Biomarkers for pancreatic cancer: Recent achievements in proteomics and genomics through classical and multivariate statistical methods

Table 1 Statistical methods adopted in the identification of biomarkers for pancreatic cancer

Type of statistical method	Method adopted
Classical mono- and multi-variate methods	Student t-test (parametric)
	Mann-Whitney U-test (non-parametric)
	T2 Hotelling
	ANOVA and MANOVA
	Bayes factors
Unsupervised pattern recognition methods	Principal Component Analysis
	Cluster Analysis
	Multidimensional Scaling
Supervised classification methods	SIMCA
	Ranking-PCA
	O-PLS
	CART
	Random Forests
Methods for determining survival outcomes	Kaplan Meyer functions
	Cox Regression
Other methods	PAM
	Metropolis algorithm and Monte Carlo simulation

PCA: Principal component analysis; SIMCA: Soft independent model of class analogy; PLS: Partial least squares; CART: Classification and regression tree.

Table 2 Studies based on serum and tissue biomarkers through non-omics techniques

Ref.	Type of marker	Markers	Sample	Study group	Analytical methods	Statistical methods	Performance
41	P	CA 19-9	S	Pretreatment CA 19-9: 115 patients from 5 German centers; 73% treated within prospective clinical trials. Median TTP: 4.4 mo; median OS: 9.4 mo. CA 19-9 kinetics during chemotherapy: 69 patients (TTP) and 84 patients (OS)	Elecsys assay	Cox proportional hazards regression; for CA 19-9 kinetics, CA 19-9 was treated as a time-varying covariate	Univariate analysis: log (CA 19-9) associated with TTP (HR = 1.24; P < 0.001) and OS (HR = 1.16; P = 0.002). Multivariate analysis: results confirmed. Log(CA 19-9) kinetics during chemotherapy: significant predictor for TTP in univariate analyses (HR = 1.48; P < 0.001) and multivariate (HR = 1.45; P < 0.001) and for OS (univariate: HR = 1.34; P < 0.001; multivariate: HR =1.38; P < 0.001)
42	P	CA 19-9, CEA, CRP, LDH and bilirubin		291 patients; 253 patients (87 %) received treatment within prospective clinical trials. Median TTP: 5.1 mo. Median OS 9.0 mo	Elecsys assay	Kaplan Meier method and Cox proportional hazards regression	Univariate analysis: pre-treatment CA 19-9 (HR = 1.55), LDH (HR = 2.04) and CEA (HR = 1.89) significantly associated with TTP. Baseline CA 19-9 (HR = 1.46), LDH (HR = 2.07), CRP (HR = 1.69) and bilirubin (HR = 1.62) significant prognostic factors for OS. Multivariate analyses: pre-treatment log (CA 19-9) for TTP and log (bilirubin) and log (CRP) for OS had an independent prognostic value
44	P	IGFs	S and P	80 patients received treatment (40 Ganitumab; 40 placebo)	Immunoassays	Kaplan Meier method and Cox proportional hazards regression	Ganitumab associated with improved OS vs placebo (HR = 0.49; 95%CI: 0.28-0.87)
45	P	TROP2	T	197 patients; subgroup of 134 patients treated surgically	Immunohistochemistry	Kaplan Meier method and Cox proportional hazards regression	TROP2 overexpression observed in 109 (55%) patients and associated with decreased OS (P < 0.01). Univariate Analysis: TROP2 overexpression correlates with lymph node metastasis (P < 0.04) and tumor grade (P < 0.01). In the subgroup of patients treated surgically, TROP2 overexpression correlated with poor progression-free survival (P < 0.01). Multivariate analyses: TROP2 is an independent prognosticator
46	P	JAM-A	T	186 patients; subgroup of 83 patients treated surgically	Immunohistochemistry	Kaplan Meier method and Cox proportional hazards regression	Low expression of JAM-A observed in 79 (42 %) patients and associated with poor OS (P < 0.01). Univariate analysis: low expression of JAM-A correlates with positive lymph node status (P = 0.02), the presence of distant metastasis (P = 0.05), and tumor grade (P = 0.04). In the subgroup of patients with surgically resected PC, low expression of JAM-A correlated with decreased progression-free survival (P < 0.01). Multivariate analysis: JAM-A was an independent predictor of poor outcome
47	P	TBX4	T	77 stage II PDAC tumors	Immunohistochemistry	Kaplan Meier method and Cox proportional hazards regression	48 cases (62.3%) expressed TBX4 at a high level. No significant correlation between TBX4 expression and other clinicopathological parameters, except tumor grade and liver metastasis recurrence. Survival of patients with TBX4-high expression significantly longer than those with TBX4-low expression (P = 0.010). Multivariate analysis: low TBX4 expression independent prognostic factor for OS. TBX4 promoter methylation status frequently observed in PDAC and normal adjacent pancreas
