Identification of spliceosome components pivotal to breast cancer survival

ABSTRACT

Cancer cells employ alternative splicing (AS) to acquire splicing isoforms favouring their survival. However, the causes of aberrant AS in breast cancer are poorly understood. In this study, the METABRIC (Molecular Taxonomy of Breast Cancer International Consortium) data were analysed with univariate feature selection. Of 122 analysed spliceosome components, U2SURP, PUF60, DDX41, HNRNPAB, EIF4A3, and PPIL3 were significantly associated with breast cancer survival. The top 4 four genes, U2SURP, PUF60, DDX41, and HNRNPAB, were chosen for further analyses. Their expression was significantly associated with cancer molecular subtype, tumour stage, tumour grade, overall survival (OS), and cancer-specific survival in the METABRIC data. These results were verifiable using other cohorts. The Cancer Genome Atlas data unveiled the elevated expression of PUF60, DDX41, and HNRNPAB in tumours compared with the normal tissue and confirmed the differential expression of the four genes among cancer molecular subtypes, as well as the associations of U2SURP, PUF60, and DDX41 expression with tumour stage. A meta-analysis data verified the associations of U2SURP, PUF60, and HNRNPAB expression with tumour grade, the associations of PUF60, DDX41, and HNRNPAB expression with OS and distant metastasis-free survival, and the associations of U2SURP and HNRNPAB expression with relapse-free survival. Experimentally, we demonstrated that inhibiting the expression of the four genes separately suppressed cell colony formation and slowed down cell growth considerably in breast cancer cells, but not in immortal breast epithelial cells. In conclusion, we have identified U2SURP, PUF60, DDX41, and HNRNPAB are spliceosome-related genes pivotal for breast cancer survival.

Introduction

The RNA splicing process is a two-step transesterification reaction exerted by a large and dynamic ribonucleoprotein (RNP) complex, the spliceosome [1]. The building blocks of the spliceosome are small nuclear ribonucleoproteins (snRNPs), which comprise Sm proteins and one or two small nuclear RNAs (snRNAs). The snRNPs form complexes with other splicing factors to accomplish a splicing event. The vast majority of pre-mRNA splicing events are implemented by 5 major snRNPs, U1, U2, U4, U5, and U6 snRNPs [1]. The snRNPs and their associated spliceosome components assemble on the pre-mRNA in a temporary manner by recognizing specific sequence elements on the pre-mRNA, including 5ʹ splice sites (5’SS), branch point sequence (BPS), polypyrimidine tract, and 3ʹ splice sites (3’SS) [2]. Moreover, auxiliary splicing factors can interact with regulatory sequences on the pre-mRNA, such as exonic/intronic splicing enhancers (ESEs or ISEs) or exonic/intronic splicing silencers (ESSs or ISSs), to alter the inclusion or exclusion of exons [1].

More than 95% of human protein-encoding genes undergo alternative splicing (AS), which is a fundamental cause for proteome diversity and a central driving force of evolution [3]. Splicing errors are one of the common reasons for genetic diseases, including cancer [4]. Once underappreciated, AS has been currently recognized to render advantages of survival, proliferation, metastasis, and therapy resistance to cancer cells [3]. However, our knowledge of AS in cancer is still limited.

Cancer cells employ multiple means to acquire the splicing isoforms that favour their survival and spread [3,5]. First, genetic variations at or near splice sites or at ESE, ISE, ESS, ISS of oncogenes or tumour suppressor genes can alter the expression of splicing isoforms and change the outcomes of a gene. Second, mutations on spliceosome components can cause a systematic change of splicing isoforms and promote cancer malignancy. The exemplary genes are the recently-discovered recurrent mutations on SF3B1 and U1 snRNA [6–8]. Third, the aberrant expression of splicing factors resulted from copy number variations, epigenetic regulation, or other reasons can induce genome-wide splicing alterations of cancer-related genes. Finally, epigenetic changes of histone or DNA modifications can impact the recruitment of splicing factors or change the transcription rate to regulate the expression of splicing isoforms.

While aberrant splicing is gaining more attention in the cancer field, it is still poorly understood how breast cancer cells acquire splicing isoforms benefiting their survival. Breast cancer is the most common malignancy in women. Based on the expression of oestrogen receptor (ER), progesterone receptor (PR), human epidermal growth factor receptor (HER2), breast cancer is classified into multiple molecular subtypes. The survival time of breast cancer patients has been extended considerably by targeting ER and HER2. However, triple-negative breast cancer (ER, PR, and HER2 negative, TNBC) is still a great challenge in cancer treatment due to the lack of targets [9,10]. Targeting RNA splicing could be a potential way to improve the survival time of this subtype of cancer patients [5,11]. Comparing with gene mutations, we focus more on the aberrant expression of spliceosome components as expression deregulation will be easier to target than mutations in practice. Therefore, we aim to identify deregulated spliceosome-related factors that are crucial for breast cancer survival in this study.

