Limits...
Protein profiling reveals consequences of lifestyle choices on predicted biological aging.

Enroth S, Enroth SB, Johansson Å, Gyllensten U - Sci Rep (2015)

Bottom Line: Lifestyle factors such as smoking or stress can impact some of these molecular processes and thereby affect the ageing of an individual.Here we demonstrate by analysis of 77 plasma proteins in 976 individuals, that the abundance of circulating proteins accurately predicts chronological age, as well as anthropometrical measurements such as weight, height and hip circumference.We found smoking, high BMI and consumption of sugar-sweetened beverages to increase the predicted chronological age by 2-6 years, while consumption of fatty fish, drinking moderate amounts of coffee and exercising reduced the predicted age by approximately the same amount.

View Article: PubMed Central - PubMed

Affiliation: Department of Immunology, Genetics, and Pathology, Biomedical Center, SciLifeLab Uppsala, Uppsala University, SE-75108 Uppsala, Sweden.

ABSTRACT
Ageing is linked to a number of changes in how the body and its organs function. On a molecular level, ageing is associated with a reduction of telomere length, changes in metabolic and gene-transcription profiles and an altered DNA-methylation pattern. Lifestyle factors such as smoking or stress can impact some of these molecular processes and thereby affect the ageing of an individual. Here we demonstrate by analysis of 77 plasma proteins in 976 individuals, that the abundance of circulating proteins accurately predicts chronological age, as well as anthropometrical measurements such as weight, height and hip circumference. The plasma protein profile can also be used to identify lifestyle factors that accelerate and decelerate ageing. We found smoking, high BMI and consumption of sugar-sweetened beverages to increase the predicted chronological age by 2-6 years, while consumption of fatty fish, drinking moderate amounts of coffee and exercising reduced the predicted age by approximately the same amount. This method can be applied to dried blood spots and may thus be useful in forensic medicine to provide basic anthropometrical measures for an individual based on a biological evidence sample.

No MeSH data available.


Related in: MedlinePlus

Protein overlap in core models.Overlaps between proteins present in each of the four core models predicting Age, Hip Circumference (HIP), Weight and Height.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4664859&req=5

f2: Protein overlap in core models.Overlaps between proteins present in each of the four core models predicting Age, Hip Circumference (HIP), Weight and Height.

Mentions: We have previously quantified abundance levels of circulating plasma proteins from cardiovascular and cancer biomarker panels using the highly sensitive protein extension assay (PEA)1021 in 976 individuals from the Northern Swedish Population Health Study (NSPHS). Seventy-seven of these protein measurements were used to build models to predict chronological age, weight, height and hip circumference. Prediction models were built using generalized linear models with penalized maximum likelihoods as implemented by the glmnet-package23 in R24 and models were optimized using a 10-fold cross-validation scheme on 75% of the observation and subsequently evaluated using the remaining 25% (see Methods for details). We repeated the process 500 times and recorded which proteins were selected in the model. As expected, individual variation in protein abundance values and the distribution of phenotypes, gave rise to some variation in the proteins selected to be part of the final model. On average 68 of the 77 proteins were included in the model predicting age (Fig. 1A, Table 1). In total, all 77 proteins were included at least once in any of the age predicting models and a core set of 29 proteins was present in all models. The models for age, height, weight and hip circumference performed well on the test and training sets (Table 1) and summary statistics (including protein inclusion statistics) for all models and traits are reported in Supplementary tables 2–5. The models predicted chronological age with an R2 = 0.83, while predicting weight (R2 = 0.48), height (R2 = 0.34) and hip circumference (R2 = 0.60) with somewhat lower correlation coefficients. An example of the correlation between chronological and predicted age for one model is shown in Fig. 1B, and the distribution of prediction errors for 500 age models in Fig. 1C. In the test sets, 95% of the average errors for each of the models were within +/− 1.23 years and there was no statistically significant difference (p = 0.52, Wilcoxon Ranked Sum test) between the distribution of errors in the training and test sets, indicating that the models were not over-fitted to the training data. In terms of accuracy, the plasma protein profile predicted chronological age within 5.0 years, weight within 6.8 kg, height within 4.7 cm and hip circumference within 5.1 cm, for 50% of the observations. Additional performance measurements for the models are shown in Supplementary Figures 1–3. We also evaluated the performance of the models when restricted to a core set of proteins that were included in all models for each trait (Table 1). Interestingly, the models based on the core set of proteins showed similar performance statistics as the models using the full set of proteins, suggesting that a smaller set of proteins can capture most of the phenotype variation. This observation was also confirmed by an analysis of the fraction of variance of the traits that can be explained by individual and combined proteins included in the prediction models (Supplementary Figure 4, Supplementary Tables 2–5). An analysis of the overlap between the proteins that were present in the four core-models showed that only 4 proteins (Fig. 2) were common between all models. These were Tissue plasminogen activator (tPA), Tumor necrosis factor receptor 1 (TNFR1), the Receptor tyrosine-protein kinase ErbB-3 (ErbB3) and Endothelial cell-specific molecule 1 (ESM-1). None of the genes coding for these proteins have been implicated in a recent GWAS for variation in human adult height25. In our material, out of the four proteins common to all models, ESM-1 explains the largest proportion of the variance seen in height (9.8%, Supplementary Table 4). ESM-1 is mainly expressed in endothelial cells in lung and kidney tissue but circulates in the bloodstream26. We have found no evidence relating ESM-1 to height in the literature but speculate that circulating levels of ESM-1 could be a reflection of lung volume, which is correlated to height27. Notably, none of the four proteins in common to the traits are among the set of proteins explaining the largest fraction of variance in the four traits (Supplementary Tables 2–5).


