375,886	94,403,262	CEP83/ hsa-CEP83_0013	−	0.084	1.95	1303	16	90	1
15	49,876,278	49,879,459	ATP8B4/ hsa-ATP8B4_0010	−	0.087	1.67	466	3	4	0
2	174,141,825	174,144,376	OLA1/ hsa-OLA1_0013	−	0.096	1.38	2552	32	172	1
10	18,596,208	18,648,645	NSUN6/ hsa-NSUN6_0004	−	0.089	1.31	702	4	14	0
8	22,498,001	22,505,769	PPP3CC/ hsa-PPP3CC_0037	+	0.082	1.07	7769	171	304	2
16	53,442,658	53,447,106	RBL2/ hsa-RBL2_0020	+	0.085	−1.05	415	5	8	0
5	142,789,430	142,792,773	ARHGAP26/ hsa-ARHGAP26_0083	+	0.091	−1.10	3344	136	244	1
3	158,122,103	158,225,733	RSRC1/ hsa-RSRC1_0010	+	0.094	−1.20	10,729	85	318	0
1	161,791,408	161,802,272	ATF6/ hsa-ATF6_0016	+	0.094	−1.32	555	35	16	0
3	196,064,309	196,065,600	TFRC/ hsa-TFRC_0049	−	0.052	−1.52	na	5	na	0
14	103,386,041	103,457,212	MARK3/ hsa-MARK3_0091	+	0.087	−1.55	1865	40	114	1
2	206,046,504	206,086,042	INO80D/ hsa-INO80D_0042	−	0.093	−1.75	1454	70	64	3
6	87,371,593	87,375,710	SMIM8/ hsa-SMIM8_0001	+	0.099	−1.76	1143	4	33	1
3	148,586,122	148,667,102	ENSG00000285557/na	−	0.060	−1.84	na	na	na	na
12	72,331,532	72,331,872	TRHDE/ hsa-TRHDE_0035	+	0.096	−1.93	341	7	1	0

Our implementation of circRNAFlow incorporated annotation from the circAtlas [23] to determine if the circRNA identified was novel or known (Table 1; Fig. 2). To assess biological processes and molecular functions altered in the whole blood transcriptomes due to CF, over-representation enrichment analyzes was performed utilizing clusterProfiler (v4.4) [22]. The human genome GRCh38 was used as the reference set. Regarding circRNAs in CF compared with controls, circRNAs were considered dysregulated when the adjusted p-value < 0.1. The host genes of the dysregulated circRNAs were used as input to clusterProfiler (Fig. 3A–D). The hypergeometric test was utilized for over-representation evaluation for the host genes of dysregulated circRNAs. The Benjamini-Hochberg method was used to calculate the adjusted p-values for the enrichment. The gene ontology (GO) biological process, GO molecular function, GO cellular component, and the biological pathway KEGG databases were examined (Fig. 3). In general terms, circRNAs were defined as up regulated when the log₂ FC > 0 and the adjusted p-value < 0.1; circRNAs were down regulated when the log₂ FC < 0 and the adjusted p-value < 0.1.

2.3. Other downstream analysis

Ingenuity Pathway Analysis (IPA) QIAGEN, USA is a commercial platform based on the IPA Ingenuity Knowledge Base was also utilized for functional annotation of host genes of the dysregulated circRNAs in CF cases homozygous for the F508delCFTR pathogenetic variant (Supplement Fig. S1). The canonical pathway analysis module of IPA was utilized to annotate pathway and biological functions. The p-value <0.05 was considered a statistically significant threshold. CRAFT (CircRNA Function prediction Tool) [24] is a computational biology tool which predicts circRNA functions. CRAFT predicts the circRNA interactions with miRNAs and RNA-binding proteins and assesses the protein coding potential of circRNAs. The CF dysregulated circRNAs highlighted in Table 1 were used as input to CRAFT.