Results

Identification of spliceosome-related factors associated with the survival of breast cancer patients

Among the 1904 patients with gene expression information in the METABRIC (Molecular Taxonomy of Breast Cancer International Consortium) dataset, 1351 patients had non-missing data in their clinicopathological features and were kept in the analysis (Tables S2 & S3). Univariate log-rank tests revealed that ER status, HER2 status, PR status, tumour stage, histologic grade, PAM50, integrative cluster, ontotree code, surgery type, and chemotherapy were closely associated with the cancer survival (Table S4). We initially chose 137 proteins, which were identified in various purified splicing complexes repeatedly. Of the genes for these 137 proteins, 15 genes were not expressed in all the examined breast cancer samples. Thus, 122 genes were added one at a time to the following analysis with a multivariable Cox regression model (Table S5). The analyses showed U2SURP, PUF60, DDX41, HNRNPAB, EIF4A3, PPIL3, SF3B4, SF3B2, NAA38, and SNRPB were genes that were significantly associated with cancer patient survival after the p-values were adjusted for covariates in the model (Table 1). We controlled the false discovery rate (FDR) at 15% with sharpened q-values, and the first six genes met the criterion for further exploration (Table 1).

Table 1.

Spliceosome components that are identified to be significantly associated with breast cancer survival probability using the METABRIC dataset. The adjusted p-values, hazard ratio (HR), sharpened q-values and the complexes the proteins associated with are shown

Gene Symbol	Gene ID	HR	p-value	Sharpened q-value	Complex
U2SURP	23350	1.13	0.0002801	0.036	E to C
PUF60	22827	1.08	0.0017118	0.115	E to C
DDX41	51428	1.10	0.0034501	0.115	B to P
HNRNPAB	3182	1.11	0.0041437	0.115	E to C
EIF4A3	9775	1.09	0.0042485	0.115	B to P
PPIL3	53938	0.90	0.0064667	0.145	B to P
SF3B4	10262	1.07	0.0184531	0.441	E to P
SF3B2	10992	1.07	0.0266696	0.622	E to P
NAA38	84316	0.92	0.03284	0.713	Pre-B to C
SNRPB	6628	1.07	0.048069	1.000	E to P

Proliferating cells produce more mRNAs and proteins to satisfy their growth requirement. Thus, the upregulated spliceosomal components might be a result of cancer cell proliferation. To correct this potential confounder, we added Ki67 expression to the Cox model. We chose the top 4 genes for this analysis and found that their p-values were larger than before but remained significant at the 5% level (Table S6). Another potential confounder is cancer cellularity. The expression profiles are from bulk tissues, which are mixtures of the tumour and stromal components. After adjusting for cellularity, the four genes remained statistically significant in the Cox model (Table S6). Therefore, after correcting potential confounders, we identified that U2SURP, PUF60, DDX41, HNRNPAB are closely associated with breast cancer survival in the METABRIC cohort.

In an effort to validate these results, a similar analysis was carried out using the TCGA (The Cancer Genome Atlas) dataset. In this replication analysis, we were able to control age, stage, ER status and AJCC nodes in the Cox regression. The top four genes, U2SURP, PUF60, DDX41, and HNRNPAB from the METABRIC analysis, were analysed, and the p-values in this Cox model were listed in Table S7. As only 4% of women in the TCGA cohort had follow-up time longer than 10 years, whereas in the METABRIC, the percentage was 48%, only PUF60 remained statistically significant at a 5% level in the TCGA cohort (Table S7). After correcting the potential confounder of Ki67, PUF60 was still the only gene that remained significant.

The expression of U2SURP, PUF60, DDX41, and HNRNPAB is associated with the clinicopathological features of breast cancer patients

Next, the top 4 genes, U2SURP, PUF60, DDX41, and HNRNPAB, were chosen for further analyses. We examined the associations of their expression with available clinicopathological features of the METABRIC data. The results showed that all the four genes were differentially expressed among the subtypes of tumours in the METABRIC dataset whenever the traditional 3-gene (ER, PR, and HER2) method or the Pam50 + claudin-low method (p < 2 x 10⁻¹⁶ in all cases) was used (Fig. 1A–D, Table S8). A similar result was observed when the TCGA dataset was used for the validation (Fig. 1E, H, p < 0.01 in all cases). The differential expression patterns agreed well in both datasets (Fig. 1A–H, Table S9). In both datasets, the expression of the four genes was generally low in the normal-like and luminal-A subtypes, which usually respond well to the therapies, but high in the HER2 and basal-like subtypes (mostly are TNBC), which typically have a poor prognosis. Moreover, the expression of U2SURP and PUF60 was highest in the basal-like subtype of breast cancer in both METABRIC and TCGA datasets (Fig. 1A, B, E, F), indicating these two factors may be crucial for basal type breast cancer. The HER2(+) type of cancer expressed the highest amount of HNRNPAB in both METABRIC and TCGA dataset, indicating HNRNPAB may be more critical in HER2(+) type breast cancer (Fig. 1D, H). Although DDX41 was differentially expressed among the breast cancer subtypes in both datasets, the differences were not as salient as the other three genes (Fig. 1C, G). To better illustrate the differences, we performed student t-tests to compare the expression between pairwise subtypes. The p-values and means were listed in Table S9.

