Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
ApoB‐Lipoprotein profile modelling to
derive cardiovascular risk markers
Albert de Graaf
Dietary fat
absorption

From: Keith Frayn, “Metabolic Regulation - A Human Perspective”.
Wiley-Blackwell, Third Edition, 2...
Exogenous pathway of
lipoprotein metabolism

From: Keith Frayn, “Metabolic Regulation - A Human Perspective”.
Wiley-Blackw...
Endogenous pathway of
lipoprotein metabolism

From: Keith Frayn, “Metabolic Regulation - A Human Perspective”.
Wiley-Black...
HDL metabolism

From: Keith Frayn, “Metabolic Regulation - A Human Perspective”.
Wiley-Blackwell, Third Edition, 2010
Forward and reverse
cholesterol transport

From: Keith Frayn, “Metabolic Regulation - A Human Perspective”.
Wiley-Blackwel...
Lipoprotein structure
Different lipoprotein classes

From: Keith Frayn, “Metabolic Regulation - A Human Perspective”.
Wiley-Blackwell, Third Edi...
Lipoprotein Distribution (LPD) measurement
Cholesterol and heart disease

From: Keith Frayn, “Metabolic Regulation - A Human Perspective”.
Wiley-Blackwell, Third Edi...
Stable isotope studies: lipoprotein classes
correspond to kinetically different pools
Multicompartmental model for apoB
me...
Our vision …

LPD kinetic model
development

Model-based LPD analysis
Lipoprotein flux ratio
biomarkers
Advantage of our concept
From one single plasma measurement to rates of processes
Process rates are closer to functional a...
Particle Profiler project overview
Background of model
Model
development

=> Particle size-dependent rate constants => fluxes of different lipoprotein
proces...
Lipolysis cascades - 1
Model
development

Best fit overall
Lipolysis cascades - 2

The ApoB LPD size range is divided into 11 cascade
fractions
Each cascade has e.g. 1000 particle s...
Lipolysis steps - decreasing particle size

apoB apoE
uptake uptake

HL - lipolysis

LPL - lipolysis

Production

Liver at...
Lipoprotein production process model

Production flux (particles / min)

Model
development

VLDL1

VLDL2

IDL
Lipoprotein ...
Lipolysis and uptake process models (single particle rates)

0.02

process rate (1 / min)

Model
development

Each subfrac...
Flux Data for model testing
Packard et al. 2000, human study
Model validation

stable isotope fluxes analyzed with multi-c...
Example output (main classes) compared
to stable isotope study
Particle Pools

Uptake Flux
0.2

400

number of particles f...
Flux Data for model testing
4 results that give credibility to our model
Model validation

Model can reproduce particle co...
concentration

Model validation

Lipoprotein
Pooled
Fluxes

Lipoprotein
Profile
particle diameter

Particle Profiler
Lipop...
What are we doing?
Derive
model-based
markers

Derive ratios between processes

For instance:
VLDL lipolysis outside the l...
Example new diagnostic for dyslipidemia
Derive
model-based
markers
Receiver operating characteristic (ROC) curves

Validate markers

Diagnostic

AUC

'VLDL performance'

0,937

'TG (mmol/l)...
Reclassification analysis
You have three risk categories
Validate markers

People are treated based on their category
Low ...
Validation CVD risk prediction
The Framingham Heart Study Offspring cohort (FOS)
Validate markers

Inclusion criteria:
no ...
Statistics
Validate markers

First using logistic regression
No improvement possible
Then SVM methodology
Best predicting ratios
Validate markers

In the VLDL size range
VLDLH – VLDL Hepatic turnover indicator

Liver
functional ...
Predictor variables
Conventional
markers

LDLp

LDLp + HDLp

LDLp + HDLp +
VLDLE + VLDLH

Age

Age

Age

Age

Age

Sex

Se...
Area under the ROC-curve statistics for general CVD

Validate markers

Framingham Offspring Study
* Significantly better t...
Reclassification analysis
Model 1

Model 2

NRI

Standard
Error

P value

% of
Events
correctly
reclassified
(n=145)

Even...
What is happening? An ‘average’ person.
A - no blood pressure medication

Validate markers

B - no blood pressure medicati...
Conclusions
Lipoprotein metabolic ratios derived from Particle Profiler
significantly improve CVD risk prediction
As measu...
But…
Clinicians & customers require an independent validation.
Different cohort e.g. MESA (Multi Ethnic Study of Atheroscl...
Lessons learned
SVM and binary variables – take care!
Introducing new diagnostics in the current clinical setting

is a al...
Towards linking carbohydrate and fat metabolism
Lipoprotein metabolism as a central node for interplay of
glucose, TG and ...
TNO can contribute datasets
Collaborators
TNO
Daan van Schalkwijk
Evgeni Tsivtsivadze
Bianca van der Werff
Ben van Ommen
Albert de Graaf
(Andreas Frei...
Upcoming SlideShare
Loading in …5
×

