Developmental Regulation of circRNAs in Normal and Diseased Mammary Gland: A Focus on circRNA-miRNA Networks

Abstract

Circular RNAs (circRNAs) have emerged as critical regulators in various biological processes including diseases. In the mammary gland (MG), which undergoes most of its development postnatally, circRNAs play pivotal roles in both physiological and pathological contexts. This review highlights the involvement of circRNAs during key developmental stages of the MG, with particular emphasis on lactation, where circRNA-miRNA networks significantly influence milk secretion and composition. CircRNAs exhibit stage-, breed- and species-specific expression patterns during lactation, which underscores their complexity. This intricate regulation also plays a significant role in pathological conditions of the MG, where dysregulated circRNA expression contributes to disease progression such as mastitis, early breast cancer (BC) stages, and epithelial-to-mesenchymal transition in BC (EMT). In mastitis, altered circRNA expression disrupts immune responses and compromises epithelial integrity. During early BC progression, circRNAs drive cell proliferation, while in EMT, they facilitate metastatic processes. By focusing on the circRNA-miRNA interactions underlying these processes, this review highlights their potential use as biomarkers for MG development, disease progression, and as therapeutic targets.

Introduction

Epigenetics represents the mechanisms by which gene expression is altered without a change in underlying DNA sequence. Five molecular mechanisms have been described in the epigenetic regulation namely DNA methylation, histone post-translational modification, chromatin remodeling, RNA modification, and the regulation of non-coding RNAs (ncRNA) [1]. One type of RNAs is circular RNAs (circRNAs) which are formed during the pre-mRNA back-splicing to normally remove the introns and generate circular RNAs [2]. CircRNAs can carry out several biological functions such as sponging microRNAs (miRNAs), sponging RNA binding proteins (RBPs), regulating transcription and translation, and acting as protein scaffolds or decoys [3]. While circRNAs were originally categorized as ncRNA, recent research has shown that some circRNAs can act as templates for protein synthesis, challenging the previous classification [4, 5]. Nonetheless, their broader function remains regulatory rather than coding.

CircRNAs are implicated in the developmental regulation of the mammary gland (MG) and in its dysregulation during disease states. By maintaining mRNA expression levels, circRNAs have been shown to be involved in MG development in multiple species of livestock and rats [6–8]. Most of this review is dedicated towards compiling the majority of published work on circRNAs in different stages of MG development from pregnancy to lactation and involution. No studies exist on the role of circRNAs in the MG during embryonic or pubertal stages, to our knowledge. As for pregnancy and involution, we were only able to find the very few studies discussed below. Through our literature search, we find that most studies investigate the expression of circRNAs during lactation and are less focused on their mechanistic role. The miRNA targets of each circRNA are often predicted but not investigated further in terms of their differential expression patterns. Nonetheless, several reports propose circRNA-miRNA networks that may be involved in MG developmental pathways [9, 10]. Few studies have moved further and investigated not only the expression of target miRNAs, but also the respective mRNA targets of each miRNA to assemble circRNA/miRNA/mRNA axis [6, 11]. For the unreported networks lacking the expression of target miRNA and consequent mRNA targets, the completion of the axes would require these two elements: expression of target miRNA and possible mRNA targets of each miRNA. Seeing as those two elements were missing in several of these studies, we searched for each identified miRNA to investigate its expression and its downstream mRNA targets. We performed this search in data published by Wu et al. The authors performed miRNA sequencing and mRNA sequencing of bovine tissue at peak lactation relative to dry period [12]. They conducted differential analyses and aligned miRNAs with their target mRNAs. As such, combining the data of circRNA/target miRNAs with the matching miRNA/mRNA data allows to complete the circRNA/miRNA/mRNA axes to better understand the signaling pathways that these circRNAs influence.

The need to understand the roles of circRNAs in the regulation of the development of the MG is also manifested in their implication in disease progression. As circRNAs are part of key pathways involved in metabolic activity and immune response, they are also involved in inflammation of the breast during mastitis. This is especially important considering that mastitis is very common, whereby it affects one in four breastfeeding women during the first 26 weeks postpartum [13]. It is also very common in livestock, where it is considered to be the most common disease leading to economic loss in dairy industries, due to loss of yield and milk quality [14]. Using animal models, literature shows the involvement of circRNAs in mastitis.

Furthermore, by regulating polarity, proliferation, milk proteins secretion and other processes, circRNAs can partake in breast cancer (BC) initiation and progression. BC has the highest incidence rate among cancers worldwide and ranks first for cancer related death causes in females [15]. A number of studies show circRNAs to be involved in the key events of BC [16–21]. In this review, we will focus on two relatively understudied phases of BC progression in the context of circRNAs, which are early cancer stages and epithelial-to-mesenchymal transition (EMT).

CircRNAs and their target miRNAs, which are released into circulation, can act as readout molecules that reflect the state of the MG, especially as the gland undergoes early dysregulation. Their potential use as liquid biomarkers carries great importance to enhance our understanding of normal developmental processes and improve both diagnosis and prognosis of disease [22, 23]. By allowing for early detection of gland dysregulation, it can give patients more treatment choices.

The aim of this review is, therefore, to elucidate the role of circRNAs in both physiological and pathological conditions of the MG, with a particular focus on circRNA-miRNA interactions. Normal physiological conditions include all states of the MG starting from embryonic development, and resuming at puberty to pregnancy, lactation, and involution. As stated above, most studies focus on lactational stages which is reflected in the content of this review. For pathological conditions, we narrowed our focus to three states: mastitis, early BC stages, and EMT. Providing an extensive overview of the involvement of circRNAs in these states contributes to the future application of circRNA/miRNA/mRNA axes as potential biomarkers.

Overview of MG Development

The MG is one of the few organs that continues to undergo developmental and cyclic changes well into adulthood. In fact, most of its growth occurs after birth. Its development is largely governed by cell–cell and cell-ECM interactions, as well as soluble mediators. A basement membrane (BM) basally underlies the mammary epithelium and consists of thick sheets of glycoproteins and proteoglycans with laminins and type IV collagen networks, supported by a stroma [24]. The stroma encompasses fibrous connective tissue proteins, as well as multiple types of cells such as fibroblasts, adipocytes, and innate immune cells. In this brief overview, only prominent events of MG development common to the species discussed in this manuscript are described. The studies referring to circRNAs in normal MG development as well as mastitis were largely in bovine, ovine, and rodent species. Meanwhile, the BC studies cited are mainly in human-derived samples or human breast cell culture models.

At birth, the MG consists of a rudimentary ductal system that emanates from the teat [25]. Prior to puberty, the epithelium grows isometrically as the rest of the body grows. At the onset of puberty, the endocrine and paracrine environment supplies growth hormone, estrogen, and insulin-like growth factor 1 (IGF1) which promote ductular morphogenesis. The ducts undergo expansive proliferation and grow into the fat pad to form a more ductular pronounced tree with secondary branches. At the ends of the growing ducts, club-shaped structures called terminal end buds are formed, which penetrate the fat pad [25]. These terminal end buds are described in humans and mice, meanwhile ruminants will have terminal duct units [26]. Terminal duct units resemble a multi-lobular terminal end bud, and it directs the growth of the mammary ducts in ruminants during puberty. Estrogen and progesterone play key roles in the cyclic regulation of the MG every estrous cycle (which shares similarities with menstrual cycle in humans). The highest proliferative rate in the estrous cycle is reported to be in the mid-luteal phase whereby both hormones are highly expressed [27].

Complete development of the MG will occur during pregnancy when tertiary branches form and alveologenesis, when milk-secreting acinar units are formed, occurs [26]. Looking closely at the terminal end buds, the leading cap cells on the bud give rise to myoepithelial cells, which in turn arrange as an outer layer that basally surrounds an inner layer of luminal epithelial cells [25, 28]. Under the control of progesterone and prolactin, each alveolar bud differentiates into a milk-secreting lobule surrounded with a network of capillaries in preparation of lactation. Polarized apical milk secretion by luminal cells occurs during lactation as milk is secreted into the lumen of the alveoli. Post-lactation, milk stasis triggers involution of the MG, due to the mechanical stretch of the alveoli resulting from milk build-up in the lumen or due to build-up of secreted factors in the milk [29]. Involution in the MG occurs over two phases. The first phase occurs upon weaning and is marked by the presence of dying cells in the alveoli lumen and neutrophil infiltration. The second phase is characterized by another wave of cell death, accompanied by ECM degradation and remodeling, as well as controlled influx of macrophages and other immune cells [29]. Cell death during involution is largely governed by a lysosomal-mediated programmed cell death pathway mainly triggered by the transcription factor STAT3 [30]. After involution, most of the alveoli have regressed and the remodeled MG morphologically resembles the virgin gland [25]. The adipocytes compartment will undergo extensive regeneration and fill the space from the regressed epithelium [31].

