SEARCH

SEARCH BY CITATION

Keywords:

  • MIC-1/GDF15;
  • all-cause mortality;
  • serum marker;
  • cytokine;
  • prospective observational cohort;
  • environmental toxicity

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. Financial disclosure
  9. Competing interests
  10. References

Macrophage inhibitory cytokine-1 (MIC-1/GDF15) is a member of the TGF-b superfamily, previously studied in cancer and inflammation. In addition to regulating body weight, MIC-1/GDF15 may be used to predict mortality and/or disease course in cancer, cardiovascular disease (CVD), chronic renal and heart failure, as well as pulmonary embolism. These data suggested that MIC-1/GDF15 may be a marker of all-cause mortality. To determine whether serum MIC-1/GDF15 estimation is a predictor of all-cause mortality, we examined a cohort of 876 male subjects aged 35–80 years, selected from the Swedish Population Registry, and followed them for overall mortality. Serum MIC-1/GDF15 levels were determined for all subjects from samples taken at study entry. A second (independent) cohort of 324 same-sex twins (69% female) from the Swedish Twin Registry was similarly examined. All the twins had telomere length measured and 183 had serum levels of interleukin 6 (IL-6) and C-reactive protein (CRP) available. Patients were followed for up to 14 years and had cause-specific and all-cause mortality determined. Serum MIC-1/GDF15 levels predicted mortality in the all-male cohort with an adjusted odds ratio (OR) of death of 3.38 (95%CI 1.38–8.26). This finding was validated in the twin cohort. Serum MIC-1/GDF15 remained an independent predictor of mortality when further adjusted for telomere length, IL-6 and CRP. Additionally, serum MIC-1/GDF15 levels were directly correlated with survival time independently of genetic background. Serum MIC-1/GDF15 is a novel predictor of all-cause mortality.


Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. Financial disclosure
  9. Competing interests
  10. References

Macrophage inhibitory cytokine-1 (MIC-1/GDF15), a member of the TGFβ superfamily, is upregulated in many epithelial malignancies (Bauskin et al., 2006). Macrophage inhibitory cytokine-1 is a prognostic marker for prostate cancer (Bauskin et al., 2005; Brown et al., 2006), and its serum levels are massively elevated in the latter stages of this disease (Welsh et al., 2003). In prostate cancer, elevated serum MIC-1/GDF15 levels were identified as a likely cause of cancer-associated cachexia being directly related to body mass index (BMI) (Johnen et al., 2007). As in cancer, cachexia is a recognized adverse prognostic feature in heart failure, chronic renal failure, rheumatoid arthritis, and in the geriatric population (Horwich & Fonarow, 2007). In each of these conditions, serum MIC-1/GDF15 levels are elevated (unpublished data). In the case of chronic renal failure and heart failure, serum MIC-1/GDF15 is an independent predictor of mortality when compared with traditional markers such as C-reactive protein (CRP). Additionally, serum MIC-1/GDF15 level is an independent risk factor for development of cardiovascular events (Brown et al., 2002b) and are predictive of recurrent myocardial infarction and death from pulmonary embolism (Kempf et al., 2007; Lankeit et al., 2008). These findings suggested that serum MIC-1/GDF15 levels might be related to all-cause mortality.

To further clarify the relationship of serum MIC-1/GDF15 levels with mortality risk, we wished to determine whether serum MIC-1/GDF15 levels could predict all-cause mortality in the general population. Moreover, we wanted to establish the independent capability of serum MIC-1/GDF15 level to predict mortality and whether this relationship substantially improved mortality prediction when compared with the established mortality marker BMI and those known to be associated with mortality such as telomere length, CRP, and interleukin 6 (IL6). Lastly, we wanted to ascertain if serum MIC-1/GDF15 levels were related to longevity independently of genetic background. To answer these questions, we determined the predictive value of serum MIC-1/GDF15 levels in two independent cohorts, one of which included mono- and dizygotic twins, the other, a group of normal males.

