Whole Transcriptome Analysis of Pre-invasive and Invasive Early Squamous Lung Carcinoma in Archival Laser Microdissected Samples

All FFPE biopsies were selected by a specialist thoracic pathologist, sectioned using a Leica RM2125 microtome and stained with Hematoxylin and Eosin. PSCC, ISCC cell clusters and normal epithelia were microdissected as follows: each FFPE biopsy was used to produce a series of a single 5μm section, transferred to a standard glass slide for diagnostic evaluation by specialist thoracic pathologists, and on average ten 10μm sections were each placed on 1mm PEN membrane slides (Carl Zeiss, Germany) for LCM using a Leica LMD6000 (Additional file 1: Figure S1). RNA was isolated using the Ambion Recover All Kit. Isolated RNA was quantified using a Qubit fluorometer (Qubit RNA assay kit) and RNA integrity was examined using an Agilent Bioanalyser. Samples with RIN values in the range 2–3 were employed for microarray analysis. All donors gave written informed consent and the conducted research was approved by the South Manchester Ethics Committee

Gene expression was analysed as previously described [16]. 50ng of total RNA was used for amplification and reverse transcription of individual samples using the Nugen Ovation FFPE WTA Kit, followed by biotin labelling and library fragmentation via the Nugen Encore Biotin Kit. Affymetrix Human Exon 1.0 ST array hybridisation, washing, staining and scanning was performed at the Molecular Biology Core Facility of the CRUK Manchester Institute.

Comparative RT-PCR was performed to validate expression changes of candidate genes between ISCC and paired normal biopsies by using the delta delta Ct method [17]. All primers were designed using Primer-BLAST (see Additional file 2: Table S1). PCR reactions were setup using Applied Biosystem’s Fast SYBR green master mix and were run in triplicates on an Applied Biosystems 7900HT Real-Time PCR system. Primers targeting 28S were used for the endogenous control in all assays [18].

Exon array quality control and outlier detection was performed using dChip (www.​dchip.​org, [19]). For normalization and expression analysis of the array data, the implementation of the RMA algorithm in Partek GS 6.6 (Copyright 2010, Partek Inc., St. Charles, MO, USA) using core probesets was used. Differential expression tests were performed with the Bioconductor package limma using paired designs [20]. The Bioconductor package QVALUE was used to calculate the corresponding q-values [21]. Principal component analysis (PCA) of normalized expression data from the included exon arrays was performed using the R package FactoMineR (https://​cran.​r-project.​org/​web/​packages/​FactoMineR/​). Gene Set Enrichment Analysis was used to calculate positional enrichment of abnormal transcription using pre-ranked list of aberrantly expressed genes in all ISCC or all PSCC and positional gene sets available from MSigDB (http://​www.​broadinstitute.​org/​gsea/​msigdb) according to previously published procedures [22]. RCircos was used to create the circus plot [23]. For clustering analysis, 1914 differentially expressed genes from the three contrasts, i.e., all ISCC versus control samples, all PSCC samples versus control samples and PSCC high-grade versus PSCC low-grade, were selected and filtered for p-value <0.05 and fold change greater than +/−1.75. For the clustering analysis, 4 data points were used; average controls, average PSCC low-grade, average PSCC high-grade and average ISCC. Averages are calculated in log base 2 and were standardised (standard deviation normalised to 1 and mean to 0). Genes were clustered according to these standardised expression levels by k-means into 12 clusters followed by ranking by hierarchical clustering using maxdView software "Super Grouper" plugin (available from http://​bioinf.​man.​ac.​uk/​microarray/​maxd/​). The functional analyses of differentially expressed genes were generated through the use of QIAGEN’s Ingenuity Pathway Analysis (IPA®, QIAGEN Redwood City, https://​www.​qiagenbioinforma​tics.​com/​products/​ingenuity-pathway-analysis/​). Microarray data has been uploaded to the ArrayExpress database (www.​ebi.​ac.​uk/​arrayexpress) under accession number E-MTAB-3950.

Immunohistochemistry on FFPE lung biopsies was performed on the Leica Bond platform, which included standard procedures for the removal of paraffin wax, section rehydration, epitope retrieval (Leica Bond Epitope Retrieval Solution 2, 20 min, Leica Biosystems, Germany), blocking of endogenous peroxidases for 5 min and blocking with 10% casein for 10 min. The primary antibody (Anti-PTTG1, Sigma-Aldrich HPA008890, produced in rabbit) was used at 5 μg/ml for 15 min followed by treatment with the Leica Bond Refine Kit (8 min, Leica Biosystems, Germany), Leica Bond DAB (10 mins, Leica Biosystems, Germany) and a counterstaining with hematoxylin prior to rehydration, clearing and coverslipping. Images were taken on a Leica SCN 400.