Figure 1.

Violin plots show that the expression of U2SURP (A, E), PUF60 (B, F), DDX41 (C, G), and HNRNPAB (D, H) was associated with breast cancer molecular subtypes. In the METABRIC dataset (A, B, C, D), the patient numbers of normal breast-like (Normal), Luminal A (LumA), Luminal B (LumB), HER2-positive (HER2), Basal-like (Basal), and Claudin-low (Claudin-low) subtypes are 140, 679, 461, 220, 199, 199, respectively. In the TCGA dataset (E, F, G, H), the patient numbers of normal breast-like (Normal), Luminal A (LumA), Luminal B (LumB), HER2-positive (HER2), and Basal-like (Basal) subtypes are 40, 559, 207, 82, 190, respectively

The results of the METABRIC dataset also showed that the expression of the four genes was tightly associated with other clinicopathological features, including neoplasm histologic grade and tumour stage (see detailed information in Table S8). The expression levels of all four genes were nicely associated with neoplasm histologic grade separately in the METABRIC dataset (Fig. 2A–D, Table S8), increasing with the histologic grade. Because there was no histologic grade information in the TCGA dataset, we used another meta-analysis (referred to as BRCT2010 thereafter) to validate these results [12]. The results showed that U2SURP, PUF60, and HNRNPAB expression also increased when the histologic grade elevated, but the expression of DDX41 was not related to the tumour grade in this dataset (Fig. 2E, H). The pairwise comparisons between tumour histologic grades were also performed, and the results were shown in Table S9.

Figure 2.

Violin plots show the associations of the expression of U2SURP (A, E), PUF60 (B, F), DDX41 (C, G), and HNRNPAB (D, H) with tumour grade in breast cancer. There are 165, 740, and 927 patients with grade 1, 2, 3 tumours in the METABRIC dataset (A, B, C, D), and 198, 534, and 312 patients with grade 1, 2, and 3 tumours in BRCT2010 data (E, F, G, H), respectively

The expression of the four genes was also significantly correlated with tumour stages in the METABRIC dataset. If not considering stage IV owing to its small sample size (n = 19), the expression of the four genes increased with the tumour stage (Fig. 3A–D, Table S9). Similar results were observed in the U2SURP, PUF60, and DDX41 genes using the TCGA dataset, but not in the HNRNPAB gene (Fig. 3E, H). We also performed the pairwise comparisons between stages and listed the p values and means in Table S9.

Figure 3.

Violin plots show the associations of the expression of U2SURP (A, E), PUF60 (B, F), DDX41 (C, G), and HNRNPAB (D, H) with tumour stage in breast cancer. In the METABRIC dataset (A, B, C, D), there are 475, 800, 115, and 19 stages I, II, III, and IV patients, respectively, and 181, 611, 244, and 19 stages I, II, III, and IV patients in the TCGA dataset (E, F, G, H), respectively

Owing to the differential expression of the four genes in the subtypes of breast cancer, the association of the four genes’ expression with cancer survival may be due to their differential presentation of subtypes. Therefore, we examined the effect of gene expression in each subtype using the multivariable Cox model, controlling for the same covariates as we did before. The p-values for subtype-specific gene expression level on overall survival are shown in Table S10. The results indicated that U2SURP expression had a significant effect on survival in both the Basal and LumA subtypes. PUF60 and DDX41 were significant in the LumB subtype. HNRNPAB was significant in the LumA subtype. However, the sample sizes for certain subtypes were small and might influence the statistical power.

We also compared the expression of U2SURP, DDX41, PUF60, and HNRNPAB between the normal tissue and primary solid tumours using the TCGA data. The results from 112 available pairs of normal tissue and matched tumours showed that the expression of PUF60, DDX41, and HNRNPAB, but not U2SURP was significantly elevated in the tumours (Fig. 4A, B).

Figure 4.

Comparisons of the expression of U2SURP (A), PUF60 (B), DDX41 (C), and HNRNPAB (D) between the tumours (Tumour) and their normal tissue counterparts (Normal). The p-values of the paired Student’s t-tests (n = 112) were shown in the graphs

U2SURP, PUF60, DDX41, and HNRNPAB are independent predictors of breast cancer survival

Subsequently, we divided the patients into 2 groups for each gene according to the medians of their gene expression values of the entire METABRIC dataset to visualize the relationship between the gene expression and breast cancer survival. Kaplan-Meier plots showed that the patients with high expression of the four genes all had a low overall survival probability (Fig. 5A–D). Because the METABRIC dataset includes information on cancer-specific death, we further analysed the relevance between the gene expression and cancer-specific survival. Our results showed that the hazard ratios were elevated in all the four cases when cancer-specific survival probability was evaluated, indicating the four genes play an essential role in breast cancer (Fig. 5E, F).