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. Financial disclosure
  9. Competing interests
  10. References

Cohort characteristics

Pertinent characteristics of the study cohorts are given in Table 1. During a median follow-up time of 5.3 years (range, 0.1–6.0 years), 102 (12%) of the 876 study participants from the male cohort died. Of these, 30 died of cancer and 43 from cardiovascular events. Among the 324 male and female study participants in the twin cohort, 214 (66%) died during a median follow-up time of 9.1 years (range, 0.03–14.4 years). Of the 174 twins that died before December 31st, 2003, 34 died of cancer and 97 of cardiovascular events.

Table 1.   Descriptive characteristics of the study cohorts
CharacteristicsMale cohort (n = 876)Twin cohort (n = 324)
Follow-up (years), median (range)5.3 (0.1–6.0)9.1 (0.03–14.4)
Deaths102 (12%)214 (66%)
Age (years), median (range)68 (45.8–80.1)78.6 (63.3–93.3)
Male876 (100%)102 (31%)
Zygosity
 MZNA175 (54%)
 DZNA149 (46%)
Smoking status
 Never330 (38%)211 (65%)
 Current or past513 (59%)113 (35%)
 Unknown33 (4%)0
BMI (Kg m−2), median (range)25.9 (16.6–43.6)24.4 (13.8–42.8)
MIC-1/GDF15, (pg mL−1), median (range)934.9 (156–9638)1393 (428–8064)
CRP, (pg mL−1), median (range)NA2.0 (0.2–61.6)
IL-6, (pg mL−1), median (range)NA2.2 (0.5–15.1)
Telomere length, (kb), median (range)NA10.0 (9.3–10.6)

MIC-1/GDF15 serum levels

Cohort characteristics stratified into normal (< 1200 pg mL−1), moderately elevated 1200–1800 pg mL−1 and highly elevated (> 1800 pg mL−1) MIC-1/GDF15 serum levels are shown in Table 2. In the male cohort, MIC-1/GDF15 serum levels were significantly associated with increasing age (P < 0.0001) and smoking history (P = 0.0002). In the twin cohort, MIC-1/GDF15 serum levels were significantly associated with increasing age (P < 0.0001), increasing serum levels of IL-6 (P = 0.0002), and decreasing telomere length (P = 0.02). No association between MIC-1/GDF15 serum level and BMI was observed in any of the study cohorts; neither were CRP levels or gender associated with MIC-1/GDF15 levels in the twin cohort. In multiple linear regression analysis, age, BMI, and smoking history independently associated with MIC-1/GDF15 levels in the male cohort. In the twin cohort, multiple regression analysis revealed age, gender, and IL-6 serum levels as independently associated with MIC-1/GDF15 serum levels (data not shown).

Table 2.   Cohort characteristics according to normal, moderately elevated, and highly elevated macrophage inhibitory cytokine-1 (MIC-1/GDF15) levels
CharacteristicsMIC-1/GDF15 (pg mL−1)
< 12001200–1800> 1800P value
  1. BMI, body mass index; CRP, C-reactive protein.

Male cohort
 Age (years)66 (45–80)73 (55–79)73 (52–79)< 0.0001
 Current or  past smoker333 (56.5)107 (67.7)73 (76.0)0.0002
 BMI (Kg m−2)25 (16–41)25 (20–43)26 (18–39)0.37
Twin cohort
 Age (years)71 (63–86)77 (64–92)84 (65–93)< 0.0001
 Male36 (29.3)35 (36.5)31 (29.5)0.46
 Current or  past smoker39 (31.7)37 (38.5)37 (35.2)0.57
 BMI (Kg m−2)24 (17–42)25 (16–34)23 (13–37)0.26
 CRP (pg mL−1)1.7 (0.2–38.9)2.3 (0.2–18.4)2.1 (0.3–61.6)0.09
 IL-6 (pg mL)1.8 (0.5–15.1)1.9 (0.6–10.7)2.8 (0.6–11.4)0.0002
 Telomere  length, (kb)10.2 (7.9–13.0)9.9 (8.2–13.2)9.8 (7.5–13.0)0.02

