Circular RNAs (circRNA) are covalently closed molecules that can play important roles in cancer development and progression. Hundreds of differentially expressed circRNAs between tumors and adjacent normal tissues have been identified in studies using RNA sequencing or microarrays, emphasizing a strong translational potential. Most previous studies have been performed using RNA from bulk tissues and lack information on the spatial expression patterns of circRNAs. Here, we showed that the majority of differentially expressed circRNAs from bulk tissue analyses of colon tumors relative to adjacent normal tissues were surprisingly not differentially expressed when comparing cancer cells directly with normal epithelial cells. Manipulating the proliferation rates of cells grown in culture revealed that these discrepancies were explained by circRNAs accumulating to high levels in quiescent muscle cells due to their high stability; on the contrary, circRNAs were diluted to low levels in the fast-proliferating cancer cells due to their slow biogenesis rates. Thus, different subcompartments of colon tumors and adjacent normal tissues exhibited striking differences in circRNA expression patterns. Likewise, the high circRNA content in muscle cells was also a strong confounding factor in bulk analyses of circRNAs in bladder and prostate cancers. Together, these findings emphasize the limitations of using bulk tissues for studying differential circRNA expression in cancer and highlight a particular need for spatial analysis in this field of research.

Significance:

The abundance of circRNAs varies systematically between subcompartments of solid tumors and adjacent tissues, implying that differentially expressed circRNAs discovered in bulk tissue analyses may reflect differences in cell type composition between samples.

Circular RNAs (circRNA) are covalently closed molecules that constitute a large class of mostly noncoding RNAs with diverse mechanisms of action and functional roles in cancer (1). Most circRNAs are produced from protein-coding host genes through a spliceosome-dependent backsplicing event linking a splice donor site to an upstream splice acceptor site (2). More unique circRNAs than the number of genes in the human genome have now been annotated (3, 4) as single genes can give rise to multiple different circRNAs depending on which exons are involved in the backsplicing.

Hundreds of studies have recently compared circRNA expression profiles in solid tumors and adjacent normal tissues using microarrays or high-throughput RNA sequencing (RNA-seq) in a search for differentially expressed circRNAs with functional roles in cancer and/or biomarker potential. However, if circRNA expression profiles within subcompartments of tumors and adjacent normal tissues are systematically distinct, something that has not been investigated previously, analyses of bulk tissues may lead to misinterpretations of the data.

Indeed, we have previously shown that the hitherto most studied circRNA in cancer, ciRS-7 (also known as CDR1as; ref. 5), turned out to be completely absent in the cancer cells of several different types of adenocarcinomas (6), despite often being more abundant in these tumors relative to adjacent normal tissues. This seeming discrepancy could be explained by a high abundance of ciRS-7 in stromal cells within the tumors (6), implicating that the in vitro and in vivo models used by many research groups suggest that it has oncogenic properties that may be physiologically irrelevant.

Whether the differential expression of other circRNAs observed in comparisons of bulk tissues has led to similar misinterpretations is currently unknown. However, it is challenging to disentangle spatial and single-cell expression patterns of circRNAs in complex tissues as these molecules lack polyA-tails; most current protocols rely on capture of polyadenylated RNAs, and sequencing methods for total RNA have limited sensitivity due to very few sequencing reads being circRNA-specific (i.e., spanning their backsplicing junctions, BSJs; refs. 7, 8).

Therefore, we decided to explore circRNA expression patterns in colon cancer within distinct subcompartments, including cancer cells and surrounding non-neoplastic stromal cells from the tumors as well as normal epithelial cells and cells from the muscularis propria from adjacent normal tissues using laser-capture microdissection (LCM) of formalin-fixed paraffin-embedded (FFPE) tissues. We then analyzed circRNA expression in these subcompartments using NanoString nCounter technology, which is suitable for accurate quantification of circRNAs in FFPE tissues (9), because it is enzyme-free and thus avoids potential overestimation of circRNA expression levels and false positives caused by rolling circle amplification and template switching of reverse transcriptases (8–12). We contrasted data from the spatial analyses with analyses of bulk tissues from the same patients and find that the vast majority of significantly differentially expressed circRNAs discovered in bulk tissue analyses are not reflecting changes occurring within the cancer cells. Instead, most circRNA expression changes are directly linked to the proliferation rates of the cells and can be explained by their high stability and slow biogenesis. Finally, we validated our findings using single-molecule in situ hybridization and GeoMx digital spatial profiling and show that our findings may have strong implications for bulk tissue analysis of circRNAs in additional cancers.

Patient samples and ethical approvals

All patients with colon cancer used for NanoString nCounter analyses were treated surgically for stage II colon cancer with tumors classified as T3 or T4 according to the tumor, node, and metastasis (TNM) classification. Patients used for RNA-seq were treated surgically for stage II–IV colon cancer with tumors classified as either T3 or T4. All patients with colon cancer had a normal mismatch repair protein expression. FFPE tissue samples were from Vejle Hospital and fresh-frozen samples were from the Danish Cancer Biobank. Patients with prostate cancer were treated by curatively intended radical prostatectomy for histologically verified clinically localized disease and patients with bladder cancer were diagnosed with localized muscle-invasive cancer. RNA-seq data from 172 prostate cancer samples were from our previous publication (13). The use of colon cancer specimens was approved by the Regional Ethical Committee (S-20170197 CSF) and the use of bladder and prostate cancer specimens were approved by the National Committee on Health Research Ethics (#1706291 and #1603542). Written consent was obtained from all participants that excess diagnostic material could be used in future research projects and the requirement for written informed consent to the specific analyses in this retrospective study was waived by The National Committee on Health Research Ethics. The study was conducted according to the Declaration of Helsinki Principles.

Cell culture

Caco-2 (RRID:CVCL_0025), HT-29 (RRID:CVCL_0320), HTC116 (RRID:CVCL_0291), COLO 320HSR (RRID:CVCL_0220) human adenocarcinoma cell lines and the human aortic smooth muscle cell line (hASMC) were maintained at 37°C in humidified air containing 5% CO2. Caco-2 cells were cultured in EMEM (Sigma-Aldrich) and 1% nonessential amino acids (Sigma-Aldrich). HT-29 and HCT116 cells were cultured in McCoy's 5A medium (Sigma-Aldrich). COLO 320HSR cells were cultured in RPMI-1640 medium (Sigma-Aldrich). All media contained 10% FBS, 2 mmol/L l-glutamine, penicillin (100 U/mL) and streptomycin (100 μg/mL). hASMC cells were cultured in Vascular Cell Basal Medium (ATCC) containing Vascular Smooth Muscle Cell Growth Kit, penicillin (100 U/mL), and streptomycin (100 μg/mL). Coating of culture dishes with Collagen I (Corning) was used for better cultivation of hASMC cells. Cell lines were either authenticated and tested negative for Mycoplasma or purchased from the ATCC.

Laser capture microdissection

Sections of 5-μm-thick FFPE tissues were mounted on membrane slides (Leica) after 30 seconds of deparaffination in Xylene, followed by rehydration in graded ethanol, staining for 2 seconds in Mayers hematoxylin and washing in sterile water. After drying, the sections were microdissected using an LMD 630 (Leica).

RNase R treatment

Five μg of RNA extracted from Caco-2 was treated with 5U RNase R (Biosearch Technologies) for 10 minutes at 37°C or mock-treated and 1 U/μL of Ribolock (Thermo Fisher Scientific) was added, followed by standard ethanol precipitation. The resulting pellets were resuspended in nuclease-free water.

RNA isolation

Total RNA from patient samples was isolated using the miRNeasy FFPE Kit (Qiagen) following the manufacturer's recommendations, including the DNase I treatment. Total RNA from cell lines were isolated using the TRIzol (Invitrogen) or RNeasy Plus Mini Kit (Qiagen). Quantitative and qualitative assessments of the RNA were performed by using a 2200 TapeStation System (Agilent) and Nanodrop One Microvolume UV-Vis Spectrophotometer (Thermo Fisher Scientific).

