Discovery of novel circular RNAs of the apoptosis-related BAX gene in breast cancer, by combining nanopore and next-generation sequencing

TL;DR

The study identifies new circular RNAs from the BAX gene in breast cancer, which may regulate cancer signaling and apoptosis.

Contribution

The discovery of 82 novel BAX-derived circRNAs and their potential regulatory roles in breast cancer signaling is a key novelty.

Findings

106 BAX-derived circRNAs were identified, with 82 being novel and exhibiting diverse biogenesis features.

Several circRNAs showed subtype-specific expression patterns in breast cancer cell lines.

Some circRNAs may act as miRNA sponges and bind RNA-binding proteins, influencing apoptosis and cancer signaling pathways.

Abstract

Circular RNAs (circRNAs) have emerged as significant regulators of cancer biology. However, the characterization and the regulatory potential of circRNAs deriving from key apoptotic genes remain poorly understood in breast cancer. We aimed to comprehensively characterize circRNAs originating from the pro-apoptotic BAX gene and predict their regulatory potential in BC signaling. Targeted amplification of circular BAX transcripts was conducted in eleven cancerous and one non-cancerous breast cell lines, followed by third-generation (nanopore) and next-generation sequencing. Finally, extensive bioinformatic analysis was conducted. Therefore, we identified 106 circRNAs, 82 of which were novel. These circRNAs exhibited diverse biogenesis features, including exon skipping, intron retention, and rare inclusion of exon 5. Expression profiling revealed subtype-specific patterns, with several circRNAs being detected only in triple-negative or luminal BC subtypes, while circ-BAX-55b was detected exclusively in the non-cancerous cell line. Many circRNAs were predicted to sponge miRNAs such as miR-152-5p, miR-4802-5p, and miR-3619-5p, potentially modulating signaling pathways including MAPK, PI3K/AKT, and NFκB. Extensive sponging of miR-152-5p, which targets BAX mRNA, suggests a feedback mechanism modulating apoptosis. Several circRNAs also contain binding sites for RNA-binding proteins (RBPs) such as RBM6 and HNRNPF, possibly dictating RNA fate. Overlapping miRNA and RBP-binding sites imply complex competitive or cooperative interactions. Additionally, multiple circRNAs, such as circ-BAX-6c, displayed translation-related features. In conclusion, these findings reveal a novel repertoire of BAX circRNAs with putative roles in BC signaling and apoptosis regulation. Moreover, they constitute a valuable resource for functional studies as well as potential biomarkers or therapeutic targets in breastcancer.

The online version contains supplementary material available at 10.1007/s00438-026-02388-1.

Figures

Genomic locus of *BAX* gene and associated transcripts. The condensed form of linear transcripts is depicted in blue; circular transcripts (circRNAs) are also depicted in condensed form (red). Novel circRNAs that were firstly identified in this study are depicted in light red color, whereas circRNAs colored in orange were previously identified in hematological malignancies, in our lab. Single-exonic circRNAs are illustrated in green

Functional interplay between identified circRNAs and particular signaling pathways. This predicted ceRNA regulatory map illustrates circRNAs (green boxes) sequestering miRNAs (pink boxes), thereby preventing the downregulation of target transcripts that contribute to specific signaling pathways (blue boxes). Solid orange lines represent validated interactions between linear *BAX* transcripts and the respective miRNA. Dashed orange lines signify the validated indirect regulation of signaling pathways by miRNAs via the modulation of pathway-contributing mRNAs

Structural characterization and coding potential of identified *BAX* circRNAs. Transcripts features including predicted m^6^A motifs, Internal Ribosome Entry Sites (IRES), and Open Reading Frames (ORFs) of the *BAX* circRNAs comprising an ORF are illustrated. The presence of m^6^A motifs and IRES elements indicates potential cap-independent translation initiation sites, while the identified ORFs highlight the coding capacity of these circular isoforms relative to their linear parental genes

Alignment of particular circRNAs comprising an open reading frame (ORF) against major *BAX* mRNAs. Transcripts features including predicted m^6^A motifs, Internal Ribosome Entry Sites (IRES), and Open Reading Frames (ORFs) are illustrated. Only *BAX* circRNAs comprising an ORF and being aligned against linear *BAX* transcripts with GenBank IDs [NM_001291428](NM_001291428) (A) and [NM_138764](NM_138764) (B) are shown

Full text

Introduction

Circular RNAs (circRNAs) have recently emerged as an important class of molecules that regulate gene expression and fundamental cellular processes, including apoptosis, cell cycle control, and response to stress. Growing evidence suggests a significant contribution of these molecules to cancer initiation and progression. However, the identification and the assessment of the regulatory potential of circRNAs originating from genes with a pivotal role in apoptosis and their impact in cellular processes, deriving from their effect in key signaling pathways, remain insufficiently explored. A deeper exploration of this topic is essential to reveal complex post-transcriptional regulatory mechanisms in cancer, and for the identification of regulatory molecules with potential biomarker or therapeutic utility.