MIC-1/GDF15 serum levels and overall mortality

Macrophage inhibitory cytokine-1 serum levels were significantly associated with poor survival in the male cohort (log-rank test, P < 0.0001; Fig. 1). While more than 90% of the individuals with normal MIC-1/GDF15 serum levels survived during follow-up, only 61% of those with the highest serum MIC-1/GDF15 levels (> 1800 pg mL−1) did, corresponding to a more than five-fold relative risk [hazard ratio (HR), 5.30; 95% confidence interval (CI), 3.37–8.35; Table 3]. In multivariate analysis that adjusted for age, BMI, and smoking history, elevated MIC-1/GDF15 serum levels remained associated with overall mortality (HR, 2.61; 95% CI, 1.53–4.45; Table 3).

image

Figure 1.  Elevated MIC-1/GDF15 serum levels are associated with increased mortality rates in normal men. Overall survival in the male cohort stratified by normal (< 1200 pg mL−1), moderately elevated (1200–1800 pg mL−1), and highly elevated (> 1800 pg mL−1) MIC-1/GDF15 serum levels. Those men with elevated serum MIC-1 levels had a significantly greater risk of mortality in the study period.

Download figure to PowerPoint

Table 3.   Cox regression analysis for overall death
Macrophage inhibitory cytokine-1 (pg mL−1)Deaths/individualsHR (95% CI) Crude analysisPHR (95% CI) Adjusted analysisPHR (95% CI) Adjusted analysisP
  1. Hazard ratios (HR) from multivariate Cox model adjusting for age at blood draw (continuous), body mass index (continuous) and smoking history (never/ever smoker).

  2. HR from multivariate Cox model adjusting for age at blood draw (continuous), body mass index (continuous), smoking history (never/ever smoker), telomere length(continuous), serum interleukin 6 levels (continuous), and serum CRP levels (continuous).

Male cohort (N = 876)
 < 120043/6101.00     
 1200–180026/1642.33 (1.43–3.79)0.00071.24 (0.71–2.16)0.45  
 > 180033/1025.30 (3.37–8.35)< 0.00012.61 (1.53–4.45)0.0004  
 P for trend  < 0.0001 < 0.0001  
Twin cohort (N = 324)
 < 120047/1231.00 1.00 1.00 
 1200–180072/962.74 (1.97–3.95)< 0.00011.68 (1.15–2.48)0.0082.07 (1.18–3.79)0.012
 > 180095/1055.84 (4.14–8.71)< 0.00012.20 (1.47–3.42)0.00022.17 (1.17–4.37)0.016
 P for trend  < 0.0001 < 0.0001 0.008

Serum levels of MIC-1/GDF15 were also significantly associated with poor survival in the twin cohort (log-rank test, P < 0.0001, Fig. 2); almost 60% of the individuals with normal MIC-1/GDF15 serum levels survived during follow-up, while no individual with highly elevated MIC-1/GDF15 levels did, yielding an almost six-fold relative risk (HR, 5.84; 95% CI, 4.14–8.71; Table 3). Concordant with the results in the male cohort, adjustment for age, BMI, and smoking history attenuated the strength of association between MIC-1/GDF15 serum levels and overall mortality; however, elevated MIC-1/GDF15 concentrations remained an independent predictor of death with a more than two-fold higher death rate among individuals with highly elevated compared to normal MIC-1/GDF15 levels (HR, 2.20; 95% CI, 1.47–3.42; Table 3). The association between MIC-1/GDF15 serum levels and mortality did not differ significantly by cohort (data not shown).

image

Figure 2.  Increasing MIC-1/GDF15 serum levels are associated with increased mortality rates in the Twin cohort. Overall survival in the twin cohort was stratified by normal (< 1200 pg mL−1), moderately elevated (1200–1800 pg mL−1), and highly elevated (> 1800 pg mL−1) MIC-1/GDF15 serum levels. Those subjects with elevated serum MIC-1 levels had a significantly greater risk of mortality in the study period.

Download figure to PowerPoint

Telomere length was available for all of the 324 participants in the twin cohort, and CRP and IL6 serum levels were available for analysis on a subset of 183 subjects. Additional adjustment for these markers did not affect the strength of association between MIC-1/GDF15 levels and mortality (Table 3). There were no significant gender differences in the MIC-1/GDF15 serum level–associated mortality risk in the twin cohort (data not shown). Considering the strong association of MIC-1 categories with mortality, we examined the relationship of serum MIC-1/GDF15 as a continuous variable with mortality, in a Cox proportional hazards model, in both cohorts. In the twin cohort, for every 1000 pg mL−1 rise in MIC-1/GDF15 serum level there was a significant increased risk of mortality (HR, 1.38; 95% CI, 1.20–1.59: P < 0.0001), when adjusted for history of smoking, age, sex, BMI, and telomere length. Similarly, in the all male cohort, for every 1000 pg mL−1 increase in serum MIC-1 level there was a significant increase in risk of mortal events (HR, 1.45; 95% CI, 1.29–1.63: P < 0.0001), when adjusted for age, history of smoking, and BMI.

