Comparative expression patterns and diagnostic efficacies of SR splicing factors and HNRNPA1 in gastric and colorectal cancer

A total of 420 fresh stomach and colon biopsy samples were obtained from 224 patients (Table 1) who had undergone surgical resection.

All tumor samples and patients were defined and diagnosed, respectively, at the Department of Pathology and Surgery, Wonkwang University Hospital. Each fresh biopsy sample was aliquoted into two or four tubes. The first sub-sample was stored frozen in liquid nitrogen until immunoblot analysis; the second was immediately fixed with formalin, paraffin embedded, then stored for hematoxylin and eosin staining and immunohistochemical staining; the third was immediately homogenized for subcellular fractionation; and for the fourth, RNA was immediately extracted and the RNA was reverse transcripted. The synthesized cDNA sample was stored frozen at −75 °C until real-time polymerase chain reaction (PCR) analysis. All paraffin-embedded tissue samples were sectioned, stained with hematoxylin and eosin and evaluated twice by two pathologists. All histological findings were consistent with the diagnosis, and all were concordant between the two pathologists. All sections of cancer tissue included at least 30 % cancer cells, and none showed light microscopically-detectable degeneration or necrosis. Aneuploidy status was determined using routine clinical laboratory diagnostic methods consisting of propidium iodide staining followed by flow cytometric analysis. The SW1116 cell line (a CEA-producing colon cancer cell line) was obtained from the American Type Culture Collection and used between passage 3 and 8 after they were obtained.

Tissue samples (40 mg) were homogenized in RIPA buffer containing protease inhibitors using a Bullet Blender homogenizer (Next Advance; Averill Park, NY, USA), and whole-cell lysate was obtained by sequential centrifugations. Proteins (~30 μg) in the whole-cell lysate were separated on 10 % sodium dodecyl sulfate–polyacrylamide gels with SW1116 cell protein (~30 μg). The proteins were transferred onto polyvinylidene difluoride membranes. The membranes were blocked with 5 % skim milk in Tris-buffered saline containing 0.1 % Tween 20 (TBS-T), rinsed, and incubated with the appropriate antibodies in TBS-T containing 3 % skim milk. Excess primary antibody was then removed by washing the membrane four times in TBS-T. The membranes were then incubated with horseradish peroxidase-conjugated secondary antibody (anti-rabbit or anti-mouse). After three washes in TBS-T, bands were visualized using Clarity Western ECL Substrate (Biorad; Hercules, CA, USA) on the FluorChem E System (Protein Simple; Santa Clara, CA, USA). The following primary antibodies were used in the immunoblot analysis of whole- or fractionated cell lysates: anti-HNRNPA1 (dilution 1:2000; cat. sc-32301; Santa Cruz Biotechnology; Santa Cruz, CA, USA); anti-SRSF1 (dilution 1:1000; cat. 324600; Invitrogen; Carlsbad, CA, USA); anti-SRSF3 (dilution 1:1000; cat. RN080PW; MBL; Nagoya, Japan); anti-SRSF5, anti-SRSF6, anti-SRSF7 (dilution 1:1000; cat. HPA 043484, HPA029005, HPA043850; Sigma-Aldrich; St. Louis, MO, USA); anti-CEA (dilution 1:3000; cat. MS-613-P; Thermo Fisher Scientific; Fremont, CA, USA); anti-ACTB (dilution 1:5000; cat. MA5-15739; Invitrogen); anti-poly (ADP-ribose) polymerase (PARP; dilution 1:1000; cat. sc-7150; Santa Cruz Biotechnology); anti-histone H3 (dilution 1:1000; cat. 9715 s; Cell Signaling; Beverly, MA, USA); and anti-prohibitin (dilution 1:1000; cat. AB28172; Abcam; Cambridge, UK).

Then, the optical density of the region of the target molecular weight (±20 %), as described by the respective antibody manufacturers, was analyzed using ImageJ (http://​imagej.​nih.​gov/​ij/​). Relative protein levels of the samples were determined after normalization to the β-actin band and calibrated using bands from SW1116 cells (a value of 10 was assigned to the SW1116 cell) on each membrane. As presented in previous reports [13‐15] and/or manufacturers’ instructions, SRSF1, SRSF6, SRSF7, HNRNPA1, and CEA showed multiple bands in each target molecular weight area. For the five proteins, multiple bands in each target molecular weight area were calculated and the sum of the distinct band/bands was then analyzed for the proteins. Values of 0.2 (HNRNPA1 and CEA), 0.7 (SRSF1), 0.8 (SRSF3, SRSF5, and SRSF7) or 1.0 (SRSF6) were assigned to the undetected bands of target regions based on band density of SW1116 cell lysate.