The circRNAs are circular single-stranded RNA molecules produced by a back-splicing event. The alternative back-splicing mechanism results in the generation of multiple circRNAs from one gene (Kristensen et al. 2019; Zhang et al. 2020). Various biogenesis models have been proposed for this type of RNA molecules producing four major categories of circRNAs including EcircRNAs, EIcircRNAs, ciRNAs, and tricRNAs (Zhang et al. 2013, 2014; Lu et al. 2015; Errichelli et al. 2017).

circRNAs serve diverse regulatory roles. An intriguing observation regarding the localization of circRNAs is that EIcircRNAs and ciRNAs localize predominantly in the nucleus regulating the transcription of their parental genes (Zhang et al. 2013; Pisignano et al. 2023). Others act as microRNA (miRNA) or RNA binding protein (RBP) sponges, or serve as scaffolds for protein interactions affecting their regulatory effect (Memczak et al. 2013). Some circRNAs may also undergo translation in a cap-independent manner (Abe et al. 2015). These functions allow circRNAs to modulate autophagy, apoptosis, cell cycle, and other key cellular processes (Du et al. 2017, 2018; Wang et al. 2021). Besides their multitudinous regulatory potential, circRNAs appear also as valuable clinical biomarkers as their covalently closed structure provides resistance to exonuclease degradation (Suzuki et al. 2006; Zhang et al. 2018b).

The BCL2 associated X, apoptosis regulator (BAX) gene is a member of the BCL2 family, located in 19q13.33, and comprises 7 exons. BAX protein is a pivotal regulator of the cell’s fate as it is a major apoptosis initiation factor (Oltvai et al. 1993). With the appropriate apoptotic signal BAX, which is cytosolic under normal conditions, translocates to the mitochondrial membrane creating pores, leading to the release of cytochrome c and apoptosis activation (Jurgensmeier et al. 1998). There are two observations worth mentioning for the linear transcripts of BAX. Firstly, intron retention between exons six and seven appears to be a common splicing event in the validated transcripts, and the fifth exon is present only in one identified non-coding transcript.

To date, only a limited number of circRNAs of BAX have been identified. Prompted by the crucial role of this gene we aimed to discover novel circular transcripts of BAX focusing on breast cancer (BC). BC is a heterogenous disease and one of the most commonly occurring, and with high mortality rate cancers. Research advances propose the significance of molecular signatures in this cancer type, with patients being categorized into multiple subcategories with substantial differences in prognosis (Waks and Winer 2019). The effect of BAX on a transcriptional or protein level has been well documented in BC. Its elevated levels are observed in normal epithelium or non-malignant breast cell lines, whereas a downregulation of its levels appears in malignant epithelium and BC cell lines. Furthermore, its overexpression appears to increase radiosensitivity in BC cell lines and to reduce tumor growth in mice (Bargou et al. 1996; Sakakura et al. 1997).

Numerous circRNAs have been proposed as biomarkers in BC (Liang et al. 2020). Furthermore, evidence about the ability of circRNAs to bind miRNAs or RBPs and subsequently affecting mRNAs and the respective proteins in BC is steadily increasing (Zhang et al. 2021; Papatsirou et al. 2022). They may affect numerous cellular processes and cancer cell properties such as invasion and migration, with a relative impact being observed in tumor growth and progression as well (Zhang et al. 2018a; Liang et al. 2019b; Huang et al. 2023). Additionally, plentiful circRNA-miRNA regulatory axes have been proposed to affect sensitivity or resistance of BC cells to medication and circRNAs have been proposed as potential therapeutic targets or molecules with beneficial effects for BC treatment (Liang et al. 2019a, b; Huang et al. 2023).

Prompted by the aforementioned information, we intended to identify novel circRNAs of BAX gene in BC, and in-silico investigate their potential regulatory roles with an emphasis on the regulation of cell signaling. More specifically, eleven BC cell lines and one normal, non-cancerous breast cell line were used for the amplification of circRNAs deriving from BAX. Long-read sequencing was performed using nanopore technology and integrated with next-generation sequencing for the identification of novel circRNAs. Sequence validation, expression analysis, and functional prediction were performed using a suite of in-house and publicly available computational tools, with an emphasis on uncovering potential regulatory roles in cell signaling pathways.

Methods

Cell line culture

Twelve established human epithelial cell lines having derived from female individuals with BC were used in the current study. The cancerous cell lines BT-20 (CVCL_0178), BT-474 (CVCL_0179), HCC70 (CVCL_1270), Hs 578T (CVCL_0332), MCF-7 (CVCL_0031), MDA-MB-231 (CVCL_0062), MDA-MB-453 (CVCL_0418), MDA-MB-468 (CVCL_0419), SK-BR-3 (CVCL_0033), T-47D (CVCL_0553), ZR-75-1 (CVCL_0588), and the non-cancerous cell line MCF-12A (CVCL_3744) were previously purchased and propagated according to the American Type Culture Collection (ATCC^®^) guidelines, in a humidified incubator at 37 °C and 5% CO_2_. All these cell lines have not been previously reported as misidentified or contaminated. All cell culture media and supplements were purchased from Biowest (Nuaillé, France), except human recombinant insulin which was obtained from Eli Lilly and Company (Indianapolis, IN, USA).

