Remodeling of the Tumor Microenvironment Predicts Increased Risk of Cancer in Postmenopausal Women: The Prospective Epidemiologic Risk Factor (PERF I) Study

In the United States, the population ages 65 years and older is expected to double by 2060 and the incidence of cancer is estimated to increase among the elderly by 67% (1, 2). Cancer is, therefore, expected to remain one of the leading causes of death among 65- to 85-year olds in the Western world (3). Early detection of cancer greatly improves the chances of survival. There is, therefore, a need for early detection of cancers, especially in the elderly. This could be accomplished by developing noninvasive biomarkers for selection of high-risk subjects eligible to enter cancer screening programs.

One of the earliest signs of cancer is an altered tumor microenvironment, which involves the appearance of cancer associated fibroblasts, recruitment of tumor-associated macrophages, enhanced extracellular matrix (ECM) deposition, and protease secretion (4). Failure to recruit a permissive tumor microenvironment is believed to be the reason some tumors remain dormant and in accordance an altered and collagen-dense microenvironment is considered a potential risk factor for development of invasive cancers (5, 6).

We recently developed noninvasive biomarkers that reflect modifications in the tumor microenvironment. These markers measure matrix metalloprotease (MMP)-degraded type I collagen (C1M), MMP-degraded type IV collagen (C4M), and citrullinated and MMP-degraded vimentin (VICM; refs. 7–9).

C1M reflects interstitial matrix remodeling and measures a type I collagen degradation fragment generated by cleavage with MMP-2, -9, and -13 (8). Type I collagen is one of the most abundant interstitial ECM proteins. It is involved in maintaining tissue architecture and serves as a barrier for migration of epithelial cells under healthy conditions (10). During cancer progression, however, type I collagen is often dysregulated and remodeled by MMPs (11). The architecture of type I collagen also changes. It stiffens and becomes cross-linked, which together with MMP-driven tissue degradation drives cell invasion and migration (12).

C4M reflects basement membrane (BM) remodeling and invasion, and measures a type IV collagen α1-chain fragment generated by cleavage with MMP-12 (7). A hallmark of the malignant process is the acquisition of an invasive phenotype that enables malignant cells to invade the BM (13). The BM is a compact specialized ECM structure that consists mainly of type IV collagen and laminin (14). Studies show that the cellular invasion through the BM in cancer is mainly driven by increased MMP activity, which allows the malignant cells to enter the interstitial matrix and spread to distant sites (15, 16).

VICM is believed to reflect intermediate filament (IF) remodeling and inflammation and measures a citrullinated fragment of vimentin generated by cleavage with MMP-2, -3, -8, -9, -12, and -13. Vimentin is used as a marker for epithelial-to-mesenchymal transition (EMT) and can be secreted from activated macrophages and endothelial cells into the microenvironment for further processing by MMPs (17–21). Furthermore, vimentin is known to undergo citrullination by protein arginine deiminases that are overexpressed in malignant tumors and associated with inflammation (18, 22).

We have previously seen that C1M, C4M, and VICM levels are elevated in serum of cancer patients compared with serum levels measured in healthy controls (21, 23, 24), but it is not known if these biomarkers are elevated prior to the diagnosis of cancer. The aim of this study was to investigate whether levels of C1M, C4M, and VICM in serum samples collected in a large prospective study of postmenopausal women could be used as predictors of increased risk of cancer.

The Prospective Epidemiologic Risk Factor (PERF I) study aims at finding risk factors associated with age-related diseases, as previously described by Dragsbæk and colleagues (25). A total of 5,855 Danish postmenopausal women ages 48 to 89 years enrolled in the PERF I study during 1999 to 2001 (baseline). Follow up registry data from the Danish Cancer Registry were collected in 2014. Women who had previously either participated in clinical randomized placebo-controlled studies or had been screened for previous studies at the Center for Clinical and Basic Research in Denmark were invited to participate in PERF I. The PERF I study was carried out in accordance with ICH-GCP and the study protocol was approved by the local ethics committees.

Registry data were collected in 2014 from the Danish Cancer Registry. The last updated data were entered in the registry in December 2012, leading to an average follow up time of 12.1 years of women enrolled in PERF I. Cancer subtypes were classified according to WHOs international Classification of Diseases 10 (ICD10), and reported in this paper as cancer of the breast (C50), digestive organs (C15–C26), respiratory organs (C30–C39), female genital organs (C51–C58), and other organs (C40–C41, C43–C44, C45–C49, C64–C68, C69–C72, C73–C75, D00–D09).