48	P	HSP27	T	86 patients	Tissue microarray (TMA) analysis	Kaplan Meier method and Cox proportional hazards regression	HSP27 expression found in 49% of tumor samples. Univariate analyses: significant correlation between HSP27 expression and survival. Multivariate Cox-regression: HSP27 expression emerged as an independent prognostic factor. HSP27 expression also correlated inversely with nuclear p53 accumulation
49	P	dCK	T	45 patients with resected PDAC received adjuvant gemcitabine based-therapy in multicenter phase 2 studies	Immunohistochemistry	Kaplan Meier method and Cox proportional hazards regression	Median follow-up: 19.95 mo (95%CI: 3.3-107.4 mo). Lymph node (LN) ratio and dCK protein expression significant predictors of DFS and OS in univariate analysis. Multivariate analysis: dCK protein expression the only independent prognostic variable (DFS: HR = 3.48, 95%CI: 1.66-7.31, P < 0.001, OS: HR = 3.2, 95%CI: 1.44-7.13, P < 0.004)
50	P	Notch3 and Hey-1	T	42 patients who underwent resection and 50 patients diagnosed with unresectable PDAC	Immunohistochemistry	Mann-Whitney U test, Wilcoxon test, Cox regression analysis, Kaplan-Meier analysis	All 3 Notch family members significantly elevated in tumor tissue. Significantly higher nuclear expression of Notch1, -3 and -4, HES-1, and HEY-1 (all P < 0.001) in locally advanced and metastatic tumors compared to resectable cancers. In survival analyses, nuclear Notch3 and HEY-1 expression significantly associated with reduced OS and DFS following tumor resection with curative intent
51	D and P	21 biomarkers	P	clinically defined cohort of 52 locally advanced (Stage II/III) PDAC cases and 43 age-matched controls	Proximity ligation assay	Combination of the PAM algorithm and logistic regression modeling. Biomarkers that were significantly prognostic for survival were determined using univariate and multivariate Cox survival models	CA19-9, OPN and CHI3L1 were found to have superior sensitivity for pancreatic cancer vs CA19-9 alone (93% vs 80%). CEA and CA125 have prognostic significance for survival (P < 0.003)
52	D	83 circulating proteins	S	333 PDAC patients; 144 controls (benign pancreatic conditions); 227 healthy controls. Samples from each group split randomly into training and blinded validation sets. Panels evaluated in validation set and in patients diagnosed with colon (33), lung (62) and breast (108) cancers	bead-based xMAP immunoassays	A Metropolis algorithm with Monte Carlo simulation (MMC) was used to identify discriminatory biomarker panels in the training set	Training set (160 PDAC, 74 Benign, 107 Healthy): panel of CA19–9, ICAM-1, and OPG discriminated PDAC from Healthy controls (SN/SP 88/90%), panel of CA 19–9, CEA, and TIMP-1 discriminated PDAC patients from Benign subjects (SN/SP = 76%/90%). Independent validation set (173 PDAC, 70 Benign, 120 Healthy): panel of CA 19–9, ICAM-1 and OPG demonstrated SN/SP of 78%/94%; panel of CA19–9, CEA, and TIMP-1 demonstrated SN/SP of 71%/89%. The CA19–9, ICAM-1, OPG panel is selective for PDAC and does not recognize breast (SP = 100%), lung (SP = 97%), or colon (SP = 97%) cancer
53	D and P	YKL-40, IL-6, and CA 19.9	P	559 patients with PC from prospective biomarker studies from Denmark (n = 448) and Germany (n = 111)	ELISA and chemiluminescent immunometric assay	Kaplan Meier method and Cox proportional hazards regression	Odds ratios (ORs) for prediction of PC significant for all biomarkers, with CA 19.9 having the highest AUC (CA 19.9: OR = 2.28, 95%CI: 1.97-2.68, P = 0.0001, AUC = 0.94; YKL-40: OR = 4.50, 3.99-5.08, P = 0.0001, AUC = 0.87; IL-6: OR = 3.68, 3.08-4.44, P = 0.0001, AUC = 0.87). Multivariate Cox analysis: high preoperative IL-6 and CA 19.9 independently associated with short OS (CA 19.9: HR = 2.51, 1.22–5.15, P = 0.013; IL-6: HR = 2.03, 1.11-3.70, P = 0.021). Multivariate Cox analysis of non-operable patients: high pre-treatment levels of each biomarker independently associated with short OS (YKL-40: HR = 1.30, 1.03-1.64, P = 0.029; IL-6: HR = 1.71, 1.33-2.20, P = 0.0001; CA 19.9: HR = 1.54, 1.06-2.24, P = 0.022). Patients with preoperative elevation of IL-6 and CA 19.9 had shorter OS (P = 0.005) compared to patients with normal levels (45% vs 92% alive after 12 mo)

Type of marker: P: Prognostic/predictive; D: Diagnostic; Sample: S: Serum; P: Plasma; T: Tissue; TTP: Time to progression.

