Identification of Novel Alternative Splice Isoforms of Circulating Proteins in a Mouse Model of Human Pancreatic Cancer

To assess the potential of tumor-associated, alternatively spliced gene products as a source of biomarkers in biological fluids, we have analyzed a large data set of mass spectra derived from the plasma proteome of a mouse model of human pancreatic ductal adenocarcinoma. MS/MS spectra were interrogated for novel splice isoforms using a nonredundant database containing an exhaustive three-frame translation of Ensembl transcripts and gene models from ECgene. This integrated analysis identified 420 distinct splice isoforms, of which 92 did not match any previously annotated mouse protein sequence. We chose seven of those novel variants for validation by reverse transcription–PCR. The results were concordant with the proteomic analysis. All seven novel peptides were successfully amplified in pancreas specimens from both wild-type and mutant mice. Isotopic labeling of cysteine-containing peptides from tumor-bearing mice and wild-type controls enabled relative quantification of the proteins. Differential expression between tumor-bearing and control mice was notable for peptides from novel variants of muscle pyruvate kinase, malate dehydrogenase 1, glyceraldehyde-3-phosphate dehydrogenase, proteoglycan 4, minichromosome maintenance, complex component 9, high mobility group box 2, and hepatocyte growth factor activator. Our results show that, in a mouse model for human pancreatic cancer, novel and differentially expressed alternative splice isoforms are detectable in plasma and may be a source of candidate biomarkers. [Cancer Res 2009;69(1):300–9]

Alternative splicing plays an important role in protein diversity without significantly increasing genome size. Aberrations in alternative splice variants contribute to a number of diseases, including cancers (1, 2). For example, Thorsen and colleagues (2) identified cancer-specific splicing events in colon, bladder, and prostate tissues, with diagnostic and prognostic implications. The several alternatively spliced sequence databases now publicly available differ in their annotation and modeling methods and contain many transcripts not present in reference resources like Ensembl or Refseq (3). The ECgene database is one of the largest, alternatively spliced sequence databases (4). Entries in this database are scored as high, medium, or low confidence, reflecting the amount of cumulative evidence in support of the existence of a particular alternatively spliced sequence. Evidence is collected from clustering of ESTs, mRNA sequences, and gene model predictions. We modified the ECgene database to include three-frame translations of the cDNA sequences (5) to determine the occurrence of novel splice variant proteins. An important development in recent years is the substantial improvement in tandem mass spectrometry instrumentation for proteomics, allowing in-depth analysis and confident identifications even for proteins coded by mRNA transcript sequences expressed at low levels (6–8).

Pancreatic ductal adenocarcinoma (PDAC) is among the most lethal of human cancers due to absence of methods for early diagnosis and chemoresistance of advanced disease. The Kras^G12D/Ink4a/Arf mouse model of PDAC was engineered with signature mutations that recapitulate the histopathologic progression of the human disease in a highly reproducible and synchronous fashion (9, 10). Here, we exploit this model to test the hypothesis that cancer-specific alternative splice variants can be identified by in-depth mass spectrometric analysis of plasma proteins from this mouse model of pancreatic cancer. In this study, we have interrogated our modified ECgene database to identify both novel and known splice variants among circulating proteins. In addition, our analysis of quantitative expression ratios reveals variant proteins that are differentially expressed in pancreatic cancer.

Our search for alternative spliced forms used data from extensive proteomic analysis of plasma from 7-wk-old male wild-type mice and KRas^G12D/Ink4a-Arf mouse model of PDAC (10). Approximately one third of the Kras Ink4a/Arf mice present with the most common pathology observed in human cases—glandular. Plasma from Kras Ink4a/Arf mice with exclusively glandular tumor areas were used in this study. The samples were processed by the Intact Protein Fractionation and Analysis System (11) protocol, which incorporates immunodepletion to eliminate the most abundant plasma proteins, thus removing 90% of the protein mass. Immunodepletion was followed by isotopic labeling of protein cysteine residues with acrylamide, heavy (D3) for mutant and light (D0) for wild-type samples. The mass difference between a D3-labeled and a D0-labeled cysteine residue is 3.01884 Da. Samples from wild-type and PDAC-bearing mice were then pooled, and the intact proteins were fractionated into 12 anion exchange (AX) fractions followed by 13 or more reverse phase (RP) fractions for each AX fraction, yielding a total of 163 fractions. Individual fractions were digested with trypsin and analyzed using a ThermoFinnigan LTQ-FT mass spectrometer. Mass spectra from the LTQ-FT experiment were acquired as RAW files. The mzXML files containing the spectral information were extracted from RAW files using ReAdW.exe program.⁶

The mzXML files were then searched against a modified ECgene database (for alternate splice variant analysis) using X!Tandem (12) software.