Sanger sequencing

Sanger sequencing across BSJs of five previously undescribed circRNA candidates was performed. Primer sequences are listed in Supplementary Table S1. Five hundred ng of total RNA from the HT29 was used for M-MLV (Thermo Fisher Scientific) reverse-transcription following the manufacturer's instructions using random primers. 1:5 of the resultant cDNA was used as template for standard PCR. PCR products were visualized on agarose gels, purified using the GeneJet PCR Purification Kit (Thermo Fisher Scientific), and Sanger sequenced using the service Eurofins Genomics.

Ki67 IHC

IHC was performed on a Ventana BenchMark Ultra instrument (Ventana Medical System). The primary antibody used was MIB-1: M7240 (DAKO; RRID:AB_2631211) diluted 1:50. After deparaffination and rehydration of the sections, antigen retrieval was done using CC1 buffer (Ventana Medical System) at 99°C for 48 minutes. Slides were treated with inhibitors to endogenous peroxidase followed by incubation with primary antibody for 32 minutes at 36°C. Amplification was done using OptiView system (Ventana Medical System). Visualization was done using OptiView DAB IHC Detection Kit 760/700 (Ventana Medical System) and Hematoxylin was used as counterstain.

Gene and circRNA expression analyses using NanoString nCounter

A custom CodeSet of capture and reporter probes was designed to target 70 circRNAs, including the top 50 and 20 most abundant circRNAs in the human colon as detected by RNA-seq (GSE77661; ref. 14) and microarray technology (GSE126094, GSE142837, and GSE138589; refs. 15–17), respectively. In addition, tissue-specific marker genes (SATB2, KRT8, CDX2, VIM, ACTA2, and DES) and reference genes (ACTB, PUM1, MRPL19, and SF3A1) were included (Supplementary Table S2). RNA samples were analyzed using the nCounter SPRINT (NanoString Technologies) according to the manufacturer's instructions using a 20 hours hybridization time. Normalization of data was performed in nSolver Analysis Software 4.0 (NanoString Technologies) using the most stable reference genes (PUM1, MRPL19, and SF3A1) and the positive control probes.

In situ hybridization

The cellular localization of circZNF91, circSLC8A1, circITGA7, and circHIPK3 was investigated by chromogenic in situ hybridization using BaseScope high-definition RED procedure (Advanced Cell Diagnostics) using the BaseScope probes (BA-Hs-ZNF91-E4-circRNA and BA-Hs-SLC8A1-circRNA-Junc-C1, BA-Hs-ITGA7-circRNA-E4-C1, and BA-Hs-HIPK3-E2-circRNA-Junc, respectively) following the manufacturer's recommendations. Data were collected using a Hamamatsu NanoZoomer XR digital slide scanner and analyzed using NDPview 2 software version 2.7.52 for Macintosh.

GeoMx digital spatial profiling

The NanoString GeoMx platform was used for digital spatial profiling (18) of ciRS-7, which was added as a custom spike-in to the GeoMx Hu WTA panel. Three 5-μm-thick FFPE sections of colon tumors [10 regions of interest (ROI) in total] and three sections of adjacent normal tissues (6 ROIs in total) were analyzed. Experiments were performed at NanoString Technologies according to the manufacturer's recommendations. Briefly, slides were baked at 37°C overnight to maximize tissue adherence before baking at 65°C for 2 hours to remove paraffin. Slides were then transferred to a Leica Bond Rx for tissue rehydration, antigen retrieval (Tris-EDTA 99°C, 20 minutes), and treatment with proteinase K (1 μg/mL, 15 minutes). After post-fixation with 10% NBF, the slides were incubated with the Human Whole-Transcriptome Atlas probes overnight at room temperature. Following stringent washes (1:1 SSC buffer and 4x formamide), slides were incubated with morphology markers SMA (1:400, cat# 53–9760–82, Thermo Fisher Scientific), PanCK (1: 400, cat# NBP2–33200DL594, Novus Biologicals), and CD45 (1:200, cat# 13917BF, Cell Signaling Technology). In addition, cells were stained with SYTO-13 to visualize the nuclei. Immunofluorescence images were used to select the ROIs, which were then segmented into areas of illumination (AOI; PanCK+, CD45+, and PanCK/CD45) and oligonucleotide tags collected when more than approximately 200 cells were present. Illumina dual-index primers were added using PCR giving each oligo a unique AOI identity. After library purification with Ampure Beads, oligos were sequenced on an Illumina NovaSeq 6000. Fastq files were converted to .dcc files with the GeoMx NGS Pipeline for further data analysis. The third quartile Q3 normalization method was used for expression analysis performed using the GeoMX DSP Analysis Suite (v. 2.5.1.143) according to NanoString guidelines. All 33 AOIs passed the quality control and a total of 18,681 genes were detected.

Manipulation of cell proliferation rates using serum starvation

To reduce cellular proliferation rates, FBS was removed from the normal growth medium of Caco-2 and hASMC. Forty-eight hours after seeding, cells were set to starving conditions for 72 or 96 hours. Furthermore, one 96 hours starving condition was recovered with normal growth medium (10% or 5% FBS) and cultured for an additional 48 hours. For each time point, cells were washed with PBS, trypsinized, and counted before RNA isolation.

Reverse transcriptase quantitative PCR analysis

cDNA for RT-qPCR expression analyses of MKI67 was synthesized from total RNA by MLV-RT (Invitrogen) according to the manufacturer's protocol using random hexamer primers. SYBR Green PCR Master Mix (Thermo Fisher Scientific) was used to perform the RT-qPCR on LightCycler 480 (Roche) using the following primers: MKI67 forward: 5′-ATACGTGAACAGGAGCCAGC-3′ and MKI67 reverse: 5′-CCTCACTCTCATCAGGGTCAGAAG-3′ and, for normalization, MRLP19 forward: 5′-GTTCTATGTTGGAAGTATTCTTCGT-3′ and reverse: 5′-TCTCTGAATGCAAATCCCCAGA-3′.

Library preparation and analyses of RNA-seq data

Ten ng total RNA isolated for each time point during the cell proliferation rate manipulation experiments described above, as well as for the fresh-frozen colon tumors and adjacent normal tissue samples, was subjected to library preparation using the SMARTer Stranded Total RNA-seq kit v3-Pico Input Mammalian (Takara), according to the manufacturer's instructions, with 12 PCR cycles for amplification. The quality of the libraries was controlled using a Bioanalyzer (Agilent) and the KAPA Library Quantification Kit (Roche Diagnostics) was used to quantify the libraries. Sequencing was performed on the NovaSeq 6000 platform (Illumina) using the S1 Reagent Kit v1.5 (300 cycles; cell proliferation rate manipulation experiments) and S4 Reagent Kit v1.5 (300 cycles; colon tumors and adjacent normal tissue samples) to generate 150 bp paired-end reads. The mean sequencing depth was 206 million filtered reads (range, 138 to 333 million reads) for the cell proliferation rate manipulation experiments and 79 million (range, 48 to 96 million reads) for the colon tumors and adjacent normal tissue samples.

The Raw sequencing data were demultiplexed and preprocessed by removing adapter sequences and low-quality bases with a Phred score below 20 using Trim Galore (RRID:SCR_011847). The filtered data were mapped against the human reference genome (hg19/GRCh37) using STAR (RRID:SCR_004463). FeatureCounts with gene annotations from Gencode release 37 was used to quantify gene expression, and circRNAs were quantified using the bioinformatic algorithm CIRI2 (19). The data were normalized by DESeq2 (RRID:SCR_000154; ref. 20) using linear counts only for quantification of gene expression and both circular and linear counts for quantification of circRNA expression.

Total RNA-seq of the samples from patients with prostate cancer was performed previously and experimental procedures have been described (13). These data were analyzed using CIRI2 for circRNA quantification and hg19/GRCh37 was used for annotation. The BSJ-spanning reads were normalized using log2 transformation using the counts per million (cpm) function from edgeR package (v3.30.3).