3. Results

3.1. Novel dysregulated circRNAs discovered in CF

The fastq files for whole blood transcriptomes from 20 CF patient samples homozygous for the F508delCFTR variant and 20 non-CF healthy controls were used as input into circRNAFlow (Fig. 1) to identify and characterize dysregulated circRNAs in CF samples. A total of 118 dysregulated circRNAs were discovered in whole blood transcriptomes from individuals with CF compared to healthy controls (Table 1; Table S1). There are 33 circRNAs up regulated and 85 circRNAs down regulated in CF samples compared to the non-CF controls (Table S1). Some genes have more than one circRNA.

Over half of the host genes harboring dysregulated circRNAs in CF cells compared to healthy controls have key cellular functions (Table S2). The dysregulated circRNAs originate from host genes known to function as transcription regulators (10.4%), transporters (11.5%), enzymes (14.6%), kinases (9.4%), peptidases (5.2%), phosphatases (3.1%) and transmembrane receptors (2.1%) (Table S2). The percentages of dysregulated circRNA host genes with known function, as per the IPA knowledge base, are provided in parenthesis. Some of these are highlighted in the context of CF. Phosphatase and tensin homolog (PTEN) is a dysregulated circRNA host gene. PTEN is a phosphatase and key regulator of cellular metabolic activity. CFTR is a chloride channel that also acts as a scaffold for many proteins including PTEN. PTEN forms complexes with CFTR at the mitochondria plasma membrane for normal succinate balance and to regulate cellular responses to infection [25]. In CF, the PTEN–CFTR complexes are functionally impaired leading to mitochondrial metabolic dysfunction and predisposition to Pseudomonas aeruginosa airway infections [25,26]. ADAM metallopeptidase domain 10 (ADAM10) [27] is one example of a dysregulated circRNA host gene that functions as a peptidase. Synaptosome associated protein 23 (SNAP23) is a dysregulated circRNA host gene with known function as a transporter. Interestingly, CFTR chloride channels are regulated by the SNAP23/syntaxin 1A complex [28,29].

The dysregulated circRNAs discovered in CF samples compared to the non-CF controls were considered together with their known circAtlas identifiers [23] and the human genome location defined by the chromosome, the start and the end coordinates, and the strand orientation (Table 1; Table S1). Six dysregulated circRNAs with the largest magnitude of change (absolute log₂ FC) and most statistically significant (adjusted p-value) are highlighted. Among the six dysregulated circRNAs, three up regulated circRNAs (hsa-TTTY10_0001, hsa-ATP8B4_0010, and hsa-OLA1_0013)(Fig. 2A–C) and three down regulated circRNAs (ENSG00000285557, hsa-MARK3_0091, and hsa-tfrc_0049) are reviewed (Fig. 2 D–F).

3.1.1. Novel up regulated circRNAs discovered in CF

CircRNA hsa-TTTY10_0001 is up regulated in CF cells compared to non-CF controls (log₂ FC=3.44; adjusted p-value=0.094) (Fig. 2A). CircRNA hsa-TTTY10_0001 comprises 1237 bases and is predicted to interact with 44 miRNAs and 32 RNA binding proteins. CircRNA hsa-TTTY10_0001 has one open reading frame (ORF) across the BSJ (Table 1).

CircRNA hsa-ATP8B4_0010 is up regulated in CF cells compared to the healthy controls (log₂ FC=1.67; adjusted p-value=0.087) (Fig. 2B). CircRNA hsa-ATP8B4_0010 comprises 466 bases and is predicted to interact with 3 miRNAs and 4 RNA binding proteins. CircRNA hsa-ATP8B4_0010 has no ORF across the BSJ (Table 1).

CircRNA hsa-OLA1_0013 is up regulated in CF samples compared to the healthy controls (log₂ FC=1.38; adjusted p-value=0.096) (Fig. 2C). CircRNA hsa-OLA1_0013 contains 2552 bases. CircRNA hsa-OLA1_0013 is predicted to interact with 32 miRNAs and 172 RNA binding proteins. CircRNA hsa-OLA1_0013 comprise one ORF across the BSJ (Table 1).