Apo B‐Lipoprotein profile modelling to derive cardiovascular risk markers share

702 views

Published on

Published in: Technology, Business
  • Be the first to comment

  • Be the first to like this

Apo B‐Lipoprotein profile modelling to derive cardiovascular risk markers share

  1. 1. ApoB‐Lipoprotein profile modelling to derive cardiovascular risk markers Albert de Graaf
  2. 2. Dietary fat absorption From: Keith Frayn, “Metabolic Regulation - A Human Perspective”. Wiley-Blackwell, Third Edition, 2010
  3. 3. Exogenous pathway of lipoprotein metabolism From: Keith Frayn, “Metabolic Regulation - A Human Perspective”. Wiley-Blackwell, Third Edition, 2010
  4. 4. Endogenous pathway of lipoprotein metabolism From: Keith Frayn, “Metabolic Regulation - A Human Perspective”. Wiley-Blackwell, Third Edition, 2010
  5. 5. HDL metabolism From: Keith Frayn, “Metabolic Regulation - A Human Perspective”. Wiley-Blackwell, Third Edition, 2010
  6. 6. Forward and reverse cholesterol transport From: Keith Frayn, “Metabolic Regulation - A Human Perspective”. Wiley-Blackwell, Third Edition, 2010
  7. 7. Lipoprotein structure
  8. 8. Different lipoprotein classes From: Keith Frayn, “Metabolic Regulation - A Human Perspective”. Wiley-Blackwell, Third Edition, 2010
  9. 9. Lipoprotein Distribution (LPD) measurement
  10. 10. Cholesterol and heart disease From: Keith Frayn, “Metabolic Regulation - A Human Perspective”. Wiley-Blackwell, Third Edition, 2010
  11. 11. Stable isotope studies: lipoprotein classes correspond to kinetically different pools Multicompartmental model for apoB metabolism in VLDL1 (Sf 60400), VLDL2 (Sf 20-60), IDL (Sf 1220), and LDL (Sf 0-12). Tracer data time range 0-250h !!! Gaw A et al. 1996 Arterioscler Thromb Vasc Biol 16:236-249
  12. 12. Our vision … LPD kinetic model development Model-based LPD analysis Lipoprotein flux ratio biomarkers
  13. 13. Advantage of our concept From one single plasma measurement to rates of processes Process rates are closer to functional activity than concentrations!  better resolution to pick up risk-associated variation in lipoprotein metabolism ?  New cardiovascular risk markers?  New diagnostic? Otherwise, determination of process rates is only possible with costly stable-isotope studies
  14. 14. Particle Profiler project overview
  15. 15. Background of model Model development => Particle size-dependent rate constants => fluxes of different lipoprotein processes: - Hepatic production Peripheral lipolysis Liver attachment (ApoE- and ApoB-mediated) Hepatic uptake (annihilation of particle) Hepatic lipolysis Applied to each e.g. 0.1 nm subclass in the range 100 – 10 nm
  16. 16. Lipolysis cascades - 1 Model development Best fit overall
  17. 17. Lipolysis cascades - 2 The ApoB LPD size range is divided into 11 cascade fractions Each cascade has e.g. 1000 particle subfractions (pools). Model development Each subsequent cascade has smaller size ranges of itself and of all its subfractions Each lipolysis step transfers particles from a given cascade to the next cascade The other processes can add or remove particles from each cascade For every subfraction a particle mass balance is set up The computer solves the 11000 mass balances for steady state simultaneously The result is a simulated LPD whose appearance depends on the process parameter settings The simulated LPD is compared to a measured LPD A parameter optimization algorithm searches the best parameter settings
  18. 18. Lipolysis steps - decreasing particle size apoB apoE uptake uptake HL - lipolysis LPL - lipolysis Production Liver attachment Model development
  19. 19. Lipoprotein production process model Production flux (particles / min) Model development VLDL1 VLDL2 IDL Lipoprotein diameter (nm) LDL
  20. 20. Lipolysis and uptake process models (single particle rates) 0.02 process rate (1 / min) Model development Each subfraction particle P flux component F is modeled as F = k(r).P with k(r) explicitly depending on particle subfraction diameter according to a 1- or 2- paramer function liver attachment extrahep. lipolysis uptake hepatic lipolysis 0.015 0.01 0.005 0 100 80 60 40 lipoprotein diameter (nm) 20
