Background: The morphology of the invasive margin in colorectal cancer can be described as either pushing or infiltrative. These phenotypes carry prognostic significance, particularly in node negative disease, and provide an excellent model for the study of invasive behaviour in vivo.
Methods: The marginal edges of 16 stage-matched tumours exhibiting these contrasting growth patterns were microdissected. The extracted mRNA was amplified and hybridised to a 9546 feature oligonucleotide array. Selected differentials were validated using real-time polymerase chain reaction and the protein product was interrogated by using immunohistochemistry.
Results: After stringent quality control and filtering of data generated, 39 genes were identified as being significantly differentially expressed between the two types of marginal edge. Several genes involved in cellular metabolism were identified as differentials including lactate dehydrogenase B (LDHB) and modulators of glucose transport.
Conclusions: The LDH expression profile differs between the invasive phenotypes. A hypothesis is proposed in which altered metabolism is a cause of contrasting invasive behaviour independent of the hypoxia-inducible factor mediated hypoxic response, consistent with the Warburg phenomenon.
Statistics from Altmetric.com
The “infiltrative” margin is a histologically identifiable tumour characteristic which confers a poor prognosis in colorectal cancer (CRC) and contrasts with the synonymous “pushing” or “expansile” margin.1 Dukes B stage adenocarcinomata have a heterogeneous prognosis. The nature of the invasive margin is an independent risk factor for survival and recurrence which has been confirmed by survival analysis and logistical regression analysis in two separate series.2 3 A recent publication has further demonstrated its prognostic significance in a stage-specific context.4 However, observer variation of this histological feature has limited its acceptance.5 6 Tumour “budding” describes a histological morphology at high-power magnifications with similarity to the infiltrative margin and is commonly utilised for prognostication in the clinical setting.7 8 Immunohistochemical studies have demonstrated correlation between tumour budding and expression of CD44v6 (a cell adhesion molecule (CAM)), the integrin ligand laminin-5 γ2 and the epithelial CAM Ep-CAM.9–12 In CRC, the invasive margin has recently been described as being an area of low proliferation relative to the tumour centre and in basal cell carcinoma, reduced proliferative activity is associated with an infiltrative growth pattern.13 14 Many studies have demonstrated that the tumour–stromal interface functions as a distinct entity owing to the dynamic interaction between epithelial and stromal compartments and may influence growth patterns.15–18 Margin morphology provides an in vivo model of contrasting invasive behaviour that has been little exploited in the laboratory. Elucidation of the differences in gene expression which underlie these contrasting growth patterns may lead to a greater understanding of the aetiology of invasion and hence metastasis, which remains the most common cause of treatment failure in the majority of visceral cancers.
Gene expression analysis of CRC has been used to investigate stage and metastasis-related profiles in recent publications.15 19–21 Laser capture microdissection (LCM) enables the study of tumour epithelium independent of stromal contamination thus simplifying the interpretation of results. The combination of these technologies has been employed to study differential expression between normal and malignant colonic epithelium as well as tumour and stromal tissues22 23 but to our knowledge this study is the first example of this technique being used to study contrasting invasive growth patterns. The aim of this study was to characterise gene expression within the tumour epithelium at the margin in the contrasting phenotypes in colorectal cancer in order to elucidate the molecular changes that orchestrate invasive growth.
Tumour sample selection and processing
A consultant histopathologist (NS) reviewed both prospectively collected and archival samples of Dukes B CRC to identify the best examples of the invasive phenotypes. The invasive margin refers to the tumour–stromal interface, typically at the deepest part of the lesion. The original definitions were used.1 All samples were Dukes B staged tumours and eight samples of each margin type were selected. Matched normal mucosa from these 16 samples was used as a pooled reference sample for microarray analysis (table 1). Tissue samples were cryopreserved within 45 min of extirpation and stored at −80°C. Exclusion criteria were applied to identify sporadic tumours in patients who had not undergone neo-adjuvant therapies since DNA damage from irradiation has the potential to confound expression data. Paraffin embedded specimens representing the same tumours were retrieved for subsequent immunohistochemical analysis.
