Alteration of tumor suppressor BMP5 in sporadic colorectal cancer: a genomic and transcriptomic profiling based study

All patients included in this study had written informed consent. DNA samples used for whole exome sequencing were extracted from 3 cases of fresh frozen tumor tissues. For cases used in expanded deep sequencing, tumor DNA and matched normal tissue DNA were extracted using QIAamp DNA FFPE Tissue Kit (QIAGEN). 28 cases of validation samples were obtained from preclinical medical teaching experiment center of Fourth Military Medical University. All samples were confirmed with no family history of colorectal cancer and the pathological information were listed in Additional file 1: Table S1.

Human colorectal cancer cells (HT-29, LoVo, HCT 116, and SW480) were purchased from ATCC (American Type Culture Collection), and immortal normal epithelial cells NCM460 were purchased from INCELL (San Antonio).

Exome capture was performed using the TruSeq Exome Enrichment Kit (Illumina) according to the manufacturer’s instructions. Sequencing was performed on the Illumina Hiseq2000 platform. The variant calling was analyzed using GATK tools as previously described [6]. We performed loss of function mutation screening strategy (Additional file 2: Figure S1). The remaining candidates are subsequently filtered out with the following criteria: 1. Reported SNVs in dbSNP130 and 1000 genomes were removed. 2. The SNVs quality score in tumor must be ≥100. 3. The sequencing depth of every SNV must be between 10 and 200. 4. The SNVs in the last 5% of gene coding region were removed. High confidence novel SNVs were resequenced in original samples and matched blood samples to confirm somatic SNVs. Primers used for expanded sequencing of BMP5 coding regions were shown in Additional file 1: Table S2.

Total RNA Miniprep Kit (Sigma, St Louis, MO) was used to extract RNA from cell lines according to the manuals. The isolated RNA was then reverse transcribed to cDNA (Takara, Japan). Quantitative real-time PCR amplifications were performed using the SYBR Premix Ex Taq™ II kit (TAKARA, Japan) by CFX96™ real-time PCR detection system. Primers used in this study can be found in Additional file 1: Table S3.

BMP5 immunohistochemistry was performed on tumor microarrays (TMAs) from Fanpu Biotech, Inc., and the pathological information can be found in Additional file 1: Table S4. For antigen retrieval, tissue sections were boiled in 0.01 M citrate solution (pH 6.0) and incubated with primary antibody to BMP5 (1:200, Proteintech, China). Tissue sections were observed with a standard light microscope (Leica). The intensity of staining ranged from negative (0) to weak positive (1–2) or strong positive (3).

Protein extraction and blotting was performed as described previously [7]. Western blots were probed with the following antibodies anti-BMP5 (Abcam, ab88064, 1:500), anti-PCNA (Immunoway, 1:5000), anti-MMP2 (Proteintech, 1:500), anti-MMP9 (Proteintech, 1:1000), anti-E-cadherin (BD Biosciences, 1:1000), anti-EPSTI1 (Proteintech, 1:500) and anti-GAPDH (Immunoway, 1:5000).

5 × 10³ SW480 or NCM460 cells were seeded into 24-well plates for cell climbing. Cells were fixed with 4% paraformaldehyde for 15 min, blocked with 1% bovine serum albumin at ambient temperature for 1 h and then incubated with rabbit anti-BMP5 primary antibody (1:100) at 4 °C overnight. Cells were incubated with goat anti-rabbit IgG secondary antibody (Abbine, USA, 1:800) at 37 °C, and with nuclear stained by DAPI (SouthernBiotech, USA) subsequently. Images were filmed and analyzed by Fluview FV10i.

The entire 3’-UTR of human BMP5 was amplified and then inserted into the pmiRGLO vector (Promega, Madison, WI, USA). To evaluate binding specificity, the sequences that interact with the seed region of miR-25, miR-32, miR-92a, and miR-655 were mutated respectively using KOD-Plus Mutagenesis Kit (Toyobo, Japan). For luciferase reporter assays, we performed as described previously [8].

The full-length cDNA of BMP5 was obtained from 293 T cDNA by PCR using the primers described in Additional file 1: Table S5. PCR products were then cloned into the BamHI/SalI sites of pEF-BOS-EX. The p. D183G mutation was introduced using KOD -Plus- Mutagenesis Kit (Toyobo, Japan). Transfection was carried out using X-tremeGENE HP DNA Transfection Reagent (Roche Applied Sciences).