The dynamic remodeling of the gland is achieved through several mechanisms including regulation by ncRNAs. This type of regulation is an essential modulator as recently published studies have shown.

CircRNAs Biogenesis and Functions

NcRNAs primarily function as regulators of gene expression whether at transcriptional or post-transcriptional stages. They include microRNA (miRNA), circular RNA (circRNA), piwi-interacting RNA (piRNA), long ncRNA (lncRNA), among others. CircRNAs are single-stranded covalently closed loops that lack the modifications found in linear mRNA, namely the 5’ cap and the 3’ poly-A tail. Their covalently closed loop structure offers them higher stability by avoiding exonuclease activity. They are diverse in terms of size and constituents between exons and introns. Most circRNAs are derived from protein coding genes ranging from exonic, exon–intron, and intronic circular RNA. They have also been reported to originate from non-coding, antisense, 3’ or 5’ untranslated region (UTR), or intergenic genomic regions [32]. Interestingly, the composition of circRNAs between exons and introns has been related to their localization and function. Exon–intron circRNAs have been found to mainly localize in the nucleus and promote the transcription of their parental genes [33]. This effect is mediated by RNA-RNA binding between the exon–intron circRNAs and U1 snRNAs which then allows the circRNA to interact with RNA polymerase II. Similar roles have been described for intronic circRNAs where they mainly accumulate in the nucleus and function to regulate transcription [34]. In contrast, exonic circRNAs are predominantly exported to the nucleus and are thought to regulate gene expression post-transcriptionally [34, 35]. The other types of circRNAs such as those originating from UTRs or intergenic regions have been less studied to be able to detect patterns in localization and/or function. A study by Long et al. reported the detection of an intergenic circRNA circ_0007379 and found it to be downregulated in colorectal cancer [36]. Circ_0007379 was found to act as a scaffold for two primary miRNA transcripts and modulate their maturation in KSRP-dependent manner. This is an interesting finding that reveals a new mechanism by which circRNAs can influence the expression of miRNAs due to base complementarity. It warrants further investigation into other types of circRNAs that may be exerting similar effects.

CircRNAs are post-transcriptionally generated from pre-mRNA splicing [32]. Their splicing is distinct from the canonical splicing mechanism that removes introns and assembles exons to form the linear RNA transcript with 5’−3’ polarity. Their generation utilizes the canonical splice sites but undergoes alternative splicing mechanism termed back-splicing i.e. head-to-tail splicing. The splicing of circRNAs has been shown to be dependent on the spliceosome machinery or catalyzed by group I and II ribozymes [37]. Consistent with suggested role of canonical splicing machinery in circRNA biogenesis, inhibition of the spliceosome by isokinetic (a general inhibitor of pre-mRNA splicing [38]) depleted both circRNA and linear mRNA levels [39]. In contrast, depletion of spliceosome protein component U2 snRNP using double-stranded RNA in Drosophila cell line model, increased the ratio of circRNAs to linear RNA levels [40]. These combined findings suggest that while the spliceosome machinery is part of the generation of circRNAs, it is possible that its depletion redirects pre-mRNA towards other pathways that favor circRNAs. Studies have also shown a role for cis-acting elements and trans-acting factors in controlling circRNA formation [32, 41–43]. Several models for circRNA biogenesis have been reported, with the most common type of model being the exon skipping or the lariat-driven circularization model. In this model, a large lariat consisting of one or more exons is synthesized before the introns are successively cleaved out, which produces exonic or exon-intronic circRNAs [32]. Another suggested model is that of intron pairing-driven circularization. Specific reverse complementary sequences within introns, which are often located near the splice donor and acceptor sites, pair with each other. The introns pairing brings into proximity the donor sites which aid in the back-splicing. Other models have been suggested for the biogenesis of circRNAs [43]. Interestingly, cancer can promote the biogenesis of novel circRNA classes, like the read-through circRNAs (rt-circRNAs) and fusion circRNAs (f-circRNAs) [44]. Cancer progression is associated with read-through circularization whereby transcription continues beyond the termination signal and into the intergenic region, allowing for the formation of circRNAs from two adjacent genes. Meanwhile, f-circRNAs are formed due to the juxtaposition of two regions of typically separate genes that favor back-splicing [44]. This is consistent with the aberrant chromosome rearrangement seen in cancer. As such, the circular structure of circRNAs and their pre-mRNA origin accordingly shape the research methods used to study them. Considering that they represent a small fraction of total RNA and have a convalently closed loop, RNAse R is commonly used to degrade linear mRNA and subsequently enrich the sample with circRNAs [45]. This can be coupled with RNA-seq or qPCR detection methods. Manipulation of circRNAs expression in model systems also requires additional consideration. Their inhibition, for instance, often requires targeting the junctional region to avoid unintended inhibition of their parental genes. The field has been moving forward to standardize and enhance research practices to study circRNAs. This methodology has been summarized and updated in several published review articles [46, 47]

Many functions have been described for circRNAs. They can act as miRNA sponges, bind to RBPs and RNA polymerase, or get translated into proteins [34, 48]. One circRNA can contain one or multiple miRNA binding sites for one or multiple miRNAs. Often, miRNA binding sites contain mismatches at intermediate positions to evade the endoribonuclease activity of argonaute 2 (Ago2) [49]. circRNAs can also target miRNA clusters with a common seed region, blocking their activity. This sponging of miRNAs would lead to a higher expression of downstream mRNA targets of miRNAs that are then relieved of inhibition. In addition to binding to miRNAs, circRNAs can interact with proteins, most commonly with RBPs. They have specific binding sites for RBPs, which is likely to have unique effects distinct from typical RNA-RBP interaction due to recent evidence on the role of the tertiary structure of the RNA in its influence on RBPs [50]. CircRNA-protein binding can have a multitude of effects on cellular processes as it can affect the protein expression and function and simultaneously influence circRNA synthesis and stability. Other functions of circRNAs that directly alter transcription include binding to RNA polymerase I and II. A smaller population of circRNAs, primarily reported to be intronic and exon-intronic circRNAs [51], remains in the nucleus, possibly performing several functions including binding to polymerases. This binding can alter the expression of genes at a pre-transcriptional level and has been reported to promote the transcription of their own parental genes [33]. Although circRNAs originally emerged as a regulatory type of RNA, recent reports suggest that few of them can also be translated into proteins. As they lack the 5’ cap, their translation adopts cap-independent mechanism such as that of m6A-mediated translation, or if they contain an internal ribosome entry site (IRES) that can recruit ribosomes [2, 34]. The two approaches can also be coupled with each other to increase efficiency of translation. Interestingly, circRNAs that contain an infinite open reading frame (ORF) can undergo a rolling circle amplification which can lead to a much higher yield than their linear counterparts [52].

Regulation of circRNAs Expression During MG Development

Very little is known about the involvement of circRNAs in the regular development of the MG, and their regulation as it progresses through the different stages of puberty, pregnancy, lactation, and involution. The studies are grouped in the below section according to stage of development, species of the model used, and type of study whether focused on expression or function. Some of these studies fully investigated circRNAs with their target miRNAs and subsequent mRNA targets. Other studies only identified the circRNA-miRNA networks but did not investigate further. As described in the introduction, Wu et al. conducted a high throughput miRNA and mRNA sequencing study of the MG tissue of bovine origin at peak lactation (day 60 of lactation) and late-lactation (day 315 of lactation), as explained by the authors in the sample preparation section [12]. The data was made available with a list of the miRNAs and mRNAs detected and their expression level and then aligned together as miRNA and their target mRNAs. Most of the studies cited below reported expression of certain circRNAs, and predicted their miRNA targets but did not look into the expression of these miRNAs. We undertook the effort ourselves and investigated the expression of the miRNAs referring to the data from Wu et al.’s study and aligned the combined findings in Table 1.