Figure 5.

Kaplan-Meier curves show that the overall and cancer-specific survival probabilities of the patients with high or low expression of U2SURP, PUF60, DDX41, and HNRNPAB. A-D: the overall survival (OS) probability of patients with high or low expression of U2SURP (A), PUF60 (B), DDX41 (C), and HNRNPAB (D) (n = 1904). E-H: The cancer-specific survival probability of patients with high or low expression of U2SURP (E), PUF60 (F), DDX41 (G), and HNRNPAB (H) (n = 1423). The high and low expression groups were divided by the medians of expression values of the corresponding genes. The log-rank test p-values and hazard ratios (Mantel-Haenszel) were shown in the graphs

Because the follow-up time of the TCGA dataset is short, we used another meta-analysis data to further confirm the role of U2SURP, PUF60, DDX41, and HNRNPAB in cancer survival. After dividing the patients into 2 groups by the medians of gene expression data of the entire dataset, the associations of the gene expression with the overall survival (OS), relapse-free survival (RFS), and distant metastasis-free survival (DMFS) were analysed (Fig. 6). The results showed that U2SURP (p < 0.0001, HR = 2.292) and HNRNPAB (p < 0.0001, HR = 2.094) expression were related to RFS (Fig. 6E, H), while PUF60, DDX41, and HNRNPAB were independent predictors of OS and DMFS (Fig. 6B–D, J–M). In the case of U2SURP, its associations with OS and DMFS were not statistically significant, but the p-values were close to 0.05 (p = 0.0913 and HR = 1.450 for OS; p = 0.1066 and HR = 1.267 for DMFS) (Fig. 6A, I).

Figure 6.

Kaplan-Meier curves show the overall, relapse-free, and distant metastasis-free survival probabilities of the patients with high or low expression of U2SURP, PUF60, DDX41, and HNRNPAB. A-D: the overall survival (OS) probabilities of patients with high or low expression of U2SURP (A), PUF60 (B), DDX41 (C), and HNRNPAB (D) (n = 357). E-H: The relapse-free survival (RFS) probability of patients with high or low expression of U2SURP (E), PUF60 (F), DDX41 (G), and HNRNPAB (H) (n = 1592). I-M: The distant metastasis-free survival (DMFS) probability of patients with high or low expression of U2SURP (I), PUF60 (J), DDX41 (K), and HNRNPAB (M) (n = 766). The high and low expression groups were divided by the medians of expression values of the corresponding genes. The log-rank test p-values and hazard ratios (Mantel-Haenszel) were shown in the graphs

Inhibition of U2SURP, PUF60, DDX41, and HNRNPAB expression impaired breast cancer cell proliferation

To further confirm the role of the above four genes in breast cancer, we used cell biological approaches to determine their role in breast cancer cell proliferation. The sgRNAs for each gene were synthesized and cloned into the plentiCRISPRv2 vector. The sgRNA-containing viral particles were used to infect MDA-MB-468 cells, a triple-negative breast cancer cell line, as well as MCF-10A cells, an immortal breast epithelial cells which are generally considered as normal epithelial cells. The results of quantitative PCRs (q-PCRs) showed that all the sgRNAs decreased the mRNA expression of their target genes (Figs. 7A and 8A). Two sgRNAs that showed the most significant repression of gene expression were selected for the further analyses. In the case of DDX41, unlike the other 2 sgRNAs, sgDDX41-3 transduction resulted in 2 peaks in the melting curve of DDX41(data not shown), so we chose sgDDX41-1 and sgDDX41-2 for the following experiments.

Figure 7.

Inhibition of U2SURP, PUF60, DDX41, and HNRNPAB expression in MDA-MB-468 cells decreased the cell colony formation capability and inhibited cell growth. A. Quantitative PCR showed the inhibition efficiencies of 3 sgRNA sequences in MDA-MB-468 cells (n = 3). B. The numbers of colonies that were formed after the MDA-MB-468 cells were transduced with different sgRNAs (n = 3). C. The images of formed colonies by MDA-MB-468 cells that were transduced with different sgRNAs. D. The growth curves of MDA-MB-468 cells that were transduced with different sgRNAs. * p < 0.05; **, p < 0.01; ^##, p < 0.001

Figure 8.