3.1.2. Novel down regulated circRNAs discovered in CF

The circRNA ENSG00000285557 is a novel circRNA that is down regulated in CF samples compared to healthy controls (log₂ FC=−1.84; adjusted p-value=0.060) (Fig. 2D). There are no CRAFT predictions computed for the interactions of this novel circRNA with miRNAs or RNA binding proteins (Table 1).

CircRNA hsa-MARK3_0091 is down regulated in CF cells compared to healthy controls (log₂ FC=−1.55; adjusted p-value=0.087) (Fig. 2E). CircRNA hsa-MARK3_0091 comprises 1865 bases and is predicted to interact with 40 miRNAs and 114 RNA binding proteins. CircRNA hsa-MARK3_0091 has one open reading frame (ORF) across the BSJ (Table 1).

CircRNA hsa-tfrc_0049 is down regulated in CF cells compared to healthy controls (log₂ FC=−1.52; adjusted p-value=0.052) (Fig. 2F). CircRNA hsa-tfrc_0049 is predicted to interact with 5 miRNAs and no RNA binding proteins. CircRNA hsa-tfrc_0049 has no ORF across the BSJ (Table 1).

3.2. Pathways of host genes of novel dysregulated circRNAs discovered in CF

The top ten GO biological processes predicted for the host genes from which dysregulated circRNA (adjusted p-value < 0.28) arise in the CF samples are intracellular transport, regulation of organelle organization, membrane organization, small GTPase mediated signal transduction, chromatin organization, positive regulation of protein-containing complex assembly, oligodendrocyte differentiation, positive regulation of response to endoplasmic reticulum stress, regulation of membrane tubulation, and negative regulation of dendritic spine development (Fig. 3A; Table S3A).

The top ten GO molecular functions predicted for dysregulated circRNA host genes discovered in CF transcriptomes (adjusted p-value < 0.13) are purine ribonucleoside triphosphate binding, ATP binding, adenyl ribonucleotide binding, adenyl nucleotide binding, zinc ion binding, protein serine kinase activity, protein serine/threonine kinase activity, protein serine/threonine/tyrosine kinase activity, protein kinase activity, metalloexopeptidase activity (Fig. 3B; Table S3B).

The top ten over-represented GO cellular components for the dysregulated circRNA host genes (adjusted p-value < 0.06) are Golgi apparatus, organelle envelope, envelope, Golgi membrane, nuclear body, nuclear envelope, nuclear membrane, specific granule membrane, PML body, phospholipid-translocating ATPase complex (Fig. 3C; Table S3C).

The three canonical pathways (DNA methylation and transcriptional repression signaling, cellular senescence and ferroptosis signaling) are the three most statistically significant pathways en riched from the Qiagen IPA canonical pathway analysis (Supplement Fig. S1). A finding borne from both the KEGG (Fig. 3D) and the Qiagen IPA canonical pathway (Supplement Fig. S1) enrichment analyzes is that cellular senescence is enriched in dysregulated circRNA host genes from CF samples. The cellular senescence pathway is enriched in the host genes of dysregulated circRNAs in CF with the clusterProfiler software and the KEGG database (adjusted p-value=0.014) (Fig. 3D). The role of cellular senescence has been previously implicated in CF [30,31]. Whilst ferroptosis is an iron-dependent mode of cell death caused by the accumulation of lipid peroxides. Ferroptosis has also been implicated in several pathologies including infection, chronic inflammation and CF [32,33].

3.3. Prediction of CF dysregulated circRNAs as proteins translated and/or interaction partners with miRNAs and RNA binding proteins

CRAFT is a software tool which predicts circRNA-specific translation potential with a focus on open reading frames (ORFs) overlapping the BSJs, which are typically absent in the linear transcripts (Fig. 4A). Approximately 50% (i.e., 10/19) of CF dysregulated circRNAs (Table 1) have one or more such ORFs with a stop codon. CircRNA INO80D has three such ORFs (Table 1). Dysregulated circRNAs in CF are calculated to interact with miRNAs and the associated miRNA targets are enriched for different signaling pathways such as cellular senescence, PTEN regulation and transcriptional regulation of TP53 (Fig. 4B, C). Dysregulated circRNAs in CF are predicted to interact with RNA binding proteins enriched for KEGG pathways such as (a) spliceome, (b) mRNA surveillance and (c) RNA degradation (Fig. 4D). CircRNA hsa-TTTY10_0001 (Fig. 4E), and other CF dysregulated circRNAs (Table 1) are predicted to interact with RNA binding proteins enriched for IGF2BP3, FGFR2 signaling pathways, metabolism of RNA, mRNA splicing, processing of capped intron containing pre-mRNA and mRNA 3′ end processing.