Protein profiling reveals consequences of lifestyle choices on predicted biological aging.

Enroth S, Enroth SB, Johansson Å, Gyllensten U - Sci Rep (2015)

Protein overlap in core models.Overlaps between proteins present in each of the four core models predicting Age, Hip Circumference (HIP), Weight and Height.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4664859&req=5

f2: Protein overlap in core models.Overlaps between proteins present in each of the four core models predicting Age, Hip Circumference (HIP), Weight and Height.
Mentions: We have previously quantified abundance levels of circulating plasma proteins from cardiovascular and cancer biomarker panels using the highly sensitive protein extension assay (PEA)1021 in 976 individuals from the Northern Swedish Population Health Study (NSPHS). Seventy-seven of these protein measurements were used to build models to predict chronological age, weight, height and hip circumference. Prediction models were built using generalized linear models with penalized maximum likelihoods as implemented by the glmnet-package23 in R24 and models were optimized using a 10-fold cross-validation scheme on 75% of the observation and subsequently evaluated using the remaining 25% (see Methods for details). We repeated the process 500 times and recorded which proteins were selected in the model. As expected, individual variation in protein abundance values and the distribution of phenotypes, gave rise to some variation in the proteins selected to be part of the final model. On average 68 of the 77 proteins were included in the model predicting age (Fig. 1A, Table 1). In total, all 77 proteins were included at least once in any of the age predicting models and a core set of 29 proteins was present in all models. The models for age, height, weight and hip circumference performed well on the test and training sets (Table 1) and summary statistics (including protein inclusion statistics) for all models and traits are reported in Supplementary tables 2–5. The models predicted chronological age with an R2 = 0.83, while predicting weight (R2 = 0.48), height (R2 = 0.34) and hip circumference (R2 = 0.60) with somewhat lower correlation coefficients. An example of the correlation between chronological and predicted age for one model is shown in Fig. 1B, and the distribution of prediction errors for 500 age models in Fig. 1C. In the test sets, 95% of the average errors for each of the models were within +/− 1.23 years and there was no statistically significant difference (p = 0.52, Wilcoxon Ranked Sum test) between the distribution of errors in the training and test sets, indicating that the models were not over-fitted to the training data. In terms of accuracy, the plasma protein profile predicted chronological age within 5.0 years, weight within 6.8 kg, height within 4.7 cm and hip circumference within 5.1 cm, for 50% of the observations. Additional performance measurements for the models are shown in Supplementary Figures 1–3. We also evaluated the performance of the models when restricted to a core set of proteins that were included in all models for each trait (Table 1). Interestingly, the models based on the core set of proteins showed similar performance statistics as the models using the full set of proteins, suggesting that a smaller set of proteins can capture most of the phenotype variation. This observation was also confirmed by an analysis of the fraction of variance of the traits that can be explained by individual and combined proteins included in the prediction models (Supplementary Figure 4, Supplementary Tables 2–5). An analysis of the overlap between the proteins that were present in the four core-models showed that only 4 proteins (Fig. 2) were common between all models. These were Tissue plasminogen activator (tPA), Tumor necrosis factor receptor 1 (TNFR1), the Receptor tyrosine-protein kinase ErbB-3 (ErbB3) and Endothelial cell-specific molecule 1 (ESM-1). None of the genes coding for these proteins have been implicated in a recent GWAS for variation in human adult height25. In our material, out of the four proteins common to all models, ESM-1 explains the largest proportion of the variance seen in height (9.8%, Supplementary Table 4). ESM-1 is mainly expressed in endothelial cells in lung and kidney tissue but circulates in the bloodstream26. We have found no evidence relating ESM-1 to height in the literature but speculate that circulating levels of ESM-1 could be a reflection of lung volume, which is correlated to height27. Notably, none of the four proteins in common to the traits are among the set of proteins explaining the largest fraction of variance in the four traits (Supplementary Tables 2–5).

Bottom Line: Lifestyle factors such as smoking or stress can impact some of these molecular processes and thereby affect the ageing of an individual.Here we demonstrate by analysis of 77 plasma proteins in 976 individuals, that the abundance of circulating proteins accurately predicts chronological age, as well as anthropometrical measurements such as weight, height and hip circumference.We found smoking, high BMI and consumption of sugar-sweetened beverages to increase the predicted chronological age by 2-6 years, while consumption of fatty fish, drinking moderate amounts of coffee and exercising reduced the predicted age by approximately the same amount.

View Article: PubMed Central - PubMed

Affiliation: Department of Immunology, Genetics, and Pathology, Biomedical Center, SciLifeLab Uppsala, Uppsala University, SE-75108 Uppsala, Sweden.

ABSTRACT
Ageing is linked to a number of changes in how the body and its organs function. On a molecular level, ageing is associated with a reduction of telomere length, changes in metabolic and gene-transcription profiles and an altered DNA-methylation pattern. Lifestyle factors such as smoking or stress can impact some of these molecular processes and thereby affect the ageing of an individual. Here we demonstrate by analysis of 77 plasma proteins in 976 individuals, that the abundance of circulating proteins accurately predicts chronological age, as well as anthropometrical measurements such as weight, height and hip circumference. The plasma protein profile can also be used to identify lifestyle factors that accelerate and decelerate ageing. We found smoking, high BMI and consumption of sugar-sweetened beverages to increase the predicted chronological age by 2-6 years, while consumption of fatty fish, drinking moderate amounts of coffee and exercising reduced the predicted age by approximately the same amount. This method can be applied to dried blood spots and may thus be useful in forensic medicine to provide basic anthropometrical measures for an individual based on a biological evidence sample.

No MeSH data available.


Related in: MedlinePlus