Inhibition of U2SURP, PUF60, DDX41, and HNRNPAB expression in MCF-10A cells did not impair cell growth. A. Quantitative PCR showed the inhibition efficiencies of the sgRNA sequences in MCF-10A cells. B. The growth curves of MCF-10A cells that were transduced with the corresponding sgRNAs. * p < 0.05; **, p < 0.01; ^##, p < 0.001

Because the mRNA expression could not reflect the levels of functional proteins thoroughly, the protein levels of U2SURP, DDX41, and HNRNPAB, whose antibodies were available, were examined next (Fig. S1). In agreement with the qPCR results, the sgRNAs reduced the amounts of the three proteins in MCF-10A cells (Fig. S1), suggesting that the antibodies were specific. Even though the decrease in mRNA levels in MCF-10A cells were moderate compared to MDA-MB-468 cells, U2SURP and HNRNPAB reduced more significantly at the protein level in MCF-10A cells than MDA-MB-468 cells (Figs. 7A, 8A and S1). In MDA-MB-468 cells, sgU2SURP-2, and sgHNRNPAB-1 did not decrease the levels of their target proteins markedly as shown in the mRNA levels (Fig. S1). This discrepancy indicated that the half-lives of the proteins might be different in the two cell lines. At the detection point, the reduction of mRNAs had not yet been translated to the reduction of proteins in MDA-MB-468 cells. However, this would not influence the results of proliferation assays since the assays lasted for 7 days after the detection point, and the sgRNAs caused irreversible genomic alterations and permanent silencing.

Inhibiting U2SURP, DDX41, PUF60, and HNRNPAB expression in MDA-MB-468 cells significantly decreased the colony formation capability of MDA-MB-468 cells (Fig. 7B, C). Furthermore, the growth curves indicated that cell proliferation was significantly blocked after inhibiting the mRNAs of U2SURP, DDX41, PUF60, and HNRNPAB in MDA-MB-468, suggesting that these genes were essential for the maintenance of the unlimited division capability of cancer cells (Fig. 7D).

In contrast, inhibition of these genes did not impair the proliferation of MCF-10A (Fig. 8B), suggesting these four genes were not essential for normal epithelial cell survival. Although the mRNA levels were reduced mildly by the sgRNAs in MCF-10A cells compared with MDA-MB-468 cells (Fig. 8A), U2SURP and HNRNPAB decreased more at the protein levels in MCF-10A cells than in MDA-MB-468 cells (Fig. S1), and DDX41 protein also decreased comparably in the two cell lines (Fig. S1). These results suggested that the growth differences between the two cell lines should not be attributed to the different mRNA knockdown levels between the two cell lines. As shown in the Fig. 8B, sgPUF60-3 even enhanced cell growth slightly while the mRNA level was down to a similar level as sgPUF60-1. Similar observation was also seen in sgDDX41-1 and sgHNRNPAB-2 experiments. Therefore, we concluded that knockdown of the four genes could not inhibit the cell growth of MCF-10A cells.

We also queried 6606 breast cancer samples through cBioPortal website (https://www.cbioportal.org/). Among the 3706 patients who had the copy number profiles of these four genes, there are 1, 6, 4, and 5 tumours having homozygous deletions of U2SURP, PUF60, DDX41, and HNRNPAB (Table S11). The existence of homozygous deletion in the patients suggests that these four proteins are replaceable in the spliceosome. The high-frequency amplification of the four gene loci may be one of the possible reasons for their high expression in breast cancer (Table S11). Adding these pieces of evidence together, we deduced that the four proteins are rather regulatory than core catalytic components in the spliceosome if their functions in breast cancer involve mRNA splicing. Therefore, targeting U2SURP2, DDX41, PUF60, or HNRNPAB has potential use for triple-negative breast cancer treatment.

Discussion

Aberrant mRNA splicing is a common phenomenon in cancer and plays a significant role in cancer progression. However, reports concerning deregulated splicing factors in breast cancer are still scarce [11], and which splicing factors promote breast cancer progression is largely unknown. In this study, we analysed the association of 122 spliceosome-related genes with cancer outcomes using the METABRIC data and identified and validated that U2SURP, PUF60, DDX41, and HNRNPAB are independent predictors for breast cancer prognosis. Furthermore, we ablated the four genes separately in basal-like, triple-negative breast cancer cells, MDA-MB-468, as well as relatively-normal breast epithelial cells, MCF-10A, proving that the four spliceosome-related genes indeed are pivotal to breast cancer survival, but not to breast epithelial cells. Concurrently, we could not proclaim the four factors exert their role in breast cancer via regulation of splicing. The dissection of their biological functions needs to be done in the future.