Laser capture microdissection and RNA extraction
Serial frozen sections (6 μm thick) of tumour and normal mucosa were obtained using a Leica CM1850 cryostat (Leica, Nussloch, Germany). Immediate fixation in 70% ethanol was performed and the slides kept on dry ice for a maximum of 60 min prior to staining. Rehydration in RNase-free water was followed by haematoxylin and alcoholic eosin staining prior to dehydration in graded ethanol for 60 s and xylene for 6 min. The slides were then air dried for 20 min. Microdissection was performed using the PixCell I microdissector (Arcturus Bioscience, Mountain View, California, USA). Approximately 40 000 cells were captured per sample (2000–4000 pulses, 15 μm beam) depending upon tissue architecture and transfer quality. Representative digital images were acquired and reviewed by a histopathologist to ensure >95% specificity of captured tumour cells (fig 1). RNA extraction was immediately performed using the PicoPure RNA extraction kit according to the manufacturer’s instruction (Arcturus Bioscience). A DNase step was incorporated (Qiagen, Crawley, UK). A 1 μl sample of elutant was retained for quality control using the Agilent 2100 BioAnalyser with the RNA 6000 Nano LabChip Kit (Agilent Technologies, Stockport, UK). Full length transcripts were confirmed using the cDNA integrity kit (KPL, Gaithersburg, Massachusetts, USA). The remaining 10 μl eluted RNA was stored at −70°C.
Reverse transcription and SMART amplification
Polymerase chain reaction (PCR)-based amplification was performed using the SMART amplification system (Clontech, Palo Alto, California, USA) as described previously.24
To a 42 μl aliquot of amplified double-stranded cDNA was added 40 μl of 2.5× Random Primer reaction buffer prior to incubation at 95°C for 5 min. The reaction was quenched on ice, after which 5 μl 10× low-C dNTP mix, 2 μl Cy-3 or Cy-5 fluorescent label, 1 μl Klenow and 10 μl RNase free water (Ambion, Austin, Texas, USA) were added and incubated at 37°C for 16 h before addition of 5 μl stop buffer and storage in foil to prevent photo-bleaching. Purification was performed using Sephadex G-50 superfine columns (Sigma, St Louis, Missouri, USA) with a MultiScreen HV 96 well plate (Millipore, Billerica, Massachusetts, USA). The purified Cy-3 and Cy-5 dye-labelled targets were then combined according to the hybridisation pairs using pooled labelled product from normal mucosa samples as a reference. To these labelled pairs of sample cDNA, 1/10 volume of 3 mol/l sodium acetate and 2.5 volumes of 100% ethanol (VWR International, Lutterworth, UK) were added. Precipitation was achieved by incubation at −70°C for 30 min followed by centrifugation. Resulting pellets were washed in 750 μl of 70% ethanol before vacuum drying and storage at −20°C.
Quantification and hybridisation
Pellets were re-suspended in 20 μl hybridisation buffer. A 1.5 μl sample of the co-precipitated suspension was removed for quantification (NanoDrop ND-1000 spectrophotometer; NanoDrop Technologies, Rockland, Delaware, USA). To the remaining 18.5 μl was added 230 μl hybridisation buffer. The targets were then heated to 90°C for 5 min followed by 50°C for 5 min. The suspension was microfuged for 5 min at 13 500 g and cooled. Hybridisation to the HGMP_Human_Hs_SGC_Av1 oligonucleotide array (ArrayExpress ID: A-MEXP-52) containing 9546 transcript specific probes spotted in duplicate on the array was performed over 16 h at 37°C in duplicate serial runs. Hybridisation co-precipitate (200 μl) was injected into a chamber of an automated slide processor (Amersham Biosciences, Amersham, UK) which was loaded with sequential microarrays from the same print run. The slides were removed after a final washing step and scanned using the Agilent DNA microarray scanner (Agilent).