Table 1.

Summary of circRNA-miRNA networks involved in different stages of normal mammary gland development. The information about circRNAs was obtained from their respective reference paper as well as their predicted target miRNAs. Target miRNAs were mined for in the supplementary data of the study by Wu et al. which contains analyzed microRNA and mRNA sequencing data of bovine tissue samples obtained during lactation in comparison to non-lactating samples [12]. Some miRNAs were matched to their target mRNA. The pattern of regulation of each miRNA was cited, and the target mRNAs were included if they were matched. Some miRNA and mRNA information were alternatively obtained from the same paper of the circRNA when it was available. Bold cells are used to denote the axes circRNA/miRNA/mRNA that were fully reported in the papers. The rest, marked in italic, were assembled using data from Wu. Et al. with separate data for the pattern of miRNAs and their target mRNAs. As for pattern of regulation of each circRNA, the reference stage was specified where peak lactation samples were compared to samples obtained during the dry period (DP) which ranged from 10 to 25 days post the cessation of lactation, late lactation (LL) which is day 250 of lactation, or non-lactation (NL) where is was specified as day 315 postpartum with no reported number for days of cessation of lactation. The rows are referred to as numbered from 1–19. (-) indicates that this information wasn’t available in the reference data

Model of study	Stage	circRNA	Pattern	Reference	Target miRNA	Pattern [12]	Target mRNA [12]	Pattern [12]
Bovine epithelial cell line	Pregnancy/ Lactation	circHIP3K	Up from No PRL	[53]	-	-	-	-
Bovine mammary tissue	Lactation	circRNA_08052	Up from NL	[10]	AC_000178.1_22528	Up	55 targets	34 Up, 21 Down
	Lactation	circRNA_08052	Up from NL	[10]	bta-miR-154b	Up	5 targets	4 Down, 1 Up
	Lactation	circRNA_08052	Up from NL	[10]	bta-miR-451	Up	-	-
	Lactation	circRNA_03706	Down from NL	[10]	bta-miR-6522	Down	-	-
	Lactation	circRNA_03706	Down from NL	[10]	bta-miR-2411-5p	Down	3 targets	2 Down, 1 Up
	Lactation	circRNA_02728	Up from NL	[10]	bta-miR-154b	Up	5 targets	4 Down, 1 Up
	Lactation	circRNA_02728	Up from NL	[10]	bta-miR-182	Up	LOC101907515	Down
	Lactation	circRNA_09048	Up from NL	[10]	bta-miR-370	Up	102 targets	64 Up, 38 Down
	Lactation	circRNA_09048	Up from NL	[10]	bta-miR-154b	Up	5 targets	4 Down, 1 Up
	Lactation	Casein circRNAs	Up From LL	[6]	bta-miR-2284 h-5p	Down	CSN1S1, CSN2, and the Ltf whey genes	-
	Lactation	Casein circRNAs	Up From LL	[6]	bta-miR-2284j	Down	CSN1S1, CSN2, and the Ltf whey genes	-
Ovine mammary tissue	Lactation	circ_022888	Down from DP	[9]	miR-493	Up	-	-
	Lactation	circ_022888	Down from DP	[9]	miR-10b	Down	-	-
	Lactation	circ_014489 circ_008952	Up from DP	[9]	miR-148a	Down	-	-
	Lactation	circ_008952	Up from DP	[9]	miR-362-3p	Down	-	-
	Lactation	circ_014489 circ_008952	Up from DP	[9]	miR-200a	Down	-	-
	Lactation	circ_015343	Down from DP	[54]	miR-200a	Down	-	-
Primary culture of goat mammary epithelial cells	Lactation model	circRNA-006258	-	[11]	miR-574-5p	Down	EVI5L	Up
Ovine mammary tissue	Lactation	circRNA-08436	Up from DP	[55]	miR-195	Down	ELOVL6	Up

Pregnancy: Bovine

Whereas no studies investigate the expression of circRNAs in the MG during pregnancy, one study offers a similar model to pregnancy using bovine mammary epithelial cells MAC-T treated with prolactin which is expressed during pregnancy and lactation. These cells differentiate when cultured on collagen gel in the presence of prolactin and secrete casein [56]. Using this model, the study by Liu et al. offers insight into the expression of some circRNAs, most prominently focusing on one circRNA circHIPK3, where its relevance to gestation as well as postpartum lactation is proposed [53]. After administering treatment of prolactin to bovine mammary epithelial cell line MAC-T, twenty circRNAs were significantly differentially regulated. CircHIPK3 was upregulated in prolactin-treated bovine as well as in mouse mammary epithelial cell HC11 (Table 1, row 1). HIPK3 (homeodomain interacting protein kinase 3), the parental gene of circHIPK3, enables protein serine/threonine kinase activity and is involved in mRNA transcription, apoptotic activity, and protein phosphorylation [57]. Investigating this circRNA using transient transfection in HC11 cells, it is reported that the prolactin treatment is possibly affecting the expression of alternative splicing factors through the STAT pathway, which alters circHIPK3 expression. CircHIPK3, in turn, promotes the proliferation of mammary epithelial cells. It is important to distinguish that prolactin upregulation is only one aspect of the change in microenvironment of mammary epithelial cells during pregnancy.

Lactation

Bovine

Liang et. al. showed that circRNAs are differentially regulated between early lactation (day 30 of lactation) and non-lactation period (day 315 postpartum, dry period length non-specified) in Holstein cows [10]. 68 circRNAs were upregulated and 19 were downregulated during lactation. As an indication of their function, Liang et al. performed an enrichment analysis on the circRNAs’ parental genes. Gene Ontology (GO) enrichment revealed that their functions included immune response, triglyceride transport, and T cell receptor signaling. Triglyceride transport is consistent with the need to keep up with increased secretion of proteins and lipids by the MG during lactation as part of milk secretion. Triglycerides are a constituent of the milk fat globules that are secreted apically [58]. Another purpose for triglyceride transport is to govern their uptake to the cells as they are used as a source for fatty acids [59]. KEGG (Kyoto Encyclopedia of Genes and Genomes) pathway analysis was performed and identified involvement in cytokine-cytokine receptor interactions, Th17 cell differentiation, fatty acid biosynthesis and the JAK-STAT signaling pathway. The JAK-STAT pathway leads to the expression of milk proteins such as WAP (Whey acidic protein) and β-casein [60]. Within enriched pathways and functions, there is a focus on immune response related terms. This falls into context in lactation as breast milk provides passive and active immunity to the infant by providing it with immunoglobulins, cytokines, lactoferrin, and even maternal leukocytes. These also serve to protect the lactating MG from infection. The miRNA targets were predicted relying on competitive endogenous RNA (ceRNA) scores. ceRNA scores are used to predict if a ncRNA and an mRNA compete for the same miRNA and therefore have a regulatory relationship. Four circRNAs were selected for qRT-PCR verification and the miRNA targets of four selected circRNAs were predicted through the miRanda algorithm. Multiple regulatory circRNA-miRNA networks were revealed having high ceRNA scores. In the case of the circRNAs, they compete with the target mRNA of the miRNA. So, when they sponge the miRNA, they are keeping it away from its target mRNA. However, the regulation between the circRNA and their target miRNAs were not elaborated on in the study. Evaluating the expression of these target miRNAs would provide information about this pathway and possibly suggest sponging effects when the expressions of circRNA and respective miRNA are reciprocal. It is also possible that they have parallel regulation or no differential regulation of the target miRNA which would suggest that other factors are affecting the network. We looked into the pattern of regulation of the miRNA targets for the four selected circRNAs that were predicted by Liang et al. The target miRNAs of the four circRNAs were almost all found to be differentially regulated during peak lactation as reported in the study by Wu et al. [12], which investigated differentially expressed miRNAs and their target mRNAs during peak lactation relative to late lactation. Interestingly however, each of the four circRNAs and their respective miRNAs targets were found to share the same trend, being both upregulated, or both downregulated. The results are assembled and summarized in Table 1 (rows 2–10). Some of these networks (For example, Table 1, rows 2–10) had parallel trends of expression between circRNA and its target miRNA. In such cases, the data suggests that the regulatory relationship is likely not the commonly described sponging mechanism [61], which would produce reciprocal expression. This, however, requires further research.