Three of the top 4 identified factors, U2SURP, PUF60, and HNRNPAB, are recruited to the spliceosome in early phases (Table 1). The step-wise assembly of spliceosome components on the pre-mRNA forms 10 phase-specific complexes, the E (the earliest), A, pre-B, B, B^act, B* (the active form for the first catalytic step), C, C* (the active form for the second catalytic step), P (post-catalytic) and ILS (intron lariat spliceosome) complexes [2]. The early-formed E and A complexes are involved in the multiple recognitions between the conserved splicing sequences of pre-mRNAs and the protein components in the spliceosome. These recognitions confer the splicing events flexibility and are crucial for the regulation of splicing events [1]. Because the core transesterification reactions need to be maintained in cancer cells to satisfy the growth requirement, it is reasonable that cancer cells utilize early-recruited regulatory splicing factor to determine whether a splicing event should happen or not, rather than employ the core catalytic components to gain advantages over the normal cells. For the case of DDX41, an RNA helicase, we are currently not clear when it starts to associate with the spliceosome. An early study showed that DDX41 is a component of C complex, a complex formed after the first catalytic step [13]. However, DDX41 was shown to interact with SF3B1 in a later study. It is well characterized that SF3B1 is recruited in the A complex and dissociates with the spliceosome after the B^act phase, a phase earlier than C complex [14]. Therefore, it is also possible that DDX41 is recruited to the spliceosome in an early phase and stays in the spliceosome until the C complex forms.

Of the four proteins, U2SURP is a U2 snRNP associated protein containing the Ser/Arg rich domain (SR domain). SR proteins bind to the enhancer sequences of pre-mRNA to regulate splicing events [1]. U2SURP interacts with RBM17 and CHERP to stabilize each other and regulate AS [15]. It was shown U2SURP expression was elevated in colon cancer, and it promoted colon tumour growth in vivo and in vitro by regulating UPF3A alternative splicing [16]. In this study, we firstly demonstrated that U2SURP is also a crucial player in breast cancer. Firstly, U2SURP expression is associated with the OS and cancer-specific survival in the METABRIC dataset and associated with RFS in the BRCT2010 dataset (Figs. 5A, E and 6E). Second, U2SURP is differentially expressed in cancer molecular subtypes with significantly high expression in the basal-like breast cancer (Fig. 1A, E), which mostly is triple-negative breast cancer and represents the most therapy-resistant breast cancer subtype [17]. Thirdly, U2SURP expression is correlated to tumour histologic grades and stages (Figs. 2A, E and 3A, E). Stage IV patients expressed lower U2SURP than other stages in the METABRIC and TCGA datasets, mostly because the sample sizes of this stage patients are small (n = 19 in both datasets). Fourthly and most importantly, the ablation of U2SURP inhibited cell growth and decreased the colony formation capability of cancer cells (Fig. 7A–D). Even though we did not see a significant difference in U2SURP expression between the normal and tumour tissue and the associations of U2SURP expression with OS and DMFS in the BRCT2010 data, our results still provided substantial evidence that U2SURP is a biomarker in breast cancer prognosis.

Another well-known regulatory splicing factor family is hnRNPs, which bind to the silencers of pre-mRNA to regulate AS [1]. HNRNPAB is a splicing regulator of cancer-related genes, such as FGFR-2, TERT, and p63 [18–20], as well as a transcriptional [21–24] and translational regulator [25]. In solid cancer, HRNRNPAB promotes the growth of prostate cancer cells [22] and the malignancy of hepatocellular carcinoma [23,24]. Here, we demonstrated that HNRNPAB is upregulated in breast cancer (Fig. 4D), especially in luminal B, HER-2(+), and basal-like breast cancer (Fig. 1D, H). Its expression is also associated with tumour histologic grade (Fig. 2D, H) and tumour stage (Fig. 3D). Patients with high expression of HNRNPAB had considerable low probabilities of OS and cancer-specific survival in the METABRIC cohort (Fig. 5D, H), and low probabilities of OS, RFS, and DMFS in the BRCT2010 cohort (Fig. 6D, H, M). At last, decreasing HNRNPAB expression in MDA-MB-468 cells inhibited cell growth and colony formation (Fig. 7A–D). Therefore, we demonstrated the role of HNRNPAB in breast cancer survival.

The exact functions of DDX41 in the spliceosome are largely unknown. The deficiency of DDX41 caused aberrant exon skipping and retention in hundreds of genes [14]. Germline mutations and genomic deletions of DDX41 frequently occurred in haematological malignancies [14,26,27], indicating DDX41 is a tumour suppressor in haematological malignancies. In solid tumours, several studies found that DDX41 is an oncogene in human colorectal cancer [28,29]. Our study first demonstrated that DDX41 expression is higher in the breast tumour tissue than the normal tissue (Fig. 4C). Although its association with cancer molecular subtype, tumour stage, and tumour histologic grade are not as salient as the other 3 genes (Figs. 1C, G, 2C, G and 3C, G), its expression predicts OS and cancer-specific survival in the METABRIC cohort (Fig. 5C, G) and is associated with OS and DMFS significantly in the BRCT2010 cohort (Fig. 6C, G, K). Consistently, the ablation of DDX41 in cancer cells significantly inhibited tumour growth and colony formation capacity (Fig. 7A–D). These results demonstrated DDX41 promotes cancer progression in breast cancer. It still needs further investigation on whether the functions of DDX41 is involved in its role in the spliceosome since it is also well characterized that DDX41 also regulates INF-β expression to control the innate immune reaction in myeloid dendritic cells [30].