Data extraction and analysis
Data were extracted from tagged image file format (TIFF) images using the ImaGene software package (BioDiscovery, El Segundo, California, USA) and imported into the microarray analysis platform GeneSpring7.2 (Silicon Genetics, Redwood City, California, USA). Standard normalisation procedures were performed (Lowess per spot and per chip) in order to enable comparison of hybridisations with different intensities and background staining and to militate against systematic sources of bias.25 Self–self hybridisations were used to determine false discovery rates. Using this method, a false discovery rate was calculated to be 0.34% at a fold change of 1.7. This level was considered to be a reliable indicator of true differential expression. Three hybridisations were excluded from further analysis owing to a high proportion of poor-quality features, leaving 29 hybridisations for analysis. In this way, 67 genes were identified as being differentially expressed and their expression profiles were further interrogated using the Welch t test for independent samples on the normalised numeric data. This method yielded a final group of 39 differentially expressed transcripts between the pushing and infiltrative tumours (tables 2 and 3).
Real-time polymerase chain reaction
Non-amplified total RNA from the original microdissected cell extractions was used to enable direct comparison of transcript expression between the methodologies. Reverse transcription was performed using Superscript II RNase H Reverse Transcriptase (Invitrogen, Carlsbad, California, USA) according to the manufacturer’s instruction. Cyclin D2 (CCND2), lactate dehydrogenase B (LDHB) and serine protease 23 (PRSS23) were chosen for validation and β-actin used as a reference. Primers were designed to anneal within 1 degree of 59°C and were between 17 and 25 base pairs long, spanning an intron–exon boundary. Amplicons were between 99 and 135 base pairs in length and optimum primer concentration was established for all primers. The SYBR Green RT-PCR Kit (Applied Biosystems, Carlsbad, California, USA) was employed for real-time PCR amplification as previously described.26 The ABI Prism 7700 sequence detector system (Applied Biosystems) was employed at 40 cycles. Experiments were performed in triplicate and data were analysed using the ABI PRISM Sequence Detector v1.7 software package. Expression data were analysed using the Mann–Whitney U test.
Paraffin-embedded tissue blocks corresponding to the tumours used for microarray analysis were sectioned at 5 μm were transferred to Superfrost Plus microscope slides (VWR). Liver parenchyma was used as a positive control while omission of the primary antibody served as a negative control. Slides were de-waxed with xylene followed by dehydration in graded alcohols. Endogenous peroxidase activity was blocked by immersion of the slides in 0.6% hydrogen peroxide in methanol. Antigen retrieval was achieved by microwave treatment for 10 min at 900 W (EMS1065; Sanyo, Osaka, Japan). An avidin and biotin blocking step was performed using the Avidin/Biotin Blocking Kit SP-2001 (Vector Laboratories, Peterborough, UK) according to the manufacturer’s instruction. Further antigen blocking was performed using a 1:20 dilution of host serum in Tris-buffered saline (TBS) with 1% w/v bovine serum albumin (Sigma) for 5 min. The primary antibody (100 μl, 1:200 dilution), sheep anti-human polyclonal antibody to lactate dehydrogenase isoenzyme V (ab9002; AbCam, Cambridge, UK), was added before incubation at room temperature for 1 h. The secondary swine anti-sheep antibody (100 μl, 1:200) was incubated at room temperature for 30 min (DakoCytomation, Ely, UK) prior to the addition of Avidin–Biotin Complex/HRP (DakoCymotation) for 30 min at the manufacturer’s recommended concentration. Visualisation was achieved using the 3,3′-diaminobenzidine (DAB) staining kit (Vector Laboratories), as per the manufacturer’s instruction, for 10 min prior to washing, haematoxylin counterstaining, dehydration and mounting in DePeX (Serva, Heidelberg, Germany). Areas of the tumour representing the invasive margin were defined and marked by consensus between the two principal investigators (CT and DJ) using sections stained with haematoxylin and eosin. Images were assessed by semi-quantitative analysis with the Lucia image analysis system27 in which staining intensity in 50 regions of the invasive margin were assessed by the software. The median staining intensity at the tumour–stromal margin was recorded for analysis (Mann–Whitney U test).
RNA extractions yielded a median of 260 ng of total RNA (inter-quartile range, 178–370 ng). RNA sample integrity was confirmed using the Agilent Bioanalyser 2100. Following amplification, a median of 11.76 μg of cDNA was generated (inter-quartile range, 6.87–12.76 μg).