Zhang et al. investigated the abundance and composition of circRNAs in bovine MG tissue during early and late lactation periods [6]. Almost half of the total number of circRNAs were common between day 90 and day 250 of lactation showing the context-dependent function of circRNAs. The parental gene of these circRNAs were enriched for a wide variety of terms related to Golgi apparatus, membrane-enclosed lumen, cellular protein catabolic process, protein localization, nucleotide binding, macromolecule biosynthetic process, and identical protein binding. Analyzing the parental gene associations can suggest what these circRNAs are functioning in. Comparing enriched GO terms between day 90 and day 250, it was revealed that vesicle, endoplasmic reticulum, and mitochondrial lumen were more enriched at peak lactation. This is consistent with higher protein synthesis for milk production and secretion as well as a higher energy demand. Interestingly, the most highly expressed circRNAs are derived from casein genes. Of the circRNAs detected, 80 were produced from the 4-casein coding genes: CSN1S1 (Alpha S1), CSN1S2 (Alpha S2), CSN2 (Beta casein), and CSN3 (Kappa casein). The top 5 expressed circRNAs were derived from casein genes and were significantly higher at lactation day 90. Next, Zhang et al. investigated the miRNA targets for 14 of the casein circRNAs and found that they had many target sites for the miR-2284 family. CSN1S1, CSN2, and the lactoferrin (LTF) whey protein mRNAs are in turn the targets of the miR-2284 family. Carrying the work further, the top 3 expressed circRNA and CSN1S1 and CSN2 mRNA showed a slight positive correlation. Casein and whey proteins are the main proteins secreted during lactation, suggesting that the circRNAs derived from casein genes are exerting a positive feedback loop or a protective function. It could be that by sponging the mir-2284, it inhibits it from binding to its casein and whey mRNA targets and inhibiting their translation. Importantly, whey proteins such as beta-lactoglobulin and alpha-lactalbumin are upregulated during lactation, meanwhile lactoferrin and transferrin undergo a decline after pregnancy and through lactation [62, 63]. This would suggest that the circRNA mentioned would probably be exerting an effect on casein and some whey proteins that is opposite to the effect on lactoferrin and transferrin. Further investigation is required to understand the regulation between the miR-2284 family and LTF expression. We investigated the expression of the miR-2284 family using the data by Wu et al. and found that bta-miR-2284j and bta-miR-2284 h-5p were significantly downregulated (Table 1, rows 11, 12) with a fold change of −3.7 and −2.8 respectively. The circRNAs and their target miRNAs thus have a reciprocal expression. Importantly, the researchers suggest that these networks have downstream effects of relieving casein mRNA from being sponged by miR-2884 family. This proposes possible Casein genes/Casein-derived circRNA/mir-2284/Casein and Whey mRNA axes that regulate protein secretion during lactation in bovine MG. Moreover, it manifests the newly described function of circRNAs to regulate their own parental genes, as a feedback loop. As described in the biogenesis section, circRNAs have been reported to promote the transcription of their own parental genes through binding to RNA polymerase [64], but this study by Zhang et al. also shows how such positive feedback loop can be achieved through circRNA-miRNA networks.

A recent study sequenced the MGs of cows with variable milk fat percentage (MFP) to reveal 309 circRNAs were expressed at a higher level in high-MFP group compared to low-MFP group [65]. Weighted gene co-expression network analysis (WGCNA) was conducted on this data, whereby WGCNA analysis is used to identify circRNA clusters (modules) that have similar expression patterns across samples. The analysis determined 18 co-expression modules, with multiple modules associated with MFP in the pink module. Four differentially expressed circRNAs from this module were highly expressed in mammary tissue and possibly act as competitive endogenous RNAs (ceRNAs). Constructing the ceRNAs networks revealed seven target genes: GNB1 (Guanine nucleotide-binding protein beta 1), GNG2 (G protein subunit gamma 2), PLCB1 (phospholipase C beta 1), ATP6V0C (ATPase H + transporting V0 subunit c), NDUFS4 (NADH ubiquinone oxidoreductase subunit S4) and PIGH (phosphatidylinositol glycan anchor biosynthesis class H) as the most probable ceRNA modulators in milk fat metabolism.

Ovine

CircRNAs derived from the 4 ovine casein-coding genes and 2 ovine whey protein-coding genes were highest peak lactation in ovine MG in a study conducted by Wang et al. [9], similar to findings by Zhang et al. on the bovine MG [6]. This suggested some conservation in circRNAs expression and function in the MG between bovine and ovine species. GO analysis of their parental gene in molecular function category revealed the terms molecular function, binding, protein binding, ATP binding, and ion binding. Five circRNAs were then chosen to establish circRNA-miRNA targets, finding that several miRNA targets have been previously described in the literature as involved in MG development [9]. Some of these miRNA targets of upregulated circRNAs were also found in Wu et al.’s data and mostly showed opposite expression level to the respective circRNA. The combined data has been assembled into Table 1 (rows 13–17).

A recent study conducted RNA sequencing of mammary tissue harvested from dairy goats during peak lactation and dry period [55]. They identified circRNA-08436 and mRNA ELOVL6 (ELOVL fatty acid elongase 6) to have significantly higher expression in peak lactation relative to dry period. Using goat mammary epithelial cells (GMECs), circRNA-08436 was found to promote triglyceride and cholesterol synthesis, and increase saturated fatty acids concentration in the cells. This effect was revealed to be mediated through the binding of circRNA-08436 to its target miRNA miR-195 and downregulating it. This binding relieves the target mRNA ELOVL6 of miR-195 and upregulates its expression. The axis circRNA-08436/miR-195/ELOVL6 has been added to Table 1 (Row 18).

Comparative circRNAs Expression Across Different Bovine and Ovine Breeds

While there seems to be some common expressions of circRNAs and their function between different species, there is breed specific expression of circRNAs in the MG within the same species. A study by Hoa et al. identified differentially expressed circRNAs between lactating MGs of two breeds of sheep that have different milk production profiles [66]. The parental genes of the 33 differentially expressed circRNAs were mostly enriched in heterocyclic compound binding, catalytic and kinase activity. This adds to our knowledge of circRNAs’ expression in context of milk yield and its components. Similar data was found when comparing circRNA expression in lactating MG of two cattle breeds [7]. These two breeds were also distinct with their milk production with the Jersey breed having higher lactation than Kashmiri cattle and had 21 differentially expressed circRNAs. Interestingly, seven of these circRNAs that were upregulated in Kashmiri cattle were derived from casein genes. This is reflective of the fact that beta-casein is upregulated in milk of Kashmiri cows relative to Jersey cows, while Jersey cows express higher kappa-casein [67]. Overall, beta-casein is more abundant than kappa casein in milk [68]. This further reaffirms how circRNAs expression is very specific, whereby it is more component-specific than yield-specific in this case. As for the GO enrichment analysis of differentially expressed circRNAs, the terms obtained were comparable to what was found in the previously mentioned studies with terms such heterocyclic compound binding and catalytic and kinase activity. KEGG pathway analyses showed that only the parental gene of circ-015003, TGFBR2 (transforming growth factor, beta receptor II), was significantly enriched in adherens junction, the TGFβ signaling pathway, and the MAPK signaling pathway [7].

Rodents

Other than livestock models, one study conducted deep sequencing of lactating MG samples of rats at day 1 and day 7 postpartum [8]. 6,824 and 4,523 circRNAs were identified on day 1 and day 7 respectively. 1,314 circRNAs were common between the two lactation stages. Taken as a whole, the GO analysis revealed enrichment in ATP and nucleotide binding, nuclear lumen, GTPase regulator activity, ion binding, and protein kinase activity. However, considering that this enrichment analysis is for all circRNAs detected, rather than differentially regulated against a non-lactating counterpart, it does not reveal much of their involvement during lactation in the MG. Rather, it seems to provide an idea of their ubiquitous expression as we would expect GTPase and kinase activity to be constantly ongoing in the cells.