Fig. 4.

CRAFT prediction output based on the CF dysregulated circRNAs in Table 1. (A) A summary plot of hsa-TTTY10_0001 with predicted miRNA (blue bars) and RBP binding sites (orange bars), and the longest ORF, with a stop codon (red line), along the circRNA sequence. (B, C) Network shows Reactome pathway enrichment analysis results on miRNA target genes for CF dysregulated circRNAs in INO80D and ARHGAP26 respectively (Table 1) (D) Plot of KEGG enrichment analyzes on RNA binding proteins potentially interacting with hsa-MARK3_0091. (E) Plot of Reactome enrichment analyzes on RNA binding proteins potentially interacting with hsa-TTTY10_0001. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

4. Discussion

CFTR is widely expressed in epithelia-lined tissues. Pathogenic CFTR variants lead to the most severe clinical manifestations within the respiratory and gastrointestinal tracts, with pulmonary manifestations being the primary cause of morbidity and mortality in people with CF. Absence of CFTR function alters ion transport across airway epithelia, resulting in dehydration of the airway surface liquid. This, combined with abnormal mucin secretion, creates an airway environment susceptible to recurring opportunistic infections and hyperinflammation, resulting in progressive lung damage [34]. The immune response plays a significant role in CF pathogenesis in the lungs. In addition, functional CFTR protein is expressed in multiple types of immune cells, including lymphocytes, monocytes [35]/macrophages, neutrophils, and dendritic cells. CFTR variants that alter CFTR protein expression/function impair immune cell function, leading to an inability to resolve recurrent infections and inflammation in the lungs [36].

Multiple proteomic and transcriptomic studies of immune cells from individuals with CF showed CFTR expression to be down regulated compared to healthy controls [10,37–41]. In addition, transcriptomic responses in plasma or whole blood to the administration of CFTR modulators showed the expression of hundreds of genes to be altered in immune cells by partial rescue of CFTR function [10,38]. Moreover, transcriptomic studies from immune cells have been used to identify molecular signatures of CF disease status in multiple organs, including the lung and pancreas [40–42], and to identify biomarkers for clinical responsiveness to CFTR modulator therapies [10,37,38]. Our current study analyzes whole blood transcriptomes of CF patients compared to healthy controls for CF dysregulation of circRNAs. As with previous transcriptomic studies, changes observed in circRNA expression in immune cells are likely due to both the indirect and direct effects of CFTR loss. Changes in circRNA expression could be caused indirectly due to loss of CFTR in lungs, intestines, pancreas, and other organs as whole blood circulates throughout all the major organs. Alternatively, circRNA dysregulation could be caused directly by lack of a fully functioning CF variant in immune cells. We predict that a better understanding of circRNA expression profiles can guide the identification of new biomarkers for both CF disease pathogenesis and therapy responsiveness.