  21. 21. Flux Data for model testing Packard et al. 2000, human study Model validation stable isotope fluxes analyzed with multi-compartment model Three groups of subjects, healthy men Large LDL peak size (>26 nm) Intermediate LDL peak size (25 – 26 nm) Small LDL peak size (<25 nm) – risk group Data on flux of lipoproteins input to our model.
  22. 22. Example output (main classes) compared to stable isotope study Particle Pools Uptake Flux 0.2 400 number of particles fl-1 min-1 Simulated Data 300 200 100 0 LDL Simulated Data 0.15 0.1 0.05 0 IDL VLDL2 VLDL1 lipoprotein class Lipolysis-Induced Influx IDL VLDL2 VLDL1 lipoprotein class Fitted Lipolysis and Uptake Rates Simulated Data 0.1 0.05 0 LDL 0.025 0.15 process rate min-1 N.B. The LPD is simulated as particle concentration profiles. These can be converted to TG and cholesterol concentration profiles number of particles fl-1 min-1 Model validation number of particles fl-1 500 LDL IDL VLDL2 lipoprotein class 0.02 Lipolysis rate Uptake rate 0.015 0.01 0.005 0 10 20 30 40 50 lipoprotein particle size 60
  23. 23. Flux Data for model testing 4 results that give credibility to our model Model validation Model can reproduce particle concentration & flux data The model qualitatively reproduces a separately measured LDL peak size shift The model can simulate genetic deficiencies The changed processes the model detected are biologically plausible Journal of Lipid Research, Vol. 50, 2398-2411, December 2009
  24. 24. concentration Model validation Lipoprotein Pooled Fluxes Lipoprotein Profile particle diameter Particle Profiler Lipoprotein Particle Fluxes VLDL1 VLDL2 IDL LDL Lipoprotein Metabolic Ratios Derive model-based markers
  25. 25. What are we doing? Derive model-based markers Derive ratios between processes For instance: VLDL lipolysis outside the liver / VLDL production See whether that helps to improve Risk prediction Better risk prediction will help to give the right therapy to the right people
  26. 26. Example new diagnostic for dyslipidemia Derive model-based markers
  27. 27. Receiver operating characteristic (ROC) curves Validate markers Diagnostic AUC 'VLDL performance' 0,937 'TG (mmol/l)' 0,900 'LDLc (mmol/l)' 0,794
  28. 28. Reclassification analysis You have three risk categories Validate markers People are treated based on their category Low risk – no treatment Medium risk – some treatment High risk – intensive treatment Reclassification analysis compares diagnostics for their classification ability For those people whom we know will NOT have an event How many move to lower risk categories? For those people whom we know WILL have an event How many move to higher risk categories?
  29. 29. Validation CVD risk prediction The Framingham Heart Study Offspring cohort (FOS) Validate markers Inclusion criteria: no history of cardiovascular disease gave written informed consent for general research use had complete NMR lipoprotein profiles recorded had a complete record of classical cardiovascular risk factors. Cardiovascular events 10 years after baseline measurement Population size Events (true positives) No event (true negatives) 1981 145 1836
  30. 30. Statistics Validate markers First using logistic regression No improvement possible Then SVM methodology
  31. 31. Best predicting ratios Validate markers In the VLDL size range VLDLH – VLDL Hepatic turnover indicator Liver functional status indicator VLDLE – VLDL Extrahepatic lipolysis indicator How well do extrahepatic tissues absorb fat?
  32. 32. Predictor variables Conventional markers LDLp LDLp + HDLp LDLp + HDLp + VLDLE + VLDLH Age Age Age Age Age Sex Sex Sex Sex Sex Cigarettes per day Cigarettes per day Cigarettes per day Cigarettes per day Cigarettes per day Blood pressure medication Blood pressure medication Blood pressure medication Blood pressure medication Blood pressure medication Systolic blood pressure (nurse) Systolic blood pressure (nurse) Systolic blood pressure (nurse) Systolic blood pressure (nurse) Systolic blood pressure (nurse) Glucose Glucose Glucose Glucose Glucose Total Cholesterol Validate markers Conventional markers without cholesterol LDL particle number LDL particle number LDL particle number HDL particle number HDL particle number HDL cholesterol VLDL Extrahepatic lipolysis indicator 32 VLDL Hepatic turnover indicator