The studies surveyed in this section so far have relied on high-throughput RNA sequencing to identify and characterize the circRNAs present in the MG at different lactation stages in different species. Similar results between bovine, ovine and rat studies indicate a level of evolutionary conservation of circRNA expression. Additionally, we gained insight into their functions and how it falls into context of lactation. However, the results were based on enrichment analysis of their parental genes. This presents two limitations. First, although the analysis of the parental genes can give an indication of the function of their derived circRNAs, circRNAs can perform functions distinct from their parental genes. The second limitation is that very little can be understood concerning the mechanism by which these circRNAs are functioning and remain predictive, especially in terms of the established circRNA-miRNA networks.

Involution

Xuan et al. performed sequencing for circRNAs in goat MG during late gestation (LG), late lactation (LL), and dry period (DP) stages [69]. A number of circRNAs were exclusive to each stage, showing stage-specific expression. The differentially expressed circRNAs for each stage, whether upregulated or downregulated, were pooled into the same group to perform functional analysis of their parental genes. Relative to DP group, genes enriched in MG development, sequestering triglycerides, and regulation of lipid storage were upregulated in LL group. Comparing between DP and LG groups, differentially expressed circRNAs had their parental genes enriched in lactation and mammary development related terms, as well as wound healing. Lactation genes-derived circRNAs were downregulated in the DP while the wound healing ones were upregulated. The circRNAs derived from CSN2, CSN3, ERBB4 (erb-b2 receptor tyrosine kinase 4; it is part of the epidermal growth factor receptor family and associated with growth inhibiting properties) [70], and PRLR (prolactin receptor which induces downstream JAK-STAT, AKT and MAPK signaling pathways when activated) [71], were downregulated during dry period relative to lactation. This is consistent with the lack of milk synthesis during the dry period. Meanwhile, most of the genes upregulated in DP were associated with wound healing, also consistent with the remodeling during involution. CircRNA-associated pathways were also predicted and showed that circRNAs can impact MG development, substance metabolism, immunity, and mammary cell apoptosis, all of which are essential for MG involution and tissue remodeling.

Functional Implication of Selected circRNAs on MG

Few studies have taken a closer look at specific circRNAs. Circ_015343 was upregulated at peak lactation in the MG tissue [54]. This circRNA is derived from the aminoadipic semialdehyde synthase (AASS) gene. Both circRNA and its parental gene were expressed in eight ovine tissue samples, with the highest expression in the MG. Comparing its expression between two different breeds of sheep, circ_015343 had a lower expression in Small Tail Han sheep which has a higher milk yield and fat composition relative to the Gansu Alpine Merino sheep, whereas its parental gene had an opposite regulation. The circRNA could be regulating milk synthesis and interfering in milk protein or fat synthesis. Conversely, AASS protein is part of a pathway that degrades lysine which is needed for milk protein synthesis. The inhibition of circ_015343 using siRNA led to high proliferation and viability rate in ovine mammary epithelial cells. This would suggest circ_015343 plays a role to inhibit proliferation and viability of mammary epithelial cells, which could be consistent with its high expression during lactation as the cells exhibit low proliferation rate at that stage [54, 72]. Using miRanda, 27 miRNA targets were predicted, and five of them were selected based on literature mention of these miRNAs in the MG development. Only miR-200a was found in the miRNA dataset by Wu et al., as shown in Table 1 (row 18). This study gives a closer insight into the expression of circ_015343 and its effect on the mammary epithelial cells of sheep. A study by Zhang et al. revealed another circRNA-mediated network and verified the predictions experimentally [11]. CircRNA-006258 sponges miR-574-5p, which relieves its mRNA target EVI5L. In the goat mammary epithelial cells, EVI5L is shown in the study to promote proliferation, increase viability, and increase milk synthesis via PI3K/AKT-mTOR pathway, which leads to the production of β-casein and triacylglycerol.

With circRNAs still relatively novel, research pertaining to their role in normal MG development is limited. As seen, most of the work has been conducted on livestock during lactation. However, it offers valuable insight into the temporal-specific, species-specific, and even breed-specific nature of the expression of circRNAs which in turn reflects functional differences. Many circRNA-miRNA-mRNA axes were constructed to understand the molecular mechanism by which circRNAs impact the different processes. It is important, however, to acknowledge that miRNA sponging is only one mechanism, and further work is needed to fully understand their effect. This is further highlighted by several circRNA-miRNA networks having the same pattern of expression, whereas sponging would result in reciprocal expression. This also applies to miRNA-mRNA networks. We found several of these networks having parallel regulation patterns (for example Table 1, Rows 2–10) which necessitate further investigation to understand their roles. It is known that miRNAs primarily inhibit their target mRNAs, but also rarely upregulate them [73, 74]. However, it seems that parallel pattern of regulation was at a higher frequency in the assembled data than expected (11 out of 17 complete networks, albeit 9 of these 11 circRNAs are reported in the same study). It is likely that other mechanisms are also playing a role. The understanding of their regulation during normal MG functioning is not only important in context of milk synthesis and breastfeeding but also to understand how they are altered during disease states such as mastitis and cancer.

CircRNAs Dysregulation in Diseased MG

Mastitis

Mastitis is inflammation of the MG. It can be divided into non-lactational and lactational mastitis, with the latter being more serious and frequent caused by bacterial entry through breaks in the skin or due to physical trauma [14, 75]. Non-lactational mastitis is categorized as periductal or idiopathic granulomatous mastitis (IGM).

Recent work, with focus on bovine models, has recognized the role of ncRNAs, including circRNAs in mastitis. RNA sequencing of MG tissue from healthy Holstein cows and Holstein cows with naturally occurring mastitis, due to Staphylococcus aureus, revealed 19 differentially expressed circRNAs [76]. Important to mention, no circRNAs have been reported in the literature in Staphylococcus aureus, so it would be assumed that the circRNAs detected are from the MG cells itself. Parental genes of the 19 circRNAs were most enriched in RNA polymerase transcription factor binding in GO terms, and in tight junction pathways in KEGG pathway analysis. The miRNA targets for 9 circRNAs were predicted, where circRNA4027, for example, possibly sponges miR04297, miR-4530, miR-5581-5p, among others. Validation of the RNA sequencing results was carried out using qRT-PCR and three circRNAs were confirmed to have significant differential expression, namely circRNA2860, circRNA5323 and circRNA402. The circRNA-miRNA networks of these three validated circRNAs were assembled into table 1 (rows 1–3). A similar study investigated differentially expressed circRNAs in MG of Holstein cows that the researchers infected with Escherichia coli [77]. 72 circRNAs were upregulated meanwhile 92 circRNAs were downregulated in E. coli infected MG compared to healthy tissue. GO enrichment analysis of their parental genes revealed an association with ras protein signal transduction, cytoplasmic vesicle part, and enzyme binding categories. On the other hand, KEGG pathway analysis shows a significant association with the phagosome pathway. The latter finding indicates a role for the circRNAs in mastitis-related immune response.