Alternative splicing can generate multiple transcripts from the same gene, which are translated to splice isoforms (also known as splice variant proteins). Our target alternative splice variant protein database, the modified ECgene database, was constructed by combining Ensembl 40 and ECgene database (mm8, build 1). Taking alternative splicing events into specific consideration, ECgene combines genome-based EST clustering and the transcript assembly procedure to construct gene models that encompass all alternative splicing events. The reliability of each isoform is assessed from the nature of cluster members and from the minimum number of clones required to reconstruct all exons in the transcript (13). The ECgene database contains a total of 417,643 splice variants. In the Ensembl 40 database, there is a total of 21,839 mouse genes with 28,110 transcripts, of which 10,922 alternative transcripts are derived from 4,651 genes.

Transcript sequences from the ECgene and Ensembl 40 databases were translated in three reading frames. Within each data set, the first instance of each protein sequence longer than 14 amino acids was recorded. The resulting proteins from both database translations were then combined and filtered for redundancy. For this filtering, proteins derived from Ensembl transcripts were preferentially recorded over those generated from ECgene records. A collection of common protein contaminant sequences was added to this set. Lastly, all sequences were reversed and appended to the set of forward sequences as an internal control for false identifications. This last step resulted in doubling the total number of entries in the modified ECgene database with a final total of 10,381,156 protein sequences.

Trans proteomic pipeline. For statistical purposes, the X!Tandem search results were postprocessed with PeptideProphet and ProteinProphet software using Trans proteomic pipeline (TPP; version 3.2).⁶ First, we analyzed search results from each fraction separately using PeptideProphet and then processed all PeptideProphet files together using ProteinProphet. Relative quantification of isotopically labeled peptides was performed using the Q3Ratio (14) and XPRESS (15) applications in TPP. The relative abundance of a peptide can be calculated by reconstructing the light and heavy elution profiles of the precursor ions and by determining the elution areas of each peak. From Q3Ratio and XPRESS analyses, we obtained the average expression ratio (mutant/wild type) for those peptides that were unique to the splice variant protein identified and had labeled cysteine residue.

Michigan peptide-to-protein integration. In addition to the TPP analysis, we implemented a second analysis based on selecting all peptide identifications with X!Tandem expect score of <0.001. We applied an integration process to these peptide identifications, which greatly reduced the set of proteins and retained all of the peptide identifications. The integration process was as follows:

1. A list of all peptide matches was created.

2. The peptide list was ordered by the number of spectra matching each peptide.

3. Peptide with largest number of matching spectra was selected.

4. A list of all proteins containing this peptide was made.

5. The protein list was sorted by the number of spectra matching peptides derived from the protein.

6. The protein with the largest number of matching spectra and which was identified by the largest number of distinct peptides was selected. When equal numbers of identifying spectra were present, Ensembl proteins were given preference because these transcripts have been independently reviewed and Ensembl identifiers can readily be linked to additional annotation.

7. All peptides contained within this protein were removed from the peptide list.

8. Steps 3 to 7 were repeated until no peptides remained in the peptide list.

Protein identifications based on a single peptide from this Michigan peptide-to-protein integration (MPPI) analysis were retained only if identified by at least 10 distinct spectra. For further analyses, we used the union of protein identifications from TPP analysis with ProteinProphet probability of >0.9 and the MPPI. To remove redundant protein identifications, we ran another round of MPPI using all peptides found in either TPP or first-phase MPPI (Fig. 1).

The peptide sequences, for which the alternative splice variant proteins were identified, were searched against the mouse genome (Build 37) using UCSC BLAT (16) browser and against NR using NCBI blastp (17). In some cases, when the peptide sequences are short, the blat or blastp programs were not able to align correctly. To handle these cases and verify search results, we further searched the peptide sequences against nr protein sequences by SQL pattern match. Proteins whose peptide sequences aligned perfectly to an existing canonical mouse coding sequence were removed from the list of alternatively spliced protein identifications. The identifications that remain after this round of blat and blastp analysis must be novel as all their peptides did not precisely match to any known mouse proteins. We subsequently checked whether any peptides of these novel identifications matched either partially or completely to any known protein sequences using blastp. If a match is observed, those identifications were considered as novel splice variants of known genes.