Ten paired [normal mucosa (NM) and cancer] samples (n = 20) were fractionated using a modification of the method described previously [16‐18]. Tissue samples (70 mg) were homogenized in 800 μL of homogenization buffer [0.25 M sucrose, 10 mM EDTA, 10 mM EGTA, 2 mM MgCl₂, 20 mM Tris–HCl (pH 7.4) and protease inhibitors]. The homogenized lysate was centrifuged at 1500 × g for 5 min at 4 °C, and the resulting pellet, containing nuclei, and the supernatant (post-nuclear supernatant) were separated. Subsequently, the pellet, containing nuclei, was resuspended in nuclear extraction buffer [2.5 % glycerol, 1 mM EDTA, 1 mM EGTA, 1.5 mM MgCl₂, 0.42 M NaCl, and 20 mM HEPES (pH 7.6)] for 1 h at 4 °C, centrifuged again at 20000 × g for 30 min at 4 °C. The supernatant was then collected and referred to as the nuclear extraction fraction. In parallel, the post-nuclear supernatant was centrifuged at 20000 × g for 30 min at 4 °C and the resulting supernatant was saved as the cytosol/microsome fraction. The pellet was resuspended in RIPA buffer for 10 min at 4 °C followed by centrifugation at 1500 × g for 5 min at 4 °C, and the resulting supernatant was collected and referred to as the membrane/organelle fraction, including intracellular membranes and some plasma membrane. Immunoblot analysis was performed on these three fractions (nuclear extract, membrane/organelle, and cytosol/microsome) as described above.

The resultant three fractions differed in terms of the total protein amounts (membrane/organelle < nuclear extract < cytosol/microsome) for all tissue samples. In addition, the subcellular fraction indicator proteins (PARP, histone H3, and prohibitin) showed inter-individual and intra-individual (cancer vs. NM) variations. Accordingly, to compare the relative SF expression levels for the respective fractions, an equal amount of total protein per lane was loaded.

Formalin-fixed, paraffin-embedded tissue was sectioned and placed on slides. Sections were stained using the Discovery XT automated immunohistochemistry (IHC) stainer (Ventana Medical Systems; Tucson, AZ, USA) and Ventana Chromo Map Kit (Ventana Medical Systems) according to the manufacturer’s instructions. The sections were deparaffinized using EZ prep solution; the antigen was retrieved in cell a conditioning solution (CC1; Ventana Medical Systems) under standard conditions (100 °C, 60 min), and endogenous peroxidase was inhibited by treatment with 3 % H₂O₂ for 4 min. Then, the sections on slides were incubated with the primary antibody that had been used in the immunoblot analysis (dilution 1:300 for anti-HNRNPA1 and anti-CEA and dilution 1:100 for other antibodies) for 60 min at 37 °C, and then with a secondary antibody (UltraMap anti-RB HRP or UltraMap anti-MS HRP; Ventana Medical Systems) for 28 min at 37 °C. The sections were incubated in diamidobenzidine and H₂O₂ for 8 min at 37 °C followed by counterstaining with hematoxylin and treatment with bluing solution. On completion of staining, sections were dehydrated in alcohol, cleared in xylene, and mounted in synthetic resin.

Primer sequences were as follows: 5'-TGGATTTGGTAATGATGGAA-3' and 5'-TCTCTGGCTCTCCTCTCCTG-3' (HNRNPA1); 5'-TGCCTACATCCGGGTTAAAG-3' and 5'-CTGCTGTTGCTTCTGCTACG-3' (SRSF1); 5'-TCTTGGAAACAATGGCAACA-3' and 5'-CTCGGGGATCTTCAAATTCA-3' (SRSF3); 5'-GAGGCTTTGGTTTTGTGGAA-3' and 5'-CGAGCCCTAGCATGTTCAAT-3' (SRSF5); 5'-AAATACGGACCACCTGTTCG-3' and 5'-CTTCACCTGCTTGTCGCATA-3' (SRSF6); 5'-CGCTGGCAAAGGAGAGTTAG-3' and 5'-CGAATTCCACAAAGGCAAAT-3' (SRSF7); 5'-GTAACCCGTTGAACCCCATT-3' and 5'-CCATCCAATCGGTAGTAGCG-3' (18S rRNA).