Wang et al. established an in vitro model to study bovine mastitis by stimulating bovine mammary epithelial cells MAC-T with lipopolysaccharide (LPS) to induce inflammation [78]. RNA sequencing revealed 71 differentially expressed circRNAs at different time points post LPS treatment. We compared the findings of Wang et al.’s study with previously discussed findings of Bai et al.’s study. We noted a considerable discrepancy in the number of differentially expressed circRNAs detected in the sequencing of MG tissue with naturally occurring mastitis (19 circRNAs), and the sequencing of LPS stimulated MAC-T cells (71 circRNAs). We attribute that to the difference in model and the type of stimulation, as well as the first study setting the fold change threshold >1 [76], while the latter to >0.5 folds [78]. The parental genes of the 71 differentially expressed circRNAs were enriched in GO terms of cellular proliferation, apoptotic processes, migration, and inflammatory response. Meanwhile, KEGG pathway analysis revealed ErbB, NOD-like receptor, MAPK, bacterial invasion of epithelial cells, and Wnt signaling pathways. The miRNA targets were predicted for the circRNAs. The two circRNAs novel_circ_0004830 and novel_circ_0003097 were highlighted as they can both bind to bta-miR-145 (Table 2, rows 4,5). Bta-miR-145 is shown to be associated with mastitis regulation through FSCN1 gene (Fascin actin-bundling protein 1) [79]. A similar approach was adopted by Liang et al. where they induced inflammation in bovine mammary epithelial cells (bMECs) by E. coli LPS and conducted RNA sequencing [80]. Inflammation was verified as IL6, IL8, NF-kB, and TLR4 were significantly upregulated upon LPS treatment. The analysis showed 841 significant differentially expressed circRNAs in LPS group. Again, this proved to be a large discrepancy between this study and the study on MAC-T cells with 841 versus 71 differentially expressed (DE) circRNAs, respectively. The possible apparent reason is the difference in cell models. The parental genes of the 841 DE circRNAs were enriched in positive regulation of G1/S transition of the mitotic cell cycle, histone methyltransferase activity, and DNA methylation. KEGG pathway analysis revealed enrichment in hippo signaling pathway, AMPK signaling pathway, and Fc gamma R-mediated phagocytosis. This suggests the role of these circRNAs in the progression of inflammation in the MG.

Table 2.

List of circRNAs involved in mastitis disease of the mammary gland in bovine models. All information (Model of study, circRNA, pattern, target miRNA, and respective target mRNA) in each row was obtained from the reference cited. Each row would thus present a circRNA/miRNA/mRNA axis. A maximum of 4 miRNA/mRNA targets were included in the respectively referenced paper in each row, with the full list in the original paper. Each circRNA was written as cited in its respective research paper. CircRNAs nomenclature varies between ID, Alias, chromosomal position, or referring to its parental gene. The rows are referred to as numbered from 1–7. (-) indicates that this information wasn’t available in the reference data

Model of study	CircRNA	Pattern	Reference	Target miRNA	Target mRNA
Bovine mammary tissue	circRNA4027	Down	[76]	miR-4297 miR-5581-5p miR-4530 miR-4297	-
	circRNA2860	Up	[76]	miR-4778-5p miR-4762-3p miR-4309 miR-5008-5p	-
	circRNA5323	Up	[76]	miR-4778-5p miR-4762-3p miR-4309 miR-5008-5p	-
Bovine mammary epithelial cell lines	novel_circ_0004830	Down	[78]	bta-miR-145 bta-miR-193a-5p bta-miR-132 bta-miR-186	FSCN1
	novel_circ_0003097	Down	[78]	bta-miR-145 bta-miR-26a bta-miR-759 bta-miR-302a	FSCN1
	circ_LOC616254	Down	[81]	miR-2305	BCL2 FCER1G MAX
	circ_LOC616254	Down	[81]	miR-1777a	ECF WNT10A

Xu et al. investigated N6-methyladenosine (m⁶A) modified circRNAs in bovine mastitis [81]. M⁶A is the most common epigenetic post-transcriptional modification associated with eukaryotic mRNA and has been described in the regulation of several physiological processes [82]. After injuring bovine mammary epithelial cells MAC-T with inactivated S. aureus and E. coli, sequencing for circRNAs was performed. Between the control uninjured and S. aureus group, 67 m⁶A methylation peaks within 63 circRNAs were significantly different. As for the E. coli group relative to the control uninjured, 192 m⁶A methylation peaks within 137 circRNA were significantly different. Interestingly, the differentially methylated circRNAs showed opposite patterns in the S. aureus group compared to the E. coli group, where they were mostly hypomethylated in the S. aureus group and hypermethylated in the E. coli group. The difference between E. coli and S. aureus induced DE circRNAs reflects the studied difference of the host innate response to each bacterial infection, whereby limited cytokine response is elicited due to S. aureus infection [83]. GO analysis of parental genes of modified circRNAs showed enrichment in several important cell activities such as protein tyrosine kinase activator activity, SMAD binding, and integrin binding. Meanwhile, KEGG pathway analysis showed enrichment in endocytosis, ubiquitin-mediated proteolysis, and focal adhesion, among other pathways. Next, the miRNA targets of the highly significance selected circRNAs were predicted along with the subsequent mRNA targets of these miRNAs. The analysis revealed several miRNAs that are reported to be implicated in inflammatory response such as miR-2305 and miR-1777a and two networks were assembled to Table 2 (rows 6,7), with many more found in the original publication [81]. This study added an additional dimension to the regulation of circRNAs in disease states, and in mastitis, where circRNAs expression is itself also controlled by epigenetic modification. Interestingly, m⁶A modification can possibly promote mRNA nuclear export and translation initiation, and has been linked to cap-independent circRNA translation which is worth further investigation of the proteins that can be synthesized from circRNAs in context of mastitis and BC [84, 85].

Breast Cancer

Early BC Stages

Understanding the role of circRNAs in early cancer events and stages transcends their potential role as diagnostic markers and poses them as tools for early detection of cancer. Furthermore, they can possibly act as risk predictors prior to cancer formation. Several studies have sought to identify circRNAs differentially expressed and involved in early cancer stages. Rao et al. performed RNA sequencing of stage I-IIA BC tissue and their adjacent normal obtained from five patients, as well as normal tissue samples from non-cancer patients [86]. Using two algorithms find_circ and DCC to identify circRNAs, 26 circRNAs in total were found to be commonly upregulated in both analyses. MicroRNA sequencing of the same set of samples was aligned to the predicted miRNA targets of circRNAs. This allowed the assembly of numerous axes, with four circRNA-microRNA pairs showing significant negative correlation and 15 downstream mRNA targets (Table 3, row 1–4). Pathway enrichment analysis revealed functions in cell cycle processes, response to stress, chromatin modification and cellular EGFR, PDGF and WNT signaling pathways, which show involvement in BC pathways. A study by Mojarad et al.found hsa_circ_0005046 and hsa_circ_0001791 are significantly upregulated in 60 BC tissue samples, with an association of the latter circRNA only with ER receptor [87]. Ten miRNA targets were predicted for each circRNA including miR-215 which is downregulated in BC and miR-383 (Table 3, row 5,6). Downregulation of both miRNAs is also associated with an activation of the PI3K/AKT pathway [87].

Table 3.

List of circRNA-miRNA involved in early stages and EMT of BC. All information (Model of study, circRNA, pattern, target miRNA, and respective target mRNA) in each row was obtained from the reference cited. Each row would thus present a circRNA/miRNA/mRNA axis. A maximum of 4 miRNA/mRNA targets were included in the respectively referenced paper in each row, with the full list in the original paper. Each circRNA was written as cited in its respective research paper. CircRNAs nomenclature varies between ID, Alias, chromosomal position, or referring to its parental gene. The rows are referred to as numbered from 1–13. (-) indicates that this information wasn’t available in the reference data

Model of study	Cancer Stage	CircRNA	Pattern	Reference	Target miRNA	Target mRNA
Human breast tissue	Stage I-IIA cancer	hsa_circ_0023990	Up	[86]	hsa‐miR‐548b‐3p	SLCO6A1 FAT2 CD46
	Stage I-IIA cancer	hsa_circ_0016601	Up	[86]	hsa‐miR‐1246	SKIL
	Stage I-IIA cancer	hsa_circ_0001946	Up	[86]	hsa‐miR‐1299	FSD2 OCA2 CCND1 MYRF
	Stage I-IIA cancer	hsa_circ_0000117	Up	[86]	hsa‐miR‐502‐5p	APOB
	Early cancer stages	hsa_circ_0005046	Up	[87]	hsa-miR-215	AKT1
	Early cancer stages	hsa_circ_0001791	Up	[87]	hsa-miR-383 hsa-miR-1236 hsa-miR-1244	-
Human epithelial cell line	Pre-tumorigenic/early cancer	hsa_circ_0077755	Down	[88]	miR-182	-
	EMT	circSCYL2	Down	[89]	hsa-miR-6804-5p hsa-miR-1270 hsa-miR-6824-5p hsa-miR-4308	FKBP1B ADAMTS1 MDGA2 HSF5
	EMT	circANKRD12	Down	[89]	hsa-miR-6721-5p hsa-miR-6849-5p hsa-miR-6889-3p hsa-miR-6514-5p	SGSM1 FRAT1 ERG WNT11
	EMT	circMYO9A	Down	[89]	hsa-miR-222-3p hsa-miR-5006-5p hsa-miR-6796-5p hsa-miR-6511a-5p	CSF1 TRAF3 RELB CASP10
TNBC tissue	EMT	circANKS1B	Up	[90]	miR-152-3p miR-148a-3p	USF1 ESRP1
TCGA samples and human epithelial cell line	EMT	circNCAPG	Up	[91]	miR-200c miR-200b	CSF-1 ZEB-1
Human breast tissue	EMT	circRNA_000554	Down	[92]	miR-182	ZFP36