PUF60 is homologous with U2AF65 and can substitute for U2AF65 during the recognition of the 3’SS [1]. PUF60 regulates specific exon usage in cells [31]. It is implicated that PUF60 may play a role in colorectal [32], gastric [33], oesophageal [34], prostate [35], ovarian [36] cancers, and acute myeloid leukaemia [37]. Also, it has been recently shown that PUF60 promoted breast cancer progression while this manuscript was prepared [38]. In our study, we found that PUF60 expression is elevated in breast tumours compared with the normal tissue (Fig. 4B). Its expression is highest in basal-like cancer in both METABRIC and TCGA datasets (Fig. 1B, F). Also, PUF60 expression increases with the tumour stage and grade in the examined datasets (Figs. 2B, F and 3B, F). High expression of PUF60 is associated with significantly shorter OS, cancer-specific survival in the METABRIC cohort (Fig. 5B, F), and associated with OS and DMFS in the BRCT2010 cohort (Fig. 6B, J). Same as the other three genes, inhibiting PUF60 expression in MDA-MB-468 cells reduced cell growth and colony formation (Fig. 7A–D). Our results further validated the critical role of PUF60 in breast cancer.

In summary, we identified U2SURP, PUF60, DDX41, and HNRNPAB are potential biomarkers and drug targets in breast cancer. We further verified these conclusions using other datasets, although some of the associations were not validated due to the intrinsic characteristics of different cohorts. More importantly, cellular function analyses confirmed that these four splicing factors are potential targets in breast cancer, especially for triple-negative cancer. Although the four factors repeatedly purified with splicing complexes, we could not concurrently assert that they fulfill their functions through regulating AS in breast cancer. Their molecular functions in breast cancer still need further investigation.

Materials and methods

Data curation and pre-processing

The Spliceosome Database (http://spliceosomedb.ucsc.edu/) summaries the current mass spectrometry data [39], and 137 spliceosome components were identified in the human spliceosome complexes of different assembly phases (Table S1). The gene expression data and clinical data of the METABRIC and TCGA studies were downloaded from www.cbioportal.org [40]. A dataset (BRCT2010) of a meta-analysis [12] was downloaded from www.kmplot.com. The METABRIC dataset was chosen to identify candidate spliceosome components that are critical for breast cancer patient survival because the patients had longer-term follow-up (median = 115 months) [40] compared with the TCGA dataset (median = 27.7 months) [41] and the treatments were homogenous with respect to clinically relevant groupings by ER and HER2 [42].

Statistical analysis

The METABRIC dataset includes patients’ clinicopathological characteristics, such as ER status, tumour stage, Neoplasm Histologic Grade, and Pam50 + claudin-low subtypes. Because each spliceosome component may be correlated with the clinical characteristics, it is important to control known clinical prognosis factors for survival. Univariate analysis was first performed to select discrete clinicopathological features based on the log-rank test, and then a multivariate Cox regression model was built to include all clinicopathological features with univariate p-value < 0.05. Continuous clinical features, including age at diagnosis, the Nottingham prognosis index, and tumour size, were used as time-varying covariates in the final Cox model. Out of the 137 spliceosome components, expression data of 15 genes were not available in METABRIC data. Gene expression of each of the 122 available genes was added to the final Cox regression one at a time to examine its statistical significance. To address the multiple testing issue (122 tests), we used the sharpened q-value based on the two-stage procedure to controll false discovery rate (FDR) [43]. A cut-off of 15% FDR was used to filter the significant genes. The top components were chosen for further analyses. The gene expression data and clinical information from the TCGA and BCRT2010 studies were used to validate the associations between the clinicopathological features and the candidate gene expression. The comparisons between 2 groups were analysed with paired or unpaired Student t-test, and comparisons among 3 or more groups were analysed with one-way ANOVA. Kaplan-Meier estimator was used to compare the survival probabilities between the high and low gene expression patients (the medians of the gene expression were used to divide the patients). Correlations between two series of continuous data were analysed with the Pearson method. R, Stata and SAS were used to analyse the data. P-value <0.05 was considered as statistically significant.