Subjects diagnosed after baseline with a malignant neoplasm were included as cases (n = 881). Excluded from the cases group were subjects with benign cancers (D10–D36), nonmelanoma skin cancer (C44), hematologic cancers (C81–C96), desmoplasia (N87), and neoplasms of uncertain behavior (D37–D48). If subjects had more than one cancer diagnosis after baseline the diagnosis closest to baseline was used in the analysis. Subjects with a cancer diagnosis solely prior to baseline were also excluded from the analysis (Fig. 1).

Subjects with no history of cancer, and no connective tissue, circulatory, respiratory, digestive, musculoskeletal, and genitourinary system diseases (ICD10 chapter II, IX, X, XI, XIII, XIV) were included as controls and are referred to as healthy women (n = 528) in this paper.

At baseline subjects reported on demographic characteristics together with current smoking status, alcohol consumption, physical activity, and level of education as well as whether they were receiving treatment for hypertension or hyperlipidemia. The baseline characteristics of the PERF I subpopulation studied in this article is representative of the entire PERF I cohort (25). Vital signs and fasting serum samples were collected at time of enrollment and serum samples were stored at −80°C for later analysis.

C1M (n = 5,629), C4M (n = 5,630), and VICM (n = 5,630) were measured blinded in serum by competitive ELISA in a CAP-certified laboratory as described by Leeming and colleagues (8), Sand and colleagues (7), and Vassiliadis and colleagues (9).

The serum samples were tested for stability and were considered to be stable after storage at −80°C. In detail, a 3-year stability study were performed, where the biomarker levels were measured in 1-year intervals. Furthermore, 10 freeze–thaw cycles were done with no significant change in biomarker levels. Finally, the median biomarker levels measured in this study was comparable to median biomarker levels measured in other similar studies with shorter storage times.

Serum samples were measured in double determinations and the coefficient of variation (CV) was <15%. The intra- and interassay variations were <10% and <15%, respectively.

Baseline characteristics of women with cancer events and healthy women were compared using a Mann–Whitney test for numerical variables and a chi-square test for categorical variables.

The levels of C1M, C4M, and VICM were compared using a Kruskal–Wallis test, where the median biomarker levels of women diagnosed with cancer <1 year after baseline were compared with the biomarker levels of the healthy women, and with women diagnosed with cancer 1 to 2 years, 2 to 4 years, and >4 years after baseline.

A Spearman correlation test was used to examine a possible relationship between days to cancer diagnosis after baseline and serum levels of C1M, C4M, and VICM.

Univariable and multivariable logistic regression analysis were performed to assess the ORs and AUC for the biomarkers and selected risk factors. Healthy women and women diagnosed with cancer <1 year after baseline were included in the analysis. C1M, C4M, and VICM levels were analyzed on a continuous scale and split into two groups based on median levels (38.6 ng/mL for C1M, 71.7 ng/mL for C4M, and 3.3 ng/mL for VICM). In the multivariable analysis, a backward selection method was used to identify the best prediction model and to analyze the independent nature of the biomarkers. To correct for overfitting, an internal validation was conducted by calculating the bootstrap optimism-corrected AUC. (The data were resampled 1,000 times with the bootstrapping method.) A covariate, defined as the time lag between the cancer diagnosis and baseline, was introduced into the model.

The statistical analyses were performed using MedCalc Statistical Software v.12 (MedCalc Software), R software (2.15.1 version; R Development Core Team, 2012), and GraphPad Prism v.6 (GraphPad Software).

From the PERF I study population two major groups were defined: women diagnosed with cancer after baseline (n = 881) and healthy women with no history of cancer, and no ECM- or inflammatory-related diseases (n = 528). The women diagnosed with cancer after baseline were further subdivided according to time of diagnosis and tumor stage (Fig. 1).

Table 1 summarizes the baseline characteristics of the total cohort and the two groups. The median age of the total population at baseline was 71.3 years, and the median age of the women diagnosed with cancer after baseline was significantly higher than that of the healthy women (71.9 years vs. 70.2 years; P < 0.0001). The entire cohort was characterized by being slightly overweight (BMI 25.5) with the healthy women having a significantly lower BMI than women with cancer events (BMI 25.2 vs. 25.8; P = 0.0007). Compared with the healthy women, the group with cancer had a significantly higher percentage of current smokers (26.2% vs. 20.7%; P = 0.02), a significantly lower proportion of physically active subjects (71.8% vs. 81.0%; P = 0.0001), a lower percentage educated to high school level (19.4 % vs. 22.3%; P = ns), and a nonsignificant higher proportion of alcohol-consumers drinking ≥7 drinks/week (34.5% vs. 31.1%; P = ns). The healthy women compared with the women with cancer had a significantly lower proportion women on hypertensive treatment (23.9% vs. 33.1%; P = 0.0003) and hyperlipidemia treatment (8.7% vs. 7.5%; P = ns).