A study by Naser Al Deen et al. established a pre-tumorigenic model using nontumorigenic HMT-3522 human mammary epithelial cell line, where they silenced gap junctional protein Cx43 which led to early transformation events with loss of polarity, and a multilayered morphology in 3D cultures [88]. Microarray profiling of Cx43-KO-S1 (pre-tumorigenic) relative to S1 (nontumorigenic) showed 121 differentially expressed circRNAs. Eighteen high confidence circRNAs were chosen for validation by qRT-PCR and revealed the same pattern and significance of differential regulation. miRNA sequencing of the same samples revealed 65 differentially regulated miRNAs. The miRNA targets of each circRNA were predicted then the microarray and sequencing data were used to align circRNA-miRNA axes with reciprocal expression patterns. Derived from Cx43-producing gene (GJA1), hsa_circ_0077755 was found to be downregulated, and one of its target miRNAs, miR-182, was found to be upregulated in sequencing data. As such, the axis Cx43/hsa_circ_0077755/miR-182 was proposed as a biomarker axis for increased risk of BC initiation (Table 3, row 7).

These studies discussed in the above section have offered valuable insight into the role of circRNAs in cancer initiation. We are interested to understand the involvement of circRNAs in the very early events that precede tumor formation and characterization, which can have major application for population with risk factors for BC [93]. Such work is still lacking in literature.

Other than transitioning from normal to tumor cells, which is better known as transformation, epithelial cells can undergo EMT. EMT is not specific to neoplastic cells, and can occur for tissue repair or organ development [94]. The areas pertaining to the biochemical processes that underlie epithelial cells acquiring an invasive phenotype have witnessed intensive research. The plasticity of the EMT program adds to its importance, as it maintains a reversible trait. CircRNAs are reported to be involved in EMT in the breast epithelium, as will be discussed next.

Breast EMT

Many studies have investigated circRNAs in breast EMT events. Some circRNAs are involved in enhancing transformation, whereas others function as suppressors of tumor progression. A study by Yuan et al. established a model that simulates EMT by administering transforming growth factor-β1 (TGFβ1) to MCF-7 and MDA-MB-231 cell lines [89]. After treatment with TGFβ1, EMT was established in the two cell lines as verified by loss of E-cadherin and upregulation of Vimentin [89]. RNA sequencing was performed for both MCF-7 and MDA-MB-231 cells with and without TGFβ1 treatment. Relying on mapping of junctional region reads which are specific to circRNAs, MCF-7 cells had 1,402 significantly differentially regulated circRNAs in EMT group relative to untreated group. While in MDA-MB-231 cells, 322 circRNAs were significantly differentially regulated in EMT group. It is worth noting that circRNAs, like other regulatory molecules, function in overlapping pathways and alter the expression of downstream circRNAs. It would be interesting to investigate the few circRNAs that are initially affected and lead to the cascade involving the other circRNAs and thus play an integral role in very early transformation events. Regardless, GO analysis of parental genes of differentially regulated circRNAs in both cell lines showed most significant enrichment in functional terms associated with cell adhesion and immune/inflammatory response namely terms such as: inflammatory response, positive regulation of α-βT cell proliferation, extracellular matrix organization, and actin cytoskeleton reorganization. KEGG pathway analysis showed tumor growth associated pathways such TNF and NF-kB signaling pathways, ECM-receptor interaction, and PI3K-Akt signaling pathway. From the differentially regulated circRNAs in the two cell lines, 5 circRNAs were selected that were common and similar in pattern of regulation as well as having a high fold-change and abundancy. CircRNA-mediated regulatory networks were predicted for the 5 circRNAs with downstream target genes such as OSR1, DRD2, MAP3K8, and TNFSF4. Three of these axes are listed in Table 3 (row 8–10). Further focus was placed on the downregulated circRNA circSCYL2 (SCYL2: SCY1 like pseudokinase 2) in the two cell lines, which showed significant downregulation in human BC tissue samples. Its overexpression in MCF-7 and MDA-MB-231 cells inhibited their migration and invasion.

A study by Li et al. adopted similar methodology where MCF10A cells were subjected to TGFβ mediated EMT [95]. qPCR experiments revealed circITGB6 as the most highly upregulated due to TGFβ treatment in MCF10A cells. The increase was also verified in colorectal cancer cell line, HCT116 and HT29. Overexpression of circITGB6 in untreated MCF10A cells yielded increased migration, N-cadherin and vimentin expression, coupled with downregulation of E-cadherin. This circRNA is unlikely to function as a miRNA sponge with two miRNA binding-sites. Interestingly, using RNA pull-down, circITGB6 was shown to act as a protein decoy to insulin-like growth factor 2 mRNA binding protein 3 (IGF2BP3). Knockdown of IGF2BP3 limited the effects of circITGB6 on EMT markers and migration. Relying on previously mentioned roles of IGF2BP3 in binding to several metastasis related mRNA cargo, it was proposed that circITGB6 promoted the interaction between IGF2BP3 and PDPN mRNA improving the stability of the latter. As such, the axis is circITGB6/IGF2BP3/PDPN which is well defined and provides an example of mechanisms other than miRNA sponging that could be at play as circRNAs act as proteins decoys.

TGFβ/SMAD pathway was found as a downstream target of circANKS1B upregulation in BC [90]. Investigating its role through overexpression in MDA-MB-231 and knockout in MCF-7, revealed a function in promotion of migratory and invasive abilities of BC cell lines. circANKS1B was found to promote EMT through its sponging of miR-148a-3p and miR-152-3p, which are known to have tumor suppressive activity by suppressing oncogenic genes translation (Table 3, row 11). Indeed, circANKS1B upregulation in MCF-7 most notably resulted in upregulation of USF1, a target of miR-148/152 family, and a transcriptional regulator of TGFβ1 and ESRP1. ESRP1 is a splicing factor that was found to promote the generation of circANKS1B. Overall, this shows that this circRNA upregulation ultimately induces EMT through an activation of TGFβ/SMAD pathway and concomitantly promotes a positive feedback loop leading to its own generation. The upregulation of circANKS1B was also consistent in tissue samples. RNA sequencing of TNBC and adjacent normal tissue samples revealed significant upregulation of circANKS1B with highest fold change of approximately 6-folds. This upregulation was consistent in TNBC when compared to luminal A, luminal B, and HER2 subtypes as well as TNBC cell lines relative to non-TNBC.

A study by He et al. found circNCAPG to be overexpressed in TCGA BC cases, and it sponges miR-200b and miR-200c [91]. These miRNAs, in turn, target CSF1 (chemokine), and ZEB1 (TF involved in EMT) as shown in Table 3 (row 12). Knocking out circNCAPG in MCF-7 cells led to decreased expression of CSF1 and ZEB1 and inhibition of BC metastasis, which suggests its role in cancer metastasis by sponging of miR-200 s.

