modeling a novel radiotracer: “a” checklist · fate of the tracer in the human body, in...
TRANSCRIPT
Modeling a novel radiotracer: “a” checklist
Ronald Boellaard
Department of Nuclear Medicine & PET Research VU University Medical Center, Amsterdam, The Netherlands
Email: [email protected]
Overview of presentation..
• Preclinical evaluations: a very short overview
• Modeling a novel tracer:– Introduction into models– A ‘checklist’– Checklist examples
Starting point
Target defined
Lead compound identified
Failed drug New molecule
Pre-PET evaluation
Workup of new molecule
• Affinity for target
Pharmacology
Ex vivo autoradiographyEx vivo autoradiography[[33H]R116301 (gerbil)H]R116301 (gerbil)
StrStr
OBOBLCLC
StrStr
CxCx
0.0025 0.01 0.04 0.16 0.63 2.5 10
0
25
50
75
100
dose (mg/kg)
NK
1 r
ecep
tor
occu
pan
cy
(% o
f co
ntr
ol)
CrbCrb
Ex vivo receptor binding Ex vivo receptor binding [[33H]subP (gerbil striatum) H]subP (gerbil striatum)
Pre-PET evaluation
Workup of new molecule
• Affinity for target
• Selectivity– Lack of affinity for other targets
• Amenable to radiolabelling with C-11 or F-18– C-11
• on-site cyclotron required• intervention studies possible
– F-18• good statistics• limited repeat studies
Pre-PET evaluation
Workup of new molecule
• Affinity for target
• Selectivity
• Amenable to radiolabelling with C-11 or F-18
• Substrate for P-gp?
• Toxicity– Tracer alone: acute toxicity– Pharmacological dose: full blown toxicity– C-11 versus F-18 labelling– Better after preclinical evaluation?
Preclinical evaluationEx vivo biodistribution in rats
• Tracer alone– Sufficient uptake in tissue (e.g. VT > 1)– Biodistribution comparable with in vitro data– Specific to non-specific signal (> 2)– Identification of labelled metabolites
• Blocking studies– Specificity and selectivity in vivo
• Animal model of disease– Proof of concept– Genetically modified animals (mice)
• Radiation dosimetry!
Preclinical evaluation
PET studies in rats
• Same issues as for biodistribution studies
Wild type mouseMDR1a(+/+)/1b(+/+)
Double gene knock-out mouseMDR1a( -/-)/1b(-/-)
Pgp function: [11C]verapamil – modulation of BBB
Preclinical evaluation
PET studies in rats
• Same issues as for biodistribution studies
• Assessment of kinetics– Reversible versus irreversible– Time required to reach transient equilibrium
• Intervention studies– Combination with in vivo microdialysis
• Kinetics of labelled metabolites– Penetration of blood-brain barrier– Synthesise labelled metabolites?
Modeling a novel radiotracer: introduction into models…..
Clinical evaluationGMP cleanroom required
Overview of presentation..
• Preclinical evaluations: a very short overview
• Modeling a novel tracer:– Introduction into models– A ‘checklist’– Checklist examples
Clinical evaluation
Tracer alone studies
• Signal?
