Background
Colorectal cancer (CRC) is one of the leading malignant neoplasms worldwide, and early diagnosis and molecular characterization is considered essential to decrease CRC-related deaths [
1]. A discriminatory set of transcripts was recently identified that proved to be suitable for distinguishing CRC from normal colon with high sensitivity and specificity [
2]. The study used 53 biopsies fresh frozen colon samples along the dysplasia-carcinoma transition containing CRC (n = 22), adenoma (n = 20) and 11 healthy colon (n = 11) specimens, and a qPCR-based expression profile (transcript set) was identified that discriminated the malignant from benign colon.
Although the best specimens for gene expression analysis are fresh frozen tissue samples stored with or without stabilization reagents (e.g. RNA Stabilization Reagent) with highly intact RNA content, difficulties and disadvantages of fresh frozen specimen collection (as requiring special logistic issues, equipment and hands-on time) can limit the number of fresh frozen samples in certain hospitals. In parallel, formalin-fixed, paraffin-embedded (FFPE) tissue samples are routinely collected for diagnostic purpose and are stored at room temperature in archives for several years. Diagnosis of colorectal diseases are commonly made on the basis of histology of stained slides from FFPE biopsies or bigger removed tissue samples taken during routine endoscopy examination or surgery. The fixation in formalin conserves the tissue structure, but holds technical disadvantages in terms of molecular sample quality, such as nucleic acid degradation and crosslink formation; this renders it a sub-optimal sample source [
3,
4]. In order to optimize sample quality from FFPE tissue samples, manual and automated nucleic acid isolation kits and FFPE-optimized protocols are continuously being developed, that can also enhance the use of FFPE samples in gene expression studies, as well [
5‐
7]. Automated nucleic acid isolation systems hold the possibility to remarkably increase the throughput of diagnostic laboratories. Furthermore, automated protocols have the advantage of standardization, minimal hands-on time that is crucial in studies with high sample numbers [
8].
Although molecular characterization of homogenized tissue samples serves information about the whole tumor mass, up- or downregulation of genes can be characteristic for some certain tumor areas or cell types [
9]. In situ analysis can broaden our knowledge by providing information about tumor heterogeneity, as well as the localization of the altered mRNA expression.
As a validation study, we aimed to extend the analysis of the previously published CRC-specific marker set on FFPE samples. The applicability of this marker set was tested on 30 independent fresh frozen biopsy samples (CRC: n = 15; healthy colon: n = 15) and also on 30 formalin-fixed, paraffin-embedded (FFPE) tissue samples (CRC: n = 15; healthy colon: n = 15) after RNA isolation with an automated method. Furthermore, in situ hybridization was performed on 9 FFPE samples (CRC: n = 3; adenoma: n = 3; healthy colon: n = 3).
Discussion
Conventional tumor classification based on histopathology is likely to be accompanied by molecular biology methods in the future. In order to determine the applicability of a previously identified colorectal cancer-specific transcript set, gene expression of these markers were analysed on independent fresh frozen and also on FFPE samples. Furthermore, as methods enhanching diagnostic approaches are required to have straightforward workflow, automated RNA isolation and automated polymerase chain reaction setup were also integrated in the study.
The present study aimed to test the discriminatory power of a set of 11 transcripts. According to our results this mRNA transcript set could be suitably applied even on FFPE samples with relatively high sensitivity and specificity. Based on the expression level of the transcript set, independent fresh frozen biopsy samples could be classified correctly in original grouped cases (100 %; 100 %, respectively), while 93.3 % of cross validated grouped cases were correctly classified. The aim of this study was to confirm the possibility of discriminating FFPE samples by using mRNA markers. Although only a relatively low sample set was analyzed, an almost clear separation of normal vs. CRC FFPE samples could be achieved with unsupervised hierarchical clustering.
Analytical parameters of total RNA samples isolated from different sample sources can be critical regarding downstream molecular analyses. In our study yield of RNA isolated from FFPE and fresh frozen samples were found to be similar, which is in accordance with literature data [
13]. In our analysis altogether 150 ng of total RNA was enough to analyze gene expression of 11 selected transcripts and an endogenous control on a custom made pre-lyophilized real-time PCR plate. As it can be expected according to previous methodical studies [
14,
15], RNA purity ratios were found to be lower in RNA eluates isolated from FFPE samples and RIN was significantly higher (p < 0.001) in fresh frozen biopsies. However, the majority of the FFPE reactions had similar results as the fresh frozen samples and only three out of 11 assays showed higher presence of failed PCR reactions with FFPE samples. The use of even shorter amplicon sizes might result in successful reactions [
16].