Total RNA extraction and cDNA synthesis

Total RNA from the aforementioned twelve breast cell lines was extracted using the TRI Reagent^®^ (Sigma-Aldrich, Merck Group, St. Louis, MO, USA) (Chomczynski and Sacchi 1987), prior to its storage at ₋80℃. Purity and concentration of total RNA were determined with a BioSpec-nano Micro-volume UV–vis Spectrophotometer (Shimadzu, Kyoto, Japan) and its integrity was assessed electrophoretically.

In the next step, 3 µg of total RNA were used for first-strand cDNA synthesis using Random primer 6 (New England Biolabs Ltd., Hitchin, UK) and M-MLV reverse transcriptase (Invitrogen, Thermo Fisher Scientific, Waltham, MA, USA) according to the manufacturers’ instructions.

Two-step polymerase chain reaction, electrophoresis, and clean-up of PCR products

The next step consisted of two PCR cycles with divergent primers to amplify circRNAs (Table S1). Amplification was performed using the KAPA Taq DNA Polymerase (Kapa Biosystems Inc., Woburn, MA, USA) according to the manufacturer’s instructions (Supplementary Methods). The PCR products deriving from the amplification of each exon were combined for each cell line and cleaned using suitable gel and PCR clean-up columns (Macherey-Nagel GmbH & Co. KG, Düren, Germany). Concentration was quantified fluorometrically using a Qubit Fluorometer (Thermo Fisher Scientific, Waltham, MA, USA).

Targeted third-generation sequencing using nanopore technology, for the identification of novel ΒΑΧ circRNAs

Third-generation sequencing was conducted for the identification of full-length circular transcripts from single reads. The pool of PCR products from each cell line was used for the construction of libraries following the Native Barcoding protocol of Oxford Nanopore (Oxford Nanopore Technologies, Oxford, UK). Libraries were sequenced with the MinION Mk1C using the Flongle adapter (Oxford Nanopore Technologies, Oxford, UK) for the assessment of the quality and the representation of the libraries. Next, the pool was sequenced in a PromethION 2 Solo sequencer (Oxford Nanopore Technologies, Oxford, UK) using the R10 chemistry.

Next-generation sequencing for the cross-validation of the novel circRNAs

The next-generation sequencing approach was performed to obtain highly accurate reads for the validation of the sequences of the novel circRNAs. The aforementioned sample that was sequenced using the third-generation sequencing approach, was modified according to the sequencing platform requirements, and sequenced with a single-end 500 approach in the MiSeq platform (Illumina, San Diego, CA, USA) using the MiSeq Reagent Nano Kit v2.

Bioinformatic analysis

For the analysis of the FASTQ files generated from the third-generation sequencing, Minimap2 was utilized in order to align the long reads to the GRCh38/hg38 assembly (Li 2018). The generated FASTQ files from next-generation sequencing were aligned to the same assembly using the STAR aligner (Dobin et al. 2013). SAMtools and BEDtools were incorporated for the generation of the relative SAM, sorted SAM, BAM, indexed BAM and BED files (Quinlan and Hall 2010; Danecek et al. 2021). Files were visualized using Integrative Genomics Viewer (IGV) (Robinson et al. 2011).

Novel transcripts were identified using ASDT (Alternative Splicing Detection Tool) v2.1 algorithm with a modified GenBank™ file and ASDT remodeler (Adamopoulos et al. 2018; Karousi et al. 2025). Manual annotation followed for the determination of the sequence of the novel circRNAs. Expression analysis followed using specific keywords and alignment with Minimap2, BWA, and Bbmap aligners (Li and Durbin 2009; Li 2018).

The novelty of the identified transcripts was validated by alignment against previously identified BAX circRNAs. Furthermore, the identified circRNAs were examined for the presence of N^6^-methyladenosine (m^6^A) motifs using the SRAMP tool, as well as for the presence of Internal Ribosome Entry Sites (IRES) using IRESite (Mokrejs et al. 2010; Zhou et al. 2016). The translational potency of transcripts was assessed using the ORF Finder tool of NCBI, alignment with Clustal Omega, and BAX-related domains were sought with PROSITE.

Aiming to identify the regulatory potential of the novel circRNAs, RNA-binding proteins that potentially bind on the identified circRNAs were discovered with the RBPmap tool (Paz et al. 2014). The identified sequences were assessed for their miRNA sponging activity with the custom prediction tool of miRDB (Chen and Wang 2020). The miRNAs with a ≥ 80% prediction score were used for further analysis. Validated mRNA targets of the aforementioned miRNAs were determined using miRTarBase (Huang et al. 2022). The potential effect of the circRNAs on three major cell signaling pathways, namely MAPK, PI3K/AKT, and NFκB, by the circRNA–miRNA–mRNA regulatory axis, was assessed by exploring the participation of identified mRNAs targets in pathways according to KEGG pathway (Kanehisa and Goto 2000). Lastly, aiming to elucidate the RNA interactome, the position in the circRNAs that miRNAs and RBPs may antagonistically sponge were determined.