FFPE biopsies used in this study originate from the Manchester Cancer Research Centre Biobank and sample interpretation was undertaken by three specialist thoracic pathologists who classified areas of normal bronchial epithelium, PSCC and ISCC. Pathologists also subdivided PSCC into low- or high-grade lesions, according to mild and moderate squamous dysplasia (low grade dysplasia) and severe squamous dysplasia and carcinoma in-situ (CIS) (high grade dysplasia). Final sample classifications were based on two concordant reviews; those without a majority agreement or technically uninterpretable were excluded. Using this approach, we obtained 15 ISCC regions (from surgical specimens) and 20 PSCC regions (from bronchial biopsies) from 25 patients (Fig. 1a, Additional file 2: Table S2). ISCC samples include 10 primary tumours (5 lymph node-negative (TN0), 5 lymph node-positive (TNx)) and 5 paired lymph node metastases (Nx) to the primary TNx samples. PSCC samples were composed of 8 high-grade and 12 low-grade lesions. The median age of the 17 male (68%) and 8 female (32%) donors at the time of sampling was 65years (range 45-79years). 72% (18/25) where treated for ISCC, 8% (2/25) for head and neck squamous cell carcinoma and another 8% (2/25) for CIS. Three patients (12%) had no treatment. A history of smoking was recorded in 52% (13/25) of the donors.

RNA libraries isolated from 15 ISCC (and 10 paired samples of control normal bronchial epithelia) and 20 PSCC (and 15 paired samples of control normal bronchial epithelia) using micro-dissection were amplified and converted into cDNA for hybridisation using Exon 1.0 ST arrays. Principal Component Analysis (PCA) was performed in order to visualize variability of the entire normalized expression data from all arrays included in this study (Fig. 1b). In this PCA, ISCC and PSCC data points show a clear tendency of being separated from each other by mainly the first principal component (explains 28.13% of the observed variability in gene expression), which suggests that the general gene expression in ISCC and PSCC is strikingly different (Fig. 1b and 95% confidence intervals therein). Controls paired to ISCC or PSCC were widely distributed across the PCA plot but did not show any significant clustering according to sample origin (surgical or bronchial biopsies) or their various locations (see Additional file 2: Table S2) (Fig. 1b). However, a PCA of low-grade and high-grade PSCC showed widely spread low-grade PSCC data points that overlap with high-grade PSCC data points (Fig. 1c), and similarly, no clear clusters of TN0, TNx or Nx ISCC data points were observed in a PCA (Fig. 1d), which suggests that the observed variability of gene expression changes in these samples do not differentiate these distinct histological categories. Furthermore, the observed gene expression was clearly different between ISCC and high-grade PSCC, which were previously reported with similar changes to chromosomal structures and which were suggested to be more prone to malignant transformations (Fig. 1d) [13, 24, 25].

Following comparison with matched normal epithelium, 1200 RNAs were differentially expressed in ISCC samples (p < 0.05 and fold change ±1.75, Additional file 2: Table S3). As might be expected, a considerably smaller number of 184 RNAs were differentially expressed in PSCC samples (Additional file 2: Table S4). However, 142 (77%) of these RNAs were abnormally expressed in both PSCC and ISCC (Fig. 1e, Additional file 2: Table S5). Further comparative analysis of the high-grade and low-grade preinvasive samples and their controls confirmed differential expression of 243 (Additional file 2: Table S6) and 183 RNAs (Additional file 2: Table S7) respectively, of which 99 RNAs (40% and 54% respectively) were differentially expressed in both low-grade and high-grade groups (Fig. 1e, Additional file 2: Table S8). Hence, the obvious difference in the amount of abnormal gene expression between ISCC and PSCC suggests a substantial molecular difference between these two conditions, but no such dramatic changes were found between low-grade and high-grade PSCC, which in the case of PSCC reasons against a process of gradually added modifications to gene transcription. Nevertheless our analysis identified abnormal gene expression shared between ISCC and PSCC that supports the assumption of a common ancestry and the possibility of molecular events that could drive the progression of squamous lesions.

We used Gene Set Enrichment Analyses (GSEA) with pre-ranked lists of differently expressed genes derived from all ISCC or PSCC samples versus paired controls using the positional gene sets provided by MSigDB in order to correlate aberrant transcription to chromosomal positions as previously reported for PSCC and ISCC [22, 24]. Significantly enriched gene up-regulation in ISCC was found for chromosomes 2p25, 3q21, 3q24, 3q26-28, 6p22, 15q11 and 18q12 (Fig. 2a). Similarly, gene down-regulation in ISCC was significantly enriched at 2q35, 9p13, 10q11, 11p14, 11q12, 12q14 and 18q21 (Fig. 2a). Interestingly, significant positional gene expression changes at the positions 2q35, 10q11, 11q12, 12q14 (down-regulation) as well as 3q21 and 18q12 (up-regulation) were detected in both ISCC and PSCC (Fig. 2a). As reported previously, this analysis confirms elevated levels of gene expression in ISCC and, to a lesser extent, PSCC at the distal part of chromosome 3q (Fig. 2a). Conversely, chromosome 3p confirmed a moderate degree of down-regulation, most evident in ISCC (Fig. 2a). This data supports and extends previous observations of structural changes to chromosomes in PSCC that are thought be important in the phenotypic transformation to invasive malignancies and further suggests the propagation of aberrant transcriptional activity during squamous carcinogenesis [11, 13, 22].