Cell culture

MDA-MB-468, MCF-10A, and HEK-293 T cells were obtained from the National Infrastructure of Cell Line Resource (Shanghai, China). MDB-MB-468 cells were maintained in L15 medium containing 10% foetal bovine serum (FBS) and 1% penicillin-streptomycin at 37ºC in a humidified incubator with 100% air atmosphere. HEK-293 T cells were maintained in DMEM medium containing 10% foetal bovine serum (FBS) and 1% penicillin-streptomycin at 37ºC in a humidified atmosphere with 5% CO₂. MCF-10A cells were cultured in DMFM/F12 medium containing 10% FBS, 20 ng/mL EGF, 0.5 mg/ml hydrocortisone, 100 ng/mL cholera toxin, 10 µg/mL insulin, and 1% penicillin-streptomycin at 37ºC in a humidified atmosphere with 5% CO₂

Lentiviral vector construction

The plentiCRISPRv2 plasmids (Addgene, USA) were digested with Esp3I (Thermo Fisher Scientific Inc., USA) first and gel purified using the Tiangen Gel Extraction kit (Tiangen Biotech Co., LTD., Beijing, China). Three different sgRNA sequences were tested for each gene (Table S12). For every sgRNA, two oligos were synthesized (Synbio Technologies, Suzhou, China) and annealed. The annealed oligos were diluted and ligated to the Esp3I-digested lentiCRISPRv2 vectors. The ligation products were transformed into Stbl3 bacteria, and the resulting clones were sequenced to confirm the correct insertion of sgRNA sequences.

Viral packaging and transduction

For viral packaging, 5 µg pMD2 (Addgene, USA), 7.5 µg pPAX2 (Addgene, USA), and 10 µg lentiviral vector plasmids were transfected into HEK-293 T cells using CaPO₄ method. After transfection, the cells were maintained at 37ºC in a humidified atmosphere with 5% CO₂ for 3 more days. The medium was then collected and filtered with 0.45 µm filters (Pall, USA). One-fourth volume of viral concentration solution (25% PEG8000/0.75 M NaCl)) was added, and the mixture was incubated at 4ºC overnight. The next day, the mixture was centrifuged at 4ºC, and the supernatant was resuspended in PBS, aliquoted, and store at −80ºC until used. 100 µl viruses were used to transduce 4 × 10⁵ MDA-MB-468 cells in a 6-well plate. After adding the viruses, and the six-well plate was centrifuged at 2000 rpm for 1 hour at 32ºC. The next day, the cells were screened with 0.5 µg/ml puromycin for 4 days.

RNA purification and quantitative reverse transcription-polymerase chain reaction

Cells were washed with cold PBS twice and lysed immediately with 1 ml TRIzol Reagent (Thermo Fisher, USA). The RNA was purified by following the manufacturer’s instructions. PrimeScript RT Master Mix (Takara, Japan) was used for reverse transcription of RNA samples as instructed by the manual. Five hundred nanograms of RNAs were reverse-transcribed for each sample. After dilution, the RT products were used for quantitative PCR (q-PCR) with the iTaq Universal SYBR Green Supermix on a CFX96 Real-Time PCR Detection System (Bio-Rad, USA). The q-PCR primers were listed in Table S13. The data were analysed with the 2^ΔΔCt method.

Western Blot

MDA-MB-468 and MCF-10A cells were harvested immediately using RIPA buffer after puromycin selection. The total protein amounts were quantified with a Bradford Assay Kit (23246, ThermoFisher) and separated in SDS-PAGE gel. After transferring to PVDF membranes, the target proteins were probed with anti-U2SURP (sc-398718, Santa Cruz, 1:500), anti-DDX41(15076S, Cell Signalling, 1:1000), and anti-HNRNPAB (sc-376411, Santa Cruz, 1:400) primary antibodies first and then the corresponding secondary antibody. The signal was developed with a ECL kit and visualized with a ProteinSimple FluoChem E system.

Colony formation and cell proliferation

For colony formation, 500 MDA-MB-468 cells/well were seeded in a six-well plate. The cells were cultured at 37ºC for 14 days. The medium was changed every 7 days. On day 14, the cells were washed with PBS twice, fixed with methanol for 30 minutes, and stained with crystal violet (Beyotime Biotechnology, Shanghai, China) for 45 minutes. After washing with distilled water, the plate was photographed, and the colony numbers were counted. The experiment was done in triplicate and repeated at least twice.

For cell proliferation analysis, 1 × 10⁴ cells/well were seeded in 96-well plates. The cell viability was measured with CCK-8 kits (Dojindo Laboratories, Japan) at day 1, 3, 5, and 7. All the values were normalized to day 1 and presented as mean ± standard deviation. The experiment was done in quadruplicate and repeated at least twice.

Funding Statement

The study was supported by the National Natural Science Foundation of China under Grant Number [81472636 and 81702609]; and the Natural Science Foundation of Heilongjiang Province under Grant Number [LC2015033 and QC2017107].

Data availability

The data were derived from the following resources available in the public domain: spliceosomedb.ucsc.edu/;www.kmplot.com; www.cbioportal.org.

Disclosure of potential conflicts of interest

The authors report no conflict of interest.

Supplementary material

Supplemental data for this article can be accessed here.