Results

Identification of novel BAX circRNAs using a targeted third-generation sequencing approach

Pooled samples (Fig. S1) were sequenced using a targeted third-generation sequencing approach. A total of 106 circRNAs of the human BAX gene were identified, with the implementation of the abovementioned approach. From these, 82 transcripts were identified for the first time in BC and 24 have been previously identified in hematological malignancies. Moreover, alignment of the identified circRNAs to known circRNAs deriving from BAX further validated the novelty of the transcripts. Lastly, the identified circRNAs cover 34.13% of BAX genomic region.

The novel transcripts have variable structure with exonic extension or truncation, intron retention, and exon skipping phenomena being observed (Fig. 1). Furthermore, eight monoexonic circRNAs were also identified. Most of the transcripts consist of four exonic regions, followed by the category with three exonic regions. The presence of all seven exonic regions is not observed. Internal exons are mostly present in the identified circRNAs in comparison to the terminal exons, with exon 4 being the most frequently present, followed by exons 6 and 3. Furthermore, exon 5 that is only present in a non-coding linear transcript of the gene, is only identified in one circRNA, circ-BAX-67b.

circ-BAX-107 and circ-BAX-88 derive from the most upstream region of BAX gene, starting at positions 7 and 15, respectively. The circRNAs circ-BAX-110 and circ-BAX-34b appear to derive from the most downstream region of BAX gene, at the relative positions 6874 and 6875, respectively. Additionally, the length of the identified circRNAs is significantly variable from circ-BAX-23 consisting of 175 nucleotides to circ-BAX-30 consisting of 795 nucleotides.

Validation of the identified circRNAs using next-generation sequencing

Sequences of the identified circRNAs were validated using a targeted next-generation sequencing approach. Back-splice sequences were determined, and the expression analysis in the eleven cancerous and the non-cancerous cell lines was performed (Table 1). It is important to note a small sequence similarity in the two regions of some circRNAs contributing to the back-splice site, meaning that the back-splice sequence could derive from both the first and the last exon of the transcript. Eight circRNAs (circ-BAX-55b, circ-BAX-67b, circ-BAX-101, circ-BAX-108, circ-BAX-109a, circ-BAX-127, circ-BAX-132, circ-BAX-133) were identified in a single cell line. The majority of the circRNAs were identified in less than seven cell lines, whereas nineteen circRNAs were identified in eight or more cell lines. The circ-BAX-48a appeared expressed in eleven cell lines and being absent in the MDA-MB-468 cell line.

Fig. 1. Genomic locus of BAX gene and associated transcripts. The condensed form of linear transcripts is depicted in blue; circular transcripts (circRNAs) are also depicted in condensed form (red). Novel circRNAs that were firstly identified in this study are depicted in light red color, whereas circRNAs colored in orange were previously identified in hematological malignancies, in our lab. Single-exonic circRNAs are illustrated in green

