Introduction

Currently, the incidence of BC ranks second-highest among that of cancers worldwide, with 2,261,419 cases every year1. The incidence of BC is increasing year by year, and the age of onset is decreasing. Exploring new molecular markers of BC is beneficial for predicting prognosis accurately and monitoring curative effects. Therefore, finding an effective, rapid, noninvasive and specific marker is urgent and is crucial for the diagnosis, prognosis evaluation and drug resistance evaluation of BC2,3.

The choice of drugs for BC patients varies according to individual circumstances4. To date, the main treatments for BC are surgery, radiotherapy and chemotherapy5. Chemotherapy is a standard method for BC treatment6. There are many commonly used chemotherapy drugs for BC, including anthracyclines (doxorubicin, epirubicin, doxorubicin liposomes, etc.), paclitaxel drugs (paclitaxel, docetaxel, paclitaxel liposomes, and nab-paclitaxel) and fluorouracil (5-FU, capecitabine). In addition, there are targeted drugs such as trastuzumab and pertuzumab7,8 for BC. However, patients are developing resistance to conventional drugs. Chemotherapy resistance is one of the main reasons for clinical treatment failure and poor prognosis in BC patients9. This resistance might be due to alterations in several main regulatory pathways, such as PI3K/AKT10,11,12. Recently, some studies have found that certain circRNAs are strongly associated with resistance to a number of anticancer drugs, ranging from traditional chemotherapy drugs to targeted and immunotherapy drugs13,14,15,16,17.

CircRNAs are endogenous RNAs characterized by a covalent ring structure. Compared with other RNAs, circRNAs are less abundant, but circRNAs exhibit the advantage of high tissue specificity18. Recently, many researchers have indicated that certain circRNAs in different tumors might play essential roles in tumor cell proliferation, metastasis and drug resistance19. Several studies have suggested that circRNAs affect the development of drug resistance and prognosis of BC patients. Upregulated or downregulated circRNAs are involved in tumor growth and drug resistance, affecting the prognosis of breast cancer patients. Liang et al. showed that circKDM4C could inhibit BC proliferation and doxorubicin resistance in vitro, and this circRNA is a tumor suppressor in BC20. Wang et al. found that miR-142 regulated the WWP1 and PI3K/AKT genes10. Circ-WAC could act as a sponge for miR-142 and decrease the inhibitory effect of miR-142 on its target WWP1. In addition, if triple-negative breast cancer patients expressed a high level of miR-142, their overall survival time was longer than that of other patients with low miR-142 expression. Additionally, Yang et al. showed that circ-CDR1as was involved in breast carcinogenesis and sensitivity to cisplatin in vivo. Knockdown of circ-CDR1as might increase the sensitivity of drug-resistant BC cells by reducing REGγ expression by eliminating the competition of miR-721. Some articles have reported an association between changes in circRNAs and changes in drug resistance status in BC11,20,21,22,23,24,25,26,27,28,29,30,31,32. However, no article has summarized the specific mechanisms and modalities of circRNAs involved in BC drug resistance.

In the related meta-analysis, the involvement of circRNAs in BC was included, and we investigated the undiscovered prognostic value of circRNAs in BC. Several studies found that the expression of certain circRNAs was associated with increased drug resistance and a poor prognosis in BC patients10,22,23,32. Preclinical and clinical observational studies have shown that circRNA expression profiles can help identify patients at possible high risk for chemotherapy-resistant BC11,33. Therefore, we attempted to conduct a comprehensive systematic review and meta-analysis of published studies on circRNA-mediated chemoresistance in BC.

Materials and methods

Registration

We have registered the protocol on PROSPERO. Our registration number is CRD42022295180.

Data search strategy

We searched all relevant articles through the PubMed, Embase, and Web of Science online databases that were published before 13 October 2022. The following entry words were used: (1) “Breast Neoplasms” or “Breast Cancer” or “BC." (2) “RNA, Circular” or “circRNAs” or “hsa circ”; (3) “Resistance, Drug” or “Drug resistance” or “chemoresistance.". The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines were used to conduct search strategies.