Table 3 Proteomic based studies

Ref.	Type of marker	Markers	Sample	Study group	Analytical methods	Statistical methods	Performance
54	D	Among 2393 unique proteins, 104 proteins significantly changed in cancer	T	5 patients; matched pairs of tumor and non-tumor pancreas	Tissues treated to obtain cytosol, membrane, nucleus and cytoskeletonfractions. Fractions separated and digested underwent LC-MS/MS	PLGEM	104 proteins significantly changed in cancer. Among these, 4 proteins validated that were up-regulated in cancer: biglycan (BGN), Pigment Epithelium-derived Factor (PEDF) Thrombospondin-2 (THBS-2) and TGF-β induced protein ig-h3 precursor (βIGH3)
57	D	Serum MALDI-TOF features	S	15 healthy (H), 24 cancer (Ca), 11 chronic pancreatitis (CP) samples	MALDI-TOF	Nonparametric	8 serum features: Ca samples differentiated from H (SN = 88%, SP = 93%), Ca from CP (SN = 88%, SP = 30%), and Ca from both H and CP combined (SN = 88%, SP = 66%). 9 features obtained from urine: differentiated Ca from both H and CP combined (SN = 90%, SP = 90%)
59	D	Serum SELDI-TOF features	S	96 serum samples from patients undergoing cancer surgery compared with sera from 96 controls	SELDI-TOF	pairwise statistics, MDS, hierarchical analysis Mann-Whitney U test, CART	Data analysis identified 24 differentially expressed protein peaks, 21 of which under-expressed in cancer samples. The best single marker predicts 92% of controls and 89% of cancer samples. Multivariate analysis: best model (3 markers) with SN = 100% and SP = 98% for the training data and SN = 83% and SP = 77% for test data. Apolipoprotein A-II, transthyretin and apolipoprotein A-I identified as markers and decreased at least 2 fold in cancer sera
60	D	Serum SELDI-TOF features	S	57 PC samples were compared to 59 controls	SELDI-TOF	Multivariate decorrelation filtering	Improved classification performances when the presented strategy is compared to standard univariate feature selection strategies
61	D	Proteins	S	Sera from patients diagnosed with PC compared with age- and sex-matched normal subjects	Protein microarrays	Rank-based non-parametric statistical testing	A serum diagnosis of PC was predicted with 86.7% accuracy, with a sensitivity and specificity of 93.3% and 80%. Candidate autoantibody biomarkers studied for their classification power using an independent sample set of 238 sera. Phosphoglycerate kinase-1 and histone H4 noted to elicit a significant differential humoral response in cancer sera compared with age- and sex-matched sera from normal patients and patients with chronic pancreatitis and diabetes
62	D	Proteins	PDAC cell lines	435 spots identified from 18 samples from 2 cell lines (Paca44 and T3M4) of control and drug-treated PDAC cells	2D-PAGE	PCA, SIMCA, Ranking-PCA	Samples were all perfectly classified. Significant proteins were further identified by MS analysis
63	D	Proteins regulating the conversion of quiescent to activated PaSC cells	rat PaSC cell line	-	SDS-PAGE and GeLC-MS/MS	QSPEC	Qualitative and quantitative proteomic analysis revealed several hundred proteins as differentially abundant between the two cell states. Proteins of greater abundance in activated PaSC: isoforms of actin and ribosomal proteins. Proteins more abundant in non-proliferating PaSC: signaling proteins MAP kinase 3 and Ras-related proteins

Type of marker: P: Prognostic/predictive; D: Diagnostic; Sample: S: Serum; P: Plasma; T: Tissue.