• Distribution
• Kinetic modeling
• Specific to non-specific ratio
• Level of non-specific binding
• Labelled metabolites
• Quality input function– Stickiness tracer– Size and accuracy parent fraction
Tracer Kinetic Modeling: clinical evaluation
Tracer Model:
Purpose:
Method:
Mathematical description of thefate of the tracer in the humanbody, in particular in the organunder study
To quantify functional entitiesgiven the distribution ofradioactivity (over time)
Divide possible distribution oftracer in a limited number ofdiscrete compartments
PET pharmacokinetic modeling
C’free’K1
k2
Cplasma
Cboundk3
k4
CPET
0
50
100
150
200
250
0 2 4 6 8 10
Tijd (min)
Blo
od a
ctiv
ity (k
Bq/
cc) 1022 keV
511 keV
manual samples
Input function
PET, TAC
Model
K1,…VT, BP
Examples of PET pharmacokinetic models
plasma input single tissue compartment model (1T-2k)
Ctissue=f+sb+nsbK1=EF
k2
Cplasma CPET
Requires-Dynamic PET scan (TAC)-Metabolite corrected plasma input function
PET pharmacokineticparameter:VT=K1/k2
C’free’
K1
k2
Cplasma
Cbound
k3
k4
CPET
Examples of PET pharmacokinetic modelsplasma input two tissue compartment model (2T-4k)
Requires-Dynamic PET scan (TAC)-Metabolite corrected plasma input function
VT=K1/k2*[1+BP]BPnd=k3/k4
Examples of PET pharmacokinetic modelsreference tissue model
C’f+ns
Cf+ns+sp
Cp
K1
k2
K’1
k’2
Target
Reference
Requires-Dynamic PET scan (TAC)-Definition of a reference region or reference TAC
Simplified methods
Model simplifications/assumptions– Logan (plasma and/or reference input)– Patlak
Model linearisation– Blomquist– Ichise (MRTM0-4)
Model basis function implementation – RPM– Spectral analysis/Basis pursuit/RS-ESA
Uptake & uptake ratio’s– SUV (standardised uptake value)– SUVr, ratio of act.conc. target/reference regions
Logan plot
Simplified methods
Why using simplified methods ?– Clinical applicability, ease of (practical) use– Computational speed– Reduce sensitivity to noise (less fit parameters)– Possibility to generate parametric images
Purpose of simplified method evaluations:Look for method that is most suitable for a given (clinical) task
Potential limitations of simplified methods
Simplified methods often do not provide information on microparameters
Model simplification & linearisationNoise induced bias
Basis function methodSelection of basis range = bias versus precision trade-off
Uptake/uptake ratioDoes not provide a pharmacokinetic parameterEffects of global and regional flow differences
Overview of presentation..
• Preclinical evaluations: a very short overview
• Modeling a novel tracer:– Introduction into models– A ‘checklist’……..finally !– Checklist examples
Modeling a novel tracer : example of a recipe
1. Start with dynamic PET studies to measure kinetics• Duration 60, 90, 120, 150 min ?• Frame durations should match kinetics, start with short frames (<5s) !• Include arterial blood sampling with metabolite analysis
2. Generate regional time activity curves (TAC)• VOI versus cluster analysis• Heterogeneity within VOI
3. Determine optimal plasma input model• Processing of plasma input: metabolite analysis weakest link !• Approaches:
• Data driven approach: try models with increasing complexity• Hypothesis drives approach: define (complex) model based on
existing (preclinical) data and then simplify.
Data driven approach:
e.g. start analyzing VOI TAC using plasma input models
• One tissue compartment model– 2 rate constants + Vb (blood volume)
• Irreversible two tissue comparment model– 3 rate constants + Vb
• Reversible two tissue comparment model– 4 rate constants + Vb
Modeling a novel tracer : example of a recipe
Modeling a novel tracer : example of a recipe
1. Start with dynamic PET studies to measure kinetics• Include arterial blood sampling with metabolite analysis
2. Generate regional time activity curves (TAC)
3. Determine optimal plasma input model• Processing of plasma input: metabolite analysis weakest link !
4. Validate reference tissue models and reference regionsagainst plasma input model/results
5. Validate (use of) simplified methods
6. Validate parametric methods
Measures used for model evaluation
Fit accuracy – how well does model fit thru PET data
Accuracy / bias – how accurate are observed parameters
Precision - reproducibility / test-retest variability
Correlation with other measures (MMSE, age, gender etc)
Discriminating ability
- individual (important when tracer is intended for diagnostic purposes)
- group level (tracer ‘only’ useful in research setting)
Model development
Multiple data sets required/optimal• Baseline data
– Does it enter brain ?– Expected distribution OK ?
• Blocking or displacement data– Ideally, data over entire range of occupancies– Same ligand or other specific & selective ligand (pref)– Specificity– Reference regions (‘free of displacement’)
• Test-retest data• Patient and healthy subject data
– collect demographic, blood, tissue, sample or other data
• Simulation studies
Overview of presentation..
• Preclinical evaluations: a very short overview
• Modeling a novel tracer:– Introduction into models– A ‘checklist’– Checklist examples
Model evaluation
processing of measured arterial input function
min
Whole blood activity
Plasma PK11195 activity
Model evaluation
processing of measured arterial input functionEffects of using different fitting routines for plasma/whole blood and/or metabolite correction of input function on observed Vd and BP in case of missing data or ‘outliers’ (Lubberink et al., abstract NRM, 2004)
VT, missing data BP, outliers
Fit accuracy – ‘incorrect’ model selection by
parent fraction fitting
NB1: ‘suboptimal’ plasma input processing may affect model selection and
results !