Statistical analyses and data visualization

Nonparametric and parametric t tests were performed to compute the P values between groups as detailed in the figure legends. Adjusted P values were calculated using the Benjamini–Hochberg method. Unsupervised hierarchical clustering distances and correlations were measured using Pearson correlation values calculated using pheatmap R package (v. 1.0.12, RRID:SCR_016418). Statistical analyses were performed using the R package ggpubr (v. 0.4.0.999) and GraphPad Prism software (v. 9.5, RRID:SCR_002798). Principal component analysis (PCA) plots were calculated using R package ggfortify (v. 0.4.15) and generated using ggplot2 (v. 3.4.0) and GraphPad Prism.

Data availability

NanoString nCounter data from patients and raw RNA-seq data from the cell line starvation experiments, are available at the NCBI Gene Expression Omnibus (GEO, RRID:SCR_005012) repository under accession number GSE233800. The RNA-seq data were obtained from a previous study (13). As the requirement for patient consent was waived in that study, the raw RNA-seq data cannot be deposited in a public repository. The raw RNA-seq data are available from the corresponding author upon reasonable request according to Danish Law and the European Union General Data Protection Regulation. Access to these data will require: (i) Ethical approval of the new research project and (ii) a data sharing agreement for the new project between the data owner (corresponding author) and the PI/collaborator of the project and their institution(s). All other raw data generated in this study are available upon request from the corresponding author.

Development of a custom NanoString nCounter panel for accurate quantification of colon-enriched circRNAs

We designed a custom panel of 70 high-abundance circRNAs based on publicly available RNA-seq and microarray data from colon adenocarcinomas and adjacent normal tissues and added linear reference and tissue marker genes (Supplementary Table S2). Using RNA extracted from bulk tissue sections of five colon tumors and five adjacent normal tissues, we detected 48/50 (96%) circRNAs selected from the RNA-seq data above background levels, defined as the average counts of the negative controls plus two standard deviations (Supplementary Fig. S1A). However, only 2/20 (10%) of the circRNAs selected from the microarray data were detected (Supplementary Fig. S1B). Thus, we decided to focus only on the circRNA candidates from the RNA-seq data, which included many well-known and previously validated circRNAs, such as circCAMSAP1, circCCDC66, circCDYL, circCSNK1G3, circFBXW7, circFNDC3B, circHIPK3, circLPAR1, circSLC8A1, circSMARCA5, circXPO1, circZKSCAN1, circZNF609, and ciRS-7. In addition, we further evaluated the circular nature of the circRNA candidates using RNase R treatment of total RNA from Caco-2 colon cancer cells (Supplementary Fig. S1C) and performed PCR using a divergent primer design and Sanger sequencing across the BSJ for circRNAs not previously validated (Supplementary Fig. S1D). Together, these analyses confirmed a circular nature of all expressed circRNAs, except for the circRNAs from abParts and RPPH1, for which Sanger sequencing data were inconclusive.

Low expression of circRNAs in bulk colon tumors relative to adjacent normal tissues

Next, we compared the NanoString nCounter data from the bulk tissue sections of colon tumors and adjacent normal tissues (Fig. 1A) and found that the overall circRNA abundance was significantly lower in the tumors (Fig. 1B). Indeed, 41 of 48 circRNAs had negative log2 fold changes and 19 of these circRNAs were statistically significant after correcting for multiple testing, whereas only one circRNA, circCAMSAP1, was significantly upregulated (Fig. 1C). Interestingly, the most downregulated circRNA was derived from the muscle-specific gene ITGA7, indicating that muscle cells could potentially be a confounding factor in circRNA analyses of bulk colon tumors and adjacent normal tissues.

Figure 1.

Bulk and spatial analyses of colon tumors and adjacent normal tissues using laser-capture microdissections and NanoString nCounter technology. A, Illustration of the bulk tissue analysis of tumor (n = 5) and adjacent normal tissues (n = 5) for which data are presented in B and C. B, Violin plots comparing the expression levels of the 48 circRNAs. Significance was determined using a two-tailed t test. C, Volcano plot comparing log2-fold changes for each individual circRNA and significance, determined using parametric t tests adjusted by Benjamini–Hochberg correction. D, Illustration of the laser-capture microdissections for which data are presented in E–H. E, Heat map and unsupervised hierarchical clustering of subcompartment-specific markers (CDX2, KRT8, and SATB2 for epithelium, DES and ACTA2 for muscle, and VIM for activated stroma). F, Principal component analysis of the 48 circRNAs. G, Violin plots comparing the expression levels of the 48 circRNAs in the laser-capture microdissected samples and in colon cancer cell lines (n = 4). H, Heat map and unsupervised hierarchical clustering of the 48 circRNAs.

Figure 1.

Bulk and spatial analyses of colon tumors and adjacent normal tissues using laser-capture microdissections and NanoString nCounter technology. A, Illustration of the bulk tissue analysis of tumor (n = 5) and adjacent normal tissues (n = 5) for which data are presented in B and C. B, Violin plots comparing the expression levels of the 48 circRNAs. Significance was determined using a two-tailed t test. C, Volcano plot comparing log2-fold changes for each individual circRNA and significance, determined using parametric t tests adjusted by Benjamini–Hochberg correction. D, Illustration of the laser-capture microdissections for which data are presented in E–H. E, Heat map and unsupervised hierarchical clustering of subcompartment-specific markers (CDX2, KRT8, and SATB2 for epithelium, DES and ACTA2 for muscle, and VIM for activated stroma). F, Principal component analysis of the 48 circRNAs. G, Violin plots comparing the expression levels of the 48 circRNAs in the laser-capture microdissected samples and in colon cancer cell lines (n = 4). H, Heat map and unsupervised hierarchical clustering of the 48 circRNAs.

Close modal

Spatial profiling of circRNAs in colon tumors reveals high expression in muscle and stromal cells

To investigate whether the many differentially expressed circRNAs detected in the bulk analysis reflect a downregulation in cancer cells relative to their normal counterpart or merely reflect differences in cell type compositions, we performed LCM of the same tissue blocks analyzed in bulk above. To this end, we separated normal epithelial cells and cells from the muscularis propria in the adjacent normal tissues as well as subcompartments of epithelial cancer cells and surrounding non-neoplastic stromal cells within the colon tumors (Fig. 1D). The individual laser-capture microdissected samples were then analyzed using the NanoString nCounter panel described above. First, we confirmed that the LCMs were successful by analyzing the expression of the tissue marker genes included in the panel (Fig. 1E; Supplementary Fig. S2). Next, we observed a separation of the different subcompartments when performing a PCA (Fig. 1F), which was driven by marked differences in circRNA abundance (Fig. 1G). The lowest circRNA abundance was observed in the cancer cells followed by normal epithelial cells. On the other hand, we observed a much higher circRNA expression in the muscularis propria and the circRNA expression levels were significantly higher in the stromal fractions within the tumors compared with the cancer cells. In addition, we found that the circRNA expression levels in colon cancer cell lines were very similar to the cancer cell fractions (Fig. 1G). These major differences in circRNA abundancy were reflected in heat map and unsupervised hierarchical cluster analyses where it was observed that approximately two thirds of the circRNAs were expressed at much higher levels in the muscularis propria compared with the other subcompartments (Fig. 1H; Supplementary Fig. S3).

Identification of differentially expressed circRNAs in subcompartments of colon tumors and adjacent normal tissues

Having observed that the overall circRNA expression levels differ systematically between the different subcompartments of colon tumors and adjacent normal tissues, we wanted to identify significantly differentially expressed circRNAs. Most interestingly, we found nine significantly downregulated and two significantly upregulated circRNAs after correction for multiple testing in the cancer cells relative to normal epithelial cells (Fig. 2A). These circRNAs are more likely to be drivers of tumorigenesis as they are not merely reflecting differences in cell-type compositions among the samples as may be the case when performing bulk analysis. Moreover, we found that all nine downregulated circRNAs were also generally lowly expressed in the colon cancer cell lines, whereas the two upregulated circRNAs (circCAMSAP1 and circXPO1) were abundant in all cancer cell lines except one (Supplementary Fig. S4). When comparing with the bulk analysis (Fig. 1AC), we found that most differentially expressed circRNAs (14 of 19) were not significantly downregulated when comparing the cancer cells directly to normal epithelial cells. When comparing the cancer cell fractions with the tumor stroma cell fractions, we observed that many individual circRNAs were significantly more abundant in the stroma and only one circRNA, circABR, was significantly more abundant in the cancer cells (Fig. 2B).

