Type XXII Collagen Complements Fibrillar Collagens in the Serological Assessment of Tumor Fibrosis and the Outcome in Pancreatic Cancer

Abstract

Circulating fragments of type III collagen, measured by PRO-C3, has shown promising results as a tumor fibrosis biomarker. However, the fibrotic tumor microenvironment consists of many other collagens with diverse functions and unexplored biomarker potential. One example hereof is type XXII collagen (COL22). In this study, we investigated the biomarker potential of COL22 by measuring this in serum. An ELISA, named PRO-C22, was developed and measured in two serum cohorts consisting of patients with various solid tumors (n = 220) and healthy subjects (n = 33) (Cohort 1), and patients with pancreatic ductal adenocarcinoma (PDAC) (n = 34), and healthy subjects (n = 20) (Cohort 2). In Cohort 1, PRO-C22 was elevated in the serum from patients with solid tumors, compared to healthy subjects (p < 0.01 to p < 0.0001), and the diagnostic accuracy (AUROC) ranged from 0.87 to 0.98, p < 0.0001. In Cohort 2, the high levels of PRO-C22, in patients with PDAC, were predictive of a worse overall survival (HR = 4.52, 95% CI 1.90−10.7, p = 0.0006) and this remained significant after adjusting for PRO-C3 (HR = 4.27, 95% CI 1.24−10.4, p = 0.0013). In conclusion, PRO-C22 has diagnostic biomarker potential in various solid tumor types and prognostic biomarker potential in PDAC. Furthermore, PRO-C22 complemented PRO-C3 in predicting mortality, suggesting an additive prognostic value when quantifying different collagens.

Keywords: ECM; PDAC; collagens; non-invasive biomarker; tumor fibrosis; type XXII collagen.

Figure 1

The PRO-C22 ELISA was specific for the target epitope. Dose-dependent signal inhibition of the interaction observed between the biotinylated coating peptide (Biotin-¹⁶¹⁶AARPGNVKGP¹⁶²⁶) and the horseradish peroxidase labeled monoclonal antibody (HRP-mAb) by the assay-specific ten and 30 amino acid (AA) standard peptides (¹⁶¹⁶AARPGNVKGP¹⁶²⁶ and ¹⁵⁹⁶GPPGPPGQCDPSQCAYFASLAARPGNVKGP¹⁶²⁶, respectively). Lack of inhibition was observed by the truncated peptide (¹⁶¹⁶AARPGNVKG¹⁶²⁵), elongated peptide (¹⁶¹⁶AARPGNVKGP¹⁶²⁶+A), or the non-sense 30 AA peptide. The non-sense coating peptide resulted in no signal generation. Three de-selection peptides corresponding to sequences from the extracellular matrix proteins and with sequence similarities to the PRO-C22 epitope showed no effect. B/B0 is the ratio between B (the signal intensity when a sample or peptide is present) and B0 (the maximum signal intensity generated when no sample or peptide is present).

Figure 2

PRO-C22 was significantly elevated in the serum from patients with solid tumors. Quantification of PRO-C22 in the serum of healthy subjects (n = 33) and bladder cancer (n = 20), breast cancer (n = 20), colorectal cancer (n = 20), gastric cancer (n = 20), head and neck cancer (n = 20), lung cancer (n = 20), melanoma cancer (n = 20), ovarian cancer (n = 20), pancreatic cancer (n = 20), prostate cancer (n = 20), and renal cancer (n = 20). PRO-C22 levels are presented individually as scatter plots with black bars marking the median. Samples measuring below the LLOQ were given the LLOQ value (of 3.9 ng/mL) determined in the validation of PRO-C22. Differences in the PRO-C22 levels between the cancer-types and the healthy subjects were calculated using a Kruskal–Wallis test followed by multiple comparisons to the controls with Dunn’s test. **** indicates p < 0.0001, *** p < 0.001, ** p < 0.01.

Figure 3

PRO-C1 and PRO-C3 are different from PRO-C22. Quantification of PRO-C1 (top) and PRO-C3 (bottom) in the serum of healthy subjects (n = 33) and bladder cancer (n = 20), breast cancer (n = 20), colorectal cancer (n = 20), gastric cancer (n = 20), head and neck cancer (n = 20), lung cancer (n = 20), melanoma cancer (n = 20), ovarian cancer (n = 20), pancreatic cancer (n = 20), prostate cancer (n = 20), and renal cancer (n = 20). PRO-C1 and PRO-C3 levels are presented individually as scatter plots with black bars marking the median. Samples measuring below the LLOQ were given the LLOQ value. Significant differences in the biomarker levels (PRO-C1 and PRO-C3) between the cancer-types and the healthy subjects were calculated using a Kruskal–Wallis test followed by multiple comparisons to the controls with Dunn’s test. *** p < 0.001, ** p < 0.01, ns indicates p > 0.05.

Figure 4

PRO-C22 and PRO-C3 were significantly elevated in patients with pancreatic ductal adenocarcinoma (PDAC). (a) Levels of PRO-C22 were significantly elevated in patients with PDAC (n = 34), compared to healthy subjects (n = 20). PRO-C22 was increased in patients with metastatic PDAC (stage IV, n = 19), compared to non-metastatic PDAC (stage I–III, n = 15). (b) Levels of PRO-C3 were significantly elevated in PDAC patients (n = 34), compared to healthy subjects (n = 20). Comparisons were based on nonparametric t-tests (Mann–Whitney test). **** indicates p < 0.0001, ** indicates p < 0.01, ns indicates a p > 0.05.

Figure 5

High levels of PRO-C22 and PRO-C3 are associated with a short overall survival (OS) in patients with PDAC. Kaplan–Meier survival plots showing the association between the OS and PRO-C22 (a) and PRO-C3 (b) levels in patients with PDAC. Orange lines shows the low biomarker levels (tertiles 1 + 2 = T1 + T2), blue lines show the high biomarker levels (tertile 3 = T3).

Figure 6

Classification and regression tree (CART) analysis to explore the additive value of PRO-C22 to PRO-C3. Dichotomization of the patients into low PRO-C3 levels (tertiles 1 + 2 = T1 + T2, orange) and high PRO-C3 levels (tertile 3 = T3, blue). The following branches of the tree represent the dichotomization of patients that additionally have low (T1 + T2, orange) or high (T3, blue) PRO-C22 levels. Median overall survival time (mOS) in months (m) and the number of patients (n) is listed in each group.