How Long Will My Mouse Live? Predictors and Biomarkers of Aging

1
How Long Will My Mouse Live? Predictors and
Biomarkers of Aging

• Richard Miller
• University of Michigan
• October, 2007

2
Three and a Half Things to Keep Straight

• Age-sensitive traits
• Risk factors for death
• Biomarkers of aging
• Predictors of lifespan

3
Biomarker A surrogate of something hard to
measure

• Glycated hemoglobin integrated glucose levels
• Height childhood IGF-I levels
• Blood cotinine cigarette consumption
• Walk on this line please, sir blood alcohol

4
Are Alice and Betty The Same Biological Age?

• Both were born on the same day
• As young adults, they were the same in strength,
  speed, and cognition
• Now, in middle-age, Alice has better hearing, no
  cataracts, runs further, better vaccine response,
  less bone loss, good at chess
• It seems fair to say that Alice is younger then
  Betty

5
Alice and Betty Three Punchlines

• Ending I Alice is a miniature poodle, and Betty
  a standard poodle

6
Alice and Betty Three Punchlines

• Ending I Alice is a miniature poodle, and Betty
  a standard poodle
• Ending II Alice is a white-footed mouse
  (Peromyscus), and Betty a house mouse (Mus)

7
Alice and Betty Three Punchlines

• Ending I Alice is a miniature poodle, and Betty
  a standard poodle
• Ending II Alice is a white-footed mouse
  (Peromyscus), and Betty a house mouse (Mus)
• Ending III Alice and Betty are both people

8
The Biomarker Challenge in a Nutshell Can
Alice and Betty be People?Can we meaningfully
estimate biological age in normal people?
9
Three and a Half Things to Keep Straight

• Age-sensitive traits
• Risk factors for death
• Biomarkers of aging
• Predictors of lifespan

10
Risk Factors for Death

• Had a heart attack last week
• Wife just died
• Poor glucose tolerance
• Tied to railroad track
• Its easy to be a great risk factor and a
  terrible biomarker of aging.

11
Age-sensitive traits

• Grip strength
• Speed over one mile run
• Hearing high-pitched tones
• T cell subset distribution
• Grey hair
• Fond memories of Elvis, live in Vegas

12
Age-sensitive traitsWhich ones might be good
biomarkers?

• Grip strength
• Speed over one mile run
• Hearing high-pitched tones
• T cell subset distribution
• Grey hair
• Fond memories of Elvis, live in Vegas

13
Age-sensitive traitsWhich ones are proven
biomarkers?

• Grip strength
• Speed over one mile run
• Hearing high-pitched tones
• T cell subset distribution
• Grey hair
• Fond memories of Elvis, live in Vegas

14
How to validate biomarker (sets) Three
strategies

• (A) It changes with age, so its a biomarker

15
How to validate biomarker (sets) Three
strategies

• (A) It changes with age, so its a biomarker
• (B) It predicts lifespan, so its a biomarker

16
How to validate biomarker (sets) Three
strategies

• (A) It changes with age, so its a biomarker
• (B) It predicts lifespan, so its a biomarker
• (C) It predicts outcome of multiple,
  age-sensitive tests, after statistical adjustment
  for age

17
Single biomarkers are easy to fool

• Walk the line test could be blind person
• Glycated hemoglobin recent blood loss?
• Speed in mile run could be tennis pro
• High pitched sounds could be rock fan
• Grey hair could be naturally good-looking

18
Three and a Half Things to Keep Straight

• Age-sensitive traits
• Risk factors for death
• Biomarkers of aging
• Predictors of lifespan

19
The Predictors of Lifespan Trap

• Alice and Betty are newborn baby humans
• Alices parents are very wealthy, wear seatbelts,
  and live in Japan or Sweden
• Bettys parents are very poor, are addicted to
  drugs, and live in an American inner city. Worse
  yet, Betty is a boy (dont ask)

20
Weight at 3 Months Predicts Life Span in UM-HET3
Mice
21
IGF-I Levels Lower in Wild-Derived Mice
22
Low IGF-I Levels in Young Adults Predict Long
Life Span
Jim Harper et al.
23
Screening age-sensitive traits for use as
biomarkers of aging some first steps

• Does it change with age?
• Does it predict lifespan in the correct
  direction?
• Does it work in multiple populations?
• Is it independent of cause of death?
• Does it predict outcome of multiple other
  age-sensitive tests in middle-aged mice?