To corroborate the identifications of novel splice variants, peptide sequences with splice variations from seven novel variants were selected for validation by reverse transcription–PCR (RT-PCR; Fig. 4). We chose these seven peptide sequences as they were identified by multiple high-scoring spectra (X!Tandem expect value of <0.001; Supplementary Table S8) and are much less annotated than splice variants of apolipoproteins, complement component proteins, hemopexin, and vitamin D–binding protein, which were also identified. The primers were designed using the Primer3 software⁷

from the cDNA sequences of these seven proteins in the ECGene database (Supplementary Table S8). Total RNA was extracted from frozen pancreatic tissue of wild-type and mutant mice using TRIzol reagent (Invitrogen) in accordance with the manufacturer's instructions. RNA purification was performed using the RNeasy Mini kit (Qiagen, Inc.) and Rnase-free Dnase digestion (Qiagen, Inc.). Qiagen One-Step RNA PCR kit (AMV) was used for RT-PCR amplification of the seven novel peptides from total RNA. Primers for amplifying γ-actin (forward, 5′-GCGCAAGTACTCTGTGTGGA-3′; reverse, 5′-ACATCTGCTGGAAGGTGGAC-3′) were used to amplify the house-keeping gene as a positive control and standard for quantification.

After incubation at 50°C for 30 min, one cycle of 94°C for 15 min and 40 cycles of 94°C 30 s, 50°C 30 s, 72°C 1 min followed by 72°C 10 min for final extension were performed using the thermal cycler MJ PTC 200 (Bio-Rad). Amplicons were run on 2% agarose gel and visualized under UV light. Images were captured using a gel imaging system (Biostep GmbH).

Four of these seven novel peptides chosen above for RT-PCR showed differential expression by TPP expression ratio analyses. These peptides include pyruvate kinase variant, malate dehydrogenase 1 variant, PCTAIRE-motif protein kinase 2 variant, and solute carrier family 19 (sodium/hydrogen) variant. The mRNA expression levels of these four novel peptides were measured by quantitative real-time PCR using a QuantiTect SYBER Green PCR (Qiagen) and ABI Prism 7000 Sequence Detector (Applied Biosystems). The expression of γ-actin was determined as an internal control. The PCR amplification reaction and the detection of PCR product was monitored by measuring the increase in fluorescence intensity caused by the binding of SYBER Green dye to double-stranded DNA. The same RT-PCR primers (Supplementary Table S8) were used in this method. PCR amplification was performed using the following profile: one cycle at 50°C for 30 min, one cycle at 95°C for 10 min, and 40 cycles at 94°C for 15 s, 60°C for 1 min. Calculations for relative quantification were performed as outlined in User Bulletin 2: ABI Prism 7700 Sequence Detection System (PE Applied Biosystems; ref. 18).

Because we had access to only a single sample of pancreatic tissue from the wild-type and mutant mice, respectively, we are unable to do biological sample replicates. However, we performed the quantitative RT-PCR thrice using the same two samples to assess technical errors.

The gene ontology (GO) annotation file for Mus musculus was downloaded from their Web site,⁸

and the GO terms associated with the parent gene symbols of the splice variants were studied. The functional annotations of the splice isoforms were done by searching the parent gene names using FuncAssociate (19). FuncAssociate is a Web-based tool that accepts as input a list of genes and returns a list of GO attributes that are overrepresented (or underrepresented) among the genes in the input list. Only those overrepresentations (or underrepresentations) that are statistically significant, after correcting for multiple hypotheses testing, are reported.