In contrast to the above-described studies, several circRNAs were found to be repressors of EMT in BC models. Using in-silico analysis, Mao et al. detected differentially expressed circRNAs in BC tissues and overlapped them with miRNA and target genes analysis [92]. hsa_circRNA_000554, also known as circ_0000376, was significantly downregulated in BC and targets miR-182 which is reciprocally upregulated in BC. This was coupled with downregulation of its target gene ZFP36 (Table 3, row 13). After establishing the axis circRNA_000554/miR-182/ZFP36, the functional implication of the upregulation of this circRNA was investigated in BC cells. Its upregulation led to the reversion of the EMT process in MCF-7 cells with increased E-cadherin, decreased N-cadherin and vimentin, and reduced cell invasion and migration. It was mediated by inhibition of miR-182 and downstream upregulation of ZFP36, which is reported to have suppressive roles of cell proliferation and promotion of apoptosis [96]. A study by Wu et al. revealed an interesting negative feedback loop from the circRNA circYap on its parental gene yes-associated protein (Yap) where Yap protein is part of the Hippo pathways involved in tumorigenesis [97]. circYap is found to be significantly downregulated in breast tumor tissues and cell lines compared to adjacent normal and non-tumorigenic cell lines, respectively. Overexpression of circYap led to the downregulation of Yap protein expression and was mechanistically revealed to be through the interaction of circYap with PABP and eIF4G, members of translation initiation machinery. circYap binding to the two proteins prevents their binding to Yap mRNA and its subsequent suppression. The overexpression of the circRNA also led to a suppression of proliferation, migration, and colony formation of the cells.

As such, circRNAs can act as promoters or suppressors of BC, whether at early stages or during EMT. This posits them as possible therapeutic targets for BC. However, the dual nature of circRNAs offers a layer of complexity, whereby the same circRNA could be a promoter of tumorigenesis in one context and a suppressor in another. This context could be on organ level, cancer subtype, or stage of cancer. For example, circRNA_0000376 (also known as circRNA_000554) was found to be a tumor suppressor in BC as previously described [92]. However, it is found to promote non-small cell lung cancer progression through three different reported pathways: miR-545-3p/PDPK1, miR-1182/NOVA2, or miR-1298-5p/KPNA4 pathways [98–100]. The same complexity can be observed on a cancer subtype level. As discussed, circANKS1B is a promoter of EMT in BC [90]. However, it’s important to note that it was found to be upregulated in TNBC relative to other BC subtypes including luminal A, luminal B, and HER2 subtypes. These are just a few of many examples of circRNAs that act as promoters or suppressors of BC progression depending on the subtype as well as the stage of cancer. This duality reflects the intricate molecular networks circRNAs participate in, where their impact depends on factors related to inter- and intra-tumor heterogeneity.

The context-dependent roles of circRNAs pose significant challenges and opportunities for their therapeutic targeting. Targeting tumor-promoting circRNAs could potentially inhibit cancer progression and enhancing the function of tumor-suppressing circRNAs might reinforce the body's tumor-suppressive mechanisms. However, the pleiotropic effects of circRNAs necessitate careful consideration, as interventions targeting one circRNA could inadvertently disrupt multiple pathways, leading to unintended consequences. Future research should prioritize understanding the specific conditions under which circRNAs exert their promoting or suppressing effects, as well as their interaction with other regulatory molecules.

Conclusion

This review aimed to investigate the involvement of circRNAs in physiological and pathological states of the MG. For physiological states, the studies discussed revealed several circRNA/miRNA/mRNA axes, mainly during lactation. We explored the pattern of regulation of target miRNAs by aligning circRNA and miRNA sequencing data together. It was expected that they would share a reciprocal pattern of regulation, due to the sponging relationship between the two. Interestingly, several circRNA-miRNA networks shared parallel patterns of regulation which suggests that those circRNAs are possibly involved in other signaling pathways that are overshadowing their sponging of target miRNAs. Nonetheless, CircRNA-miRNA networks suggest that these circRNAs play roles in key signaling pathways, including JAK-STAT, MAPK, TGFβ, and Wnt.

Understanding the expression and function of circRNAs during normal development is crucial to understand their alteration during diseased states. Dysregulation of aforementioned signaling pathways disrupts the stringent control on cellular processes during cyclic MG remodeling as well as immune response to inflammation. CircRNAs are thus found to be involved in mastitis as well as BC progression. We focused our attention on stages of BC that are less studied; mainly BC initiation and EMT, as they serve to help us understand better how circRNAs are part of the transitioning of normal MG cells towards malignancy. To summarize, we constructed an overview figure portraying the different circRNA-miRNA networks involved in each state of the MG as discussed (Fig. 1).

Fig. 1.

Overview of circRNA-miRNA networks differentially expressed across physiological and pathological states of the mammary gland. This figure highlights the prominent circRNA-miRNA networks described in the text and summarized in Tables 1, 2, 3. The left panel depicts physiological states of the mammary gland, represented in diagrams, for pregnancy (alveologenesis at terminal end buds), lactation (alveologenesis complete), and involution (alveoli regressed). The righthand panel depicts diseased states: mastitis, early breast cancer and EMT. Each circRNA is listed alongside its predicted target miRNA(s) and according to the nomenclature used in their respective studies. Arrows indicate the differential expression levels of each circRNA (upregulated ↑ or downregulated ↓) relative to controls as reported in the respective referenced studies (in text and in Tables 1, 2, 3). The differential expression of the listed circRNAs is associated with pathways and processes in the mammary gland. They are outlined in the boxes beneath each state

Driven forward, identifying the very early circRNA/miRNA/mRNA axes, implicated in early BC initiation events, can be utilized for the early detection/prediction of BC. CircRNAs and miRNAs combined have an additional advantage in this application to other RNA molecules because they are released into circulation and are stable for appreciable periods. As such, they can serve as biomarkers, indicative of BC initiation, that are accessible through non-invasive sampling.

Several studies have investigated the potential of circRNAs as clinical biomarkers, highlighting their stability in bodily fluids such as blood and saliva [101, 102]. This intrinsic stability, conferred by their covalently closed-loop structure, makes circRNAs attractive candidates for liquid biopsy applications. In addition to their stability, circRNAs are shown to have tissue-specific patterns of expression [103, 104]. This could offer an advantage if we were able to identify circRNAs specific to the MG. While there are currently no active clinical trials for the use of circRNAs as biomarkers for BC early detection (https://clinicaltrials.gov/), there is a recruiting study for pancreas adenocarcinoma (ClinicalTrials.gov search for identifier: NCT03334708). This clinical trial, led by Memorial Sloan Kettering Cancer Center, will investigate many potential biomarkers including circRNAs and circulating tumor DNA (ctDNA) in patient blood samples. The purpose of the study is to utilize blood biomarkers not only for early detection, but to also monitor response to treatment. It highlights the potential of circRNAs for both diagnosis and prognosis. There are several other registered clinical trials for circRNA biomarkers in different types of cancer that are still in their early stages as the field gains interest and offers promising outcomes. We explored the applications of circRNAs as biomarkers for early detection of BC in a recently published review, referenced here for a more extensive discussion [105].

It remains that the application of circRNAs as therapeutic targets, as previously discussed, requires vigorous consideration to comprehend the plethora of signaling pathways that they are often involved in. Several studies have manipulated the expression of a target circRNA in-vivo using BC xenograft experiments in mice models [17, 106–108]. Fewer studies have gone further to investigate the application of circRNAs using RNA-based therapies in BC models. Zhou et al. identified an oncogenic circRNA, named cSERPINE2 (based on its parental gene), and it was significantly overexpressed in BC [109]. Using PLGA-nanoparticles carrying cSERPINE2-targeted siRNA, they show that the systematic knockdown of cSERPINE2 suppresses mammary tumor growth in mice and suppresses lung metastasis. Further investigation utilizing similar methodology is required to fully realize the potential applications of circRNAs in BC.

To conclude, this review summarizes and adds to our current knowledge of circRNAs in normal MG development and in diseased states as mastitis and BC, with the future outlook of utilizing circRNAs as biomarkers for early BC detection and as therapeutic targets.

Authors’ Contributions

SM read the relevant literature and drafted the manuscript. NM assisted with drafting and critical reading of the manuscript. MA provided feedback on the conceptual framework and writing of the paper. RT mentored SM throughout the conceptualization and writing processes and critically revised all the drafts. All authors contributed to the final manuscript and approved it for submission.

Funding

The authors gratefully acknowledge the funding from the University Research Board at the American University of Beirut (URB-AUB; Grant ID: 26748) and the Interdisciplinary Research Funding from the Faculty of Arts and Sciences at the American University of Beirut (FAS-AUB).

Data Availability

No datasets were generated or analysed during the current study.

Declarations

Ethics Approval and Consent to Participate

Not applicable.

Competing Interests

The authors declare no competing interests.