The serum C1M and C4M levels, but not VICM (3.5 ng/mL vs. 3.1 ng/mL), were significantly higher (41.1 ng/mL vs. 38.0 ng/mL and 72.7 ng/mL vs. 71.0 ng/mL; P = 0.0001 and 0.02) in women diagnosed with cancer after baseline compared with the healthy women. When age matching the cohort C1M and C4M levels continued to be significantly different in women diagnosed with cancer after baseline than healthy controls (data not shown).

Patients diagnosed with cancer after baseline were divided into four groups: women diagnosed <1, 1 to 2, 2 to 4, and >4 years after baseline. The levels of C1M, C4M, and VICM were all elevated in serum from women diagnosed <1 year after baseline compared with women diagnosed >4 years after baseline and healthy women.

In detail, C1M levels were significantly elevated in serum from women diagnosed <1 year after baseline compared with women diagnosed >4 years after baseline and healthy women (P = 0.019 and P ≤ 0.0001, respectively; Fig. 2A). C4M levels were significantly elevated in serum from women diagnosed <1 year after baseline compared with women diagnosed 1 to 2, 2 to 4, >4 years after baseline and healthy women (P = 0.02, 0.03, 0.04, and 0.002, respectively; Fig. 2B). VICM levels were significantly elevated in serum from women diagnosed <1 year after baseline compared with women diagnosed >4 years after baseline and healthy women (P = 0.0075 and 0.0071, respectively; Fig. 2C). When the biomarker levels in women diagnosed with cancer <1 year after baseline were compared with the levels of all women without a cancer diagnosis (n = 4015), C1M, C4M, and VICM continued to be significant (data not shown).

C1M, C4M, and VICM levels all correlated to time of cancer diagnosis up to 1,000 days after baseline (r = −0.14, P = 0.047; r = −0.17, P = 0.011 and r = −0.18, P = 0.0067, respectively; Supplementary Fig. S1A–S1C). The significant P-value may, however, be due to the large power in this study and not necessarily reflect a real correlation between days and biomarker levels.

Together, these findings indicate that changes in C1M, C4M, and VICM could be useful for prediagnostic risk assessments as elevated levels of the biomarkers indicate risk of a cancer diagnosis within a year.

The women diagnosed with cancer up to 1 year after baseline were divided into three groups: women diagnosed with localized cancers, lymphovascular invasions, or metastases. The median C1M, C4M, and VICM levels were highest in serum from women diagnosed with metastasized cancers. C1M levels were significantly elevated in patients diagnosed with metastasis compared to healthy controls (P = 0.01; Fig. 3A). C4M levels were significantly elevated in patients diagnosed with metastasis compared with subjects diagnosed with localized cancers and healthy controls (P = 0.002 and 0.03; Fig. 3B). Finally, VICM levels were significantly elevated in patients diagnosed with metastasis compared with healthy controls (P = 0.05; Fig. 3C).

Univariable logistic regression was used to assess C1M, C4M, VICM levels as well as risk factors of cancer development (age, smoking, BMI, physical inactivity, alcohol consumption, education level, hypertension, and hyperlipidemia) in women diagnosed with cancer up to 1 year after baseline compared with healthy women. Biomarkers were analyzed on a continuous scale as well as split into two groups based on median levels.

C1M, C4M, and VICM, age, smoking, exercise, and hypertension were all individual predictors of cancer up to 1 year prior to diagnosis. In detail, women with high levels of C1M and VICM were 2.2 and 2.3 times more likely to develop cancer within the first year after blood draw than women with low levels of C1M and VICM (OR = 2.2, AUC = 0.60, P = 0.0006, OR = 2.3, AUC = 0.60, P = 0.0003, respectively). Women with high levels of C4M were 1.6 times more likely to develop cancer within the first year after blood draw than women with low levels of C4M (OR = 1.6, AUC = 0.57, P = 0.031; Table 2).

To test if C1M, C4M, and VICM were independent predictors of cancer, a multivariable logistic regression model was made. A backward selection of the markers yielded in C1M and VICM levels, age, smoking, and exercise being statically significant, and the bootstrapped AUC was 0.70. This suggests that both C1M and VICM, but not C4M, levels up to 1 year prior to diagnosis were independent predictors of cancer in this cohort of elderly women (Table 3). The statistically significant association between C1M, VICM, and cancer diagnosis was slightly attenuated (OR = 1.8 vs. 2.2 for C1M and OR = 1.9 vs. 2.3 for VICM) when the markers were included in the multivariable logistic regression model.

