Ethics Statement

All participants gave written consent, and the Uppsala University ethics committee approved the study.

Study Sample

We used the Uppsala Longitudinal Study of Adult Men (ULSAM), a prospective cohort study to which all 50-year-old men living in Uppsala County, Sweden, in 1970–1973 were invited (www.pubcare.uu.se/ULSAM). Out of 2,841 invited, 2,322 accepted. The cohort was re-examined at circa age 60 and age 71, the latter examination (in 1991–1995) defining the baseline cohort of the present study. Of the 1,681 71-year-old men alive, 1,221 (73%) participated. A further 100 men were excluded because of lack of matrix biomarker measurements, and another 39 due to lacking covariates, rendering a sample of 1,082 men eligible for the present study. A subsample of 818 men without cardiovascular disease (myocardial infarction [ICD-9 code 410, ICD-10 code I21], stroke [ICD-9 codes 430–433 & 436; ICD-10 codes I60–I64], heart failure [chart review-validated cases], or electrocardiographic Q-waves or left bundle-branch block [Minnesota codes 1.1 and 7.1]) or cancer (ICD-9 codes 140–209; ICD-10 codes C00–C99) was also investigated in order to minimize influence of reverse causation.

Baseline Investigations

At the age 71 examination, all participants answered self-administered questionnaires, underwent a medical examination, blood sampling after an overnight fast, supine blood pressure measurements, and anthropometric measurements as described previously. [18]

Smoking experience was classified into four levels: never smoking, current smoking, or two levels of previous smoking (above or below median [33] of pack-years of previous smoking). Diabetes at baseline was defined as fasting plasma glucose > = 7.0 mmol/L or use of oral hypoglycemic agents or insulin. Office systolic and diastolic blood pressures were measured twice in the right arm to the nearest even number with the subject in the supine position after resting for 10 minutes, and the means of the two measurements were used. Use of antihypertensive or lipid-lowering drugs was classified using questionnaires. Cholesterol concentration was analyzed by an enzymatic technique. High sensitivity C-reactive protein and cystatin C were measured using latex enhanced reagent (Dade Behring, Deerfield, IL, USA) using a Behring BN ProSpec analyzer (Dade Behring). Glomerular filtration rate was estimated from serum cystatin C by the formula eGFR  = 77.24*[cystatin C]-1.2623.

MMP-9 and TIMP-1 were measured in fasting serum samples frozen shortly after sampling and then stored in -70 degrees C awaiting analysis. The samples were thawed and analyzed in early 2007 using commercial ELISA assays provided by R&D Systems Europe Ltd (Abingdon, United Kingdom). The samples had when thawed been frozen for a mean of 13.5 years (extremes 11.6–15.4 years), herein denoted “freezer time”.

Follow-Up and Outcome Parameters

For the primary outcome, all-cause mortality, participants were followed from the baseline examination in 1991–1995 to October 4^th, 2009, with a maximum of 18.1 years of follow-up (median 14.7 years; 13,682 person-years at risk). For all other outcomes, follow-up data was available until December 31, 2007, rendering up to 16.4 years of follow-up. We investigated six outcomes: all-cause mortality; cardiovascular mortality (ICD-9 codes 390–459; ICD-10 codes I00-I99); non-cardiovascular mortality (all other deaths); cancer mortality (ICD-9 codes 140–209; ICD-10 codes C00-C99); fatal or non-fatal myocardial infarction (ICD-9 code 410, ICD-10 code I21); and fatal or non-fatal stroke (ICD-9 codes 430–433 & 436; ICD-10 codes I60–I64). The primary outcome was defined using the Swedish census registry, which is updated within a few days after a person's death. All other outcomes were defined using the Swedish national Cause-of-Death and In-Patient registries, which are updated yearly with some lag. All three data sources include all Swedish citizens, and none of the study participants emigrated during follow-up; therefore we had zero loss to follow-up.

Statistical Analysis

Initially, distributional properties of all variables were examined (Figures S1 and S2). We thereafter examined relations of MMP-9 and TIMP-1 to the six outcomes (all-cause mortality, CVD mortality, non-CVD mortality, cancer mortality, fatal or non-fatal myocardial infarction, and fatal or non-fatal stroke), each matrix biomarker and outcome in a separate set of models, using Cox proportional hazards analyses.

