Proteogenomic characterization and comprehensive integrative genomic analysis of human colorectal cancer liver metastasis

The study was examined and approved by the Ethics Committee of the Shanghai Tenth People’s Hospital, Tongji University School of Medicine (SHSY-IEC-PAP-16-24). This study was registered with ClinicalTrials.​gov, number NCT02917707. Each participant provided their written informed consent to participate in this study. The inclusion criteria included: age ≤ 75 years with histologically proven CRC, no severe major organ dysfunction, WHO performance status of 0 or 1, or no prior cancer chemotherapy. The exclusion criteria included: age ≥ 76, severe major organ dysfunction, World Health Organization (WHO) performance status of > 1, or prior cancer chemotherapy. The morphology of primary CRC and paracarcinoma normal colorectal tissues was confirmed by two independent pathologists using cryostat frozen sections stained with hematoxylin and eosin. Follow-up data and statistics were recorded for all patients through Dec. 31, 2017.

Using a co-isolation protocol, DNA and RNA were purified simultaneously using the QIAGEN All Prep DNA/RNA Micro Kit (Qiagen, CA, USA) according to the manufacturer’s instructions. The nucleic acid concentration was determined using a Nanodrop1000 spectrophotometer (Thermo Fisher Scientific; Waltham, MA, USA), and the RNA purity was verified using 1.5% denaturing agarose gels.

Fresh CRC tissues and para-tumor normal colorectal tissues (PN) were used for proteogenomic analysis. Three different parts of the same lesions for every sample were compared for data analysis and measurement of the variation caused by random biological effects. The samples were cut into small pieces (about 1 mm³) and rinsed in PBS to remove the blood. Then the tissues were homogenized in 4% SDS and 0.1 M DTT in 0.1 M Tris-HCl, pH 7.6 on ice, sonicated 10 times (80 w; 10 s sonication/5 s suspension), incubated for 3 min at 95 °C, and briefly sonicated. The protein concentrations of clarified lysates were determined using a fluorescence assay and then 200 μg of clarified lysates were proteolyzed on a 10 kDa filter (PALL Life Sciences, Shanghai, China) using the filter-aided sample preparation method [7]. The peptide samples were then desalted onto a solid-phase extraction cartridge. The lyophilized peptide mixture was re-suspended in water with 0.1% formic acid (v/v), and its content was estimated by ultraviolet light spectral density at 280 nm [8]. Then, 3 μg of the digest sample was analyzed by nano-liquid chromatography-tandem mass spectrometry on a LTQ Orbitrap Velos Pro mass spectrometer as previously described [9].

The acquired data from mass spectrometry runs were combined and searched against the UniProt Human database (05/2016, 153,652 entrys) using Maxquant software (version 1.3.0.5; http://​maxquant.​org/) as described [10]. Proteins were identified using the Andromeda peptide search engine integrated into the Maxquant platform. Trypsin-digested fragments were analyzed, allowing for a maximum of 2 missed cleavages. Carbamidomethyl cysteine was set as a fixed modification, with protein N-acetylation and methionine oxidation as variable modifications. Precursor ion tolerances were 20 ppm for first search and 6 ppm for a second search. The MS/MS peaks were de-isotoped and searched using a 20 ppm mass tolerance. The required minimum peptide length for identification was 7 amino acids, and the false discovery rate at the protein level, peptide level and site were set to 0.01. The normalized spectral protein intensity (label-free quantification) values were calculated for each protein group.

The Maxquant peptide and protein quantification result files were imported into Perseus software (version 1.5.1.6) to identify the differentially expressed proteins. After importing the quantitative data from ProteinGroups.txt into Perseus, a filtering criterion is set to keep the identified proteins with the quantified values of all ten reporter ions (no missing value) in the final identification list. The protein intensities are log₂-transformed and normalized by subtracting the median intensity in each column/sample. Principal component analysis is performed based on protein intensities to differentiate groups. Two-samples tests coupled with Benjamini–Hochberg (FDR cutoff of 0.05) correction are performed to identify the differentially expressed proteins.