Ingenuity pathway analysis (IPA) was employed to identify canonical pathways linked to genes that were differentially expressed in ISCC, total PSCC, high-grade PSCC and low grade PSCC (Fig. 3 and Additional file 2: Table S9). The widespread changes in genes expression in ISCC samples were associated with 40 significantly affected pathways (P: <=0.05) including those characteristic for invasive malignancies such as ‘Cell Cycle: G2/M DNA Checkpoint Regulation’ (p-value: 2.18e-6, z-score: −1.414), ‘ATM signaling’ (p-value: 0.00173, z-score: −1.134), ‘p53 Signaling’ (p-value: 5.75e-5, z-score: −0.905) and ‘GADD45 Signaling’ (p-value: 9.12e-6) [25‐27]. IPA further suggested ‘Cell Cycle: G2/M DNA Checkpoint Regulation’, pathways concerned with DNA and RNA metabolism and inflammation-related processes to be affected in all contrasts, which is line with the finding of differential gene expression shared between ISCC and PSCC. Overall, this pathway analysis further supports the existence of shared aberrant gene expression in the analysed samples with potential to affect cellular functions that may drive squamous carcinogenesis across pre-invasive and invasive stages.

A cluster analysis was performed using 1914 genes that are differentially expressed in either of the low-grade PSCC, high-grade PSCC or ISCC samples in order to identify genes that share a particular expression pattern across these samples. Among the 12 identified clusters, three (clusters 4, 6 and 12) showed a gradual increase of expression of 702 genes (36.7% of 1914 genes) across controls, PSCC and ISCC (Fig. 2b). Signalling pathways associated with these genes by IPA cover functional themes of DNA damage response and cell cycle regulation, including ‘cell cycle: G2/M DNA checkpoint regulation’ (z-score: −0.447), ‘p53 signaling’, ‘GADD45 signaling’ and ‘mitotic roles of Polo-like kinases’, and other pathways associated with tumor development, such as ‘Rac signaling’ (z-score: 2.449), ‘RhoA signaling’, ‘ERK/MAPK signaling’ (z-score: 2.236) and ‘PI3K/AKT signaling’ (z-score: 2.449) (Fig. 2b and Fig. 3, see Additional file 2: Table S10 for full details) [28]. Two clusters (3 and 5) of 596 genes (31.1% of 1914 genes) were found gradually down regulated from control to PSCC and then ISCC. These included several cytochrome P450 genes (CYP4B1, CYP2B6, CYP4Z1) that indicates a reduced metabolism of potential carcinogens (see Additional file 2: Table S10). In addition, we identified 243 RNAs in cluster 11 (12.7% of 1914 genes) that demonstrated an increase in expression levels only in ISCC that were linked to changes in the extracellular matrix and an increase in inflammation-related processes (‘IL-17 signaling’, ‘IL-8 signaling’, ‘agranulocyte/granulocyte adhesion and diapedesis’, see Additional file 2: Table S10). Hence, using cluster analysis and IPA we were able to sort genes with similar expression profiles across PSCC and ISCC (see Additional file 2: Table S11) and to identify biological themes associated with these profiles, which refines our understanding of the molecular pathology that underlies squamous carcinogenesis. As summarised in Fig. 3, IPA pathways associated with ISCC are mainly, but not exclusively, linked to a gene up-regulation.

To validate the microarray data, we undertook RT-PCR to confirm changes in gene expression of four candidate genes, PTTG1, FN1, HIF1alpha and ITGAV that have been reported to drive tumorigenesis. As a result of the availability of independent samples, this could only be performed in ISCC samples. As with the microarray data, RNA expression of these four genes was found to gradually increase across the tested histology and to peak in ISCC samples. Thus, the observed upregulation of these four genes was confirmed by RT-PCR: PTTG1 (P: 0.0015), FN1 (P: 0.0011), HIF1α (P: 0.0017) and ITGAV (P: 0.0082) in ISCC samples with fold changes comparable to the results obtained from the Human Exon arrays (Fig. 4a). In addition, we performed cross-validation of PTTG1 up-regulation using FFPE tissue microarrays of 45 unrelated ISCC and paired normal bronchial epithelia by immunohistochemistry. The obtained antibody stainings were quantified by an experienced pathologist using the H-score system, which revealed a significant up-regulation of PTTG1 in the cytoplasmic (mean in tumor: 124.1, mean in control: 79.78, P: 3.97e-08) and nuclear (mean in tumor: 125.4, mean in control: 30.66, P: 4.35e-16) compartments of tumor cells in comparison to cells in paired pseudo stratified epithelia that were used as controls. As suggested by the H-scores, a prevalent nuclear localization of PTTG1 protein was noted in ISCC (Fig. 4b and c). Hence, we detected a significant up-regulation of PTTG1 in malignant squamous cells from an independent set of ISCC that confirms the obtained array data and is in line with previously published PTTG1 expression in lung and other cancers [29].