Data from hybridisations were initially submitted to unsupervised hierarchical clustering. This demonstrated a clear separation of normal mucosa from tumour samples and close pairing of technical replicates (data not shown). Using genes identified as differentially expressed, all but one of the pushing tumours clustered together whilst the infiltrative tumours clustered in two distinct groups (fig 2).
Differentially expressed genes
Seven hundred and ninety-four genes were identified as being differentially expressed between malignant and normal colonic epithelium. Thirty-nine genes were found to be differentially expressed between the phenotypes (tables 2 and 3) using the filtering steps and quality control measures described.
Validation by real-time polymerase chain reaction
The expression of lactate dehydrogenase B (LDHB), cyclin D2 (CCND2) and serine protease 23 (PRSS23) were interrogated by real-time PCR in order to validate the microarray results. Transcript expression identified using these two modalities are displayed in fig 3 and tables 2 and 3. Expression distributions are seen to be closely matched and retain significant differential expression after quantitative analysis confirming the overall accuracy of the microarray expression data.
Protein expression of lactate dehydrogenase
The enzyme lactate dehydrogenase is a tetramer composed of subunits coded for by the lactate dehydrogenase A (LDHA) and LDHB genes. A high proportion of LDHA subunits leads to the formation of the LDHV isoenzyme, which has a higher catalytic affinity for the conversion of pyruvate to lactate and hence results in the complete glycolysis of hexose sugars.28 Results from the microarray analysis demonstrate a significant relative reduction in the expression of LDHB in the infiltrative tumours, while the relative expression of LDHA transcripts is equal between the phenotypes (fig 4). As demonstrated in table 4, this results in the generation of LDHV. Immunohistochemistry demonstrated a non-significant trend towards increased expression of LDHV in the infiltrative tumours by the semi-quantitative method (fig 5).
Several of the differentially expressed genes identified in this study have established associations with invasion and prognosis, consistent with the tumour phenotypes under study, including endothelin 1 (EDN1),29–35 homeobox B7 (HOXB7),36–38 v-src (SRC),39–42 dipeptidylpeptidase 4/CD26 (DPP4)43–48 and beta-parvin (PARVB).49 In addition, genes involved in cell cycle regulation were upregulated in the infiltrative phenotype, including retinoblastoma (RB1) (1.9-fold), JARID1A (formally retinoblastoma binding protein) (1.9-fold) and cyclin D2 (CCND2) (2.9-fold), suggesting a modulation in proliferative activity between the growth patterns, again implicated in metastatic spread and tumour invasion.50–52 However, the differential expression of LDHB and of certain related transcripts will form the remainder of this discussion.
Lactate dehydrogenase (LDH) exists as a tetramer with five isoforms formed by the random association of LDHA and LDHB subunits. Therefore, over-expression of LDHA or suppression of LDHB favours the formation of cathodic LDH (isoenzymes IV and V) and generation of lactate (table 4). Over-expression of LDHV is associated with aggressive histological features in colonic polyps, advanced tumour grade, poor prognosis, upregulation of vascular endothelial growth factor (VEGF), and hypoxia.53–57
Expression of LDHV isoenzyme has been considered to be primarily orchestrated by hypoxia-inducible factor (HIF)-modulated LDHA transcript upregulation in acute hypoxia.58 59 This model, however, has been criticised for being poorly representative of the chronic and cyclical hypoxic tumour environment in vivo and does not take into account other factors such as concomitant low extracellular pH which inhibits hypoxia-induced LDHA expression.60–64 Moreover, LDHA was not identified amongst upregulated proteins in hypoxic prostatic carcinoma cells65 and was also not found to be upregulated in a microarray analysis of hypoxia and dimethyloxaloylglycine (DMOG)-inducible transcripts in breast cancer and hepatoblastoma cell lines.66
Whilst the influence of LDHA in the regulation of LDH isoforms is well established, the role of LDHB has yet to be fully elucidated. However, LDHB suppression has been identified as the primary mediator of LDH isoenzyme regulation in vivo during the murine physiological adaptation to hypoxia.67 LDHB suppression by hypermethylation has been observed in gastric cancer in vitro and prostate cancer in vivo and its expression is inversely proportional to metastatic progression.68 69 LDHA knockdown mice demonstrate a reduction in tumour proliferation rate under hypoxic conditions and orthotopic tumour development is less aggressive.70 The relative suppression of the LDHB transcript observed in the infiltrative phenotype favours the accumulation of cathodic isoforms and hence lactic acid. The development of an acidic extracellular compartment generates a pro-invasive environment, inducing apoptosis in untransformed cells and enhancing extracellular matrix degradation.71