Six specimens from 2 patients (including 2 triplet sets of primary MT, matched CLM and PN) were used for chromosome microarray analysis and whole exome sequencing analysis. DNAs and cRNAs were hybridized to the Affymetrix CytoScan HD Array as described and recurrent genomic regions with DNA copy gain and loss were identified using GISTIC, version 2.0 [11]. Genomic DNA was enriched for exonic regions using the SureSelect Biotinylated RNA Library. The sequences of captured libraries were generated as 90-bp pair-end reads on an Illumina Hiseq2000. Sequencing reads were processed and mapped to the reference GRCh37/hg19 human genome assembly using the Burrows-Wheeler Aligner as described [12]. Further processing, including duplicate removal, local realignment, base quality recalibration, and filtering, as well as the identification of SNVs and indels, was performed using the Genome Analysis Toolkit [13], SAM [14], and Picard tools (http://​picard.​sourceforge.​net). Then, filters were applied to obtain higher confidence, and annotation and classification were performed using ANNOVAR [15]. The variant collection was excluded from positions reported in the 1000 Genomes Project and dbSNP. The mean sequencing depth in the target regions was 80.28× (range 71.5 to 92.85).

The reliability of the exome analysis and somatic variant identification strategies was assessed using PCR and Sanger sequencing. PCR was performed using the GeneAmp PCR System 9700 (Applied Biosystems, Foster City, CA, USA). About 20 ng template DNA from each sample was used per reaction. The products were sequenced, and all sequences were analyzed with the Sequencing Analysis Software Version 5.2 (Applied Biosystems).

A panel comprising 120 positive sites identified by Sanger sequencing were selected. These single nucleotide polymorphisms were located within genes of different functional categories. For the PCR amplification and single base extension reaction, primer pairs and extension primers were designed using Assay design suite v2.0. These primers were multiplexed and genotyped using the Sequenom MassARRAY platform integrating the iPLEXSBE reaction and MassARRAY technology (Agena Bioscience, San Diego, CA, USA) based on the MALDI-TOF MS assay [16].

Hierarchical clustering was performed using MEV software (http://​mev.​tm4.​org/​, v4.7.0, TIGR). The matrix was presented graphically by colouring each expression result on the basis of measured colour range: lower limit ‘0.0’ was coloured green, upper limit ‘369.5’ was coloured red and midpoint value ‘37.5’ was coloured black. Pearson correlation was used as distance metric and the complete linkage method was used. To identify genes/proteins that are specifically dysregulated in CLM, we fixed the cutoff at 2-fold with a P value less than 0.05. Dysregulated genes/proteins were subjected to GO analysis and KEGG pathway analysis by DAVID (http://​david.​ncifcrf.​gov). Pathway analysis is used to map genes to KEGG pathways. The P value denotes the significance of the pathway correlations (P value < 0.05 is recommended).

We downloaded RNA-sequencing data from 379 CRC patients from TCGA portal (https://​cancergenome.​nih.​gov/), 12 of which had liver metastasis at the time of diagnosis or during the five-year follow-up period, and 367 of which had CRC without metastasis to the liver. The mRNA expression levels were investigated in 379 CRC tissues and 32 PN tissues in TCGA datasets by Illumina HiSeq 2000 RNA Sequencing Version 2 analysis and normalized by the RSEM algorithm. Whole-exome sequencing mutation datasets were downloaded from TCGA data set to create a customized CRC mutation database. The clinical information recorded, including the patient’s characteristics, tumor characteristics, and overall and progression-free survival was assessed.

Human CRC cell line SW480 were purchased from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China) and cultured in DMEM media (Invitrogen, Carlsbad, USA) and supplemented with 10% (v/v) fetal bovine serum, 100 U/ml penicillin, and 100 mg/ml streptomycin. SW480 cell lines were routinely tested for mycoplasma contamination, and have been authenticated with short-tandem repeat analysis. Cell culture was conducted at 37 °C in a humidified 5% CO₂ incubator. For COL1A2 and BGN over-expression, the human full length cDNA with or without point mutation were cloned into the pMSCV-hygro vector. The SW480 cells with stable over-expression were polyclonal derivatives with hygromycin selection to avoid clonal variations in functional assays.

The human CRC SW480 cells were conducted at 37 °C in a humidified 5% CO₂ incubator and cells were grown into confluency in 6-well plates. The monolayer was artificially injured by scratching across the plate with a 200 ul pipette tip. The wells were washed 3 times to remove detached cells or cell debris. After 12 h, digital images were captured using a camera-equipped, inverted microscope (Carl Zeiss, Inc., Thorwood, NY, USA) and wound width measurements were subtracted from wound width at time zero to obtain the net wound closure.