Summary of protein identifications. The mass spectra derived from plasma protein analysis were searched against the modified ECgene database using X!Tandem. The empirical false discovery rate (FDR) calculated based on peptide identifications from reverse protein sequences was <0.5% for all peptide identification scoring bins with X!tandem expect score of <0.001 (Supplementary Fig. S1). Protein and peptide identifications from the X!Tandem search and postsearch TPP/MPPI analysis are summarized in Table 1. Figure 1 summarizes the analysis workflow. TPP analysis of the search results yielded 1,150 proteins with ProteinProphet probability of ≥0.9 (FDR for proteins, <1.5%). In parallel, using our MPPI method (expect <0.001), we identified a total of 1063 proteins (FDR for proteins, <1.5%). There were 870 biological protein identifications, and three reverse sequences were found in common by both approaches. Thus, the empirical FDR for proteins identified by both methods is <0.3%. The union of the TPP and MPPI methods resulted in 1,340 protein identifications. There were 467 biological proteins and 22 reverse sequences identified by either TPP or MPPI, but not by both. Thus, the empirical FDR for proteins identified by either, but not both methods, is <5%. The reverse sequence proteins identified by TPP and MPPI analyses were different; because we took the union of protein identifications from both methods, the false discovery rate at the protein identification level increased.

Taking the union of the two methods introduces some redundancy in identifications, so we applied a second round of MPPI on these 1,340 proteins, yielding 1,278 distinct protein identifications with 25 proteins from reverse sequence. Of the 1,278 protein identifications, 420 were splice variant proteins (Supplementary Table S1).

Novel alternative splice variant protein identifications. To identify novel spliced sequences, we searched the peptide sequences of the 420 alternative splice proteins against the mouse genome using blat and against NR database using blastp (see Materials and Methods); 92 of 420 alternative splice proteins were identified by unique peptides that did not align completely with any known mouse coding sequence (Fig. 1; Supplementary Table S2). These proteins are considered novel identifications. Of these 92 identifications, 84 were novel alternative splice variants of known genes. An example is the identification of novel variant for peptidase inhibitor 16 (pi16) by the unique peptide ATEAPSSRETGAENPEK. Figure 2 shows the alignment of this peptide to the known pi16 sequence by blastp. The alignment suggests a partial deletion (229 amino acids) of the exon sequence due to a splicing event resulting in the novel variant.

For eight identifications from the 92 novel proteins, neither the complete protein sequence nor the peptides identified matched even partially to any known mouse proteins by blastp (parent gene name and gene symbol is given as null in Supplementary Table S2). However, we found that the complete protein sequences of these eight proteins matched perfectly to various mouse genomic regions (Supplementary Table S3).

Both known and novel variant forms of many genes were found in the 420 protein set. For example, we found seven variants of α-2-HS-glycoprotein, including three known and four novel variant forms. A particularly interesting example is the identification of known and novel isoforms of muscle pyruvate kinase; the novel alternative splice variant “M9C5102_13_s4154_e5807_1_rf0_c1_n0|” was identified by the unique peptide “CLAAALIVLTESGR” from five different spectra. This peptide sequence did not align to the coding sequence of annotated mouse muscle pyruvate kinase protein. However, it aligned to human, rat, and chicken muscle pyruvate kinase M1 isoform (Fig. 3A). The previously known mouse gene product for muscle pyruvate kinase (pkm2) gene, which is homologous to the M2 isoform of human and rat pkm2, was identified by 25 distinct peptides, 22 of which were shared by the novel mouse variant protein. Figure 3B shows identifications of unique peptides of a known isoform mouse malate dehydrogenase protein and its novel splice form from different clusters of RP fractions; the isoform exhibits faster mobility by reversed phase chromatography.

Tumor-related novel alternative splice variant proteins. By combining the Q3Ratio and XPRESS analyses, we calculated expression ratios for the cysteine containing unique peptides with splice variation from tumor-bearing mice relative to controls for 28 of 92 novel splice variant proteins (Table 2). The unique peptides from 16 variants were identified by multiple spectra, which enabled us to calculate the 95% confidence intervals for their expression ratios. Splice variants of malate dehydrogenase 1, muscle pyruvate kinase, proteoglycan 4, complement component factor H, major urinary protein, vitamin D–binding protein, and α-2-HS-glycoprotein had the 95% confidence intervals for their expression ratios of >1 (Table 2). However, expression ratios of unique peptides of 12 variants were calculated from single spectrum. Six of these proteins showed >2-fold expression ratios; splice variants of high mobility group box 2 protein and minichromosome maintenance complex component 9 are in this group. The other six proteins include novel variants of α-fetoprotein, peroxiredoxin 2, and solute carrier family 19 member 1.

α-Fetoprotein, apolipoprotein E, ceruloplasmin, fibronectin, glyceraldehyde 3 phosphate dehydrogenase, hemopexin, peptidylprolyl isomerase, and tubulin α are among proteins previously identified as involved in pancreatic cancer (20–22). It is noteworthy that in our 92 novel proteins, we identified the presence of additional novel alternative splice variants of these proteins.