Discriminative capacity of MIC-1/GDF15

Because MIC-1/GDF15 showed similar effects on mortality risk in both populations, we pooled the cohorts in exploring the discriminative capacity of MIC-1/GDF15 serum level to predict death. The area under the curve (AUC) at end of follow-up for MIC-1/GDF15 alone was 0.79 (95% CI, 0.76–0.82). Combining age, BMI, and smoking history in a logistic regression model yielded an AUC of 0.86 (95% CI, 0.84–0.89); also including MIC-1/GDF15 in the logistic model significantly (P = 0.01) increased the AUC to 0.87 (95% CI, 0.85–0.90).

MIC-1/GDF15 serum levels and cause-specific mortality

To characterize the association between MIC-1/GDF15 serum levels and cause-specific mortality, the male and twin cohorts were pooled (Table 4). After adjustment for age, BMI, and smoking history, individuals with highly elevated MIC-1/GDF15 levels, compared to individuals with normal levels, were at increased risk of cardiovascular death (OR, 1.93; 95% CI 1.25–2.99), cancer death (OR, 2.74; 95% CI 1.70–4.22), as well as death caused by other reasons (OR, 3.20; 95% CI 2.07–5.04).

Table 4.   Cox regression analysis for cause-specific death†
Cause of deathMIC-1/GDF15 (pg mL−1)
1200–1800> 1800
  1. †Hazard ratios from multivariate Cox model adjusting for age at blood draw (continuous). Logarithmically transformed macrophage inhibitory cytokine-1 (MIC-1/GDF15) serum level was modeled as a continuous variable.

Cardiovascular1.62 (1.00–2.63)0.0481.93 (1.13–3.29)0.016
Cancer1.30 (0.64–2.62)0.472.74 (1.36–5.49)0.0046
Other1.49 (0.89–2.49)0.133.20 (1.80–5.70)< 0.0001

Heritability of MIC-1/GDF15 levels

Twin similarity in serum MIC-1/GDF15 levels was assessed as within-pair correlations in monozygotic and dizygotic twins. The correlation within monozygous twin pairs was moderate (rs = 0.47, P = 0.0001), while in dizygous pairs the correlation was less pronounced (rs = 0.24, P = 0.05), but not significantly different from monozygous twin pairs (P = 0.30). In variance component analysis of genetic, shared environmental and unique environmental effects, the best model included genetic and unique environmental effects only (AE; Table 5). In this model, the heritability of MIC-1/GDF15 serum levels was estimated to be 0.48 (95% CI, 0.01–0.61) while unique environmental factors explained the remaining variance. This indicates that genetic and environmental factors contribute about equally to variation in serum MIC-1/GDF15 level.

Table 5.   Variance component analysis of MIC-1/GDF15† serum levels
ModelA: additive genetic effectsC: shared environmental effectsE: unique environmental effectsP value
  1. Age and gender adjusted macrophage inhibitory cytokine-1 (MIC-1/GDF15) serum levels.

Full (ACE)0.48 (0.01–0.61)0 (0.00–0.37)0.52 (0.38–0.70)
AE0.48 (0.30–0.61)0.52 (0.39–0.70)1.00
CE0.37 (0.22–0.49)0.63 (0.51–0.78)0.05

MIC-1/GDF15 serum levels predict death independent of genetic influences

To determine whether genetic factors contribute to the association of MIC-1/GDF15 serum levels with mortality, we examined the correlation between within twin-pair differences in serum MIC-1/GDF15 levels (age and sex adjusted) and survival time in the monozygous twins. The correlation between within-pair differences was moderate (rs = −0.42, P = 0.0004) indicating that the twin with the highest MIC-1/GDF15 levels had the shortest survival time. Thus, the observed association of serum MIC-1/GDF15 with mortality was not confounded by genetic factors.