Corning Costar Transwell 24-well plates with 8-um-pore-size polycarbonate membrane filters (Costar, Cambridge, MA) coated with BD Matrigel matrix (Becton Dickinson, Bedford, MA) were maintained for 1 h at 37 °C, followed by the addition of 1 × 10⁵ transfected cells suspended in 200 μl medium with 1% serum into the top of each well insert. Normal growth medium was added to the bottom wells. The cells were allowed to migrate for 24 h at 37 °C. The migrated cells were fixed with 10% methanol for 15 min. The invading cells on the lower surface of the membrane were stained with 0.5% crystal violet for 5 min at room temperature. Random fields were photographed and the stained cells were counted under a microscope (Nikon Corporation).

Data were expressed as means ± standard deviations. Categorical data were reported as numbers and percentages. F tests were used to assess the equality of variances for comparable groups. Paired t test, One-way analysis of variance (ANOVA), Kruskal-Wallis test, and χ² tests were used to analyze mRNA expression. Forty four paired fresh CRC and PN tissues were used for survival analysis (Table 1). OS was measured from the date the patient underwent surgery until the date of death resulting from any cause or last known follow-up for patients still alive. DFS analysis was measured from the date the patient underwent surgery to the date of disease recurrence, death from any cause (ie, noncancer deaths were not censored), or until last contact with the patient. For time-to-event analyses, survival estimates were calculated by the Kaplan-Meier analysis, and groups were compared with the log-rank test. Clinical variables that were considered for single variable analyses were previously identified as confounding variables with impact on the prognosis of patients with colorectal cancer: age at diagnosis (continuous), sex, primary site (colon vs. rectum), pathological differentiation (well to moderate vs. poor), completeness of colorectal resection (R0 vs. R1), tumor size (≥ 5 cm vs. < 5 cm), number of primary foci (multiple vs. single) and necrosis (yes vs. no). The Spearman’s correlation coefficient was used to test the relationship of two independent groups. To identify genes/proteins that are specifically dysregulated in CLM, we fixed the cutoff at 2-fold with a P value less than 0.05. All calculations were performed with SPSS 20.0 software (SPSS Inc., Chicago, IL, USA).

Genomic features and proteomic analyses of CRC have been characterized; however, the primary genetic basis of CLM has not been fully elucidated, which is essential for discovering metastasis-specific molecular biomarkers and for devising a better therapeutic approach for this disease. To address these issues, we performed a nano-liquid chromatography-tandem mass spectrometry (LC-MS/MS)-based shotgun proteomics profiling of 2 sets of triplet samples comprised of para-tumor normal colorectal tissue (PN), primary CRC tissue (MT), and synchronous matched liver metastatic tissue (LM) (Fig. 1). Three different parts of the same lesions for every sample were compared for data analysis and measurement of the variation caused by random biological effects. A total of 596,234 spectra were used in the Andromeda engine search, and 26,375 unique peptides were identified in an assembly of 4198 protein groups with a protein-level false discovery rate of 1.0%. Ingenuity pathway analysis with all 4198 identified proteins showed that about 51% of the proteins were from the cytoplasm, 26% were from the nucleus, 9% were from the plasma membrane and 5% were from the extracellular space, whereas 9% of proteins remained unclassified (Fig. 2a). The random predicted cellular distribution of the proteins supports the quality of the sample preparation.

A scatter plot of protein abundance (label-free quantification intensity) between CRC and PN tissues showed that there was a great variation between the MT or LM tumors and PN tissue (Fig. 2b and c). However, the protein expression between the MT and LM group was positively correlated (R² = 0.78) (Fig. 2d). These results suggest that liver metastasis focus and primary focus share similar protein profiles and that there are common molecular alterations at each stage of tumor development.

To identify proteins that are specifically dysregulated in CLM, we fixed the cutoff at 2-fold with a P value less than 0.05. Among the 4198 proteins, a total of 1041 proteins were significantly altered between MT and PN tissue, 636 (61.09%) of which were down-regulated and 405 (38.91%) of which were up-regulated (Fig. 2a, left bars). There were 754 proteins with significantly difference in LM tissues when compared with PN tissues and 632 proteins with significantly difference in LM tissues when compared with MT tissues (Fig. 2d). Among that, 656 (47.33%) of which were down-regulated and 730 (52.67%) of which were up-regulated (Fig. 2a, middle and right bars) and 198 significantly differential expression proteins (DEproteins) (14.28%) involved in metabolic pathways (P = 3.28E-14) (Fig. 2e). These results suggest that metabolism-related pathways may play important roles in the liver metastasis of CRC.