  33. 33. Area under the ROC-curve statistics for general CVD Validate markers Framingham Offspring Study * Significantly better than conventional, no cholesterol p<0.05 ** Significantly better than LDLp, p<0.05 † Significantly better than LDLp+HDLp, p<0.05 Model AUC SE AUC improvement from random % incremental AUC improvement from random Conventional, no cholesterol 0.759 0.0204 0.259 0.0 Conventional 0.795 0.0193 0.295 (*) 12.2 LDLp 0.791 0.0192 0.291 (*) 11.0 LDLp + HDLp 0.797 0.0192 0.297 (*) 12.8 LDLp + HDLp + VLDLE + VLDLH 0.812 0.0192 0.312 (*,**, †) 17.0
  34. 34. Reclassification analysis Model 1 Model 2 NRI Standard Error P value % of Events correctly reclassified (n=145) Event Pvalue % of Nonevents correctly reclassified (n=1836) Nonevent P-value Conventional no cholesterol Conventional 0.1111 0.0418 0.0088 8% 0.0455 3% 0.002 Conventional no cholesterol LDLp 0.0830 0.0386 0.0327 6% 0.1441 3% 0.0021 Conventional LDLp + HDLp 0.0080 0.0339 0.8135 1% 0.8348 0% 0.8802 Conventional LDLp + HDLp + VLDLE + VLDLH 0.0902 0.0366 0.0143 5% 0.1779 4% <.0001 LDLp LDLp + HDLp 0.0356 0.0307 0.2489 3% 0.2513 0% 0.8751 LDLp LDLp + HDLp + VLDLE + VLDLH 0.1178 0.0376 0.0020 8% 0.0411 4% <.0001 LDLp + HDLp LDLp + HDLp + VLDLE + VLDLH 0.0828 0.0330 0.0127 4% 0.2008 4% <.0001 Validate markers
  35. 35. What is happening? An ‘average’ person. A - no blood pressure medication Validate markers B - no blood pressure medication 0.35 male female 0.25 10 year risk 10 year risk 0.3 0.4 0.2 0.15 0.1 male, low-med LDLp female, low-med LDLp male, med-high LDLp female, med-high LDLp 0.3 0.2 0.1 0.05 0 50 100 150 200 250 0 300 0 20 40 LDLp C - with blood pressure medication 100 0.25 male female male female 0.2 10 year risk 0.8 10 year risk 80 D - no blood pressure medication 1 0.6 0.4 0.2 0 -14 60 HDLp 0.15 0.1 0.05 -12 -10 -8 -6 VLDLE -4 -2 0 0 -6 -5 -4 -3 VLDLH -2 -1
  36. 36. Conclusions Lipoprotein metabolic ratios derived from Particle Profiler significantly improve CVD risk prediction As measured by area-under-the-ROC-curve 4% of the total population is positively reclassified by these markers In Framingham Offspring Cohort If we assume Framingham is a good representation of the US population (it is not, it is generally healthier) then: Given that approx. 32 million people in the US use statins, this could help more than 1 million people improve their treatment regime in the US alone. “ “
  37. 37. But… Clinicians & customers require an independent validation. Different cohort e.g. MESA (Multi Ethnic Study of Atherosclerosis) data would be needed MESA has slightly different data type (e.g. “Diabetes Y/N” instead of “plasma glucose”) Simulating this different data type in Framingham did not result in better risk prediction using Particle Profiler- derived flux ratio markers MESA was not analyzed and the project was stopped by TNO (we might have considered EPIC, but didn’t)
  38. 38. Lessons learned SVM and binary variables – take care! Introducing new diagnostics in the current clinical setting is a all-or-nothing business. Nothing is certain until the last statistical analyses come in, the final success is highly unpredictable. Because this is a high-risk enterprise, companies do not easily invest in early development stages, but are willing to collaborate with data sharing etc.
  39. 39. Towards linking carbohydrate and fat metabolism Lipoprotein metabolism as a central node for interplay of glucose, TG and NEFA metabolism VLDL Prod Chylo microns - VLDL LDL glucose
  40. 40. TNO can contribute datasets
  41. 41. Collaborators TNO Daan van Schalkwijk Evgeni Tsivtsivadze Bianca van der Werff Ben van Ommen Albert de Graaf (Andreas Freidig) LACDR Jan van der Greef TUFTS university Boston Laurence D. Parnell José M. Ordovás Financial support The official Framingham investigators were not involved in the analysis nor commented on the conclusions drawn

×