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. Financial disclosure
  9. Competing interests
  10. References

Our results show that serum MIC-1/GDF15 level is a novel and powerful predictor of all-cause mortality in general populations independent of several markers associated with mortality risk including age, BMI, smoking history, IL-6 and CRP serum levels, and telomere length. Furthermore, serum MIC-1/GDF15 level, like telomere length, predicted longevity independently of genetic background. While the relationship of serum MIC-1/GDF15 level to mortality prediction for cardiovascular disease (CVD) and cancer was expected, serum MIC-1/GDF15 level was also related to deaths from other causes in both male and twin cohorts.

The reasons underlying the association between serum MIC-1/GDF15 levels and noncardiovascular, noncancer deaths is unclear, but may be related to the aging process and its impact on serum MIC-1/GDF15 levels, which are known to rise with age (Brown et al., 2006). Biologic age has been related to several serum markers that reflect cumulative oxidative stress, protein glycation, inflammation, hormonal changes, and replicative senescence (Simm et al., 2008). Many of these stresses induce MIC-1/GDF15 expression via either p53 or egr-1 transcription factors to which MIC-1/GDF15 is responsive (Bauskin et al., 2006). For example, p53 is induced in cell senescence, which is also related to telomere length (Simm et al., 2008), is correlated with serum MIC-1/GDF15 levels. Telomere length is also associated with oxidative stress (Carrero et al., 2008), indicating that serum MIC-1/GDF15 levels may be responsive to environmental factors. This notion is supported by our variance component analysis, suggesting MIC-1/GDF15 levels are significantly affected by environmental factors, independently of genetic background.

The relative contribution of genetic and environmental factors in humans is best determined by twin studies. We found that genetic factors explain about half of the variation in MIC-1/GDF15 levels and that the remaining variance could be attributed to individual-specific environmental influences. However, there was no evidence for involvement of shared environmental factors. Additionally, despite a genetic component, presumably dictating baseline levels of MIC-1/GDF15, differences in longevity in monozygotic twin pairs were strongly correlated with differences in serum MIC-1/GDF15. As MZ twins have identical genotypes, differences in environmental factors affecting the individual seem to be the major determinant of the differences in serum MIC-1/GDF15 level that relate to mortality risk. However, to confirm this, a longitudinal study would have to show an altered rate of rise of MIC-1/GDF15 in genetically identical twins. Additionally, investigation of the cohort of twins indicated that unique, rather than shared environmental factors influenced serum MIC-1 levels independent of genetic background (Table 5). This might indicate that serum MIC-1/GDF15 levels reflect environmental factors associated with aging. For example, oxidative stress is associated with premature aging (Salmon et al., 2010) and upregulates p53 and HIF-1α (Bianchi et al., 2006), both capable of upregulating MIC-1/GDF15 expression (Bauskin et al., 2006; Saletta et al., 2010). Other environmental toxins directly upregulate egr-1 (Harrill et al., 2008; Nielsen et al., 2009), potentially regulating MIC-1/GDF15 expression (Bauskin et al., 2006). These data, combined with our findings in the twin cohort, add weight to the possibility that serum MIC-1/GDF15 levels might reflect environmental toxicity. Given these findings and the relationship of serum MIC-1/GDF15 level with age, the question as to whether the rate of longitudinal increases of serum MIC-1/GDF15 might be a better marker of mortality risk is raised.