Inclusion and exclusion criteria

The selection criteria for inclusion in the literature were as follows: (1) studies that involved the effect of circRNAs on drug resistance or drug sensitivity in BC; (2) studies that collected clinical samples or involved in vitro preclinical analysis; and (3) studies that involved the effect of circRNA on the prognosis of BC. The criteria for exclusion were as follows: (1) duplicate studies; (2) reviews, editorials, opinions, case studies, and reports; unpublished materials, uninterpretable data, conference proceedings, or theses; (3) articles without complete information; (3) studies that did not indicate whether circRNA expression was upregulated or downregulated; (4) studies that did not include specific drug resistance changes; and (5) studies in languages other than English.

Data extraction

Two researchers (Z.Z. and H.J.) extracted the data independently. When necessary, divergences were resolved by a third investigator (J.X.). The extracted information was as follows: (1) first author, publication year, circRNA, number of patients, detection methods for circRNAs, HR, CI; (2) follow-up time and outcomes; and (3) clinicopathological features, including TNM stage and T classification. When the results were not directly shown in the articles for HRs and 95% CIs, survival data were extracted from Kaplan‒Meier plots using Engauge Digitizer 4.1 software. The Excel program file of Tierney et al34. was then be used to calculate the HRs and 95% CIs.

Quality assessment

Two independent investigators (Z.Z. and H.J.) used the Newcastle Ottawa Scale (NOS) to assess the quality of the articles for meta-analysis. If one study had a total score of > 6 points, it was considered high quality. When necessary, divergences were resolved by a third investigator (J.X.).

Statistical analysis

Statistical analysis was performed using Stata 15.0. Data in the form of Kaplan‒Meier survival curves were converted to HRs and 95% CIs. Pooled outcome data were generated for forest plots to assess the prognostic value of circRNAs in BC. Heterogeneity tests were obtained by Cochran's Q test and Higgins I2. According to the rule, a random-effects model was used to generate pooled results if the I2 value was > 50%, and a fixed-effects model was used if the I2 value was <  = 50%. A p value < 0.05 was used to determine statistical significance.

Results

Selection of studies

Figure 1 shows a flowchart for study selection. Through the search strategy, 520 articles were identified from PubMed, Embase, and Web of Science. After deduplication, screening criteria were used to review 291 potentially eligible studies. A careful selection of 137 articles was made by finalizing 34 full-text studies with available information according to PRISMA guidelines. Of these 34 articles, four studies were excluded since they were about other cancers. After multistep screening, the remaining 30 articles were used for systematic reviews10,11,12,20,21,22,23,24,25,26,27,28,29,30,31,32,35,36,37,38,39,40,41,42,43,44,45,46,47, of which nine were used for meta-analysis.

Figure 1
figure 1

Flowchart of trial selection.

Full size image

Study characteristics and quality assessment

All included studies were collected until 13 October 2022 (the details of the description of the 30 included studies are shown in supplemental appendix 1). The 11 chemotherapy drugs used in the studies included 5-FU, lapatinib, adriamycin, doxorubicin, paclitaxel, cisplatin, monastrol, tamoxifen, docetaxel, trastuzumab, and oxaliplatin. Of these, paclitaxel is the most commonly used chemotherapeutic agent in clinical practice, while lapatinib and oxaliplatin are the least used. A total of 2077 BC tissue samples were included in the analysis. Seven of 30 studies documented clinical stage, including 187 in stage I, 464 in stage II, and 211 in stage III. Thirty studies used reverse transcription-polymerase chain reaction (RT‒qPCR) to detect circRNA, and only one study used the raw sequencing reads. Nine studies with survival curves were included in the meta-analysis, containing a total of 1962 individuals.

Preclinical and clinical investigation of circRNA expression

A total of 17 different cell lines were used in 30 studies to explore circRNA expression and its association with drug resistance and associated pathways or proteins/axes. MCF-7 was the most commonly used cell line, while the U343 and U251 cell lines were the least commonly used. The experimental methods used in these studies include western blot, transfection and vector construction, flow cytometry, Transwell, ELISA, cytotoxicity assay, dual‑luciferase reporter assay, RNA pull‑down, RIP assay, IHC assay, RNase R treatment assay, 5-ethynyl-2′-deoxyuridine (EdU) assay, fluorescence in situ hybridization (FISH) assay, exosome tracing and blockade of exosome secretion.