A metabolic phenotype characterised by upregulation of glycolytic pathways is postulated to occur early in tumour development resulting from growth beyond the diffusion limit of oxygen.72 73 A preference for glycolytic metabolism in tumour cells, independent of local hypoxia, was first described by Warburg and may be referred to as “aerobic glycolysis” or “pseudohypoxia”.74 Elevated glucose uptake has been correlated with poor prognosis although not independently of hypoxia.75 76 A predilection for glycolysis rather than a response to hypoxia may account for the observed difference in LDHB expression between infiltrative and pushing growth patterns. In this case, infiltrative tumours would be anticipated to upregulate genes supportive of glycolysis in the absence of HIF1A mediated transcript expression. Twenty-three HIF1A regulated transcripts were represented on the array of which only two were differentially expressed: EDN1 was upregulated by 2.7 and the transferrin receptor (TFR) was downregulated by 1.9 in infiltrative cancers (tables 2 and 3). The HIF1A transcript was not differentially expressed and is regulated at a post-transcriptional level. The hypoxic marker carbonic anhydrase-9 was not differentially expressed between the phenotypes. These findings support the hypothesis that local tumour hypoxia is not responsible for the differential expression of LDH transcripts.
No differential expression of the glycolytic enzymes was observed between the phenotypes although phosphoglycerate kinase and glyceraldehyde-3-phosphate dehydrogenase expression were elevated in tumour samples compared with normal mucosa. The glucose transporter family (GLUT), were not represented on the array; however, two modulators of GLUT4, insulin receptor substrate 2 (IRS2) and flotillin 1 (FLOT1) were upregulated in the infiltrative phenotype (table 2). GLUT4 expression is positively correlated with stage in ovarian cancer and, interestingly, is not associated with hypoxia.77 78 IRS2 contributes to the translocation and activation of the GLUT4 glucose transporter, activates the insulin receptor and contributes to invasion.79 80 FLOT1 facilitates GLUT4 translocation to the cell membrane.81 Increased glucose transport is normally achieved by HIF-mediated upregulation of GLUT1 and GLUT2 whereas GLUT4 provides a hypoxia independent mechanism.
Control of LDHB expression remains poorly characterised. MYC is a transcriptional regulator of LDHB. To determine whether MYC transcriptional activity might be involved in the generation of the infiltrative phenotype, the microarray data were interrogated for known MYC transcriptional targets. Seven of the 39 (18%) differentially expressed genes described are known transcriptional targets of MYC and their expression in the phenotypes was consistent with the influence of this modulator in all cases in which it has been described.82 Differential MYC activity between the phenotypes may therefore provide a HIF/hypoxia independent mechanism to account for the differential expression of LDHB.
The employment of LCM and the restriction to stage-matched tumours minimises confounding variables to provide a robust methodology for the study of the molecular characteristics of invasive growth. The infiltrative phenotype in colorectal cancer is associated with increased transcript expression of modulators of glucose transport in addition to a putative shift in LDH expression to the cathodic isoform, modulated by suppression of the LDHB gene. This metabolic switch appears to be independent of the local oxygen environment, evidenced by a paucity of HIF target differential expression and therefore might represent the Warburg phenomenon.
We thank Dr T Freeman and his team, formerly based at the Rosalind Franklin Centre for Genomic Research, Hinxton, Cambridge, for invaluable assistance in sample amplification, hybridisation and data analysis; and Dr R Banks and her team at the department of Clinical and Biomedical Proteomics Research, University of Leeds, for the use of LCM hardware and advice.
Funding: I would like to thank the Medical Research Council who provided the microarrays through the Rosalind Franklin Centre for Genomic Research, Hinxton, Cambridge, UK.
Competing interests: None.
Ethics approval: Approval for this study was given by the Leeds Regional Ethics Committee on 21 January 2003 (reference LHA 02/318). Patient consent was obtained.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.