Table 4 Genomic based studies

Ref.	Type of marker	Markers	Sample	Study group	Analytical methods	Statistical methods	Performance
68	P	K-ras mutation status and subtypes	endoscopic ultrasound-guided fine-needle aspiration specimens	242 patients	RT-PCR	Kaplan Meier method and Cox proportional hazards regression	Multivariate analysis: CA19-9 C 1000 U/mL (HR = 1.78, 95 %CI: 1.28-2.46, P < 0.01), metastatic stage (HR 2.26, 95 % CI 1.58-3.24, P < 0.01) and mutant-K-ras (HR 1.76, 95 %CI: 1.03-3.01, P = 0.04) negative prognostic factors. Among patients with K-ras mutation subtypes: CA19-9 C 1000 U/mL (HR 1.65, 95%CI: 1.12-2.37, P < 0.01), metastatic stage (HR 2.12, 95%CI: 1.44-3.14, P < 0.01), and G12D or G12R mutations (HR = 1.60, 95%CI: 1.11-2.28) negative prognostic factors for OS
69	P	MicroRNA-21	T	82 resected Korean PDAC cases. Subgroup of patients treated with adjuvant therapy (n = 52)	Protein expression by immunohistochemistry microRNA expression by qRT-PCR	Cox proportional hazards model	Subgroup with adjuvant therapy: lower than median miR-21 expression associated with lower HR for death (HR = 0.316, 95%CI = 0.166-0.600, P = 0.0004) and recurrence (HR = 0.521, 95%CI = 0.280-0.967, P = 0.04). MiR-21: single most predictive biomarker for treatment outcome. No significant association in patients not treated with adjuvant therapy. Independent validation cohort of 45 frozen PDAC tissues from Italian cases treated with adjuvant therapy: lower than median miR-21 expression confirmed to be correlated with longer OS and DFS
71	P	13 putative PDA biomarkers from the public biomarker repository		A survival tissue microarray was constructed comprised of short-term (cancer-specific death ,12 mo, n = 58) and long-term survivors (30 mo, n = 79) who underwent resection for PDA (total, n = 137)	Immunohistochemical analyses; survival tissue microarray (s-TMA)	Wilcoxon rank sum test	Multivariate model: MUC1 (OR = 28.95, 3+ vs negative expression, P = 0.004) and MSLN (OR = 12.47, 3+ vs negative expression, P = 0.01) highly predictive of early cancer-specific death. Pathological factors (size, lymph node metastases, resection margin status, and grade): ORs < 3 and none reached statistical significance. ROC curves used to compare the 4 pathological prognostic features (ROC area = 0.70) to 3 univariate molecular predictors (MUC1, MSLN, MUC2) of survival group (ROC area = 0.80, P = 0.07)
1	P	MTA2 mRNA and protein expression	T	123 PDAC samples and 40 control tissues	qRT-PCR and immunohistochemistry	Kaplan-Meyer curves and Cox analysis	MTA2 mRNA and protein expression levels up-regulated in PC. MTA2 correlated with poor tumor differentiation, TNM stage and lymph node metastasis. Patients with high expression levels of MTA2 showed lower OS. MTA2: independent prognostic factor.
72	D	Leukocyte DNA Methylation	Blood	Phase I: 132 never-smoker PaC patients and 60 never-smoker healthy controls. Phase II: validation of 88 of 96 phase I selected CpG sites in 240 PaC cases and 240 matched controls	DNA array	Wilcoxon Rank Sum tests and likelihood penalized logistic regression models	Significant differences found in 110 CpG sites (FDR < 0.05). Phase II: 88 of 96 phase I selected CpG sites validated in 240 PaC cases and 240 matched controls (P < 0.05). Prediction model: 5 CpG sites (IL10_P348, LCN2_P86, ZAP70_P220, AIM2_P624, TAL1_P817) discriminated PaC from controls (C = 0.85 in phase I; 0.76 in phase II). One CpG site (LCN2_P86) could discriminate resectable patients from controls (C = 0.78 in phase I; 0.74 in phase II). 3 CpG sites identified (AGXT_P180_F, ALOX12_E85_R, JAK3_P1075_R) where the methylation levels were significantly associated with SNPs (FDR < 0.05)
73	D	cell-surface targets	T	28 PC specimens and 4 normal pancreas tissue samples. Expression in normal tissues evaluated by array data from 103 samples representing 28 organ sites as well as mining published data	Complementary assays of mRNA expression. Immunohistochemistry. qRT-PCR	-	170 unique targets highly expressed in 2 or more PC specimens and not expressed in normal pancreas samples. Two targets (TLR2 and ABCC3) further validated for protein expression by tissue microarray based immunohistochemistry have potential for the development of diagnostic imaging and therapeutic agents for PC
74	D	Differentially expressed genes	Blood	25 patients diagnosed with PC and diabetes, 27 patients with PC without diabetes, 25 patients with diabetes mellitus > 5 yr, and 25 healthy controls. Results further validated for 101 blood samples	Microarray and qRT-PCR	-	58 genes found to be unique in patients with cancer-associated diabetes, including 23 up-regulated and 35 down-regulated genes. 11 up-regulated genes further validated by RT-PCR; 2 of these (VNN1 and MMP9) selected for logistic regression analysis. The combination of VNN1 and MMP9 showed the best discrimination of cancer-associated diabetes from type 2 diabetes. The protein expression of MMP9 and VNN1 was in accordance with the gene expression

Type of marker: P: Prognostic/predictive; D: Diagnostic; Sample: S: Serum; P: Plasma; T: Tissue; PDAC: Pancreatic ductal adenocarcinoma; VNN1: Vanin-1; MMP9: Matrix metalloproteinase 9.