Table 1. Expression analysis^a^ of BAX circRNAs in the 12 cell lines having derived from breast tissuecircRNAGenBank™ accession #MCF-7T-47DZR-75-1BT-474SK-BR-3BT-20HCC70Hs 578TMDA-MB-231MDA-MB-453MDA-MB-468MCF-12A^b^Luminal ALuminal BHER2-positiveTriple- negativecirc-BAX-6aOQ139913++−−+++++−−+2014circ-BAX-6bOQ348784+−++++++−+−−2114circ-BAX-6cOQ348785−+−−++++++++1016circ-BAX-7aOQ348786+−−+++++++−−1115circ-BAX-7bOQ348787++−+++−+++−+2114circ-BAX-8aOQ139914−−+−−+−−−−+−1002circ-BAX-8cOQ260040−−−−++++−+−+0014circ-BAX-8dOQ139921−+−−−+++++−−1005circ-BAX-8fOQ348833−−+−−+++−+++1005circ-BAX-10OQ139916−−−−−+−+−−−+0002circ-BAX-13aOQ348788+−−++++++−−+1114circ-BAX-13bPV654371++−−+++−+−−+2013circ-BAX-14OQ348789+−+−−+++−−−+2003circ-BAX-15bOQ348790−−−−++−−−−−−0011circ-BAX-15cOQ348791−+−−−+++++−−1005circ-BAX-16OQ348792−−−−++−−−−−−0011circ-BAX-17OQ348793−−−+−+−−+−−−0102circ-BAX-18OQ348794++−+−+++++++2106circ-BAX-19OQ348795−+++−+−++−−−2103circ-BAX-20OQ348796−+++−+++−−++2104circ-BAX-21OQ348797−−−−−−−−−++−0002circ-BAX-22OQ348798−−−+−+−+−−−−0102circ-BAX-23OQ348799−−+−−++++++−1006circ-BAX-25OQ348801−+−+−+−+−−−−1102circ-BAX-26OQ348802−−+−−++++−−−1004circ-BAX-28OQ348803+−+−−+−−+++−2004circ-BAX-29OQ348804−−−+−++−−−−−0102circ-BAX-30OQ348805−+−+−+−+++−−1104circ-BAX-33OQ260045−−+−++−−−−+−1012circ-BAX-34bPV654360−+−−+++−++−+1014circ-BAX-36OQ348800+−−+−++−−−−−1102circ-BAX-37aPV654349+−−−+++++−−−1014circ-BAX-37bOQ348829−−−+−+++−−−+0103circ-BAX-40OQ260049−+−++++−+−−−1113circ-BAX-41PV654341−−−+−+−−−−−+0101circ-BAX-44aOQ260052−−−−−++−−+−−0003circ-BAX-45OQ260054−−−−−++−−+−−0003circ-BAX-47bOQ348806−−+−−++++−+−1005circ-BAX-48aOQ260056++++++++++−+3115circ-BAX-52PV654363+−−+++−−−−−−1111circ-BAX-55aOQ260060++−++++++−−+2114circ-BAX-55bPV654351−−−−−−−−−−−+0000circ-BAX-55cPV654366+++++++++−−+3114circ-BAX-56aOQ260061+−−−−++−−+++1004circ-BAX-56bPV654350−−−−−+−−−+++0003circ-BAX-57OQ260062−−−−−+−−−−−+0001circ-BAX-58aOQ260063−+−+++++++−+1115circ-BAX-58bOQ348807+−−+−+−++−−−1103circ-BAX-59OQ260064−+−−−++−−+−+1003circ-BAX-60aOQ348808++++−++−++−+3104circ-BAX-60bPV654355+−−+−+++−−−−1103circ-BAX-61OQ348809−−−+−+−−+−−−0102circ-BAX-65OQ348810++−++++++−−+2114circ-BAX-67bOQ260070−+−−−−−−−−−−1000circ-BAX-69OQ348811+−+−−+−−−−+−2002circ-BAX-71OQ348812−−−+++−−−−−−0111circ-BAX-72OQ348813−−−+−−+−−−−−0101circ-BAX-73OQ348814+−−+−−−++−−−1102circ-BAX-74OQ139927++−−++++++++2016circ-BAX-75OQ348815−−−++++−++−+0114circ-BAX-77OQ348816−−+−++−−−−+−1012circ-BAX-79OQ348817−−−−−+−+−−−−0002circ-BAX-80OQ348818−−+−++−+−−+−1013circ-BAX-81OQ348819+−−+−++++++−1106circ-BAX-83aOQ348821−−++−+−−−−+−1102circ-BAX-83bOQ348822−−−−++−−+−−−0012circ-BAX-84OQ348823−−++−++−−+−+1103circ-BAX-85OQ348824−−−−−++−+−−+0003circ-BAX-86OQ348825−−−+−++−−−−−0102circ-BAX-87OQ348826−−++−++−−−−−1102circ-BAX-88OQ348827−−+−−+++−−+−1004circ-BAX-89OQ348828−−−+−++−−−−−0102circ-BAX-90OQ348830−−−−−+−+−−−−0002circ-BAX-91OQ348831−−+−−+−+−−+−1003circ-BAX-92OQ348832−−++−+−+−−+−1103circ-BAX-95OQ348834−−−+−+++−−−+0103circ-BAX-96OQ348835+−−−−+−−−−−+1001circ-BAX-97aOQ348836−−−+−+−−−−−+0101circ-BAX-98OQ348837−−−+−+−++−−+0103circ-BAX-101PV654339−−−−−−−−+−−−0001circ-BAX-102PV654354−+−−−+−−+−−−1002circ-BAX-103PV654365−−+−++−+−−+−1013circ-BAX-104PV654369−−+−−+−+−−−−1002circ-BAX-105PV654367++−+−+++−−−+2103circ-BAX-106PV654342−−++−++++−+−1105circ-BAX-107PV654338−−++−+−+−+−+1103circ-BAX-108PV654368−−−−+−−−−−−−0010circ-BAX-109aPV654348−−−−−−+−−−−−0001circ-BAX-109bPV654370−−−−−−++−−−−0002circ-BAX-110PV654347−++−−+−+−−+−2003circ-BAX-111PV654362−+++−++++−++2105circ-BAX-112PV654337−−+++−−+−−−+1111circ-BAX-114PV654391+++−−+++++−+3005circ-BAX-117PV654374−−−−−++−−−−−0002circ-BAX-118PV654375−−−−−++−−−−−0002circ-BAX-120PV654358−−−−−++−−−−−0002circ-BAX-122PV654343−+−−−−+−−+−−1002circ-BAX-123aPV654383++−−++−+++++2015circ-BAX-124PV654376−−−−−−−−++−−0002circ-BAX-127PV654356−−−−−−+−−−−−0001circ-BAX-129PV654372−−−−−++−−−−−0002circ-BAX-132PV654352−−−−−+−−−−−−0001circ-BAX-133PV654345−−−−+−−−−−−−0010circ-BAX-135aPV654384−−−−++−−−−−+0011circ-BAX-137PV654361−−+−+−−−−+−−1011circ-BAX-138PV654364−−−+−+−−−−−+0101^a^The numbers indicate the number of cell lines in which each circRNA was detected^b^Non-cancerous breast cell line