Figure 2.

Differentially expressed circRNAs in subcompartments of colon tumors and adjacent normal tissues. A, Volcano plot comparing the log2-fold changes and significance, determined using parametric t tests adjusted by Benjamini–Hochberg correction, of each individual circRNA between the laser-capture microdissected cancer cell fractions (n = 5) and normal epithelial cell fractions (n = 5). B, Volcano plot comparing the log2-fold changes and significance, determined using parametric t tests adjusted by Benjamini–Hochberg correction, of each individual circRNA between the laser-capture microdissected cancer cell fractions (n = 5) and tumor stroma cell fractions (n = 5). Host genes of differentially expressed circRNAs are indicated.

Figure 2.

Differentially expressed circRNAs in subcompartments of colon tumors and adjacent normal tissues. A, Volcano plot comparing the log2-fold changes and significance, determined using parametric t tests adjusted by Benjamini–Hochberg correction, of each individual circRNA between the laser-capture microdissected cancer cell fractions (n = 5) and normal epithelial cell fractions (n = 5). B, Volcano plot comparing the log2-fold changes and significance, determined using parametric t tests adjusted by Benjamini–Hochberg correction, of each individual circRNA between the laser-capture microdissected cancer cell fractions (n = 5) and tumor stroma cell fractions (n = 5). Host genes of differentially expressed circRNAs are indicated.

Close modal

Muscle content is a strong confounding factor in bulk analyses of circRNAs when comparing colon tumors and adjacent normal tissues

The unexpected finding that most of the differentially expressed circRNAs in the bulk analysis were not found when comparing the cancer cells directly with their normal counterpart, prompted us to further investigate this, and we hypothesized that the high abundance of many circRNAs in muscle cells could be responsible. Indeed, we found that 17 of the 19 significantly downregulated circRNAs in the bulk analysis (Fig. 1C) were among the muscle-enriched circRNAs (Fig. 1H). Thus, if the normal tissue sections contained more muscle cells compared with the tumor sections, this could explain why all but five circRNAs were downregulated only in the bulk analyses and not when comparing cancer cells directly with normal epithelial cells. To explore this, we quantified the muscle content in corresponding hematoxylin and eosin–stained sections and analyzed the expression of DES, a muscle-specific gene, in the samples. These analyses showed that the adjacent normal tissues had a much higher muscle content than the tumors (P < 0.0008; Fig. 3A and B). Moreover, we found a significant correlation between DES and the muscle-enriched circRNAs (Pearson's R = 0.34, P < 0.0001; Fig. 3C and D), whereas no association was observed for the colon epithelium marker, CDX2 (Pearson's R = −0.09, P = 0.11; Fig. 3E). To further explore how differences in muscle content may confound bulk tissue analyses, we performed total RNA-seq of six fresh-frozen colon tumors and paired normal tissues for which the muscle content varied substantially (Fig. 4AD). Again, we observed a strong downregulation of most circRNAs in tumors compared with normal tissues that contain relatively more muscle cells and, like in the NanoString nCounter analysis on bulk tissues, circITGA7 was the most downregulated circRNA (Fig. 4E and F). On the other hand, when comparing tumors with normal tissues that did not contain relatively more muscle cells, much fewer circRNAs were downregulated (Fig. 4G). When analyzing linear RNAs (including mRNAs and long noncoding RNAs), many genes were found up- and downregulated. This was most evident in the comparison, including the normal samples with high muscle content (Fig. 4H and I). For the comparisons, including the samples with high muscle content, the ratio of down- versus upregulated transcripts was much higher for circRNAs than for the linear RNAs (21.2 vs. 3.1). Finally, when considering the mean expression of the most abundant transcripts, the contrast between the sample groups was higher for circRNAs than for linear RNAs (Fig. 4J and K). Together, these analyses imply that muscle cell content is a strong confounding factor in bulk analyses in colon tumors and adjacent normal tissues.

Figure 3.

Muscle content in colon tumors and adjacent normal tissues. A, Hematoxylin and eosin (H&E) staining showing different content of muscle tissue within the five tumor (left) and adjacent normal (right) tissue samples. Estimated area of muscle content was measured and annotated as percentages of the entire tissue areas. B, Scatter plot showing the muscle content in the paired tumors and adjacent normal tissues. Significance was determined using a paired two-tailed t test. C, Scatter plot comparing the expression levels of the muscle-enriched circRNAs (n = 32) with levels of the muscle-specific marker, DES, for each of the tumor and adjacent normal tissue samples. Linear regression was performed to assess potential correlation by determining the Pearson correlation coefficient (R) and an unpaired t test to assess the significance. D, Box plots from normalized NanoString nCounter counts comparing the muscle-enriched circRNAs with DES. Significance was determined using nonparametric t tests using the values from sample T1 (the sample with the lowest muscle content) as reference. E, Scatter plot comparing the expression levels of the muscle-enriched circRNAs (n = 32) with levels of the epithelial marker, CDX2, for each of the tumor and adjacent normal tissue samples. Linear regression was performed to assess potential correlation by determining the Pearson correlation coefficient (R) and an unpaired t test to assess the significance.

Figure 3.

Muscle content in colon tumors and adjacent normal tissues. A, Hematoxylin and eosin (H&E) staining showing different content of muscle tissue within the five tumor (left) and adjacent normal (right) tissue samples. Estimated area of muscle content was measured and annotated as percentages of the entire tissue areas. B, Scatter plot showing the muscle content in the paired tumors and adjacent normal tissues. Significance was determined using a paired two-tailed t test. C, Scatter plot comparing the expression levels of the muscle-enriched circRNAs (n = 32) with levels of the muscle-specific marker, DES, for each of the tumor and adjacent normal tissue samples. Linear regression was performed to assess potential correlation by determining the Pearson correlation coefficient (R) and an unpaired t test to assess the significance. D, Box plots from normalized NanoString nCounter counts comparing the muscle-enriched circRNAs with DES. Significance was determined using nonparametric t tests using the values from sample T1 (the sample with the lowest muscle content) as reference. E, Scatter plot comparing the expression levels of the muscle-enriched circRNAs (n = 32) with levels of the epithelial marker, CDX2, for each of the tumor and adjacent normal tissue samples. Linear regression was performed to assess potential correlation by determining the Pearson correlation coefficient (R) and an unpaired t test to assess the significance.

Close modal
Figure 4.

Bulk analysis on fresh-frozen colon tumors and adjacent normal tissues using RNA-seq. A, Scatter plot showing expression levels of DES in six pairs of colon tumors and adjacent normal tissues. B and C, Groupwise comparisons of DES (B) and KRT8 (C) expression levels; significance levels were determined using an unpaired two-tailed t test. D, Principal component analysis of the 500 most abundant circRNAs. E, Heat map and unsupervised hierarchical clustering of the 500 most abundant circRNAs. F–I, MA plots comparing log2-fold changes between colon tumors and adjacent normal tissues for the circRNAs detected with an average of at least 5 reads (n = 1,271; F and G) and for linear RNAs (H and I), for patients with high muscle content in adjacent normal tissues (F and H) and patients with low muscle content in adjacent normal tissues (G and I). Significance levels were determined using unpaired t tests. J and K, Groupwise comparisons of the 500 most abundant circRNAs (J) and 5,000 most abundant linear RNAs (K). Significance levels were determined using unpaired t tests. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

Figure 4.