NB2: metabolite may enter brain (may need to use parent + metabolite as input)
0.00
10.00
20.00
30.00
40.00
50.00
60.00
70.00
80.00
AIC 1T2k SCH 1T2k AIC 2T4k SCH 2T4k AIC 2T3k SCH 2T3k
Strategies/measures used for model evaluation
Accuracy of fit
- WRSE=(weighted) residual square error=how well does fit go thru data
- AIC = Akaike Information Criterion = which model fits best using least number of fit parameters
- Number of outliers = how frequent does a model provide results outside physiological realistic range
Accuracy / bias
Precision / reproducibility / test-retest variability
Discriminating ability
- individual
- group level
Evaluation of model based on ‘fit accuracy’ using WRSE & AIC
Example for PK11195 (Kropholler et al. JCBFM 2005)
0
20
40
60
80
100
0-0.025 ml 0.025-3 ml >3 ml
ROI gray matter size (ml)
Akai
ke p
refe
rred
mod
el (%
)
1TC2TC-3k2TC-4k
‘Standard’ models show a preference for the 2TC-4k model for all ROI sizes. Reducing ROI size (=increasing noise) increases preference for models with lower number of parameters.
“Fit accuracy” - comparison of SRTM with FRTM - PIB
Chart Titley = 0.9708xR2 = 0.8544
0
0.2
0.4
0.6
0.8
1
1.2
1.4
0 0.2 0.4 0.6 0.8 1 1.2 1.4BP STRM
BP
FRTM
FRTM BP
Linear (FRTM BP)
0
10
20
30
40
50
60
70
80
AIC SRTM SCH SRTM AIC FRTM SCH FRTM
Pref
eran
ce (%
)
0
5
10
15
20
25
30
35
SRTM BP FRTM BP
Out
liers
(%)
-AIC indicate preference for FRTM-SRTM less outliers-SRTM correlates well with FRTM-SRTM is preferred model
Measures used for model evaluation
Fit accuracy – how well does model fit thru PET data
Accuracy / bias – how accurate are observed parameters
Precision - reproducibility / test-retest variability
Correlation with other measures (MMSE, age, gender etc)
Discriminating ability
- individual
- group level
Evaluation of model accuracy – clinical datavalidation of reference tissue model
Correlation with ‘gold standard’ (may be experimental data as well)
PK11195, BPSRTM versus DVR2T-4k -1
y = 0.9973x + 0.0125R2 = 0.8373
-0.1
-0.05
0
0.05
0.1
0.15
0.2
-0.1 -0.05 0 0.05 0.1 0.15 0.2
BP SRTM
DV
R-1
, pla
sma
inpu
t
Kropholler et al., JCBFM 2007, Schuitemaker et al., JCBFM, 2007
Correlation between DVR-1 and BP-SRTM improves from 0.004 to
0.62 when using metabolite input corrected kinetic models
Effects of metabolites on DVR-1 and BP-SRTM
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
-0.2 -0.1 0 0.1 0.2 0.3 0.4
BP STRM
DVR
-1
PMfvdm2T4k
Measures used for model evaluation
Fit accuracy – how well does model fit thru PET data
Accuracy / bias – how accurate are observed parameters
Precision - reproducibility / test-retest variability
Correlation with other measures (MMSE, age, gender etc)
Discriminating ability
- individual
- group level
Example of test-retest variability - PIB
Subjects: 6 AD patients and 3 age matched healthy controls
Repeat dynamic scans (23 frames; 90 minutes total) in 3D
acquisition mode, following bolus injection of 370 MBq [11C]PIB.
Test and retest scans were performed on the same day.