The circ-BAX-55b was identified only in the non-cancerous cell line MCF-12A. Ninety-one circRNAs appear expressed in the BT-20 cell line, which corresponds to the triple negative subtype and a slightly higher number of circRNAs, in comparison to other cell lines was identified in the HCC70 and Hs 578T triple negative cell lines. Taking into consideration the number of cell lines that were studied and the relative molecular subtypes, a higher number of circRNAs were identified in the triple negative and luminal B molecular subtypes. Furthermore, circ-BAX-67b was identified only in luminal A subtype, circ-BAX-21, circ-BAX-44a, circ-BAX-45, circ-BAX-79, circ-BAX-90, circ-BAX-101, circ-BAX-109a, circ-BAX-109b, circ-BAX-117, circ-BAX-118, circ-BAX-120, circ-BAX-124, circ-BAX-127, circ-BAX-129 and circ-BAX-132 only in the triple negative subtype, and circ-BAX-108 and circ-BAX-133 only in the HER2 positive subtype.

Part of the reads that were acquired included circRNA structures with minor differences in the back-splice junction. These differences refer to a small number of nucleotides being different. The high number of reads acquired with these slight differences lead to the recognition of isoforms, where two or more circRNAs have the same structure except for the back-splice junction. Thirty-six isoforms were identified, corresponding to twenty circRNAs with similar structure, and a nomenclature with different letters was implemented. Lastly, most isoforms (four) that were identified in BC belonged to circ-BAX-8.

Analysis of the miRNA sponging activity of BAX circRNAs and predicted regulation on signaling pathways

The miRNA sponging activity of the circRNAs was assessed using the miRDB custom prediction tool (Table S2). Seventy-six of the identified circRNAs are predicted to sponge miR-152-5p and twenty-two are predicted to sponge miR-4802-5p, imposing a significant role of these two miRNAs in BC microenvironment. Other miRNAs, including miR-211-3p, miR-491-5p, and miR-651-3p appear to be sponged by fewer circRNAs, such as circ-BAX-30 the first two and circ-BAX-52 the last, indicating a more specific regulatory potential of these circRNAs.

For the miRNAs with a high binding prediction score (≥ 80) to the circRNAs, target prediction analysis was performed, with a focus on genes contributing to cell signaling pathways. Specific genes that are known to contribute to the MAPK, PI3K/AKT, or NFκB pathways were incorporated in this analysis. Predicted regulatory axes were constructed, including circRNAs, miRNAs, and signaling pathways (Fig. 2 and Table S3). As a result, the majority of the identified circRNAs by sponging miR-152-5p are predicted to affect the MAPK and the PI3K/AKT cell signaling pathways. The second most sponged miRNA, miR-4802-5p, is predicted to regulate only the PI3K/AKT pathway.

Fig. 2. Functional interplay between identified circRNAs and particular signaling pathways. This predicted ceRNA regulatory map illustrates circRNAs (green boxes) sequestering miRNAs (pink boxes), thereby preventing the downregulation of target transcripts that contribute to specific signaling pathways (blue boxes). Solid orange lines represent validated interactions between linear BAX transcripts and the respective miRNA. Dashed orange lines signify the validated indirect regulation of signaling pathways by miRNAs via the modulation of pathway-contributing mRNAs

Investigation of RBPs sponging activity of the identified circRNAs, and investigation of multiple RNA interactions

The RBPs sponging activity of circRNAs was assessed using the RBP map tool (Supplementary ".txt" file and Description of Supplementary file). Two RBPs, HNRNPF, RBM6 appear to be sponged by all the circRNAs, whereas others such as YBX2 and ZNF638 appear to be sponged by a single circRNA circ-BAX-59. This observation further clarifies the broad or a more specific regulatory potency of the identified transcripts.

Aiming to identify axes of RNA interactome the miRNAs and corresponding RBPs that are predicted to be antagonistically sponged by each circRNA were assessed (Table S4). This clarified the potency of 27 circRNAs to antagonistically sponge miR-152-5p and two RBPs, NOVA1 and RBM6. Numerous other antagonistic relations were observed, with miR-651-3p and miR-5590-5p antagonizing with more RBPs for the relative positions on the identified circRNAs, in comparison to other miRNAs.

Exploration of potential modifications in the identified circRNAs and assessment of their translational potency

Potential presence of significant modifications on circRNAs was assessed (Fig. 3 and Table S5). Predicted IRES sites were identified in circ-BAX-30, circ-BAX-65, circ-BAX-67b, circ-BAX-71, circ-BAX-72, circ-BAX-73, circ-BAX-88, circ-BAX-91, circ-BAX-107. Regarding m^6^A modification, 18 circRNAs are predicted to be modified, including isoforms of circ-BAX-6, circ-BAX-55, and circ-BAX-58.