24
CD4M Predicts Life Span In Both Males and Females
25
CD4M Predicts Life Span in Six Populations of
Lab Mice, Wild Mice, and Hybrids
Credit Jim Harper
26
Four T Cell Subsets Predict Longevity When
Measured at 18 Months of Age in Mice
27
Immune Factor 1 (At 18 Months) Predicts Life Span
for Multiple Causes of Death
28
Immune Factor 1 (At 18 Months) Predicts Life Span
for Multiple Causes of Death
29
Immune Factor 1 (At 18 Months) Predicts Life Span
for Multiple Causes of Death
30
Immune Factor 1 (At 18 Months) Predicts Life Span
for Multiple Causes of Death
31
Factor 1 Score (At 18 Months) Predicts Longevity
for Three Major Causes of Death
32
T cell subsets, plus early life weight
James Harper
33
T cell subsets, plus early life weight
James Harper
34
Factor 1 Score (At 8 Months) Also Predicts
Longevity for Three Major Causes of Death
35
T cell subsets at 8 months, combined with early
weight
36
Conclusions T Cell Subsets as Biomarkers

• Mice with old-looking immune systems die young
• Association is significant for each of three
  causes of death
• The association is demonstrable as early as 8
  months of age, i.e. at lt 1/3rd of full life span

37
Conclusions T Cell Subsets as Biomarkers

• Mice with old-looking immune systems die young
• Association is significant for each of three
  causes of death
• The association is demonstrable as early as 8
  months of age, i.e. at lt 1/3rd of full life span
• Hypothesis I poor immunity leads to early death
• Hypothesis II rapid immune aging is an index of
  overall rapid aging, which leads to early death

38
Machine Learning Approach A First StabWilliam
Swindell

• 750 (heterogeneous) mice 29 variables
• Predict lifespan quartile using 90 of the mice
  test with the other 10
• Repeat this 10,000 times
• Random guess would be right 25 of the time
• Compare 19 machine learningalgorithms
• Best (Nearest Shrunken Centroid) has 32.9
  success
• Confidence interval 32.8 - 32.95

39
Just How Good is Nearest Shrunken
Centroid?Distributions of 10,000 simulation
runs
Bill Swindell
40
Survival curves for Quartiles 1 and 4 picked by
Nearest Shrunken Centroid
Bill Swindell
41
We need to do a better job of picking variables
that add new information
Bill Swindell
42
Biomarkers The View from the Pet Store
43
Biomarkers The View from the Pet Store
44
Biomarker (and gene mapping) research Kicking
the lifespan habit

• Mapping genes for lifespan is likely to find
• Genes for heart attacks
• Genes for cancer
• Genes for diabetes
• Genes for Alzheimers
• An alternate approach find genes that
  distinguish Alice from Betty at age 55

45
Not good enoughChallenges for biomarkerologists

• Do biomarkers (or combos) predict outcomes of
  other age-sensitive tests in middle-aged mice?
• Can the machine learning approach predict
  outcomes of other age-sensitive tests in
  middle-aged mice?
• Can we solve the Biomarker Equation
• Validated Biomarkers Dataset1.6 x Stats x
  Money

46
The Credit Slide

• Jim Harper
• Bill Swindell
• Clarence Chrisp
• Support National Institute on Aging (dba Shock
  Center and Pepper Center) VA Medical Center

47
Sit down and see if there are any questions