For BMP5 overexpression, full length CDS of BMP5 was inserted in pLV5 vector. Lentiviral stocks were prepared in HEK-293 T cells. Viral supernatant mixed with 5 μg/ml polybrene was used to infect HT-29 or SW480 cells for 24 h. Cells were then selected by puromycin for stable BMP5 overexpression.

The sequences of the two siRNAs used are listed in Additional file 1: Table S6. SW480 cells were transfected using HiPerFect transfection reagent (Qiagen, Valencia, CA, USA) at a confluency of 50%. The final concentration of siRNAs in SW480 cells was 20 nM.

For proliferation assay, Cell Counting Kit-8 (Dojindo Laboratories, Japan) was used according to the manufacturer’s instructions. The number of viable cells was measured at a wavelength of 450 nm.

Cells were collected and washed twice with PBS after 48 h post transfection. For cell cycle analysis, the cells were fixed with 70% ethanol overnight at 4 °C, and then washed with PBS, resuspended with 500 μl buffer and then incubated with 100 μg/ml RNaseA and 50 μg/ml propidium iodide (PI) (7sea biotech) for 30 min at 37 °C. After incubation, the cells were subjected to DNA content analysis using a FACSCalibur (Beckman Coulter, Fullerton, CA) and the results were analyzed with the Summit v4.3 software. For apoptosis assay, the method was described previously [7].

Cell migration or invasion ability was evaluated using Transwell chambers (Corning, MA, USA) without or with Matrigel (BD Biosciences, Bedford, MA, USA). Cells were treated with Mitomycin C (10 μg/ml) to inhibit cell proliferation before plating. 10⁵ SW480 cells were suspended in 100 μl serum-free medium and then plated in the upper chamber of the 24-Well, with 700 μl medium containing 10% FBS in the lower chamber. The cells were incubated at 37 °C for 48 h. Detail method was described previously [7].

We adhered to standards articulated in the Animal Research: Reporting of In Vivo Experiments (ARRIVE) guidelines [9]. Male balb/c nude mice (3–5 weeks old) were purchased from Animal Center at Medical College, Xi’an Jiaotong University. All mice were housed and maintained under specific pathogen-free conditions. 2 × 10⁶ cells (BMP5 wild or mutant type stable expressing HT-29 cells) were resuspended in 200 μl serum-free medium and then injected to the subcutaneous of the right axilla of nude mice (n = 6 mice/group). The tumors’ dimensions were monitored with vernier calipers for a total period of 25 days, and the tumor volume was calculated using the following formula: 0.5 × a × b², where a represents the longer diameter and b represents the corresponding perpendicular shorter diameter.

The data were shown as Mean ± S.E.M or Mean ± SD. Paired or unpaired t-test, Wilcoxon test, Chi-square test, Pearson’s correlation and Kaplan-Meier survival analysis were used to compare the differences among groups. P value < 0.05 was considered as statistically significant. *P < 0.05, **P < 0.01, ***P < 0.001. ****P < 0.0001.

Exome capture and sequencing was performed on three cases of sCRC tumor DNAs. We focused on the loss of function (LoF) mutations that may greatly disrupt protein translation (Additional file 2: Figure S1). A total of 1301 LoF mutations were observed which was in line with the results previously reported in Asian samples (Fig. 1a) [10]. To determine which of these alterations were somatic (sCRC-specific), we resequenced the high confidence novel SNVs using Sanger sequencing, filtered the mutations that observed in the matched normal blood samples. In total 72 somatic mutations were identified, of which 20 were nonsense, 18 at canonical splice site, and 40 frameshift (Fig. 1b and Additional file 1: Table S7).