In this study, we showed that C1M and VICM, but not C4M, are independent predictors of increased risk of cancer in postmenopausal women. In detail, we found that postmenopausal women with high levels of C1M, C4M, and VICM were more likely to develop cancer within the first year after baseline than women with low levels of the biomarkers. Furthermore, we found that women had significantly increased C1M, C4M, and VICM levels up to 1 year prior to cancer diagnosis.

These findings correspond well with the literature where studies show that vimentin, increased MMP expression, BM degradation, and increased collagen density are markers of poor prognosis (6, 18, 26, 27). Studies showed that vimentin is overexpressed in majority of cancers and is generally associated with a metastatic phenotype and poor prognosis (18). Similarly, MMP-2 and MMP-9 are often found elevated in tumor tissues (9, 28) and MMP-mediated collagen degradation has been shown to be predictive of cancer mortality in studies in elderly women (25, 29). This indicates that subjects with high levels of collagen remodeling have increased chances of both developing and dying from cancer.

It is possible that increased circulating C1M and C4M levels mirror a more active ECM remodeling response. If a cancer spreads and invades distant tissues, it encounters two kinds of ECM barriers: the dense endothelial BM and the underlying more porous interstitial matrix. It is likely that C4M mirrors BM remodeling and C1M, interstitial matrix remodeling. During cancer progression, both ECM barriers are remodeled resulting in a loss of epithelial polarity and uncontrolled growth (30, 31).

Similarly, it is possible that increased circulating VICM levels mirror a more active inflammatory response. During cancer progression, innate immune cells are recruited to the tumor microenvironment where they contribute to cancer development, mostly due to their secretion of inflammatory mediators that regulate proliferation, angiogenesis, and tissue remodeling (14, 32, 33).

Results suggest that inflammatory conditions and ECM remodeling are present before a malignant change occurs (34, 35). Whether inflammation and ECM remodeling are sufficient for the development of cancer is still unclear. However, it is well accepted that modification of the tumor microenvironment is one of the earliest signs of cancer. This could explain why high C1M, C4M, and VICM serum levels were observed prior to diagnosis of cancer in this study cohort.

There are several limitations to this study. We saw that the biomarker levels increased with tumor stage. It, therefore, seems likely that the markers are secreted directly from the tumor tissue. The number of subjects in each category is low and from this study we cannot establish causality. It is possible that the high levels of C1M, C4M, and VICM in the blood are results of a systemic effect where subjects with a general increased remodeling of the ECM and IF have a higher likelihood of developing cancer. Furthermore, all markers have been found elevated in a number of diseases characterized by ECM remodeling and tissue inflammation and are, therefore, not specific to cancer (7–9, 36–39). Finally, it needs to be elucidated if the results of this study can be applied to populations of other ages and ethnic backgrounds and if the biomarkers are more useful for certain types of solid cancer.

Despite these limitations, C1M and VICM may play a future role as predictors of cancer risk in combination with age, exercise, and smoking status. There is a great need to identify predictors of cancer risk for implementation of cancer screening programs. Screening entire populations is expensive and may cause patients unnecessary worry or harm, such as from overdiagnosis, false-positives, and radiation risk (40, 41). For successful implementation of national screening programs, it is therefore necessary to identify high-risk subjects to maximize the harm-benefit and cost-effectiveness ratios (40, 41). In addition, currently screening trials favor younger adults, but as life expectancy increases it will become more important to include elderly subjects in screening and clinical trials (1).

In conclusion, we saw that C1M, C4M, and VICM were elevated up to 1 year prior to cancer diagnosis. C1M and VICM were both independent predictors of increased risk of cancer in this study of postmenopausal Danish women. Measurement of modification in the tumor microenvironment could be useful for prediction of increased risk of cancer.

No potential conflicts of interest were disclosed.

Conception and design: C.L. Bager, N. Willumsen, A.-C. Bay-Jensen, M. Karsdal

Development of methodology: C.L. Bager, N. Willumsen, A.-C. Bay-Jensen, D.J. Leeming, M. Karsdal

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K. Dragsbæk, J.S. Neergaard, M. Karsdal

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C.L. Bager, N. Willumsen, A.-C. Bay-Jensen, D.J. Leeming, M. Karsdal

Writing, review, and/or revision of the manuscript: C.L. Bager, N. Willumsen, S.N. Kehlet, A.-C. Bay-Jensen, D.J. Leeming, K. Dragsbæk, J.S. Neergaard, C. Christiansen, E. Høgdall, M. Karsdal

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C.L. Bager, S.N. Kehlet, H.B. Hansen, A.-C. Bay-Jensen, M. Karsdal

Study supervision: A.-C. Bay-Jensen, E. Høgdall, M. Karsdal

Other (sample analysis): S.N. Kehlet

We would like to acknowledge the Danish Research Foundation (Den Danske Forskningsfond) for funding the PERF I study.

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.