The open reading frame (ORF) of circRNAs with at least one of the abovementioned characteristics was assessed using the ORF Finder tool. The presence of at least one of the aforementioned characteristics is essential for a circRNA to be potentially translated. Twenty-seven of the identified circRNAs with the abovementioned characteristics appear to possess at least one ORF. Aiming to identify similarities with BAX proteins deriving from linear transcripts, an alignment of circRNAs to the Coding Sequence (CDS) of coding transcripts was performed. Ten transcripts align to the CDS of NM_001291428 from which BAX isoform 1 is translated (Fig. 4A). Furthermore, circ-BAX-6c, circ-BAX-58a, and circ-BAX-58b appear to possess the same ORF as the linear transcript from which the BAX isoform alpha is translated, and the respective proteins are predicted to have the same length and include the BH1, BH2, and BH3 domains. Even though a shorter polypeptide is predicted to be translated from circ-BAX-65, it appears to have the same ORF starting position and include BH1 and BH3 domains. These observations seem significant, as they raise a higher probability for the translation of these transcripts. Moreover, other circRNAs align to the CDS of NM_138764 from which BAX isoform sigma is translated (Fig. 4B). The circ-BAX-91 and circ-BAX-107 also possess the same ORF starting position as the linear transcript, producing however a shorter polypeptide.

Fig. 3. Structural characterization and coding potential of identified BAX circRNAs. Transcripts features including predicted m^6^A motifs, Internal Ribosome Entry Sites (IRES), and Open Reading Frames (ORFs) of the BAX circRNAs comprising an ORF are illustrated. The presence of m^6^A motifs and IRES elements indicates potential cap-independent translation initiation sites, while the identified ORFs highlight the coding capacity of these circular isoforms relative to their linear parental genes

Fig. 4. Alignment of particular circRNAs comprising an open reading frame (ORF) against major BAX mRNAs. Transcripts features including predicted m^6^A motifs, Internal Ribosome Entry Sites (IRES), and Open Reading Frames (ORFs) are illustrated. Only BAX circRNAs comprising an ORF and being aligned against linear BAX transcripts with GenBank IDs NM_001291428 (A) and NM_138764 (B) are shown

Discussion

Aiming to massively identify and validate a wide repertoire of circRNAs deriving from BAX gene, a known pro-apoptotic effector, we used an integrative sequencing approach that combined third-generation long-read nanopore sequencing with high accuracy short-read Illumina sequencing. The long reads allowed to resolve full-length transcript structures, including back-splice junctions, and the high accuracy of short-read data deriving from next-generation sequencing was used to verify sequence fidelity and expression across cell lines. Therefore, the complementary application of multiple platforms enhances the validity of these novel circRNAs.

In total, 106 BAX-derived circRNAs were identified, with 82 being novel and identified for the first time in BC. The remaining circRNAs have been identified in hematological malignancies and for the first time in BC. The structural diversity of cirRNAs, which includes exon skipping, intron retention, exon extension/truncation, and monoexonic forms, indicates a significant alternative splicing machinery acting on BAX transcripts.

Interestingly, internal exons are more frequently present in the identified circRNAs, with exon 4 being identified more frequently. This observation is extremely significant as the BH1-3 domains which are essential for the function of BAX appear in the exons 3, 4, and 6, further highlighting the potential significance of these circRNAs. Only one circRNA circ-BAX-67b appeared to include exon 5, which is only present in the non-coding variant epsilon of linear BAX. This highlights the rarity of its presence and may indicate a unique splicing event or a specific functional role that warrants further examination.

Significant variations in circRNA quantity and variety were found by expression profiling of eleven BC cell lines that represented several molecular subtypes (luminal A, luminal B, HER2+, and triple-negative) as well as one non-cancerous line, MCF-12A. Numerous transcripts showed subtype-specific expression, indicating that circRNA synthesis is not stochastic but might be impacted by splicing machinery or transcriptional programs particular to a subtype. This finding is consistent with accumulating evidence that circRNA expression in BC can be subtype-specific, potentially contributing subtype’s distinct molecular characteristics (Nair et al. 2016). For instance, the diversity of BAX circRNAs was higher in the triple-negative subtype, which is frequently characterized by higher genomic instability and unregulated splicing (Derakhshan and Reis-Filho 2022). This is in line with recent research showing that aggressive BC subtypes generate more circRNAs, either as an adaptive regulatory mechanism or as a result of unregulated splicing (He et al. 2021). The non-cancerous MCF-12A cell line was the only one to express circ-BAX-55b, suggesting that it may have a tumor-suppressive role. This finding suggests this circRNA as a candidate molecule to assess its biomarker utility for BC screening.

The identification of circRNA isoforms that varied within their back-splice junctions added another degree of complexity. circ-BAX-8a, circ-BAX-8c, circ-BAX-8d, and circ-BAX-8f are examples of these isoforms, which may have minor but significant impacts on interactome dynamics, stability, and localization. Minor variations in the back-splice site may change the secondary structures or binding affinities of RBPs and miRNAs, changing their regulatory effect in the process. Such micro-heterogeneity is consistent with the more general understanding that circRNAs are dynamic molecules that exist in a variety of isoforms with possibly distinct roles.