A related intriguing question is whether elevated serum levels of MIC-1/GDF15 are directly damaging or whether these levels represent the body’s attempt to mitigate the damage caused by an injurious insult. At least in advanced cancer, elevated serum levels appear to be detrimental and therapeutic reduction in circulating MIC-1/GDF15 serum levels has been proposed for the treatment of cachexia (Johnen et al., 2007), where serum MIC-1/GDF15 levels are inversely related to BMI (Johnen et al., 2007). Reduced BMI below the normal range is a risk factor for mortality in cancer (Johnen et al., 2007), renal failure, rheumatoid arthritis, and geriatric populations (Horwich & Fonarow, 2007), all of which have elevated serum MIC-1/GDF15 levels (Brown et al., 2002b, 2007; Welsh et al., 2003; Johnen et al., 2007). However, serum MIC-1/GDF15 level and BMI were not related in the cohorts we studied. Only after adjustment for age and smoking history, a weak positive relationship between MIC-1/GDF15 and BMI was observed in the male cohort. The relationship between BMI and serum MIC-1/GDF15 was defined in cohorts with diseases that directly impact BMI, i.e. cancer and renal failure (Johnen et al., 2007). Therefore, in normal populations, the relationship between serum MIC-1/GDF15 may not be so pronounced. While the current evidence does not support or dismiss a direct causal relationship between serum MIC-1/GDF15 levels and increased all-cause mortality, serum MIC-1/GDF15 levels predict mortality risk independently of age, sex, BMI, genetic background, and inflammatory factors. Therefore, it is possible that disease-specific therapeutic interventions that reduce serum MIC-1/GDF15 levels might also confer a decreased risk of death and increase longevity.

In animal models, longevity is mediated by a number of genes associated with metabolic and cell cycle pathways (Fontana et al., 2010). The role of the cell cycle regulator p53 in longevity has been extensively studied in mice (Donehower, 2009). While p53 null mice have a significantly shortened lifespan (Tyner et al., 2002), cells from mice with genetic abnormalities that are associated with premature aging have increased levels of p53 (Rudolph et al., 1999). In these mouse models, eliminating p53 expression by breeding with p53 mutant mice reverses the premature aging phenotype (Chin et al., 1999; Lim et al., 2000) establishing a role for p53 in the regulation of aging. As MIC-1/GDF15 is a prominent target of p53 which upregulates its transcription (Bauskin et al., 2006), these mice undergoing premature aging might be expected to have high circulating MIC-1/GDF15 levels reflecting p53 pathway activation. Other models of increased longevity in eukaryotes, from yeast to primates, have highlighted the importance of metabolic pathways in determining longevity (Fontana et al., 2010). In humans, an increased frequency of IGF-1 receptor mutations are found in centenarians (Suh et al., 2008) and reduced concentrations of IGF-1 are associated with longevity (Bonafe et al., 2003). In mice, we have shown that MIC-1/GDF15 are associated with IGF-1 levels (Johnen et al., 2007). Additionally, in one small human study, serum MIC-1 levels were also associated with IGF-1 levels (Dostalova et al., 2010). While this association does not necessarily indicate a direct role for MIC-1/GDF15 in the aging process, its documented effect on other molecules might. For example, MIC-1/GDF15, like IGF-1, is capable of inducing AKT phosphorylation (Subramaniam et al., 2003; Yamaguchi et al., 2004; Wollmann et al., 2005; Pang et al., 2007), which is reduced in most, if not all, eukaryote models of increased longevity induced by nutrient restriction (Fontana et al., 2010). Induction of another aging-related transcription factor HIF-1α leads to nutrient restriction–independent increased longevity in Caenorhabditis elegans (Fontana et al., 2010). HIF-1α binds to the MIC-1/GDF-15 promoter inducing transcription (Saletta et al., 2010) indicating MIC-1/GDF15 could play a protective role in aging. However, induction of HIF-1α in other animal models failed to increased longevity (Fontana et al., 2010). While these studies suggest MIC-1/GDF15 could directly mediate processes in aging, any such role for MIC-1/GDF15 will have to be defined in future animal and human studies and cannot be inferred from results of this study.

This study is limited in its design, in that the two cohorts were dominated by different genders and differed significantly in smoking status, BMI, and age range. The twin cohort examined in this study had lower BMI, less smokers, were older, and had more females. All these differences are likely to interact with the measured outcome, mortality. However, despite these differences, both cohorts showed remarkably similar relationships between serum MIC-1/GDF15 level and all-cause mortality, defining serum MIC-1/GDF15 level as an important newly identified independent marker of raised all-cause mortality risk that potentially links inflammation and mortality with body weight regulation (Johnen et al., 2007). Additionally, for disease risk determination, serum MIC-1/GDF15 level appears to add additional information, independent of other markers such as CRP. Clearly, further investigation to ascertain whether MIC-1/GDF15 serum estimation will identify additional at-risk individuals years before the onset of clinically overt disease is warranted.