Bulk analysis on fresh-frozen colon tumors and adjacent normal tissues using RNA-seq. A, Scatter plot showing expression levels of DES in six pairs of colon tumors and adjacent normal tissues. B and C, Groupwise comparisons of DES (B) and KRT8 (C) expression levels; significance levels were determined using an unpaired two-tailed t test. D, Principal component analysis of the 500 most abundant circRNAs. E, Heat map and unsupervised hierarchical clustering of the 500 most abundant circRNAs. F–I, MA plots comparing log2-fold changes between colon tumors and adjacent normal tissues for the circRNAs detected with an average of at least 5 reads (n = 1,271; F and G) and for linear RNAs (H and I), for patients with high muscle content in adjacent normal tissues (F and H) and patients with low muscle content in adjacent normal tissues (G and I). Significance levels were determined using unpaired t tests. J and K, Groupwise comparisons of the 500 most abundant circRNAs (J) and 5,000 most abundant linear RNAs (K). Significance levels were determined using unpaired t tests. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

Close modal

Cellular proliferation rates directly modulate circRNA expression levels

Having observed that most circRNAs were highly abundant in muscle cells and relatively highly expressed in the tumor stroma but lowly expressed in the normal epithelial cells and the least abundant in the cancer cell fractions, we hypothesized that the proliferation rates of the cells directly impact on circRNA abundancy. In line with this, circRNAs are known to be very stable (21, 22) but slowly generated (23), which may result in an accumulation in quiescent cells and a dilution in rapidly proliferating cells. In support of this hypothesis, we found a negative correlation between Ki67-staining intensities and circRNA abundance in the different subcompartments of colon cancer and adjacent normal tissues (Supplementary Fig. S5). However, to test the hypothesis directly, we manipulated the proliferation rates of cancer cells and smooth muscle cells grown in culture by serum starvation (Fig. 5A) and measured circRNA expression levels using RNA-seq. First, we validated that the serum starvation indeed slowed down the proliferation of the cells without affecting their viability (Fig. 5B and C; Supplementary Fig. S6) and, in support of the hypothesis, we found a global accumulation of the circRNAs when preventing the cells from proliferating and that the circRNA levels were diluted again once we allowed the cells to continue proliferating (Fig. 5DK). Only one circRNA did not follow this pattern among the top 50 most abundant circRNAs in the RNA-seq data (Fig. 5F); however, when inspecting it in the UCSC Genome Browser, it became apparent that this circRNA is a false positive caused by nearby homologous genes (CDK11A and CDK11B) giving rise to a linear splicing that is misinterpreted as a circRNA by the bioinformatic algorithms as previously described (24). Of note, these changes in global circRNA expression were largely unrelated to changes in the cognate linear host genes (Fig. 5G and K). Together, these data show that circRNAs are diluted in fast proliferating cells and accumulate in quiescent cells.

Figure 5.

Cellular proliferation rates directly modulate circRNA expression levels. A, Illustration of the experimental setup using colon cancer cells (Caco-2) and human aortic smooth muscle cells (hASMC). B and C, Measurements of cell numbers (B) and MKI67 expression levels using RNA-seq (C) throughout the experiments. D–G, RNA-seq data from the different time points for the experiments on Caco-2 cells. D, Heat map and unsupervised hierarchical clustering of circ-to-linear ratios (z-scores) of the top 50 most abundant circRNAs. E, Violin plots comparing circ-to-linear ratios of the top 500 most abundant circRNAs. Significance was determined using Wilcoxon tests. F, Scatter plot comparing circ-to-linear ratios of the top 50 most abundant circRNAs. G, Scatter plot comparing the expression levels of the cognate linear host genes of the top 50 most abundant circRNAs. Decreased circ-to-linear ratios and host gene expression levels upon starvation are indicated by the red lines. H–K, RNA-seq data from the different time points for the experiments on hASMC cells. H, Heat map and unsupervised hierarchical clustering of circ-to-linear ratios (z-scores) of the top 50 most abundant circRNAs. I, Violin plots comparing circ-to-linear ratios of the top 500 most abundant circRNAs. Significance was determined using Wilcoxon tests. J, Scatter plot comparing circ-to-linear ratios of the top 50 most abundant circRNAs. K, Scatter plot comparing the expression levels of the cognate linear host genes of the top 50 most abundant circRNAs. Decreased circ-to-linear ratios and host gene expression levels upon starvation are indicated by the red lines. ****, P < 0.0001.

Figure 5.

Cellular proliferation rates directly modulate circRNA expression levels. A, Illustration of the experimental setup using colon cancer cells (Caco-2) and human aortic smooth muscle cells (hASMC). B and C, Measurements of cell numbers (B) and MKI67 expression levels using RNA-seq (C) throughout the experiments. D–G, RNA-seq data from the different time points for the experiments on Caco-2 cells. D, Heat map and unsupervised hierarchical clustering of circ-to-linear ratios (z-scores) of the top 50 most abundant circRNAs. E, Violin plots comparing circ-to-linear ratios of the top 500 most abundant circRNAs. Significance was determined using Wilcoxon tests. F, Scatter plot comparing circ-to-linear ratios of the top 50 most abundant circRNAs. G, Scatter plot comparing the expression levels of the cognate linear host genes of the top 50 most abundant circRNAs. Decreased circ-to-linear ratios and host gene expression levels upon starvation are indicated by the red lines. H–K, RNA-seq data from the different time points for the experiments on hASMC cells. H, Heat map and unsupervised hierarchical clustering of circ-to-linear ratios (z-scores) of the top 50 most abundant circRNAs. I, Violin plots comparing circ-to-linear ratios of the top 500 most abundant circRNAs. Significance was determined using Wilcoxon tests. J, Scatter plot comparing circ-to-linear ratios of the top 50 most abundant circRNAs. K, Scatter plot comparing the expression levels of the cognate linear host genes of the top 50 most abundant circRNAs. Decreased circ-to-linear ratios and host gene expression levels upon starvation are indicated by the red lines. ****, P < 0.0001.

Close modal

Verification of spatial expression patterns using single-molecule in situ analyses and GeoMx digital spatial profiling

We performed single-molecule in situ hybridization for four circRNAs (circHIPK3, circITGA7, circSLC8A1, and circZNF91) with different expression patterns. As expected, circSLC8A1 was almost exclusively expressed in the muscle cells whereas circZNF91 was specific for cancer cells and normal epithelial cells (Fig. 6A and B). circITGA7 was only detectable in muscle cells whereas circHIPK3 was present at relatively high levels in all compartments, in line with the NanoString data (Supplementary Fig. S7A and S7B). In addition, we performed GeoMx digital spatial profiling using a custom-designed spike-in for ciRS-7 (Fig. 7), which was possible as ciRS-7 is relatively abundant and do not have a protein coding linear counterpart (25). The GeoMx data confirmed that ciRS-7 is most abundant in the tumor stroma and further indicated that within the stroma it is more abundant in CD45-negative cells than positive cells (Fig. 7A and B). Moreover, we found that ciRS-7 was absent in normal epithelial cells and present in muscle cells, in line with previous findings (Fig. 7C and D; ref. 6).

Figure 6.

Single-molecule in situ hybridization for circZNF91 and circSLC8A1 in colon adenocarcinomas and adjacent normal tissues. A and B, Brightfield and fluorescent microscopy images at ×200 magnifications. A, A strong signal from circSLC8A1 can be observed in muscle cells and a weak signal can be observed in the tumor stroma, whereas the epithelium and the cancer cells were negative. B, A signal from circZNF91 can be observed in the epithelium and cancer cells, whereas most muscle and stromal cells were negative.

Figure 6.

Single-molecule in situ hybridization for circZNF91 and circSLC8A1 in colon adenocarcinomas and adjacent normal tissues. A and B, Brightfield and fluorescent microscopy images at ×200 magnifications. A, A strong signal from circSLC8A1 can be observed in muscle cells and a weak signal can be observed in the tumor stroma, whereas the epithelium and the cancer cells were negative. B, A signal from circZNF91 can be observed in the epithelium and cancer cells, whereas most muscle and stromal cells were negative.

Close modal
Figure 7.