It is well known and documented, that colorectal cancer development is accompanied with gene expression alteration of several genes [
17]. Many studies aimed to explore whole genome gene expression profiles of CRC cases attempting to identify clinically useful markers [
18‐
20] and part of these projects focused on the discovery of altered gene expression patterns between normal colon mucosa and tumor tissue samples revealing different CRC-specific transcript sets [
21‐
24]. Among the firsts, in 1999 Alon and collaborators identified a set of 2000 transcripts clustering 22 normal and 30 colorectal cancer tissue samples with highest minimal intensity across the samples [
25]. Later, it was followed by subsequent studies reporting more restricted potential marker sets. For example, on the basis of Zou and collaborators’ results analyzing normal (n = 8) and CRC (n = 9 CRC) fresh frozen samples 250 differentially expressed transcripts could be identified [
26]. Lin and colleagues examined normal (n = 20), adenoma (n = 9) and CRC (n = 11) fresh frozen tissue samples and could define 427 discriminatory markers [
27]. Friederichs and collaborators identified 330 transcripts on the basis of gene expression data of CRC (n = 25) and normal (n = 6) colorectal fresh frozen tissue samples [
28].
The comparison of gene expression profiling studies focusing on colorectal cancer development remains challenging. According to a recent systematic review of 23 independent expression analyses revealed only 54 genes, reported from at least two studies, showing the same expression alteration direction [
29]. Furthermore, despite of the known molecular complexity underlying this disease, studies have been largely focused on the identification of single markers (e.g. upregulated IL8, AXIN2, CXCL12, CDK8; downregulated CA2, COL1A2, FABP1, IGFBP7) [
30].
However, in order to achieve higher sensitivity and specificity multimarker sets should be analyzed, these are considered to be superior to single marker discriminations (Grabowski [
31]). The sensitivity and the specificity of the transcript set [
2] published by Galamb and colleagues in 2012 were assessed by discriminant analysis revealing 93.3 % sensitivity and 86.7 % specificity in fresh frozen biopsies and 96.7 % sensitivity and 70 % specificity in FFPE samples.
On the basis of Garcia-Bilbao and collaborators work published in 2012, relatively higher discriminating power of non-tumoral vs. CRC fresh frozen biopsy samples could be achieved on the basis of a set of 7 transcripts (ENC1, ACAT1, TMEM132A, CMTM7, FAM60A, MADCAM1, DDX55) with RT-qPCR with 90.9 % sensitivity and 100 % specificity [
32]. Another study reported 83.3 % specificity and 70.8 % sensitivity in tumor vs. normal FFPE tissue classification by a prognostic marker set (with not only mRNA markers: neuroendocrine differentiation, overexpression of the sialyl-Lex antigen, overexpression of the peripheral benzodiazepine receptor (PBR), BAX protein expression and p53 mutational status) [
31].
In situ analysis of two transcripts of the analyzed set showed the same gene expression alteration tendency, as found in RT-PCR experiments. CA-7 was found to be downregulated showing weak CA7 ISH signal in normal colon and no detectable signal in the analyzed adenomas and colon cancer FFPE tissue samples. In contrast, chemokine (C-X-C motif) ligand 1 was found to be upregulated according to our RT-PCR results. With in situ mRNA hybridization CXCL1 was also found to be an upregulated transcript in colorectal cancer FFPE tissues, with no CXCL1 ISH signal in the healthy and adenomatous colon tissue specimens. Interestingly, CXCL1 transcript could be localized in stromal cells towards the ulcerous areas of the tumors.
The ratio of failed reactions with FFPE samples were more abundant with some assays (MMP3, CA7, IL8) (Additional file
1: Table S1), that might be due to technical reason, e.g. relatively longer amplicon sizes (MMP3: 110 bp; IL8: 92 bp; Table
1) compared to the other assays or the presence of contaminants in RNA samples isolated from FFPE tissue samples. Furthermore, failed reactions might indicate biological reason in the background, e.g. possible downregulation of CA7 in CRC samples and low MMP3 expression in normal colon samples.