To further narrow the genes screened, we used data from GEO, TCGA, HPA, and KEGG to perform a systems-level analysis of 71 mutated genes (Fig. 1c), aimed to identify novel markers that can affect CRC initiation, progression and patients’ survival. Our hypothesis is normal tissue-enriched genes may be tissue-specific, and loss of which could promote CRC tumorigenesis. In order to find colorectal-enriched genes that may be colorectal tissue-specific, we analyzed gene expression level in normal tissues using HPA and TCGA database. The gene expression in normal colon or rectum ranked top 7 in 37 tissues was identified as colorectal-enriched in this study. 11 genes are not detected in colon or rectum, while BMP5, REP15, ATP8B1, ELF3, and RASSF6 are relative high in colorectal tissue, and significantly downregulated in colon or rectum adenocarcinomas (Additional file 1: Table S8). We further analyzed differentially expressed genes in early events of CRC (normal tissue – adenoma or polyps) using GEO datasets (GSE8671, GSE71187, and GSE41258). We found 13 genes were differentially expressed in at least 2 series, of which 7 were downregulated (BMP5, PDE2A, ZNF175, CTSA, SVEP1, ATP6V0D2, and AHNAK) (Additional file 1: Table S9). Kaplan Meier survival analysis identified 33 genes may affect patients’ survival, including two favourable prognostic markers (REP15, ATP8B1) and one unfavourable prognostic marker (GPSM1) reported in HPA database [11]. In addition, high expression of BMP5 and RASSF6 were correlated with a longer patients’ survival outcome (Additional file 1: Table S10). To find disease-related genes, we further used KEGG database together with literature queries, 19 genes were involved in pathway networks, including APC, TCF7L2 (Wnt signaling), BMP5 (TGFβ signaling), and RASSF6 (Hippo signaling) (Additional file 1: Table S11). Taken together, BMP5 satisfied all criteria we tested in multi-omics. We confirmed BMP5, a novel gene that has never been investigated in sCRC, was top rank gene in our study. From the cBioPotal database, we found 30.4% of BMP5 mutated samples without APC mutation, and still 17.4% of BMP5 mutated samples without APC, KRAS, or TP53 mutation. As is well known, most sCRCs occur through chromosomal instability pathway, which is characterized by APC mutations. This result may indicate that in addition to well-studied diver genes, the alteration of BMP5 may play a role in the oncogenesis of sCRC.

Subsequent Sanger sequencing analysis in expanded individuals showed BMP5 was mutated in 7.7% of patients and 37.5% of these mutations were LoF. The distribution of 8 mutations identified in BMP5 is shown in Fig. 1d, Table 1, and Additional file 2: Figure S2. Notably, we also found a missense mutation p. D183G in In-2 patient. All missense mutations found were possibly damaging and pathogenic analyzed by PolyPhen-2 and FATHMM-MKL algorithm [12, 13]. The affected residues of BMP5 are highly conserved evolutionarily (Additional file 2: Figure S3), thus these mutations are rare in normal but of high penetrance in sCRC. Examination of publicly available databases revealed that BMP5 mutation is also found in several other tumor types (Fig. 1e). Truncating mutation frequency of BMP5 is highest in CRC, while copy number amplification could be found in all tumors but not in CRC. These results showed the characteristic of BMP5 alteration is different from that of other type of tumors.

To test whether loss of BMP5 is common in cancer, we firstly verified the protein level of BMP5 in seven tumor types using immunohistochemistry (IHC). Our results showed that BMP5 expression was negative in 31.0% (40 of 129) of the sCRC tumor cases while 10.1% (13 of 129) of the normal tissues (Chi-Square tests P < 0.0001, Fig. 2a and Table 2), and BMP5 expression showed no correlation to age, sex and clinical grade of 129 colorectal adenocarcinomas (Additional file 1: Table S12). However, BMP5 expression showed no significant difference between tumor and normal tissues in hepatic, esophagus, gastric, pancreatic, and lung, while in breast cancer, BMP5 positive was significantly higher in Invasive ductal carcinoma than normal control (52.5% vs 12.5%, P = 0.0003, Table 2). This may demonstrate loss of BMP5 may be a unique event in sCRC.

