Pan-Tissue methylation aging clock: Recalibrated and

a method to analyze and interpret the selected

features

Karthikeyan Vijayakumar
Developmental biology lab
Dept of Biology
Introduction

 An increasing body of evidence suggests that

many manifestations of aging are epigenetic.

 The epigenomic landscape varies markedly

across tissue types and many age-related
changes depend on tissue type.

https://medium.com/prime-movers-lab/aging-clocks-6f3d8af715d1
 Timeline of different Epigenetic clocks

Hannum’s blood clock, Weidner’s 3 CpG clock Pheno Age Grim Age DeepMage
Horvath’s multi tissue • Uses clinical markers • Measures the time • Deep learning
clock along with methylation of death model to predict
• One of the first models data the methylation
and highly used age
models.

2013 2014 2018 2019 2021

The Clock – model making workflow
 Datasets used in the model
S.NO GEO ID Tissue No of Samples
Peripheral blood
1 GSE147740 1032
(mononuclear cells)
2 GSE87571 Whole blood 729
3 GSE74193 Prefrontal cortex 418
4 GSE88883 Breast tissue 100
5 GSE134379 Middle temporal gyrus 173
6 GSE134379 Cerebellum 177
7 GSE119078 Saliva 59
14 different tissues
8 GSE132547 Broncho alveolar lavage fluid 24
9 GSE137716 Bronchial brush 16
10 GSE151042 Umbilical cord blood 454
11 GSE90060 Uterine Endometrium 32
12 GSE59065 CD8+ T cells 96
13 GSE59065 CD4+ T cells 95 21 different datasets
14 GSE129428 Hippocampus 32
15 GSE157341 Liver 35
16 GSE107038 Liver 40
Dorsolateral prefrontal
17 GSE125895 196
cortex,
Hippo, Cerebellum, entorhinal
18 GSE125895
cortex
19 GSE154566 Buccal swab 900
20 GSE79100 Human Kidney 31
21 GSE138307 Bone 32
Total samples 4671
 Training the Clock model

Measuring
Training model
• 1/3 rd of the data was • Grid Search
test data • L1 ratio = 0.5
performance
• Data split to • Elastic net regression • Alpha value = 0.1 • R2
accommodate even age • 10 fold cross validation • RMSE
range in training and test • MSE
data
Hyper-
Data preparation parameter
tuning

Dataset N Mean age 1st Quartile Median age 3rd Quartile Age range

Training 3114 36.5 18 36 53 0-103

Test 1557 36.9 18 36 54 0-102

 Performance of the clock

Dataset Correlation(R2) RMSE MAE

Training 0.99 2.3 years 1.62 Years

Test 0.97 4.4 years 2.8 Years

Performance Compared with Horvath clock

Correlation( Correlation( MAE

Clock RMSE MAE Test
R2) Training R2) Test Training

Our Clock 0.99 0.97 2.3 years 1.62 Years 2.8 Years

Horvath
0.97 0.96 - 2.9 Years 3.6 Years
clock

Better accuracy with lower error

Performance in different tissues
Performance in tissues that was not used for the clock training

Cells and tissues which were related to the tissues used in the clock model performed better.
 Methylation pattern of the clock CpG sites

Development Aging

 Average methylation of each 6,761 clock CpGs for every age category was calculated.

 Methylation change was constantly increased after the age of 30.

 Centenarians showed lower methylation age.

 Pathway Enrichment

Genes annotation Pathway

enrichment

• 6,761 CpGs were • G-Profiler online

annotated with tool was used to
their respective generate
gene location. enriched
• Resulting in pathways list.
4,120 genes • Gene ontology
biological
process terms
were used

Most of the enriched pathways were development related

 Differentially Methylated Regions

 Multiple regression model was used to find the highly differentially methylated regions in the clock CpGs.

 50% were highly differentiating when the p value was set to ≤ 0.0001.
Comparing with known aging related genes

 Known genes were curated from

GenAge and CellAge databases.

 97 genes from our model was

common with known genes.

 Among them 62 were common

with the highly significant
probes.
 Known aging related genes

 80 CpGs sites located in the known genes.

 Enriched pathways showed the consequences of aging process.

 Antagonistic pleiotropy theory of aging

 Postulated by Williams in 1957.

 A gene regulates a trait positively during the early age and negatively regulates the
trait in the later life.

 mTOR is responsible for the cell division and development during the early life, but
it gets turned off in the later ages leading to cellular senescence.
 Conclusion
Biologicall
y every
individual
age
differently

We devised
a clock
which
measures
methylatio
n age

Evidence
for
antagonisti
c
pleiotropy
theory

https://newsroom.ucla.edu/releases/ucla-scientist-uncovers-biological-248950
감사합니다