After excluding duplicate circRNAs, our systematic review included a total of 30 different circRNAs, 28 of which were associated with increased drug resistance and a poor prognosis in breast cancer patients when their expression was upregulated, while only 2 were associated with increased drug resistance and a poor prognosis in breast cancer patients when their expression was downregulated.

BC chemoresistance and drug‑regulated genetic pathways

In these 30 studies, a total of 32 circRNAs were reported, and excluding duplicate circRNAs, a total of 30 circRNAs were reported, and these circRNAs led to resistance to 11 drugs through 28 pathways or associated proteins/axes. (Table 1).

Table 1 Genetic pathways, proteins or axes involved in BC drug resistance.
Full size table

Findings of prognosis analysis

Nine circRNAs were used for meta-analysis. Seven circRNAs were upregulated, and two were downregulated (Table 2 demonstrates details of prognostic research). The Newcastle–Ottawa Scale (NOS) was used to evaluate the quality of the included research (Table 3). The results showed that they all qualified for meta-analysis. The results of the meta-analysis showed that both the upregulated and downregulated groups were at risk for poor prognosis (HR = 1.37, 95% Cl: 0.80–2.36, I2 = 63.7%). There was significant heterogeneity between the studies. Therefore, we classified all circRNAs into "enhanced resistance"-related circRNAs and "attenuated resistance"-related circRNAs according to the expression of circRNAs. Subgroup analysis was performed according to the upregulation or downregulation of circRNAs. Interestingly, the heterogeneity was significantly reduced after performing a subgroup analysis (Fig. 2), which suggested that circRNAs could be used to determine the prognosis of BC patients (upregulated circRNAs (HR = 2.24, 95% Cl: 1.34–3.75, I2 = 0%) and downregulated circRNAs (HR = 0.61, 95% Cl: 0.45–0.83, I2 = 0%) were associated with poor BC prognosis.). All four circRNAs in the upregulated group were highly expressed in tumor tissues, and they affected gene pathways that promoted drug resistance in breast cancer cells, while the two circRNAs in the downregulated group were expressed at low levels in tumor tissues and affected gene pathways that inhibited proliferation, metastasis and drug resistance in breast cancer cells.

Table 2 Basic features of studies for prognostic analysis.
Full size table
Table 3 Quality assessment of included studies using the Newcastle Ottawa Scale checklist.
Full size table
Figure 2
figure 2

Pooled HRs for the overall survival of patients in the included studies.

Full size image

Sensitivity analysis and publication bias

We also performed a sensitivity analysis for OS. No significant changes were observed compared to previous results after each study was removed (Fig. 3). In addition, we used funnel plots to assess publication bias. Each dot represents one study. Nine studies fell within the 95% confidence interval. The reason for the poor symmetry may be due to the inconsistent effect of circRNAs in the upregulated and downregulated groups (Fig. 4). Finally, we performed Begg's test, which showed P = 0.004 (< 0.05), and Egger's test suggested publication bias, which may be because far more circRNAs were upregulated than downregulated among the nine circRNAs (Fig. 5).

Figure 3
figure 3

Sensitivity analysis for the involved studies.

Full size image
Figure 4
figure 4

Funnel plot of publication bias related to the association between the expression of circRNAs and the prognosis of patients with BC.

Full size image
Figure 5
figure 5

Egger's test for publication bias.

Full size image

Discussion

Studies have shown that abnormal circRNA expression is important in tumor cell proliferation, metastasis and cancer recurrence in BC patients10,11,12,20,21,22,23,24,25,26,27,28,29,30,31,32,35,36,37,38,39. Many studies have also confirmed that several specific circRNAs are consistently expressed in human tissues and blood. Therefore, circRNAs have the chance to be excellent biomarkers for BC diagnosis, prognosis and drug resistance assessment3,33,48,49.