We further explored the correlation of BMP5 expression with clinicopathological features using data from GEO and The Cancer Genome Atlas (TCGA). We analyzed GEO datasets that recruit normal tissue and adenomas (polyps) and found BMP5 downregulation is an early event in CRC (Fig. 2b). TCGA data showed BMP5 was significantly downregulated in CRC (Wilcoxon test, P = 7.797e^− 28) (Fig. 2c), and early stage (Stage I) patients showed higher expression than other stage patients (unpaired t test, P < 0.05). Furthermore, paired samples analysis revealed that 93.8% (30 of 32) cases showed downregulatiton of BMP5 (Fig. 2d, paired t test, P < 0.0001), which was also confirmed in our validation group (Fig. 2e, paired t test, P < 0.0001). Low expression of BMP5 was correlated with patients’ recurrence (Fig. 2f and Additional file 1: Table S13) as well as poor survival (Fig. 2g, Log rank P = 0.0071). Notably, survival analysis in other 6 tumors we tested did not show this significance (Additional file 2: Figure. S4). Microsatellite instability (MSI) is a favourable prognostic marker for CRC, and we found five-year survival is higher in MSI patients with high BMP5 expression than other groups (Fig. 2h). Moreover, low BMP5 expression is correlated with poor survival independent of MSI status.

For the endogenous expression analysis in cell lines, BMP5 showed the highest level in normal NCM460 cells. Among CRC cell lines, BMP5 was relatively high in SW480 cells and barely expressed in HT-29 cells (Fig. 2i). In NCM460 and SW480 cells, BMP5 is mainly localized to vesicles, which is similar to its homolog BMP7 (Additional file 2: Figure S5).

To clarify the factor for decreased level of BMP5 in addition to loss of function mutations, we utilized bioinformatic analyses to find potential epigenetic changes in BMP5. CpG island was not found in BMP5 promoter using MethPrimer tools [14], and studies about hypermethylation of BMP5 was not reported yet. Meanwhile, prediction of BMP5 3’-UTR binding miRNAs using TargetScan, miRanda, and together with TCGA expression analysis helped lead to the selection of previously reported oncogenic miRNAs (miR-32, miR-92a) or upregulated miRNAs (miR-25, miR-655, let-7f-2, and miR-4766) in CRC (Fig. 3a). In CRC samples of TCGA, BMP5 expression was only negatively correlated with miR-32 and miR-655 expression, mainly in stage IV patiens (Additional file 2: Figure S6). We finally selected four miRNA for binding analysis, among which miR-32, miR-25, and miR-92 share the same seeding match sequence (Position 51–57 of BMP5 3’ UTR). Dual luciferase assay demonstrated these four miRNAs could significantly repress the reporter activity (wild-type 3’-UTR of BMP5) (Fig. 3b). Further Western blotting showed these miRNAs could inhibit the expression of BMP5 in varying degrees (Fig. 3c). We also found miR-32 and miR-655 could partially reverse the tumor inhibition ability by stable BMP5 expression in SW480 cells (Fig. 3d).

As previous studies reported that BMP5 inhibits cell proliferation in several tumor entities [15‐17], we investigated the influence of BMP5 on colorectal cancer cell growth. BMP5 overexpression and knockdown efficiency were validated by western blots (Additional file 2: Figure S7). Both WT and MUT type BMP5 could decrease cell growth rate than control (pEF-BOS-EX vector-transfected cells) (HT-29 cells: WT P < 0.0001, MUT P < 0.01. SW480 cells: WT P < 0.05. Fig. 4a). while knockdown of BMP5 in SW480 cells promoted cell proliferation (Fig. 4b). The blots of tumor proliferation biomarker PCNA confirmed BMP5 could affect HT-29 and SW480 cell growth (Additional file 2: Figure S7). Flow cytometry showed cell growth inhibition effect of BMP5 is mainly mediated by cell cycle alterations. BMP5 could promote G1 phase cell cycle arrest (Fig. 4c-d) but had no effect on CRC cell apoptosis (Additional file 2: Figure S8). To evaluate the effect of BMP5 on tumorigenicity in vivo, we engineered HT-29 cells (BMP5-deficient cells) to stably express WT or MUT BMP5 by lentiviral infection and puromycin selection. Implant of HT-29 cells in mice revealed that expression of WT BMP5 reduced the tumor size and tumor weight as compared to BMP5-deficient control group (P < 0.0001), whereas the tumor inhibition effect weakened in MUT BMP5 group though the difference is statistically significant (P < 0.001) (Fig. 4e-g).