4.1. Discovery of novel circRNAs in CF

One of the goals of this study was to discover whether dysregulated circRNAs exist in CF cells. Our study identified 118 dysregulated circRNAs which exist in CF whole blood transcriptomes compared to healthy controls (Table S1). The six most dysregulated circRNAs, three up regulated circRNAs (hsa-TTTY10_0001, hsa-ATP8B4_0010, and hsa-OLA1_0013) (Fig. 2A–C) and three down regulated circRNAs (ENSG00000285557, hsa-MARK3_0091, and hsa-tfrc_0049) are discussed (Fig. 2D–F). circRNA hsa-TTTY10_0001. Little is known about the Testis-Specific Transcript, Y-Linked 10 (TTTY10). TTTY10 is a long non-coding RNA (lncRNA) located on the human male sex chromosome. It is widely expressed in male tissues but not in female tissues. The expression data source is from GTEx Analysis Release V8 (dbGaP Accession phs000424.v8.p2). The cohorts for CF cases and the non-CF controls comprise 55% and 65% females, respectively [10]. In this study, consideration is given to two groups, the cases, and the controls, with no adjustments made for sex or age. Because TTTY10 is expressed only in male tissues, this signifies a possible role for TTTY10 contributing to sex differences observed in CF [43–46]. In human CF disease, females have worse prognoses and worse health outcomes compared to males. Genes like TTTY10 could potentially contribute to the “gender gap” in CF survival, CF pulmonary outcomes, and CF disease progression [43,44–46]. Genes on the male sex chromosome, like TTTY10, may either offer some protective and clinical benefits to males with CF or could contribute to male specific CF, for example vas deferens absence, which leads to male infertility [47]. TTTY10 appears to be human specific as it has not been found in any other species. Lncbook2 [56,57] shows an overlapping long intergenic non-coding (linc) gene in the negative strand (antisense direction) of TTTY10. The linc gene overlaps the intergenic region of the TTTY10 lncRNA. This circRNA (hsa-TTTY10_0001) is predicted to interact with miRNAs. Thus, this gene has potential to form part of key competing endogenous RNA interaction networks. circRNA hsa-ATP8B4_0010. ATP8B4 is a component of the P4-ATPase flippase complex. P4-ATPases are sub family members of the P-type ATPase superfamily which translocate membrane lipids from the exoplasmic or luminal leaflet to the cytoplasmic leaflet to regulate trans-bilayer lipid asymmetry [48,49]. Mammalian P4-ATPases localize to the various subcellular organelles or the plasma membrane where they translocate various lipids. Lipid translocation is an important process for many cellular activities [48,49]. While recent advances in biochemical and molecular structural studies of P4-ATPases have improved the current understanding of lipid transporting machinery, the mechanism of substrate specificity and the regulatory mechanism of the enzymes remain largely unknown [48,49]. The substrates of ATP8B4 remain to be determined [49] and are predicted based on homology.

Abnormal lipid levels are observed in the tissues of CF patients [50,51]. ATP8B4 is a component of the P4-ATPase flippase complex with functions in lipid translocation. This suggests that ATP8B4 may play some role in dysregulated lipids in CF. Whilst the role of ATP8B4 in context of CF is unclear, an important para-log of AT8B4, ATP8B1 has an essential role in facilitating the surface localization of CFTR in the epithelial cells of the intestines and the lungs [52]. ATP8B1 deficiencies cause progressive familial intrahepatic cholestasis type 1 (PFIC1) disease. A significant proportion PFIC1 patients develop extrahepatic symptoms typical of CF phenotypes (e.g., pulmonary infection, sweat gland dysfunction, dyslipidemia, and failure to thrive). The CFTR levels are down regulated in the livers of PFIC1 patients which is also characteristic of CF. ATP8B1 [52] is essential for correct apical localization of CFTR in human intestinal and pulmonary epithelial cells. Impaired CFTR localization underlies some of the extrahepatic phenotypes observed in ATP8B1 deficiency. P4-ATPase deficiencies are linked to a number of complex medical disorders including diabetes and reduced fertility [53], these are conditions which also affect a high proportion of CF patients. circRNA hsa-OLA1_0013. OLA1 (Obg Like ATPase 1) [54] has an important negative role in cell adhesion and cell spreading, partly through the regulation of focal adhesion kinase expression and cofilin phosphorylation [55]. Manipulation of OLA1 may lead to significant changes in cell adhesion and cell spreading [54,55]. circRNA ENSG00000285557. Not only is this circRNA novel to CF, but it is also novel to circAtlas as this circRNA ENSG00000285557 is absent from circAtlas [23] and thus has no circAtlas identifier. ENSG00000285557 is a novel lncRNA gene. From public repositories, this lncRNA is expressed in the pancreas. LncBook2 [56,57] shows ENSG00000285557 as mammalian-specific as this lncRNA is absent from the four other vertebrate classes, birds, fish, reptiles, and amphibians. Interestingly, this gene is predicted (by LncBook2) to interact with miRNA including hsa-mir-145–5p [58] and other miRNAs [59] known to bind to the CFTR gene. The length of circRNA ENSG00000285557 is an outlier at 80Kb (Table 1). circRNA hsa-MARK3_0091. Microtubule Affinity Regulating Kinase 3 (MARK3) encodes for a protein kinase activated by phosphorylation, which in turn, phosphorylates the tau proteins MAP2 and MAP4. circRNA hsa-tfrc_0049. The host gene transferrin receptor (TFRC) [60] is a protein-coding gene encoding a cell-surface receptor necessary for ion uptake via receptor-mediated endocytosis. The host gene functions in cellular iron ion homeostasis. A pathogenetic variant affecting the iron import receptor TFRC is the cause of a human primary immunodeficiency, elucidating the important role of TFRC and iron in immune cell function [61].