Some studies have concentrated on the effects of circRNAs on chemoresistance in breast, cervical50, colorectal51,52, gastric53,54, lung55, oral56, ovarian57, pancreatic58 and prostate59 cancers. In this study, we collected relevant articles before 25 October 2021 and conducted a systematic review and meta-analysis, hoping to find clues about the value of circRNAs as biomarkers for BC prognosis. In the systematic review, studies incorporating 30 circRNAs, including 28 upregulated circRNAs and two downregulated circRNAs, were included. Most studies investigated only one circRNA, while only one study focused on more than one circRNA12. Our systematic review focused on pharmacological modulation pathways, including MAPK, PI3K/AKT, AKT and AGE-RAGE, in BC chemotherapy resistance and sensitivity.

Several studies have shown that target genes of upregulated circ-00006528 play a role in the MAPK and PI3K/AKT gene pathways. Further validation showed that the expression of circ-0006528 showed a negative correlation with miR-7-5p in adriamycin resistance. Another study showed a significant increase in both phosphorylated and total AKT protein in some circ-AMOTL1-overexpressing cells, suggesting that AKT might be a key factor in adjusting the resistance effect. Thus, circ-AMOTL1 affected the expression of proapoptotic (BAX and BAK) and antiapoptotic (BCL-2) factors associated with AKT59. This suggested that circ-AMOTL1 might be important in paclitaxel resistance in BC cells by affecting the AKT pathway, promoting antiapoptotic proteins and inhibiting proapoptotic proteins. In addition, data from a study showed that circ-ABCB1, circ-cEPHA3.1 and circ-EPHA3.2 might sponge several significantly expressed miRNAs related to drug resistance through the PI3K-AKT and AGE signaling pathways and lead to doxorubicin resistance59. They also found that the expression of RNA molecules transcribed from this region might be due to DNA amplification in doxorubicin-treated cells. These results are beneficial for subsequent research on the mechanisms of drug resistance in BC.

Nine circRNAs related to prognosis were included in the meta-analysis and were critical to the development of drug resistance. Among them, seven were upregulated (circ-MMP11, circ-WAC, circ-UBE2D2, circ-0006528, circTRIM28, circCDYL2, circ-0001598), and two were downregulated (circ-LARP4, circ-KDM4C). Certain cancer-related genes could increase susceptibility to breast cancer, leading to poorer survival rates. In our analysis, the results showed that the overall HR (95% CI) of upregulated circRNAs was 2.24 (1.34, 3.75), and that of downregulated circRNAs was 0.61 (0.45, 0.83), suggesting that both upregulated circRNAs and downregulated circRNAs could predict poorer cancer prognosis. It is worth noting that if circRNAs could be used as prognostic biomarkers of breast cancer, their clinical application prospects would be very broad. Other typical clinical indicators of tumor status are susceptible to change, but the expression of circRNAs is stable60. Steps should be taken to comprehensively assess the role of circRNAs as biomarkers for BC prognosis and drug-resistance assessment.

Some shortcomings must be acknowledged. First, in our study, all of the samples we collected were from Asian populations. Samples collected from a single source may not be able to distinguish between regional and racial differences and ethnic differences. Second, the method used to detect circRNAs was RT‒PCR, except for one study that used raw sequencing reads. The relative homogeneity of the methods used to detect circRNAs may affect the value of the assay results. Third, the meta-analysis involved a relatively small sample size and was limited by the number of available articles. In addition, the number of studies included in the meta-analysis was so low that the results of publication bias using funnel plots may not be meaningful.

Overall, circRNAs, as stably expressed molecules, are expected to be biomarkers for breast cancer prognosis. The relationship between circRNA expression and breast cancer features needs to be further investigated, and the practical value of circRNAs in evaluation BC drug resistance and prognosis needs to be further explored.

Conclusion

Currently available evidence suggests that circRNAs might be considered potential prognostic biomarkers for BC patients and that there is a significant association between the expression of circRNAs and the prognosis of breast cancer patients. We anticipate that our findings might contribute to BC treatment.