We investigated relations per standard deviation (SD) and by quartiles of MMP-9 and TIMP-1 to all outcomes. Freezer time was a priori decided upon as a covariate in all models, as the effects of long-term freezing on these biomarkers is not known. Both MMP-9 (r = 0.07, p = 0.02) and TIMP-1 (r = −0.10, p = 0.002) levels were statistically significantly albeit weakly related to freezer time. Age was adjusted for in all models by using age as the timeline. We considered three sets of models, in a hierarchical fashion: A) models adjusted for age and freezer time; B) models additionally adjusted for systolic blood pressure, antihypertensive treatment, lipid-lowering treatment, diabetes mellitus, body mass index, total cholesterol, estimated glomerular filtration rate, and smoking status; and C) models additionally adjusted for C-reactive protein (because inflammation has been identified as an important MMP stimulus). Model A is the primary model, and models B and C are viewed as mechanistic models – if substantial parts of the effects of cardiovascular disease risk factors or inflammation on risk are mediated by matrix remodeling (as reflected in altered levels of matrix markers), the matrix marker estimates will be lower in these models than in models A.

Schoenfeld's tests and inspection of Schoenfeld residuals and Nelson-Aalen curves were used to confirm proportionality of hazards, and Martingale residuals and multivariable-regression spline models with 5 degrees of freedom for the matrix markers were inspected to rule out deviations from linearity, in models A and B. We investigated additive interactions between the matrix markers (4^th vs. 1^st quartiles) and the covariates by calculating the relative excess risk due to interaction (RERI),[19] and multiplicative interactions between the matrix markers (on their continuous scale) and the covariates were tested using product terms. Stata 10.1 was used for all calculations.

Comparisons with Previous Studies

Earlier studies of associations of levels of circulating biomarkers of extracellular matrix turnover with mortality risk have so far mainly been made in groups of patients with manifest cardiovascular disease or cancer. In previous studies of patients with atherosclerotic cardiovascular disease, TIMP-1 has been observed to predict cardiovascular[11], [12] and all-cause mortality[11] and cardiovascular events,[20] and MMP-9 has been associated with risk of cardiovascular death in one study[13] and a combined endpoint of stroke or cardiovascular death in another,[21] but no relation was observed in one study.[11] In one study of men, TIMP-1 was not related to prognosis,[22] and in another population-based study selected based on left ventricular geometry and with limited number of cases, TIMP-1 was related to mortality risk but not risk of cardiovascular events, and MMP-9 measured with an assay with limited sensitivity was not related to any outcome.[23] Studies in samples of heart failure patients have been inconsistent, with reports of relations of these markers with prognosis in some[24] but not other studies.[25] MMP-9 level has been related to more severe neurological deficits[26] and increased risk of hemorrhagic transformation of infarcts in stroke patients[27], [28], [29], but relations of TIMP-1 to stroke outcomes are unknown. Relations of these markers to incident stroke risk in the general population are also unknown. In one recent large nested case-control study, relations of MMP-9 to myocardial infarction and stroke were observed in unadjusted models, but not in multivariable-adjusted ones.[30]

The present study is in line with previous observations in patient samples of associations of TIMP-1 with all-cause and cardiovascular mortality, but extends these to a population-based setting and identifies a robust relation to stroke risk. The weaker associations of MMP-9 than TIMP-1 levels to cardiovascular mortality in the present study are also in line with the previous literature, but are now demonstrated in the general population.

Results of studies in small samples of cancer patients have been inconsistent, but higher circulating MMP-9 and TIMP-1 levels have indicated worse prognosis in studies of several cancer types.[14], [15], [16], [17] No previous studies have reported associations of levels of matrix biomarkers to cancer incidence. The present study does not support strong such associations for major cancer types, but cannot exclude modest relations or relations to less frequent cancer types.

Potential Mechanisms

Relations of TIMP-1 to cardiovascular disease risk may have a variety of explanations. TIMP-1 is the main inhibitor of MMP-9, but also inhibits several other MMPs. Discrepant findings between MMP-9 and TIMP-1 is therefore not unexpected.

The investigated proteins may be causally involved in atherosclerosis development. In experimental settings, increased TIMP-1 expression has led to decreased atherosclerosis development in some[31]–[32] but not other studies,[33] and decreased MMP-9 expression has also led to a reduced atherosclerotic burden.[34]

MMP-9 and TIMP-1 may also be causally involved in cardiac dysfunction. In animal models of heart failure of different origin, pharmacological MMP inhibition reduces left ventricular dilatation and preserves cardiac systolic function[35], [36] and transgenic animal models also support the notion of detrimental effects of MMP-9 and beneficial effects of TIMP-1 after myocardial infarction.[37], [38] In general population-based samples, circulating MMP-9 and TIMP-1 levels have been related to left ventricular hypertrophy,[7], [8], [9] and TIMP-1 levels inversely to left ventricular systolic function.[7], [8] One mechanism behind the relations to heart failure development may be intracellular cleaving of myosin filaments by MMP-9.[39] The sources of the matrix markers in this study are unknown, but relations of circulating and myocardial levels of some matrix markers are substantial.[6]

One explanation behind the observed robust relations of TIMP-1 to stroke risk may be that higher TIMP-1 levels signify higher left ventricular mass.[7], [8] Left ventricular hypertrophy is a powerful risk factor for stroke, possibly because it may lead to atrial fibrillation.