Arterial whole blood sampling together with plasma parent tracer and
metabolite concentrations measurements
Example of test-retest variability - PIB
• T-RT depends on VOI size (noise)• SRTM better TRT than PI 2T-4k• SRTM better TRT for AD than for control subjects
Tolboom et al. JNM 2008
Assessment of accuracy and precision with
simulations
Why using simulations ?– Simulations are useful when ‘gold standard’ is not available or when parameters
cannot be changed in a controlled way (patients as opposed to animal studies)– Simulations are useful to study e.g. effect of noise level, flow and flow differences,
blood volume fraction etc on bias and precision (‘sensitivity analysis’)
Simulation study design / setup– Generate TAC using a typical input function and PET pharmacokinetic parameters– Make 10000 noisy realisations– Fit each ‘noisy’ TAC with model or simplified method
(noise model, weighting factors, optim.alg. Yaqub et al.PMB 2006)– Bias = ratio or difference between average (or median) observed parameter value
over simulated ‘true’ valuePrecision = SD or COV of observed parameter value
Evaluation of parametric methods – PK11195
simulations
0
0.08
0.16
0.24
0.32
0.4
0 10 20 30
Noise (%)
BP
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0 10 20 30
Noise (%)
SD
B C D
E F G
Logan (DVR-1) ICH1 ICH2
R.Logan RPM1 RPM2 (*)
Schuitemaker et al. JCBFM 2007
RefLogan
ICH1
ICH2
RPM1
RPM2
DVR-1
BP-SRTM
Parametric analysis & sims using PPET. Boellaard et al. NRM2006
0
0.04
0.08
0.12
0.16
0.2
0 10 20 30
Noise (%)
BP
Evaluation of parametric methods – PK11195
simulations – effect of regional flow difference on SRTM and ref.par.methods
Schuitemaker et al.JCBFM 2007
0
0.05
0.1
0.15
0.2
0.25
0.3
0 10 20 30
Noise (%)
BP
RefLogan
ICH1
ICH2
RPM1
RPM2
DVR-1
BP-SRTM
“Normal flow” ‘Reduced’ flow
Measures used for model evaluation
Fit accuracy – how well does model fit thru PET data
Accuracy / bias – how accurate are observed parameters
Precision - reproducibility / test-retest variability
Correlation with other measures (MMSE, age, gender etc)
Discriminating ability
- individual (important when tracer is intended for diagnostic purposes)
- group level (tracer ‘only’ useful in research setting)
[11C]PIB versus [18F]FDDNP
AD 1
PIBFDDNP
AD 2
PIBFDDNP
CONTROL
PIBFDDNP
Receptor Parametric Mapping; cerebellum = reference
Courtesy of B.N.M. van Berckel and N. Tolboom
Measures used for model evaluation
Fit accuracy – how well does model fit thru PET data
Accuracy / bias – how accurate are observed parameters
Precision - reproducibility / test-retest variability
Correlation with other measures (MMSE, age, gender etc)
Discriminating ability
- individual
- group level
Discrimination between subject groups
FP-B-CIT
Parametric methods evaluation
discrimination ability
< 0.001NSLeft frontal
0.043NSRight frontal
0.002NSLeft lateraltemporal lobe
< 0.001NSRight lateral temporal lobe
< 0.0010.1Left thalamus
< 0.0010.1Right thalamus
BPVdRegion
< 0.001NSLeft frontal
0.043NSRight frontal
0.002NSLeft lateraltemporal lobe
< 0.001NSRight lateral temporal lobe
< 0.0010.1Left thalamus
< 0.0010.1Right thalamus
BPVdRegion
Effects of plasma versus reference input parametric kinetic methods on SPM analysis, young versus old & AD (Schuitemaker et al., NeuroImage 2007.)
Region/p-values
Modeling a novel tracer - summary
1. Start with dynamic PET studies to measure kinetics• Include arterial blood sampling with metabolite analysis
2. Generate regional time activity curves (TAC)
3. Determine optimal plasma input model• Processing of plasma input: metabolite analysis weakest link !
4. Validate reference tissue models and reference regionsagainst plasma input model/results
5. Validate (use of) simplified methods
6. Validate parametric methods
7. Use simulations to perform ‘sensitivity’ analysis for allmodels and methods considered
Model development –requires a lot of data collection/studiesMultiple data sets required/optimal• Baseline data• Blocking or displacement data
– Ideally, data over entire range of occupancies– Same ligand or other specific & selective ligand (pref)– Specificity– Reference regions (‘free of displacement’)
• Test-retest data• Patient and healthy subject data
– collect demographic, blood, tissue, sample or other(imaging) data
• Simulation studies
Acknowledgements
VUMC, Amsterdam, NLAdriaan LammertsmaMaqsood YaqubMark LubberinkBart van BerckelOtto HoekstraNelleke TolboomMarc KrophollerUrsula KlumpersBert WindhorstGert LuurtsemaAlie Schuitemaker