4.2. Discovery of new CF circRNAs with pathways and host genes previously associated with CF via other studies

4.2.1. Cellular senescence

Cellular senescence is a reversible, non-proliferating, cellular state due to cell cycle arrest [31]. Cellular senescence pathways have been associated with CF [30,31]. In the context of CF, cellular senescence has been described as a double-edged sword [31]. On the one hand, cellular senescence is thought to suppress bacterial infection, facilitate tissue repair and prevent the proliferation of potential cancer cells [62,63]. On the other hand, [30,31] cellular senescence promotes a pro-inflammatory environment causing damage to tissues, leading to a chronic disease and perpetuation of CF-related chronic inflammation [31]. CF is a chronic multiorgan disease characterized by neutrophilic-induced inflammation. CF chronic inflammatory processes are characterized by significant release of cytokines and chemokines, together with redox oxidative stress and proteases (peptidases such as elastases), which can accelerate cellular senescence [31].

Importantly, our results corroborate the role of cellular senescence among the pathways regulated by many of the host genes of the newly discovered dysregulated circRNAs. From our KEGG and IPA canonical over-represented pathways in CF, the cellular senescence pathway is enriched (Fig. 3D, Supplement Fig. S1). Six dysregulated circRNA host genes from the cellular senescence pathway were identified and these are: E74 like ETS transcription factor 2 (ELF2 [64]); protein phosphatase 2 regulatory subunit B alpha (PPP2R2A); protein phosphatase 3 catalytic subunit gamma (PPP3CC); phosphatase and tensin homolog (PTEN [25,26]); RB transcriptional corepressor 1 (RB1); RB transcriptional corepressor like 2 (RBL2).

4.2.2. Ferroptosis signaling pathway

Maniam et al. [32] established that CF airway epithelial cells are prone to cell death by the iron-dependent ferroptosis pathway. Ferroptosis is a type of programmed cell death that depends on iron and is characterized by the abnormal accumulation of intracellular iron, depletion of glutathione, deactivation of glutathione peroxidase, and abnormal levels of lipid peroxidation [32]. Maniam et al. 2019 suggest that whilst lipid peroxides and iron levels are detected in the lungs of CF patients, it remains unclear the extent to which the ferroptosis pathway contributes to the initiation, maintenance, and progression of CF disease [32]. They suggest that a better understanding of how ferroptosis contributes to the pathophysiology of CF lung disease is necessitated for the development of new therapeutic interventions. Literature reviews covering the role of ferroptosis in lung diseases provide support for this premise [32,33,65,66]. Our study suggests that circRNA (e.g., hsa-tfrc_0049) and the associated ceRNA interaction network may have a role in modulating the ferroptosis pathway in CF. Following up on such leads with future directed studies is likely to improve the understanding the role of ferroptosis in CF disease, which in turn, could lead to the development of better therapeutics and biomarkers for CF disease progression. There is a medical unmet need to find improved strategies to identify CF patients with early lung disease and to identify those at risk for more progressive lung disease, allowing earlier intervention [67].