Our in silico functional predictions suggest that BAX-derived circRNAs play a significant role in miRNA binding, sponging mostly miR-152-5p and miR-4802-5p. For example, it has been demonstrated that the tumor-suppressive miR-152-5p specifically targets linear BAX mRNA and is known to regulate important oncogenes (Gillen et al. 2016). miR-152-5p is a known tumor suppressor with reported effects on apoptosis, proliferation, and drug sensitivity in various cancers, including BC (Chen et al. 2017; Pang et al. 2019). The broad projected association between miR-152-5p and BAX circRNAs suggests a competitive endogenous RNA (ceRNA) network in which circRNAs may sequester this miRNA, thus affecting apoptotic susceptibility and indirectly modifying BAX protein levels. Thus, circRNAs may be able to fine-tune important signaling pathways and relative cellular processes in BC.

The projected regulatory axes, which connect circRNAs to miRNAs and their downstream mRNA targets in the MAPK, PI3K/AKT, and NFκB pathways, emphasize how important these circRNAs are in regulating the pathways that control inflammation, cell survival, and proliferation. Notably, miR-4802-5p sponging seems to have a special effect on the PI3K/AKT pathway, a major contributor to oncogenic signaling and treatment resistance. These results pave the way for functional research to confirm whether targeting particular circRNAs changes treatment response or pathway activity, particularly in resistant BC subtypes.

Numerous circRNAs have promising features that may lead to their translation in addition to their non-coding functions. The idea that some BAX circRNAs may produce short peptides is supported by the presence of recognizable ORFs and internal ribosome entry sites (IRES) and/or m^6^A RNA modifications, either of which appear essential for cap-independent translation (Qin et al. 2022; Wen et al. 2022; Pisignano et al. 2023). It is noteworthy that several circRNAs share ORFs with linear BAX transcripts, even though encoding shorter versions. Furthermore, circ-BAX-6c, circ-BAX-58a, and circ-BAX-58b are predicted to include the BH1-BH3 domains, which are crucial for BAX function. This is a particularly intriguing finding, as circRNA-derived peptides might retain, modulate, or even antagonize the apoptotic function of BAX protein isoforms deriving from mRNAs, offering a novel layer of translational regulation which should be further investigated.

Potential interactions of circRNAs with RBPs were also investigated, and several were shown to bind across the majority of circRNAs, including HNRNPF and RBM6, that bind antagonistically with miR-152-5p in specific circRNAs. For example, RBM6 protein appears significantly lost in metastatic breast tumors, and the identified circRNAs may contribute to this phenomenon (Machour et al. 2021). This implies a widespread involvement of the identified novel circRNAs in RNA interactome.

The identified BAX-derived circRNAs represent attractive candidates for translational research due to their stability, variety, and regulatory potential. Their distinct expression profiles across various cell lines and subtypes highlight their potential as biomarkers, while their involvement in key signaling pathways suggests significant therapeutic relevance. A crucial next step is the experimental validation of the identified circRNAs using RT-qPCR across panels of cell lines and subsequently in clinical specimens by combining RNase R treatment, aiming to eliminate linear counterparts, and enrich circRNA abundance. Furthermore, assays including the functional knockdown and dual-luciferase reporter assays will reveal the regulatory potential of the identified circRNAs and will allow the validation of promising circRNA-miRNA-mRNA axes.

Beyond their regulatory role, large-scale studies using BC tissue, peripheral blood, and circulating tumor cells may unravel the clinical utility of these circRNAs in patient stratification and disease progression. The circular structure of these transcripts renders them resistant to exonuclease degradation, making them more robust and reliable candidates for non-invasive diagnostics than their linear counterparts. To establish their screening and prognostic value, future studies must correlate circRNA expression levels with clinical outcomes of patients, such as overall survival and disease-free survival. Other statistical analyses, including the Mann-Whitney U test and the receiver operating characteristic curve analysis would allow to discriminate between conditions and determine the sensitivity and specificity of potential biomarkers.

Lastly, the BAX-derived circRNAs could potentially inhibit the parental pro-apoptotic BAX gene or enhance the stability and translation of linear BAX transcripts by sponging miRNAs, therefore potentially acquiring oncogenic or anti-oncogenic characteristics. Nevertheless, the translational research of these novel circRNAs may reveal previously unknown aspects of BAX, imparting a unique significance to this gene.

Conclusion

Our study provides a detailed characterization of BAX-derived circRNAs in BC, identifying a diverse repertoire of transcripts with unique structural identity and set of regulatory features. These circRNAs may modulate apoptosis, influence treatment response, and function as molecular sponges or translational units, highlighting their potential impact on BC biology. Their distinct expression across cell lines and subtypes suggest their potential utility as biomarkers for disease stratification. Furthermore, their predicted involvement in key signaling pathways points to their ability to regulate important cellular processes. Future studies should focus on functional validation of highly expressed and subtype-specific circRNAs, investigation of circRNA–miRNA–mRNA regulatory axes, and exploration of their clinical utility in patient samples. Overall, BAX-derived circRNAs appear as promising molecules for mechanistic studies and clinical applications.

Supplementary Information

Below is the link to the electronic supplementary material.

Supplementary Material 1

Supplementary Material 2

Supplementary Material 3

Supplementary Material 4

Supplementary Material 5

Supplementary Material 6

Supplementary Material 7

Supplementary Material 8

Supplementary Material 9