RT-PCR validations. The identifications of novel alternative splice variants were further validated by checking the mRNA expressions of seven peptide sequences with splice variations (Supplementary Table S8) by RT-PCR. The results were concordant with our proteomic analysis. All seven novel peptide coding sequences were amplified in both wild-type and mutant samples (Fig. 4A). Amplified sequences ranged in size from 63 to 75 bp (Fig. 4A). In addition, we did the RT-PCR on four peptides that showed differential expression by TPP analysis (Fig. 4B). The expression profile for novel variants of pkm2 and pctk2 were similar to what we observed by TPP analysis. However, the expression profiles of novel variant of mdh1 and sol19a1 were opposite to what we observed from TPP analysis. We saw decreased mRNA expression of the novel mdh1 peptide in mutant sample versus the wild type. Although the unique peptide for sol19a1 variant was identified by multiple spectra, the expression ratio was calculated by Q3Parser from a single spectrum.

Functional annotations of novel alternative splice variant proteins. GO terms associated with the parent gene symbols of the novel splice variants of known proteins are given in Supplementary Table S4. As expected, many proteins were extracellular. However, we observed splice variants of such nuclear proteins as high-mobility group box 2, DEAD box polypeptide 6, nuclear factor interleukin 3 regulated, and peptidylprolyl isomerase like 2. In the molecular function category, more splice variants were associated with metal ion binding and endopeptidase inhibitor activity. Under biological processes, many variants were involved in transport. Supplementary Table S5 summarizes the list of genes with overrepresented GO attributes using FuncAssociate.

Known alternative splice variant proteins. As mentioned earlier, 420 of the 1,278 protein identifications were alternative splice variant proteins. By sequence analysis of the peptides identified from these 420 proteins, 92 were found as novel splice proteins. The peptides from the remaining 328 protein identifications matched to sequences from known mouse coding sequences. Variant proteins of many genes previously reported as involved in pancreatic cancer are in this list (Supplementary Table S6a). In addition, variants of genes that were reported by other studies to be involved in other types of cancers are found in this set (Supplementary Table S6b).

Relative expression analysis by Xpress and Q3Proteinparser produced expression ratios for 81 proteins of 328 known splice variants (Supplementary Table S7). Forty eight of them had >2-fold expression at 95% or more confidence interval in mutant samples relative to wild type. According to the analysis, the unique peptide that identified secreted and transmembrane 1A (sectm1a) showed the highest fold change in mutant samples. This protein was reported to have potential importance in hematopoietic and/or immune system processes (23).

Ten possible pancreatic cancer biomarkers were chosen by a parallel study on the same mouse model plasma samples (24). Variants of three of these genes are found in our list of 420 splice variant proteins: lipocalin 2 (lcn2), regenerating islet-derived 3 (reg3a) and tumor necrosis factor receptor superfamily member 1A (tnfrsf1a). According to our expression analysis, tnfrsf1a showed >2-fold change in expression in tumor samples compared with wild type.

Alternative splicing allows a single gene to generate multiple mRNAs, which can be translated into functionally and structurally diverse proteins (25). One gene can have multiple variants coexisting at different concentrations, so estimating the relative abundance of each variant may be important for the study of its biological role (25). As reported here, 1,278 distinct proteins (Fig. 1) were identified from the union of TPP and MPPI analyses; 92 proteins were novel proteins. Successful RT-PCR amplifications of all seven selected novel peptide coding sequences from both wild-type and mutant samples validated our protein identifications (Fig. 4).

In 28 of the 92 novel variants identified, expression ratios were quantified for mutant and wild-type plasma samples (Table 2). The main reasons why expression ratios could not be quantified in other cases are the small proportion of cysteine-containing peptides (site of labeling) in the 92 splice variants identified; 774 of 2,513 peptides from 92 splice variants contain cysteine residues. Furthermore, the XPRESS and Q3Ratio quantification software reported only heavy or light expression for many peptides; to quantify the relative expression ratio, both values are needed. The quantifications of mRNA expression by RT-PCR were concordant with protein expression for two of the four peptides verified. The inverse correlation observed mRNA-protein expression in the other two peptides might be due to numerous factors, including negative feedback by the protein variant on mRNA synthesis, posttranscriptional control of protein translation, and protein modifications. There are several studies showing varying correlation between mRNA and protein abundance ratios (26, 27). Guo and colleagues (28), in their correlation analyses on mRNA-protein expression in freshly isolated human circulating monocytes, found some genes with moderate and varied correlations, suggesting that mRNA expression might be sometimes useful but not reliable in predicting protein expression levels.