4.2.3. Other CF-associated host genes and pathways

Some of the dysregulated circRNA identified in this study are associated with host genes and pathways previously associated with CF. In our circRNA study, we identified a circRNA in ATF6 dysregulated in CF. The activating transcription factor 6 (ATF6) host gene is known to be associated with the endoplasmic reticulum stress response observed with onset of CF [68]. Thus, our findings with ATF6 circRNA and its associated host gene support the results of previous studies using different biochemical techniques and strategies. This is the first time these pathways are highlighted in relation to the host genes of dysregulated circRNAs in CF patient cells. This finding provides support that circRNAs are likely to function in previously well-characterized molecular pathways associated with CF. These results also show that our data corroborates previous CF studies, strengthening our claims about the importance of circRNAs in CF. In context of CF research, circRNA should be studied as widely as the other classes of RNA such as mRNA, miRNA and the other lncRNA classes.

4.2.4. Possible CF-related diabetes candidates

Over 50% of adult CF patients have CF-related diabetes [69], studies are needed to decipher if circRNAs are associated with the onset of CF-related diabetes. Nineteen dysregulated circRNAs were among the most highly differentially expressed circRNAs in this study (Table 1). Six of the 19 CF dysregulated circRNA host genes (NCR1 [70], ATP8B4 [53], RBL2 [71], TFRC [72], INO80D [73], ATF6 [74]) are also associated with various aspects of diabetes etiology. The design of additional experiments is warranted to determine if these circRNAs are implicated with onset or progression of CF-related diabetes.

4.3. Prediction of CF dysregulated circRNAs with potential to translate into proteins and/or interact with miRNAs and RNA binding proteins

CF dysregulated circRNAs were computed to have one or more of three different functions such as: (a) potential to translate into proteins; (b) interacting with miRNAs; and (c) interacting with RNA binding proteins [24] (Table 1; Fig. 4). Common themes for pathway enrichment of miRNA targets related to the CF dysregulated circRNAs (Table 1) were cellular senescence, transcriptional regulation of TP53 and regulation of PTEN (Fig. 4B,C). Via this study, TP53 related pathways are highlighted in having a role in CF. TP53 related pathways are previously implicated in CF [75]. As mentioned, cellular senescence has been associated with CF [30,31]. Common themes for pathway enrichment of RNA binding proteins predicted to interact with the CF dysregulated circRNAs (Table 1) were RNA spliceome, mRNA surveillance and pathways modulating RNA stability such as RNA degradation (Fig. 4D,E). Via this study, RNA degradation is highlighted as having a role in CF. RNA degradation is an under-explored research area in CF disease.

4.4. Caveats and limitations of the study

We used the sequencing of total RNA with ribosomal RNA depleted libraries comprising 60–80 million paired end reads per transcriptome. Whilst the transcriptomes are deeply sequenced, if the sequencing had been done in more depth (for example 200 million paired end reads per transcriptome [76]) more circRNA discoveries could have been made [7]. The analysis is adequately powered for transcriptome analysis with suitable numbers of cases (n = 20) and controls (n = 20).

We assume that the CF-related changes in the circRNAs may be due to altered expression and/or function of their host genes and their corresponding pathways, which may apply for some of the circRNAs, but likely not for all. It is possible that the discovered dysregulated circRNAs could be working on completely different pathways than their host genes. Our study results will need future verification via biological experiments, and this is the scope of future studies.

5. Conclusions

These results identify and suggest potential roles for dysregulated circRNAs in CF. The results suggest that studying circRNAs, a new emerging class of long non-coding RNA, may provide a more complete picture of molecular characterization of CF. It is likely that dysregulated circRNAs have a role as circulating biomarkers as well as modulation of CF phenotype. Our study is the first to identify and characterize dysregulated circRNAs in CF. The study highlights underappreciated and underexplored associations of circRNAs in CF. Additionally, the development of computational biology pipelines, like circRNAFlow, enabled the discovery of dysregulated circRNA in whole blood transcriptomes of CF patients, thus emphasizing the importance of software development in biomedical research.

Availability for circRNAFlow code and configuration files are stored on GitHub (https://github.com/yjkedwards/circRNAFlow).