Experimental procedures

  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. Financial disclosure
  9. Competing interests
  10. References

Study populations

The present analyses are based on two independent study populations: (i) a group of 876 Swedish men aged 35–80 years and (ii) 324 Swedish male and female twins aged 63–93 years. From here on, these study populations will be referred to as the ‘Male cohort’ and ‘Twin cohort’, respectively. All study participants granted informed consent at the time of enrollment and the ethics committees at Karolinska Institutet and Umeå University approved the studies.

The male cohort comprises randomly selected and frequency-matched (in 5-year age groups and geographic area) population controls from a case–control study of prostate cancer (Lindmark et al., 2004). At the time of recruitment (between 2001 and 2002), all men donated blood and self-reported on life-style habits. The only exclusion criterion applied in recruiting these controls was a history of prostate cancer. Of the 1697 randomly selected control subjects invited, 945 (56%) agreed to participate by blood donation of which 69 had no sample available. For the present study, 876 individuals with available blood samples were available for analysis.

The twin cohort, previously explored in studies of aging (McClearn et al., 1997; Bakaysa et al., 2007), was ascertained from the Swedish Twin Registry (Lichtenstein et al., 2002). Zygosity for twin pairs was determined by self-report and later confirmed by either restriction fragment length polymorphism or serological testing and microsatellite markers. For the present study, whole blood and self-reported life-style habits were available from 324 twins comprising 154 complete twin pairs (70 monozygotic and 84 same-sexed dizygotic twin-pairs) and 16 unrelated twin individuals, which were included in analyses where zygosity was not examined. In total, 222 (69%) of the 324 twins were women.

Follow-up assessment

Vital status was followed through January 15, 2008 (male cohort) or December 31, 2006 (twin cohort) through record linkage to the Swedish Population Registry. Cause-specific survival until December 31st, 2005 (male cohort) or until December 31st, 2003 (twin cohort) was obtained through linkage with the national Cause of Death Registry. For the male cohort, review of death certificates established cause of death for individuals deceased after December 31st, 2005. Causes of death were coded according to the International Classification of Diseases (ICD). In addition to overall mortality, we evaluated primary causes of death because of cancer (ICD10 C00 to D48) and CVD (ICD10 I10 to I99).

Laboratory analyses

In both cohorts, MIC-1/GDF15 serum concentrations (pg mL−1) were determined using a sensitive in house sandwich ELISA (Brown et al., 2002a), as previously described. All samples were assayed in at least duplicate, and the coefficient of variation between samples was < 12%.

In the twin cohort, telomere length was assessed by terminal restriction fragment analysis, which relies on restriction enzyme digestion and Southern blot hybridization of a minimum of 105 cells to measure the average length of telomeres (Bakaysa et al., 2007). Serum levels of IL-6 were analyzed using the Quantikine high-sensitvity ELISA commercial kit by R&D systems (Minneapolis, MN, USA), and high-sensitivity CRP was determined using Beckman reagents on Synchron LX20 automated equipment (Beckman Coulter, Fullerton, CA USA).

Statistical analysis

Unless otherwise noted, statistical analyses were performed using R version 2.6.1 (Team, 2006). Unpaired t-test and chi-square analysis were used for continuous and categorical variables, respectively, and correlations between biomarkers were calculated by Spearman’s rank test. For our main analysis, overall mortality, we performed time-to-event analysis using death from any cause as the outcome with survival time censored at end of follow-up for individuals still alive. In sub-analyses, cause-specific mortality was assessed in time-to-event analysis using death from CVD, death from cancer, or death from causes other than CVD and cancer as the endpoints. In these analyses, survival time was censored at time of death for patients dying from causes other than the specific cause explored or at end of follow-up for individuals still alive. Association between MIC-1/GDF15 serum level and mortality was assessed in Cox regression analysis with serum levels categorized into three prespecified groups (< 1200, 1200–1800 and > 1800 pg mL−1) with the lowest category used as reference. These limits correspond to normal, moderately elevated, and highly elevated levels of MIC-1/GDF15 in a Swedish cohort of 429 apparently healthy elderly individuals (Kempf et al., 2007). To explore the independent effect of MIC-1/GDF15 serum levels on mortality, multivariate Cox regression models including attained age, BMI, and smoking history, in addition to MIC-1/GDF15 serum levels, were performed. For the twin cohort, we also performed analyses additionally adjusted for the prognostic markers telomere length, IL6, and CRP serum levels.