To explore the functions of proteins that are dysregulated in CLM, we used DAVID analysis software to classify the Gene Ontology of the 1386 significantly altered proteins in LM tissues according to their molecular functions and cellular components and ranked them according to their biological processes (Fig. 2f). The top-ranked biological function was metabolites and energy, organonitrogen compound, cellular respiration, ATP metabolic process, localization, which suggests that metabolism-related biological function is associated with CLM (Fig. 2b).

Next, we performed RNA sequencing to identify differentially expressed mRNAs (DEmRNAs) that are specifically dysregulated in CLM. Unsupervised hierarchical clustering of the expression data showed that the MT and LM tissues had closely related expression profiles when compared to para-tumor normal tissue or primary CRC tissue from CRC patients without liver metastasis (NM), suggesting clonal and genetic similarity for these pairs (Fig. 3a). We identified a total of 2136 genes significantly changed in LM (Fig. 3b) when compared to PN or NM groups, in which 462 genes (21.6%) were enriched in metabolism pathways (P = 1.58E-17) (Fig. 3c). Among them, 256 (55.41%) of which were down-regulated and 206 (44.59%) of which were up-regulated.

Moreover, analysis from TCGA sequencing dataset identified a total of 6585 significantly changed genes in CLM (Fig. 3d). A total of 5632 genes were significantly changed (3788 down-regulated and 1844 up-regulated) in the MT group compared to the PN group; and 1709 genes were significantly changed (1697down-regulated and 12 up-regulated) in the MT group compared to the CRC tumor samples without liver metastasis (NM) group. Among that, 1254 genes were in common with DEmRNAs identified in our study (58.7%).

When compared with the 1386 DEproteins identified by LC-MS/MS, 362 DEmRNAs showed significant positive mRNA-protein correlation (Fig. 3e, left and middle panels). To determine whether the concordance between protein and mRNA variation is related to the biological function of the gene product, we performed KEGG enrichment analysis, which indicated that among the 362 significantly deregulated genes/proteins, 48 are enriched in metabolic pathways (P = 4.59E-5) (Fig. 3e, right panel). These findings further verify the role of metabolic pathway genes in CLM.

We further performed global copy number variation (CNV) analysis to identify likely gene targets of focal alterations and to explore the impact of CNVs on mRNA and protein abundance and the potential correlation with LM. PN samples displayed scarcely any gains or losses, however, relative to the PN group, the LM and MT groups had 321 regions of significant focal amplification and 209 regions of significant focal deletion (Fig. 4a, b). In addition to several previously well-defined arm-level changes associated with carcinogenesis of CRC¹⁷, gains of 2q, 5p, 6p, 10q, 11p and 16p/q and deleted 18p/q were identified to contain the mRNA abundance variation (Fig. 4a, b). When compared with the correlation of the protein-CNV correlation (Spearman’s correlation coefficient 0.41; P < 0.05), the correlation between protein level and mRNA expression was much stronger (Spearman’s correlation coefficient 0.53; P < 0.01) (Fig. 4c). These results suggest that the mRNA transcript abundance is a relatively reliable predictor of protein abundance differences, but that copy number alterations showed little consistency with the protein level.

To further examine the potential role of CNV, we calculated the number of genes/proteins that also had CNV alteration. When compared with the 362 significantly changed genes/proteins, 112 were found with changed copy-number (Fig. 4d). Among those, chromosomes 1, 6, 7, 8, 12, 16, 19 and X contained the strongly hot spots driving global mRNA abundance variation (Fig. 4d), which highlights the importance of these regions in CLM.

We next evaluated the clinical significance of 286 CRC patients from TCGA database for the 112 CNV-mRNA-protein correlated molecules. Our results showed that 4 up-regulated genes (HSP90AB1, COL1A2, FABP5 and BGN), which located in CNV hotspots (located in 6p21.1, 7q21.3, 8q21.13 and Xq28, respectively) were associated with prognosis of CRC patients (Fig. 5a). Kaplan-Meier survival analysis confirmed that high expression of HSP90AB1, COL1A2, FABP5 or BGN was significantly associated with a shorter overall survival (P < 0.05) (Fig. 5b). Among those, high expression of COL1A2 and BGN was extremely significantly associated with a shorter overall survival (P < 0.01) (Fig. 5b). Moreover, high expression of COL1A2 or BGN was positively associated with disease-free survival (P < 0.05) as determined by Kaplan-Meier survival analysis.