All group values were non-normally distributed in the Kolmogorov-Smirnov test (p < 0.05). Thus, the values were compared using the Mann–Whitney U-test (between two groups) or Kruskal-Wallis test (among more than two groups). If values were significantly different in the Kruskal-Wallis test, Conover’s post-hoc tests were performed. Ratios were compared using the chi-square test for comparison of incidences between GC and CRC. Spearman correlations were used to assess relationships among the levels of SF proteins or CEA and levels of SF mRNAs. The DAs of each SF for discriminating between cancer and NM were obtained by constructing receiver operating characteristic curves. The cut-offs for defining cancer were determined by the respective highest Youden indexes. The statistical differences between the DAs were determined using the DeLong method (for same-sample-derived comparison) or a comparison of the areas under independent receiver operating characteristic curves (for two independent sample-derived comparisons).

In a comparison of relative immunoblot band intensity (Fig. 1a and b) of stomach samples, the median levels of HNRNPA1 (2.2 vs. 0.5; 4.4-fold difference; p < 0.001), SRSF7 (4.8 vs. 3.0; 1.6-fold difference; p = 0.006), and CEA (2.6 vs. 0.6; 4.3-fold difference; p < 0.001) were significantly higher in GC samples than in gastric NM (GNM) samples (Fig. 1b). Other SF protein median levels were not significantly (p > 0.05) different between GC and GNM samples.

In immunoblot analyses of CR samples (Fig. 1a and b), the median levels of all six SFs and CEA were significantly higher in CRC than in CRNM. Also, the median levels of all SFs except SRSF1 were significantly higher in CR-adenoma than in CRNM. HNRNPA1 and SRSF3 median levels in CRC were most markedly higher than in CRNM samples among the SFs. Median levels of the six SFs were not lower or even higher in CR-adenoma samples compared with CRC samples, but the median level of CEA was lower in CR-adenoma than in CRC.

When GC and CRC were compared (Fig. 1b), median levels of all SFs except SRSF1 were significantly higher in CRC than in GC samples.

In an association analysis between the clinicopathologic factors and SF proteins (Table 1), none of the SF protein levels were significantly different based on patient age, gender, or cancer location in both cancers. Most GCs and CRCs were adenocarcinomas; in gastric adenocarcinoma, the median level of SRSF7 (7.0 vs. 2.9; p = 0.049) was significantly higher in intestinal-type than in diffuse-type. In CR-adenocarcinoma, the median level of HNRNPA1 was significantly higher in well-differentiated type (7.0 vs. 5.0; p = 0.020), no lymph node metastasis (7.0 vs. 4.3; p = 0.003), or low (≤II)-stage (7.0 vs. 4.3; p = 0.003) groups than in the other groups. Median levels of other SF proteins, except HNRNPA1 and SRSF7, were not significantly different based on tumor status, lymph node metastasis, TNM stage, or aneuploidy status in both gastric and CR adenocarcinoma.

In paired sample (cancer/NM from each same patient) comparison (Table 2) of stomach samples, only HNRNPA1 showed >50 % upregulation (cancer/NM >2) incidence (UI), followed by the other UIs in the order of HNRNPA1 (52 %), SRSF7 (42 %), and other SF proteins (22 %–33 %). In CRC, all SFs except SRSF1 and SRSF5 showed UIs of >50 % in the order of HNRNPA1 (88 %), SRSF3 (74 %), and other SFs (31 %–57 %). When UIs in GC and CRC were compared, the UIs of all SFs were lower (SRSF3, HNRNPA1, and SRSF6) or tended to be lower in GC than in CRC; the differences between GC and CRC were greatest for SRSF3 (40 %; p < 0.001), followed by HNRNPA1 (36 %; p < 0.001), SRSF6 (24 %; p = 0.004), and the other SFs. The UIs of CEA were >50 % in both cancers; it was also lower in GC (53 %) compared with CRC (92 %), and the results were similar to HNRNPA1 values.