Multivariate logistic regression analysis was used to assess if MIC-1/GDF15 serum levels, when used in combination with the prognostic markers age, BMI, and smoking history, improved the predictive capacity of overall mortality. Predicted probabilities of death were applied to calculate receiver operating characteristic curves using the AUC with 95% CI as measure of predictive performance. The AUC of the model including both MIC-1/GDF15 and established markers was compared with the AUC of the model including only the established markers using the method described by Hanley and coworkers (Hanley & McNeil, 1983), which accounts for the fact that the AUCs are derived from the same sample of patients, as implemented in STATA version 9.1 (Stata Corp, College Station, TX, USA).

To account for the relatedness among twins, bootstrapping (Efron, 1979) was applied to generate empirical distributions for determination of CI and P-values. Survival curves were computed by the Kaplan–Meier method and compared among risk-stratification groups using the log-rank statistic. All P values reported were based on two-sided hypothesis.

Variance partitioning and heritability

Intraclass correlations were computed as indicators of twin similarity. The variance of MIC-1/GDF15 was partitioned into additive genetic effects (A), shared environmental effects (C), and unique environmental effects (E) through structural equation modeling as implemented in the Mx statistical program (Neale et al., 2006). Because the distribution of MIC-1/GDF15 was skewed, log transformation was applied which yielded an approximately normal distribution. Among the twins, logMIC-1/GDF15 was found to be significantly associated with age and gender. We therefore adjusted MIC-1/GDF15 values for age and sex by linear regression. Residuals standardized into Z-scores were subsequently used as the phenotype for twin modeling. Scripts downloaded from the GenomEUtwin Mx-script library (http://www.psy.vu.nl/mxbib/) were used after modification. The principle of parsimony was implemented to determine which nested models were to be preferred (e.g. ACE or AE) in case of nonsignificant differences.

To explore whether genetic factors contribute to the association between MIC-1/GDF15 levels and mortality, within twin-pair analysis using the twin-difference design was performed. Pearson correlation of within-pair difference in MIC-1/GDF15 (age and sex adjusted log MIC-1/GDF15) and within-pair difference in survival time was investigated among monozygotic twin pairs to determine whether the relationship of MIC-1/GDF15 with survival is confounded by genetic factors.

Acknowledgments

  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. Financial disclosure
  9. Competing interests
  10. References

We thank all study participants in the CAPS and twin studies. Additional thanks go to Professor Matthew Law for critical review of the manuscript. Brown and Wiklund had full access to all the data and take full responsibility for the integrity of and accuracy of the data analysis.

Financial disclosure

  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. Financial disclosure
  9. Competing interests
  10. References

This work was supported by grants from the National Health and Medical Research Council of Australia (NHMRC), a New South Wales Health Research and Development Infrastructure grant, Prostate Cancer Foundation Australia, and the Swedish Cancer Society (HG and HOA). This study was also partially supported by an NCI grant R01CA105055 (Xu). The collection of the twin materials was supported by grants from the NIH (AG 04563, AG10175 and AG 08861). Brown is a Biomedical Postdoctoral Fellow of the NHMRC and a SpinalCure Australia Senior Fellow. PK, EM are sponsored by the Swedish Heart and Lung Foundation. Measurements of IL6 and CRP were supported by grants from the Loo och Hans Ostermans Foundation to AB who is also funded by the Karolinska Institute Board of Research. The funding sources had no direct or indirect involvement in the design and conduct of the study, nor the collection, management, analysis, and interpretation of the data, nor in the preparation, review, or approval of the manuscript.

Competing interests

  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. Financial disclosure
  9. Competing interests
  10. References

Brown and Breit are named inventors on patents held by St Vincent’s Hospital Sydney covering intellectual property regarding the use of serum MIC-1/GDF15 determination in cancer. Wiklund, Adami, and Grönberg declare no conflicts of interest.

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. Financial disclosure
  9. Competing interests
  10. References