Stromal and muscle enrichment of ciRS-7 captured by digital spatial profiling. A and B, Fluorescent antibodies were used for choosing specific AOIs for each specific ROI selected from colon cancer (A) and adjacent normal (B) tissues. PanCK+ (blue), CD45+ (green), and CD45PanCK (pink) stainings were used for the colon cancer tissues, whereas SMA+ (pink) was used together with PanCK+ and CD45+ for the adjacent normal areas. C, Bar plot showing the ciRS-7–normalized counts for each AOI across the ROIs defined in the colon cancer samples. D, Bar plot showing the ciRS-7–normalized counts for each AOI across the ROIs defined in the adjacent normal tissue samples. Unpaired t tests were performed to assess statistical significance.

Figure 7.

Stromal and muscle enrichment of ciRS-7 captured by digital spatial profiling. A and B, Fluorescent antibodies were used for choosing specific AOIs for each specific ROI selected from colon cancer (A) and adjacent normal (B) tissues. PanCK+ (blue), CD45+ (green), and CD45PanCK (pink) stainings were used for the colon cancer tissues, whereas SMA+ (pink) was used together with PanCK+ and CD45+ for the adjacent normal areas. C, Bar plot showing the ciRS-7–normalized counts for each AOI across the ROIs defined in the colon cancer samples. D, Bar plot showing the ciRS-7–normalized counts for each AOI across the ROIs defined in the adjacent normal tissue samples. Unpaired t tests were performed to assess statistical significance.

Close modal

Muscle content impacts bulk tissue analyses of prostate and bladder cancer

Having observed that nearly all differentially expressed circRNAs in the bulk analysis of colon tumors and adjacent normal tissues were explained by differences in muscle content and a dilution in the fast-proliferating cancer cells, we speculated that this aspect of our findings could have strong implications for the analysis of other tumors that may invade and destroy surrounding muscle cells, like gastric, prostate, esophageal, and bladder cancers. To investigate this, we compared the mean expression of the 32 muscle-enriched circRNAs with the levels of DES in bulk tumor tissues from a cohort of 172 patients with prostate cancer and found a strong correlation, which could be observed for all different stages (Fig. 8A). The same could be observed for another marker of smooth muscle cells (ACTA2; Fig. 8B), whereas no correlation was observed when comparing with KLK3, NKX3–1, and KRT8 (markers of prostate epithelial cells; Fig. 8CE). Thus, our analyses underscore that the high content of circRNAs in muscle cells is not only a confounding factor in bulk analyses of colon cancer, but also highly relevant for analyses of prostate cancer. Indeed, there are several examples of circRNAs, which are highly abundant in muscle cells that have been shown to be downregulated in muscle invading cancers. One example is circSLC8A1, which we show is specific for muscle cells. This circRNA was previously found to be downregulated in prostate tumors relative to adjacent normal tissues (26) and, additionally, in late-stage relative to early-stage bladder cancer (27). However, we speculate that this may reflect the relative proportions of muscle cells in the samples, rather than a selective downregulation within cancer cells. Accordingly, using single-molecule in situ hybridization, we found circSLC8A1 to be absent in prostate and bladder cancer cells, whereas it is highly abundant in adjacent muscle cells (Supplementary Fig. S8 and S9). Moreover, circSLC8A1 was strongly correlated with markers of muscle cells but not cancer cells in the RNA-seq data from patients with prostate cancer (Supplementary Fig. S10).

Figure 8.

Muscle and cancer marker correlations to circRNAs in patients with prostate cancer. A–E, Scatter plots comparing the mean expression levels of the muscle-enriched circRNAs (n = 32) with levels of the muscle-specific markers DES (A) and ACTA2 (B), and the epithelial markers KLK3 (C), NKX3.1 (D), and KRT8 (E) for each of the 172 prostatic tumor tissue samples. Linear regression was performed to assess potential correlation and the Pearson correlation coefficient (R). Type refers to tissue sample type where 0 indicates adjacent normal, 1 Indicates localized prostate cancer, and 2 Indicates primary tumor sample from patient with metastatic disease.

Figure 8.

Muscle and cancer marker correlations to circRNAs in patients with prostate cancer. A–E, Scatter plots comparing the mean expression levels of the muscle-enriched circRNAs (n = 32) with levels of the muscle-specific markers DES (A) and ACTA2 (B), and the epithelial markers KLK3 (C), NKX3.1 (D), and KRT8 (E) for each of the 172 prostatic tumor tissue samples. Linear regression was performed to assess potential correlation and the Pearson correlation coefficient (R). Type refers to tissue sample type where 0 indicates adjacent normal, 1 Indicates localized prostate cancer, and 2 Indicates primary tumor sample from patient with metastatic disease.

Close modal

Hundreds of studies have recently analyzed solid tumors and adjacent normal tissues to discover differentially expressed circRNAs. However, none of these studies investigated circRNA expression systematically within different cell types and subcompartments within these complex tissues, most likely because single-cell analyses and spatial transcriptomics are particularly challenging when studying circRNAs (7, 8). To allow for spatial profiling of circRNAs, we performed LCM followed by NanoString nCounter analysis within different subcompartments of colon tumors and adjacent normal tissues. These analyses revealed striking differences in circRNA expression between the different subcompartments, which are not limited to differential expression of individual circRNAs, as we also observed major differences in overall circRNA abundance. Importantly, we found that quiescent smooth muscle cells have a very high content of circRNAs, whereas fast proliferating cancer cells have much lower levels of circRNAs. By contrasting the data from the laser capture microdissected tissues with data from bulk analyses of the same tumors and adjacent normal tissues, we found that the vast majority of the differentially expressed circRNAs in the analyses of bulk tissues were a result of a relatively lower amount of muscle cells in the tumors. These conclusions were supported by total RNA-seq data from six fresh-frozen colon tumors and paired normal tissues for which the muscle content varied substantially.

Of note, differences in muscle cell content among samples may not only confound the bulk analyses of colon tumors as we also observed strong correlations between markers of muscle cells and circRNA expression in prostate cancer. This is particularly problematic as fewer muscle cells are observed in samples representing the later stages of this disease and, therefore, the data can easily be misinterpreted. One example of this is circSLC8A1, which has been shown to be downregulated in prostate tumors relative to adjacent normal tissues (26) and in late-stage bladder cancer (27) in analyses based on bulk tissues. However, here we show that this circRNA is abundant in smooth muscle cells and correlates strongly with muscle cell markers, while being undetectable in colon, bladder, and prostate cancer cells using single-molecule in situ hybridization. Therefore, our data emphasize the limitations of using bulk tissues for studying differential circRNA expression in solid tumors and suggest that in situ analyses should always supplement bulk analyses before turning to functional studies based on overexpression in cell lines and xenograft models, which may not have any immediate physiological relevance.

The overall differences in circRNA abundance that we observed between the different subcompartments, in part, reflect that circRNAs are tissue and cell type-specific (28), which can be mirrored by tissue-specific expression patterns of their host genes as exemplified in our data by SATB2 and circSATB2, which are both absent in the muscularis propria. However, we show directly, for the first time, that most of the overall differences in circRNA abundancy are explained by a dilution effect where the very stable (21, 22) but slowly generated circRNAs (23) accumulate in quiescent cells while not reaching steady-state levels in rapidly proliferating cells, in line with what have previously been suggested (29).

Although many studies using RNA-seq have found a global downregulation of circRNAs in cancer relative to control tissues (13, 29–35), this is rarely the case in studies using microarrays. We speculate that these discrepancies result from most microarray studies using quantile normalization, which tend to cancel out gross differences between sample groups.