To investigate the biological role of these CNV-mRNA-protein correlated genes, which associated with the prognosis in CRC progression and liver metastasis, we established the CRC cell line SW480 to stably overexpress COL1A2 or BGN to perform the gain-of-function studies in vitro (Fig. 5c). We then tested the effect of cell migration by COL1A2 or BGN overexpression via wound-healing assay and observed significant improvement of cell motility by COL1A2 or BGN (P < 0.01) (Fig. 5d). By two-chamber transwell assays, we also showed that forced expression of COL1A2 or BGN markedly enhanced the transwell invasiveness of SW480 (P < 0.01) (Fig. 5e).

To provide a comprehensive understanding of genetic abnormalities occurring in CLM, we used massively parallel paired-end sequencing technology to perform whole-exome solution-based hybrid capture sequencing of 2 triplet sample sets. The mean sequencing depth in the target regions was 80.28× (range 71.5 to 92.85). Analysis of the whole-exome sequencing data identified 27,778.5 mean point mutations (range 26,323 to 29,126). There were a variety of types of mutations identified, with T > A transversion being the most common nucleotide substitution (Fig. 6a). The distribution of CLM-related SNVs is shown in Fig. 6b. After filter analysis and exclusion of synonymous mutations, the numbers of indels and non-synonymous SNVs were calculated (Fig. 6c). In addition to some previously reported mutations, such as those in TP53, APC, KRAS, and PIK3CA [5], we identified 97 MT and LM-shared point mutations and 701 point mutations only existed in MT (Fig. 6d).

We further assessed the somatic gene mutations in an extended validation group of 44 paired normal colorectal tissues and CRC tissues by Sanger sequencing and nucleotide polymorphism genotype analysis. Subsequently, 175 nonsynonymous mutations within 171 genes were further verified.

In addition to the expected APC, TP53, SMAD4, PIK3CA and KRAS mutations, we found some new mutations that have not been reported to be involved in CLM including the TLL2^A302S mutation, which was identified in both the MT and LM groups (Fig. 6e), and the KLF11^D19N mutation, which was specific for the LM cohort (Fig. 6f). Moreover, our single nucleotide polymorphism genotype analysis revealed that 2 sites (FABP5^A2T and HSP90AB1^E299N) were frequent mutated only in the MT cohort, with mutation rates of 4.55% (2/44) (Fig. 6g) and 15.9% (7/44) (Fig. 6h), which suggest their potential roles in CLM.

A fundamental goal of proteogenomics is to identify protein coding alterations that are expressed at the protein level. However, standard database search approaches cannot identify variant peptides from MS/MS data. Therefore, we created a customized mutation database to search for SAAVs in CRC. A SAAV library was prepared using 113,844 mutated sites in CRC tissues from cBioport and our whole exome sequencing data, and 16,581 mutated proteins were identified, which constitute 82.08% of 20,201 proteins in the CRC standard protein library (Fig. 7a).

We determined the total numbers of mutated and non-mutated peptides and tumor-specific mutant peptides and found that mutated peptide numbers in MT samples were significantly increased (Fig. 7b), which indicates that the mutated peptide number has potential predictive value for CRC liver metastasis. Among those, 12 SAAVs in 8 proteins occurred only in NM patients (Fig. 7c; Table 2); and 13 proteins in 18 MT patients had 26 SAAVs; of which, 26 SAAVs in 6 proteins occurred in both NM and MT samples (Fig. 7d; Table 3), and 11 SAAVs in 5 proteins only occurred in MT samples (Fig. 7e; Table 4).

To further evaluate the potential role of SAAVs, we examined the expression levels of the proteins with SAAVs. The expression of 6 NM & MT-shared (Fig. 8a) and 8 MT-specific mutated proteins was upregulated in CRC. The sites of the most frequently mutated three proteins, MYH9^A769T, HSPA9^K555N and CCT6A^I423N, are shown (Fig. 8c). Furthermore, high MYH9 and CCT6A expression were each associated with shorter overall survival and disease-free survival (P < 0.05; Fig. 8d), which indicates that they have potential predictive values for CRC liver metastasis.