The majority of the transcripts (10 from the 11 markers) analyzed in this study was found to be upregulated along colorectal cancer formation also in fresh frozen and in FFPE tissue samples. On the other hand, gene expression alterations observed in our RT-PCR experiments of the 11 transcripts have already been reported in the literature. Interleukin 1 beta (IL1B) is proven to promote the invasiveness of malignant cells and this pro-inflammatory cytokine is required for angiogenesis [
33]. The -31 T/C polymorphism of this gene is associated with colorectal cancer risk, as IL-1B-511 heterozygotes and T carriers has reduced risk of colorectal carcinoma development [
34] and it is associated with the recurrence of CRC [
35]. IL-1B is involved in the survival and proliferation of remnant cancer cells after tumor resection in colorectal carcinoma [
36]. IL1RN polymorphism is also associated with CRC [
37].
Interleukin 8 (IL8) overexpression has been documented in CRC; it is involved in proliferation, metastasis and angiogenesis, furthermore its knock down can inhibit colorectal cancer cell growth and metastasis formation [
38,
39]. Gremlin -1 overexpression has been detected in various cancer types including pancreas, breast, kidney, ovary and colorectal cancer [
40]. Secretion of gremlin-1 protein can enhance tumorigenesis via enhancing cell proliferation [
41]. CXCL2 (alias Gro2) shows elevated expression in colorectal cancer already in premalignant lesions and in CRC [
42] and it could accelerate tumor cell growth by inducing cell proliferation in a mouse implantation model [
43]. Collagen type XII (COL12A1) was recently identified as a marker of myofibroblastic differentiation in CRC in the desmoplastic invasion fronts of the tumors [
44]. Its overexpression could be detected in CRC-associated CAF (cancer-associated fibroblast) cells isolated from a mouse model of human sporadic cancer [
45]. Chitinase 3-like 1 (CHI3L1, alias HCgp39 or YKL-40) is a secreted glycoprotein that is not synthesized physiologically, but in inflammatory and cancerous states [
46,
47], it proved to have an important role in macrophage recruitment and angiogenesis during colorectal cancer formation [
48]. According to a recent report, this marker can be usefully utilized in monitoring CRC-patients in the follow-up phase, as elevated serum YKL-40 levels is associated with short survival time [
49]. Solute carrier family 7, member 5 was found to be upregulated 3-fold in the adenoma and 5-fold in CRC cases on the protein level [
50]. Carbonic anhydrase 7 (CA7) plays role in acid-base balance, bone restoration, respiration, calcification and catalyze zinc
2+ ion dependent hydration of CO
2 [
51]. According to a recently published manuscript from Yang and collaborators analyzing FFPE (n = 379) and fresh frozen CRC (n = 84) tissue samples, CA7 was frequently downregulated in tumor samples both on mRNA and protein levels. Furthermore, authors could correlate decreased CA7 expression levels with shorter disease-free survival time [
52]. Downregulation of CA7 transcript along tumor formation and the presence of a large CpG island in the gene’s promoter region raises the theoretical possibility of its epigenetic gene regulation by DNA methylation; further studies are needed to confirm this hypothesis. According to the in situ hybridization results the extent of CA7 expression level alteration was lower than observed in RT-PCR experiments. However, CA7 expression levels were confirmed to be different between healthy and diseased colon samples according to the RT-PCR results.
CXCL1 (Groα) gene encodes a secreted interleukine-like molecule binding specifically to G-protein-coupled receptor CXCR2 [
53] and potentially plays role in tumor-associated angiogenesis in non-small cell lung cancer [
54]. Its overexpression has been described in melanoma [
55] and in CRC on mRNA level with well-correlating immunohistochemical results [
56,
57]. In 2015, le Rolle and colleagues found that CXCL1 overexpression is a poor prognostic marker on metastatic CRC, furthermore, CXCL1 inhibition could suppress tumor cell growth of KRAS mutant CRC cells [
58]. On the basis of Oladipo and collaborators’ immunohistochemistry results of CRC FFPE samples (n = 254), CXCL1 protein level was significantly elevated in tumor tissue samples compared to normal adjacent tissue samples [
59]. Matrix-metalloproteinase-3 (MMP3) is highly expressed in colorectal carcinomas [
60], the encoded protein plays role in tumor invasion, lymph node involvement and metastatic spread [
61].
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
AK, OG, SS, KT, KL and BBK performed gene expression analyses, WB made statistical analysis, BSN carried out in situ hybridization experiments, ZST and BM contributed to the design and critical review of the manuscript. All authors read and approved the final manuscript.