Together, our data implicate that the expression levels of many circRNAs are related to a passive accumulation in quiescent cells or a dilution effect in fast proliferating cells rather than an active selection due to tumor-suppressor or oncogenic properties within the cancer cells. However, individual circRNAs could still have malignant driver functions. Here, we found that circCAMSAP1 and circXPO1 were more abundant in the cancer cells relative to normal epithelial cells and that they are expressed at relatively high levels in most colon cancer cell lines. This occurred despite these circRNAs being subject to dilution in fast proliferating cells, indicating that they may be actively selected in cancer cells and have oncogenic potential. Indeed, both circRNAs have been previously described as oncogenic in various tumors (36–41), and circCAMSAP1 was recently shown to promote colorectal cancer growth and to be significantly correlated with poor survival and advanced TNM stage (42). Moreover, we speculate that the relatively high abundance of circRNAs observed in the tumor stroma could be functionally relevant in the pathobiology of cancer. This subcompartment contains many different cell types, including cancer-associated fibroblasts (CAF), endothelial cells and various immune cells, such as T and B lymphocytes, together encompassing the tumor microenvironment (TME). The cellular composition of the TME orchestrates tumor progression and metastasis, and CAFs have, for instance, been shown to restrict the accumulation of T cells near cancer cells (43). Here, we show that ciRS-7 is abundant in tumor stroma using GeoMx digital spatial profiling, in line with our previous data (6), and further show that, whereas it appears to be present in some immune cells, it is more abundant in the CD45-negative fractions, which mostly include CAFs. Thus, future studies should investigate the potential roles of ciRS-7 and other circRNAs within the TME using more complex models such as knockout mice and organoids rather than cell lines and xenografts.

In conclusion, the findings presented here unveil a robust distinction of circRNA expression patterns within subcompartments of colon tumors and adjacent normal tissues. Namely, that muscle cells and the activated stroma account for the highest abundance of circRNAs, a fact that has been overlooked until now, which have strong implications for differential expression analyses of circRNAs in bulk tissues and may lead to misinterpretations unless supported by spatial data or controlling for differences in muscle content.

J.L. García-Rodríguez reports grants from Købmand i Odense Johann og Hanne Weimann Født Seedorfds Legat during the conduct of the study. U. Korsgaard reports grants from Dansk Kræftforskningsfond during the conduct of the study. L. Dyrskjøt reports grants from Ferring, Natera, C2i Genomics, Photocure, and AstraZeneca, as well as personal fees from Ferring, MSD, UroGen, AstraZeneca, Pfizer, Roche, and BioXpedia outside the submitted work. L.S. Kristensen reports grants from The Lundbeck Foundation, Riisfort Fonden, Magda Sofie Og Aase Lütz's Mindelegat (Fond), and The Novo Nordisk Foundation during the conduct of the study. No disclosures were reported by the other authors.

J.L. García-Rodríguez: Data curation, formal analysis, funding acquisition, investigation, visualization, methodology, writing–original draft, writing–review and editing. U. Korsgaard: Data curation, formal analysis, validation, investigation, methodology, writing–review and editing. U. Ahmadov: Data curation, formal analysis, visualization, writing–review and editing. M.T. Jarlstad Olesen: Data curation, visualization, writing–review and editing. K.-G. Dietrich: Data curation, formal analysis, methodology, writing–review and editing. E.B. Hansen: Data curation, formal analysis, writing–review and editing. S.M. Vissing: Data curation, writing–review and editing. B.P. Ulhøi: Resources, validation, writing–review and editing. L. Dyrskjøt: Resources, validation, writing–review and editing. K.D. Sørensen: Resources, validation, writing–review and editing. J. Kjems: Funding acquisition, investigation, writing–review and editing. H. Hager: Conceptualization, resources, funding acquisition, validation, investigation, methodology, writing–review and editing. L.S. Kristensen: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

The authors thank biotechnicians Birgit Roed Sørensen and Mariana Semenova for excellent technical assistance, Dr. S. Seeler for designing qPCR primers, Dr. A. Færch Nielsen for a careful and critical reading of the article, Dr. L. Alonso-Herranz for kindly providing the hASMC cell line, and Dr. Antoine de Morree for valuable scientific input and discussion. The Danish Cancer Biobank is acknowledged for biological material and for data regarding handling and storage. This work was supported by grants (to L.S. Kristensen) from the Lundbeck Foundation (R307–2018–3433), Riisfort Fonden, and Magda Sofie Og Aase Lütz's Mindelegat (Fond), the Novo Nordisk Foundation ODIN program (NNF20SA0061466), and the Danish Cancer Society (R304-A17698), from the Danish Cancer Society (R269-A15768 to J. Kjems), from Dansk Kræftforskningsfond (to U. Korsgaard), and from the Købmand i Odense Johann og Hanne Weimann Født Seedorffs Legat (to J.L. García-Rodríguez).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.

Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).