In the DA analysis (Table 2) for GC, only DA-HNRNPA1 (74 %) was acceptable (>70 %), while the other SFs presented poor (≤70 %) DAs. The DAs of all SFs were not significantly different between low- and high-stage gastric adenocarcinoma. In CRC, all SFs except SRSF1 and SRSF5 presented acceptable DAs in the order of HNRNPA1 (90 %), SRSF3 (84 %), and the other SFs (62–76 %). When the DAs for GC and CRC were compared, the DAs of all SFs were lower (SRSF3, HNRNPA1, and SRSF6) or tended to be lower in GC than in CRC; the differences between GC and CRC were greatest for SRSF3 (32 %; p < 0.001), followed by HNRNPA1 (17 %; p < 0.001), SRSF6 (14 %; p = 0.006), and the other SFs. The DAs of all SFs were not significantly different based on the stage of CR-adenocarcinoma, and the greatest difference between GC and CRC being in SRSF3 was consistently observed regardless of the stage. In the comparison with CEA (Table 2), DA-HNRNPA1 was not different from DA-CEA in both low- and high-stage gastric adenocarcinoma, whereas in CRC, DA-HNRNPA1 was not different from DA-CEA in low-stage CR-adenocarcinoma, but it was significantly lower than DA-CEA in high-stage CR-adenocarcinoma.

Previously, several studies showed SF distribution using enzymatic IHC [7].

Generally, this method provides information about both cellular morphology and tissue structural changes, but it is limited when used for a clear discrimination of organelles (the major location of translation regulatory molecules) from cytosol. So far, no study has demonstrated the SF subcellular distribution in primary cancer cells using biochemical fractionation and immunoblotting. Using the biochemical fractionation approach (Fig. 1c), we were able to more precisely distinguish the extranuclear fractions. HNRNPA1 and SRSF6 in both cancers frequently (≥50 %) showed different distribution patterns from those of the other SFs. When HNRNPA1 and SRSF6 were upregulated in whole cancer-cell lysates, they were predominantly distributed in nuclear and/or cytosol/microsome fractions; the upregulated SFs in the cytosol/microsome fraction were detectable in >50 % of upregulated cases in both cancers. Alternatively, when SRSF1, 3, 5, and 7 were upregulated in whole cancer-cell lysates, they were predominantly distributed in nuclear and/or membrane/organelle fractions; the upregulated SFs in the membrane/organelle fraction were detectable in >50 % of upregulated cases in both cancers. There were no remarkable differences in subcellular distributions between GC and CRC. With regard to methodological principle and analytical points, biochemical fractionation followed by immunoblotting differed from IHC. The former was normalized using equal amounts of total protein per lane, while IHC images were analyzed directly. During the former method process, the nuclear and extranuclear fractions were transiently treated in a different buffer, which may affect the affinity of the antibody, and a certain amount of the protein in the border zone of the different fractions was discarded to prevent contamination of a adjacent fraction. Regarding these concerns, we only were able crudely to match the nuclear and extranuclear fraction quantities, and were not able to precisely match the distribution of four SF proteins (SRSF 1, 3, 5, and 7) in the membrane/organelle fraction to the corresponding fraction of the enzymatic IHC analysis (Fig. 1d). Nonetheless, the two methods showed an approximate match with regard to whole-cell SF protein intensities.

In the SF mRNA level analysis (Fig. 2a), none of the six SF median levels significantly differed between GC and GNM. In contrast, all SFs except SRSF5 were significantly higher in CRC than in CRNM.

None of the SF mRNA median levels were significantly different based on histological features, lymph node metastasis, or TNM stage in gastric or CR-adenocarcinoma (Table 1).

In paired sample comparisons of mRNA (Table 2), all SF mRNAs showed <50 % UI in GC, and only HNRNPA1 showed >50 % UI in CRC.

In DA analysis of mRNA (Table 2), including DA-HNRNPA1 (69 %) for CRC, which was the highest SF mRNA DA, all of the SF mRNAs showed poor DAs for both cancers. The UIs and DAs of all of the SF mRNAs were not significantly different between GC and CRC (Table 2).

In correlation analysis (Fig. 2b), all SF proteins were significantly correlated with each other and all SF mRNAs were correlated with each other in both stomach and CR samples.

None of the SF protein levels, except SRSF6, were correlated with their respective mRNA levels in stomach samples. Whereas in CR samples, levels of three SF proteins (HNRNPA1, SRSF3, and SRSF6), which had relatively high DAs, were correlated with their respective mRNA levels.