Explanations behind the observed relation of MMP-9 to non-cardiovascular death are less substantiated. The matrix markers were not important risk factors for cancer mortality in the present study, but previous associations of these markers with prognosis in patients with malignant disease have been explained by an MMP-independent growth-promoting capacity of TIMP-1,[40] and effects of MMP-9 on tumor growth, metastasis and angiogenesis.[41]

MMP-9 has been demonstrated to regulate leukocyte function through a number of mechanisms, including activation of pro-IL-1β into IL-1β and truncation of IL-8 into a more active form and, as degranulation of MMP-9 from neutrophils is stimulated by IL-8, thus creating a positive feedback loop.[42] MMP-9 may therefore potentially play a role in the course of infectious diseases; and possibly in pulmonary disease.

Altered extracellular matrix turnover is a known cause of osteoporosis and fractures, which in some cases can lead to fatalities. This could also explain part of the non-cardiovascular mortality.

All relations of MMP-9 to outcomes were attenuated by adjustment for covariates, which is consistent with the notion that MMP-9 levels may be on a causal pathway between the covariates and the outcomes. Relations of TIMP-1 level to stroke risk were independent of these covariates. The discrepancy between relations of MMP-9 and TIMP-1 to cardiovascular events was not just due to differences in measurement precision, as the point estimates were lower for MMP-9 relations. The fact that TIMP-1 is the inhibitor also of other MMPs than MMP-9 (notably MMP-1), and these MMPs may also be important in the pathophysiology of cardiovascular disease, may partly explain why TIMP-1 was a stronger marker than MMP-9 of cardiovascular disease risk in this and other studies.

Inflammation is a known trigger of MMP expression and secretion. The observation that the relation of TIMP-1 to stroke risk was independent of C-reactive protein levels, as well as other cardiovascular risk factors, may indicate that extracellular matrix metabolism is an important process per se. Similarly, TIMP-1 predicted cardiovascular events independently of C-reactive protein, interleukin-6 and white cell count in another study of a patient sample.[20] Extracellular matrix metabolism should therefore be viewed as an entity separate from cytokine-mediated inflammation, and its best measurement and prognostic value determined.

Strengths and Limitations

Strengths of this study include the prospective cohort design, the relatively large number of participants, and the general population sample. The most obvious limitation is that it only comprises elderly men. Generalizability to women, younger persons, and non-Caucasians is unknown. Possible pre-analytical errors of TIMPs and MMPs have been a matter of debate. Falsely elevated serum levels due to release from leukocytes and thrombocytes during handling have been suggested.[43] On the other hand, this release may be a physiologically important property. One study reported decreasing levels of MMP-9, but not TIMP-1, in plasma over time in spite of frozen storage,[44] but the effects of freezing on serum levels are unknown. In this study, we accounted for time in the freezer in all statistical analyses. Because a smorgasbord of different MMPs and TIMPs exist, other MMP and TIMP species than the ones studied may be important. The cause of death is based on national registry data, and misclassification between causes of death is possible, but likely random. The investigation of multiple outcomes and multiple models is motivated in an etiological analysis such as the present, but may increase the risk of type-I error.

Conclusions

In this study, higher levels of circulating MMP-9 or TIMP-1 were associated with higher risk of death. In particular, higher TIMP-1 levels were related to higher risk of stroke and cardiovascular death.

An altered extracellular matrix metabolism is indicted in several detrimental pathways. Based on these observations, the most important of these pathways from a mortality perspective may be primarily cardiovascular disease-related ones. Circulating MMP-9 or TIMP-1 levels may be promising markers of those matrix metabolism alterations. We encourage investigation of these associations in other general population samples, and investigation of relations of other markers of extracellular matrix metabolism to hard endpoints in the general population. This is a novel group of mortality risk factors reflecting pathways for which development of drugs is ongoing.[45] Before such drugs can be useful in clinical practice, tools for identification of target groups and for monitoring treatment need to be developed. Our findings may be an important step in that direction.