Many of the known protein variants that we identified have been reported as involved in pancreatic or other types of cancer, some with significant differential expression, such as acyl coA acetyltransferase, chromogranin b, granulin, insulin-like growth factor binding protein 2, and regenerating islet-derived 3α (Supplementary Table S6 for citations and Supplementary Table S7 for expression ratios).

The focus of our work is on the novel variants, of which we annotate eight. First, we highlight three novel splice proteins with >2-fold increase in mutant plasma compared with wild type, which may have potential role in progression of pancreatic cancer: pyruvate kinase (pkm2), proteoglycan 4 (prg4), and malate dehydrogenase 1 (mdh1). The 95% confidence intervals for their >2-fold expression ratios exclude unity.

Pyruvate kinase, a glycolytic enzyme, catalyzes transfer of a phosphoryl group from phosphoenolpyruvate to ADP, generating ATP. Most adult tissues express the M1 isoform of pyruvate kinase, whereas embryonic tissues and tumors express the M2 isoform. Christofk and colleagues (29, 30) showed that the M2 isoform of muscle pyruvate kinase accounts for the long known distinctive Warburg effect of shift to aerobic glycolysis in tumor cells. The novel mouse peptide sequence CLAAALIVLTESGR, with which we identified the novel splice variant of muscle pyruvate kinase, was from five different spectra, authenticating the occurrence of this variant. In addition, our study shows that the mRNA and peptide expressions of this novel variant were positively correlated for this gene and show increased expression in mutant samples compared with wild type. The peptide sequence is identical to the human, rat, and chicken pyruvate kinase M1 variant (Fig. 3A). A second variant of pyruvate kinase, M2 type pyruvate kinase, was also identified in our 420–alternative splice isoform list. All the 25 peptides identified in this protein aligned completely with known mouse coding sequence. Studies by Kumar and colleagues (31) and Ventrucci and colleagues (32) report tumor type pkm2 as a metabolic marker specifically for pancreatic cancer.

Prg4 has been identified as megakaryocyte-stimulating factor and articular superficial zone protein. Prg4 has characteristic motifs including somatomedin B and hemopexin domains, a chondroitin sulfate-attachment site and mucin-like repeats (33). The specific role of prg4 in the articular joint has previously been reported; however, the presence of multifunctional motifs and the identification of a novel variant suggest that prg4 functions in several distinctive biological processes. Faca and colleagues (24) observed an increased gene expression of isoform E of prg4 in human pancreatic cancer samples.

Mdh1 is an enzyme in the citric acid cycle that catalyzes the conversion of malate into oxaloacetate and vice versa. It also plays a role in gluconeogenesis. Impaired oxidative metabolism may contribute to malignant growth (34). Mdh1 has been reported as overexpressed in thyroid oncocytic tumors (35). Overexpression of the novel variant of mdh1 in our study suggests that cellular energy pathways may play an important functional role in tumor progression.

The novel peptide CELAEQMTSLLER was identified by eight different spectra. Both mRNA and protein expression of this peptide were down-regulated in mutant samples when compared with that of wild type. Blasting this peptide sequence against NCBI-nr database showed that part of the peptide ELAEQMT aligned perfectly with mouse PCTAIRE-motif protein kinase 2 (pctk2). Valladares and colleagues (36) showed overexpression of pctk2 in breast cancer samples from Mexican women. The PCTAIRE protein kinases comprise a distinct subfamily of the cell division controller 2 (cdc2)–related serine/threonine-specific protein kinases. Deregulation of cell-cycle control is thought to be a crucial event in malignant transformation. Guo and colleagues (37) show an important role of cdc25b in cell-cycle progression, raising the possibility that inhibition of cdc25b may have therapeutic potential in pancreatic cancer.