1.
Kristensen
LS
,
Jakobsen
T
,
Hager
H
,
Kjems
J
.
The emerging roles of circRNAs in cancer and oncology
.
Nat Rev Clin Oncol
2022
;
19
:
188
206
.
2.
Kristensen
LS
,
Andersen
MS
,
Stagsted
LVW
,
Ebbesen
KK
,
Hansen
TB
,
Kjems
J
.
The biogenesis, biology, and characterization of circular RNAs
.
Nat Rev Genet
2019
;
20
:
675
91
.
3.
Wu
W
,
Ji
P
,
Zhao
F
.
CircAtlas: an integrated resource of one million highly accurate circular RNAs from 1,070 vertebrate transcriptomes
.
Genome Biol
2020
;
21
:
101
.
4.
Dong
R
,
Ma
XK
,
Li
GW
,
Yang
L
.
CIRCpedia v2: an updated database for comprehensive circular RNA annotation and expression comparison
.
Genomics Proteomics Bioinformatics
2018
;
16
:
226
33
.
5.
Hansen
TB
,
Kjems
J
,
Damgaard
CK
.
Circular RNA and miR-7 in cancer
.
Cancer Res
2013
;
73
:
5609
12
.
6.
Kristensen
LS
,
Ebbesen
KK
,
Sokol
M
,
Jakobsen
T
,
Korsgaard
U
,
Eriksen
AC
, et al
.
Spatial expression analyses of the putative oncogene ciRS-7 in cancer reshape the microRNA sponge theory
.
Nat Commun
2020
;
11
:
4551
.
7.
Nielsen
AF
,
Bindereif
A
,
Bozzoni
I
,
Hanan
M
,
Hansen
TB
,
Irimia
M
, et al
.
Best practice standards for circular RNA research
.
Nat Methods
2022
;
19
:
1208
20
.
8.
Verboom
K
,
Everaert
C
,
Bolduc
N
,
Livak
KJ
,
Yigit
N
,
Rombaut
D
, et al
.
SMARTer single-cell total RNA-sequencing
.
Nucleic Acids Res
2019
;
47
:
e93
.
9.
Dahl
M
,
Daugaard
I
,
Andersen
MS
,
Hansen
TB
,
Gronbaek
K
,
Kjems
J
, et al
.
Enzyme-free digital counting of endogenous circular RNA molecules in B-cell malignancies
.
Lab Invest
2018
;
98
:
1657
69
.
10.
Conn
V
,
Conn
SJ
.
SplintQuant: a method for accurately quantifying circular RNA transcript abundance without reverse transcription bias
.
RNA
2019
;
25
:
1202
10
.
11.
Szabo
L
,
Salzman
J
.
Detecting circular RNAs: bioinformatic and experimental challenges
.
Nat Rev Genet
2016
;
17
:
679
92
.
12.
Kristensen
LS
.
Profiling of circRNAs using an enzyme-free digital counting method
.
Methods
2021
;
196
:
11
16
.
13.
Hansen
EB
,
Fredsøe
J
,
Okholm
TLH
,
Ulhøi
BP
,
Klingenberg
S
,
Jensen
JB
, et al
.
The transcriptional landscape and biomarker potential of circular RNAs in prostate cancer
.
Genome Med
2022
;
14
:
8
.
14.
Zheng
Q
,
Bao
C
,
Guo
W
,
Li
S
,
Chen
J
,
Chen
B
, et al
.
Circular RNA profiling reveals an abundant circHIPK3 that regulates cell growth by sponging multiple miRNAs
.
Nat Commun
2016
;
7
:
11215
.
15.
Chen
Z
,
Ren
R
,
Wan
D
,
Wang
Y
,
Xue
X
,
Jiang
M
, et al
.
Hsa_circ_101555 functions as a competing endogenous RNA of miR-597–5p to promote colorectal cancer progression
.
Oncogene
2019
;
38
:
6017
34
.
16.
Ma
Z
,
Han
C
,
Xia
W
,
Wang
S
,
Li
X
,
Fang
P
, et al
.
circ5615 functions as a ceRNA to promote colorectal cancer progression by upregulating TNKS
.
Cell Death Dis
2020
;
11
:
356
.
17.
Yang
H
,
Li
X
,
Meng
Q
,
Sun
H
,
Wu
S
,
Hu
W
, et al
.
CircPTK2 (hsa_circ_0005273) as a novel therapeutic target for metastatic colorectal cancer
.
Mol Cancer
2020
;
19
:
13
.
18.
Merritt
CR
,
Ong
GT
,
Church
SE
,
Barker
K
,
Danaher
P
,
Geiss
G
, et al
.
Multiplex digital spatial profiling of proteins and RNA in fixed tissue
.
Nat Biotechnol
2020
;
38
:
586
99
.
19.
Gao
Y
,
Zhang
J
,
Zhao
F
.
Circular RNA identification based on multiple seed matching
.
Brief Bioinform
2018
;
19
:
803
10
.
20.
Love
MI
,
Huber
W
,
Anders
S
.
Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2
.
Genome Biol
2014
;
15
:
550
.
21.
Memczak
S
,
Jens
M
,
Elefsinioti
A
,
Torti
F
,
Krueger
J
,
Rybak
A
, et al
.
Circular RNAs are a large class of animal RNAs with regulatory potency
.
Nature
2013
;
495
:
333
8
.
22.
Jeck
WR
,
Sorrentino
JA
,
Wang
K
,
Slevin
MK
,
Burd
CE
,
Liu
J
, et al
.
Circular RNAs are abundant, conserved, and associated with ALU repeats
.
RNA
2013
;
19
:
141
57
.
23.
Zhang
Y
,
Xue
W
,
Li
X
,
Zhang
J
,
Chen
S
,
Zhang
JL
, et al
.
The biogenesis of nascent circular RNAs
.
Cell Rep
2016
;
15
:
611
24
.
24.
Kristensen
LS
,
Okholm
TLH
,
Veno
MT
,
Kjems
J
.
Circular RNAs are abundantly expressed and upregulated during human epidermal stem cell differentiation
.
RNA biology
2018
;
15
:
280
91
.
25.
Barrett
SP
,
Parker
KR
,
Horn
C
,
Mata
M
,
Salzman
J
.
ciRS-7 exonic sequence is embedded in a long non-coding RNA locus
.
PLos Genet
2017
;
13
:
e1007114
.
26.
Wang
D
,
Yan
S
,
Wang
L
,
Li
Y
,
Qiao
B
.
circSLC8A1 Acts as a tumor suppressor in prostate cancer via sponging miR-21
.
Biomed Res Int
2021
;
2021
:
6614591
.
27.
Lu
Q
,
Liu
T
,
Feng
H
,
Yang
R
,
Zhao
X
,
Chen
W
, et al
.
Circular RNA circSLC8A1 acts as a sponge of miR-130b/miR-494 in suppressing bladder cancer progression via regulating PTEN
.
Mol Cancer
2019
;
18
:
111
.
28.
Salzman
J
,
Chen
RE
,
Olsen
MN
,
Wang
PL
,
Brown
PO
.
Cell-type–specific features of circular RNA expression
.
PLos Genet
2013
;
9
:
e1003777
.
29.
Bachmayr-Heyda
A
,
Reiner
AT
,
Auer
K
,
Sukhbaatar
N
,
Aust
S
,
Bachleitner-Hofmann
T
, et al
.
Correlation of circular RNA abundance with proliferation–exemplified with colorectal and ovarian cancer, idiopathic lung fibrosis, and normal human tissues
.
Sci Rep
2015
;
5
:
8057
.
30.
Vo
JN
,
Cieslik
M
,
Zhang
Y
,
Shukla
S
,
Xiao
L
,
Zhang
Y
, et al
.
The landscape of circular RNA in cancer
.
Cell
2019
;
176
:
869
81
.
31.
Das Mahapatra
K
,
Pasquali
L
,
Sondergaard
JN
,
Lapins
J
,
Nemeth
IB
,
Baltas
E
, et al
.
A comprehensive analysis of coding and non-coding transcriptomic changes in cutaneous squamous cell carcinoma
.
Sci Rep
2020
;
10
:
3637
.
32.
Dahl
M
,
Husby
S
,
Eskelund
CW
,
Besenbacher
S
,
Fjelstrup
S
,
Côme
C
, et al
.
Expression patterns and prognostic potential of circular RNAs in mantle cell lymphoma: a study of younger patients from the MCL2 and MCL3 clinical trials
.
Leukemia
2022
;
36
:
177
88
.
33.
Rickert
D
,
Bartl
J
,
Picard
D
,
Bernardi
F
,
Qin
N
,
Lovino
M
, et al
.
Circular RNA profiling distinguishes medulloblastoma groups and shows aberrant RMST overexpression in WNT medulloblastoma
.
Acta Neuropathol
2021
;
141
:
975
8
.
34.
Ahmadov
U
,
Bendikas
MM
,
Ebbesen
KK
,
Sehested
AM
,
Kjems
J
,
Broholm
H
, et al
.
Distinct circular RNA expression profiles in pediatric ependymomas
.
Brain Pathol
2021
;
31
:
387
92
.
35.
Song
X
,
Zhang
N
,
Han
P
,
Moon
BS
,
Lai
RK
,
Wang
K
, et al
.
Circular RNA profile in gliomas revealed by identification tool UROBORUS
.
Nucleic Acids Res
2016
;
44
:
e87
.
36.
Chen
H
,
Zhang
P
,
Yu
B
,
Liu
J
.
The circular RNA circXPO1 promotes tumor growth via sponging MicroRNA-23a in prostate carcinoma
.
Front Oncol
2021
;
11
:
712145
.
37.
Huang
Q
,
Guo
H
,
Wang
S
,
Ma
Y
,
Chen
H
,
Li
H
, et al
.
A novel circular RNA, circXPO1, promotes lung adenocarcinoma progression by interacting with IGF2BP1
.
Cell Death Dis
2020
;
11
:
1031
.
38.
Jiang
Y
,
Hou
J
,
Zhang
X
,
Xu
G
,
Wang
Y
,
Shen
L
, et al
.
Circ-XPO1 upregulates XPO1 expression by sponging multiple miRNAs to facilitate osteosarcoma cell progression
.
Exp Mol Pathol
2020
;
117
:
104553
.
39.
Chen
Z
,
Xu
W
,
Zhang
D
,
Chu
J
,
Shen
S
,
Ma
Y
, et al
.
circCAMSAP1 promotes osteosarcoma progression and metastasis by sponging miR-145–5p and regulating FLI1 expression
.
Mol Ther Nucleic Acids
2021
;
23
:
1120
35
.
40.
Wang
Y
,
Yan
Q
,
Mo
Y
,
Liu
Y
,
Wang
Y
,
Zhang
S
, et al
.
Splicing factor derived circular RNA circCAMSAP1 accelerates nasopharyngeal carcinoma tumorigenesis via a SERPINH1/c-Myc–positive feedback loop
.
Mol Cancer
2022
;
21
:
62
.
41.
Wang
Y
,
Li
X
,
Wang
H
,
Zhang
G
.
CircCAMSAP1 promotes non–small cell lung cancer proliferation and inhibits cell apoptosis by sponging miR-1182 and regulating BIRC5
.
Bioengineered
2022
;
13
:
2428
39
.
42.
Zhou
C
,
Liu
HS
,
Wang
FW
,
Hu
T
,
Liang
ZX
,
Lan
N
, et al
.
circCAMSAP1 promotes tumor growth in colorectal cancer via the miR-328–5p/E2F1 axis
.
Mol Ther
2020
;
28
:
914
28
.
43.
Joyce
JA
,
Fearon
DT
.
T-cell exclusion, immune privilege, and the tumor microenvironment
.
Science
2015
;
348
:
74
80
.
This open access article is distributed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0) license.

Supplementary data