As reported previously, BMP5 has effect on cell migration and invasion, but the results were controversial [17, 18]. In SW480 cells, both WT BMP5 and MUT BMP5 led to reduced cell migration, whereas MUT BMP5 failed to suppress invasion (Fig. 5a; WT: P < 0.01, MUT: P < 0.05 for migration, and WT: P < 0.05, MUT: P = 0.25 for invasion, paired t-test). Furthermore, knockdown of BMP5 by two siRNAs significantly promoted both cell migration and invasion in SW480 cells (Fig. 5b). Migration indicator MMP2 showed inhibition of BMP5 enhanced the ability of cell mobility (Additional file 2: Figure S7).

Increasing evidence has confirmed the vital role of epithelial-mesenchymal transition (EMT) process in tumor initiation and metastasis [19]. Cell phenotypes change during the EMT process. Firstly, we investigated whether loss of BMP5 could promote cell motile abilities. Silencing of BMP5 indeed switched the epithelial cells NCM460 and SW480 from a polygonal and condensed pattern to a spindle and scattering like pattern (Fig. 5c). Addition of cobalt chloride (CoCl₂) could induce EMT in vitro, causing loss of E-cadherin and increased level of ZEB and TWIST [20], and we then asked whether the expression of BMP5 changed during this process. In our study, BMP5 was downregulated in a time course after addition of CoCl₂ in SW480 cells (Fig. 5d). Functionally, BMP5 expression could attenuate SW480 cell motility ability induced by CoCl₂ (Fig. 5e).

In clinical samples, we found BMP5 expression was positively correlated with epithelial markers, and negatively correlated with mesenchymal markers (Fig. 5f and Additional file 2: Figure S9A). Further analysis in paired samples found only the fold change (Log₂(Tumor/Normal)) of E-cadherin was correlated with that of BMP5 (Fig. 5g and Additional file 2: Figure S9B). What is more, we did not find this correlation in other 6 solid tumors tested (Additional file 2: Figure S9C). In HT-29 and SW480 cells, both overexpression and siRNA-mediated inhibition of BMP5 could affect E-cadherin expression (Fig. 5h and i). E-cadherin plays an essential role in cell-cell adhesion in normal cells, and the coexpression pattern of BMP5 and E-cadherin may reflect the importance of BMP5 in maintaining normal epithelial cell homeostasis, which confirms our hypothesis that loss of BMP5 is an early event in CRC.

In this study, we sought to characterize the downstream gene and pathway changes induced by BMP5. Transcriptome-wide analysis found 79 upregulated and 287 downregulated genes, with ≥1.5-fold change as compared to control group (Fig. 6a and Additional file 1: Table S14). The differentially expressed genes (DEGs) are mainly on interferons, interferon-stimulated genes (ISGs), chemokines and some EMT markers, which enriched in cytokine-cytokine receptor interaction, TNF signaling, chemokine signaling, and Jak-Stat signaling pathway (Fig. 6b and c). Interestingly, genes involved in BMP/Smad signaling pathway were not differentially expressed in this study, indicating BMP5 may signal through Smad-independent pathways. We further presented gene coexpression networks to reveal the interactions among genes (Additional file 1: Table S15 and Additional file 2 Figure S10). EPSTI1 (Epithelial Stromal Interaction 1), an interferon stimulated gene, is the core regulatory gene in BMP5-expressing HT-29 cells. We also found IL-28A (IFNL2), the type III interferon, can signal through the Jak-Stat pathway to activate EPSTI1 expression [21]. In this study, genes involved in Jak-Stat signaling pathway, including IFNL1, IFNL2, IFNL3, IL2RG, IL15RA, IL23A, STAT2 and CSF3 were downregulated by overexpression of BMP5.

We utilized TCGA database and found BMP5 was positively correlated with its receptor, BMPR1A, and BMPR2 (Fig. 6d), but showed no correlation with downstream effector Smad2, Smad3, and Smad4. However, in lung, breast, liver, and pancreatic cancer, BMP5 could correlated with at least one or more Smad effectors (Additional file 2: Figure S11). In CRC, BMP5 was negatively correlated with STAT2, and EPSTI1, especially in Stage IV samples (Fig. 6d). Together with the coexpression network, we found a novel pathway that BMP5 signals through. Expression of BMP5 inhibits interferons, especially IL28A, thus to block the activation of STATs. Phosphorylated STAT dimers bind to STAT-responsive elements in the promoter of EPSTI1, resulting in modulation of its transcription (Fig. 6e-f).