Another novel variant with increased expression in our peptide expression analysis is minichromosome maintenance complex component 9 variant. MCM proteins are well-known regulators of the cell cycle. Mcm9 binds to chromatin in an ORC-dependent manner and is required for the recruitment of the Mcm2-7 helicase onto chromatin (38). Furthermore, MCM family members 2 and 5 are markers of aggressive cervical and other cancers (39). Mcm9 and mdh1 have not been reported as involved in pancreatic cancer. It would be interesting to investigate potential roles of these novel variants in pancreatic cancers.

A splice variant of high mobility group box 2 (hmgb2) showed >2-fold increase in expression in mutant samples. Hmgb2 is involved in DNA repair. It has not been reported as up-regulated in pancreatic cancer, but this gene is located at 4q32-34, a region identified as a potential locus for familial pancreatic cancer in a large American family (40).

Another interesting protein is the novel splice variant of hepatocyte growth factor activator (hgfa), a serine protease that converts the inactive single-chain precursor of hepatocyte growth factor (hgf) to the active heterodimer in response to tissue injury (41). Results from Lee and colleagues (42) suggest that extracellular receptor kinase activation by hgf might be important in pancreatic cancer metastasis.

We found many novel splice variants to glyceraldehyde-3-phosphate dehydrogenase (gapdh), another critical enzyme in glycolysis. Gapdh mRNA levels are overexpressed in many cancers; relative levels correlate with pathologic stage in prostate cancer (43–45). Chen and colleagues (21) observed >2-fold increase of gapdh in pancreatic cancer tissue using isotope-coded affinity tag (ICAT) technology followed by mass spectrometry.

The ECgene⁹

database uses the ECgene gene prediction algorithm to identify splice variants by EST clustering. Three levels of confidence (high, medium, and low) were set on the gene models based on mRNA evidence (13). The custom-built ECgene database used for our study has another layer of complexity, which is the three-frame translation of ECgene sequences of predicted gene models, which makes it feasible to identify novel proteins. We found eight proteins, where no peptides matched to any known mouse protein sequences by blastp analysis (Supplementary Table S3). Three proteins (M14C7677_1_s356_e431_1_rf1_c1_n0|, M5C5274_1_s2_e95_1_rf0_c1_n0|, and M1C10834_9_s770_e824_1_rf2_c1_n0|) had differential expression in mutant samples relative to wild type. Further investigation of these novel proteins in pancreatic cancer seems warranted.

One focus in cancer biomarker discovery is driven by the approach that the most promising plasma biomarkers will be secreted proteins (46). GO analysis of the novel splice variants by FuncAssociate showed extracellular as a top-ranking attribute (Supplementary Table S5). Interestingly, the unique peptide with which we identified the novel protein M5C6460_1_s86_e1988_1_rf0_c1_n0| had a 71% sequence similarity to the peptide sequence of the extracellular protein, α-fetoprotein, a gold standard in the field of tumor markers (47). According to FuncAssociate analysis, endopeptidase inhibitor/endoproteinase inhibitor/proteinase inhibitor activity (GO 0004866) is a top-ranking GO attribute. Imbalanced protease activity has long been recognized in progression of cancers and inflammation (48). Other high-ranking biological processes are glucose metabolism and lipid transport. There are many studies linking impaired glucose metabolism and pancreatic cancer (49, 50). Supplementary Table S4 shows aldolase c, gapdh, mdh1, and pkm2 as genes involved in glucose metabolism. The differential expression for novel alternative splice variants of gapdh, and mdh1 suggests involvement of these splice variants in tumors.

The combined proteomic and bioinformatic approach in this study has identified many novel splice variant proteins, including many with differential expression in plasma from mutant mice with Kras^G12D/Ink4a/Arf-mediated pancreatic cancer versus wild-type mice. These differentially expressed splice variant proteins may influence many cancer-related mechanisms. The data suggest that alternative splice variant proteins are a potentially rich source of candidate biomarkers for cancers.

No potential conflicts of interest were disclosed.

Grant support: National Cancer Institute (NCI)/SAIC contract 23XS110A on Mouse Models of Human Cancers, MTTC GR 687 for Proteomics Alliance for Cancer Research, R01 LM008106 Automated Genome Annotation, MMHCC U01 CA84313, NCI PDAC P01 IPOICA117969-01, U54 DA021519 National Center for Integrative Biomedical Informatics, P41 RR018627 National Resource for Pathways and Proteomics, and NCI Early Detection Research Network.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We thank Dr. Thomas Blackwell for giving helpful suggestions in statistical analysis.