Pediatric Pharmacology
Population Pharmacokinetics of Nusinersen
in the Cerebral Spinal Fluid and Plasma of
Pediatric Patients With Spinal Muscular
Atrophy Following Intrathecal
Administrations
The Journal of Clinical Pharmacology
2017, 00(0) 1–11
C 2017, The American College of
Clinical Pharmacology
DOI: 10.1002/jcph.884
Kenneth T. Luu, PhD1 , Daniel A. Norris, PhD1 , Rudy Gunawan, PhD1 ,
Scott Henry, PhD2 , Richard Geary, PhD3 , and Yanfeng Wang, PhD1
Abstract
Nusinersen is an antisense oligonucleotide intended for the treatment of spinal muscular atrophy. The pharmacokinetics of nusinersen, following
intrathecal administrations, in the cerebrospinal fluid (CSF) and plasma of 72 pediatric patients (3 months to 17 years) with spinal muscular atrophy
across 5 clinical trials was analyzed via population-based modeling. With sparse data in the CSF and profile data in the plasma, a linear 4-compartment
model simultaneously described the time-concentration profiles in both matrices. The typical population parameters were: Qp = 0.572 L/h, QCSF =
0.069 L/h, CLp = 2.50 L/h, CLCSF = 0.133 L/hr, VCSF = 0.441 L, Vp = 32.0 L, Vsystemic_tissue = 429 L, and VCNS_tissue = 258 L. A full covariate modeling
approach identified baseline body weight to be a statistically and clinically relevant covariate on VCSF , Vp , and CLp . The model predicted that the
CSF volume of distribution increased steadily with age from 0 to 2 years but became relatively steady for children >2 years. Simulations from the
final model showed that age-based dosing in children under 2 years ensured a more comparable exposure (peak concentration and area under the
concentration-time curve) across subjects in the population relative to a fixed dosing scheme. However, because no dose-limiting toxicity has been
reported in any of the trials, a fixed-dose scheme (12 mg across all age groups) was recommended. The median terminal half-life of nusinersen in the
CSF was determined from the model to be 163 days, which supported infrequent dosing, once every 4 to 6 months in pediatric patients with spinal
muscular atrophy.
Keywords
nusinersen, cerebral spinal fluid, intrathecal administration, spinal muscular atrophy, Spinraza, population pharmacokinetics
Spinal muscular atrophy (SMA) is an autosomal recessive neuromuscular disease characterized by degeneration of the motor neurons in the anterior horn of
the spinal cord, resulting in atrophy of the voluntary
muscles of the limbs and trunk. SMA is a debilitating
disease, rendering patients unable to walk, eat, or
breathe in its most severe form. SMA has an incidence
of 1:6000 to 1:10,000 live births; it is the most common
monogenetic cause of infant mortality and a major
cause of childhood morbidity due to weakness.1 SMA
is caused by loss of survival motor neuron (SMN) protein due to a homozygous recessive deletion or mutation
in the SMN1 gene on chromosome 5q11-q13.2 Humans
have a duplication of the chromosome region where
SMN1 is found, resulting in a nearly identical copy
of the gene, SMN2. In SMN2, a C-to-T nucleotide
substitution in exon 7 results in an alternative splicing
event such that the majority of transcripts produced
(90%) lack exon 7, and a defective truncated protein
(often referred to as 7) is produced.3 Low levels of
SMN protein result in the degeneration of spinal motor
neurons and cause muscle weakness that is followed
by symmetric limb paralysis, respiratory failure, and
death.1,4,5
Nusinersen is a uniformly modified 2 -O-(2-methoxyethyl) antisense oligonucleotide in development
for SMA. These so-called “uniformly modified”
antisense oligonucleotides can interfere with RNA
function and metabolism (e.g., translation into protein,
splicing, polyadenylation).6 Nusinersen corrects a
splicing defect in the SMN2 pre-mRNA, rather
than to promote degradation of a targeted mRNA
as is a typical case for antisense oligonucleotides. It is
designed to bind to a specific sequence in the intron
1 Pharmacokinetics
and Clinical Pharmacology, Ionis Pharmaceuticals,
Carlsbad, CA, USA
2 Nonclinical Development, Ionis Pharmaceuticals, Carlsbad, CA, USA
3 Clinical Development, Ionis Pharmaceuticals, Carlsbad, CA, USA
Submitted for publication 25 October 2016; accepted 31 January 2017.
Corresponding Author:
Kenneth Luu, PhD, Ionis Pharmaceuticals, 2855 Gazelle Ct, Carlsbad, CA
92010
Email: kenneth.t.luu@gmail.com
2
downstream of exon 7 of the SMN2 pre-mRNA
transcript, thus promoting the increase in inclusion
of exon 7 in SMN2 mRNA.7 The result is increased
production of full-length SMN protein, equivalent to
SMN protein produced by the SMN1 gene. Because the
number of SMN2 gene copies and resulting amount of
SMN protein is correlated with disease severity and age
of onset, a therapeutic approach predicted to benefit
SMA patients is to increase the levels of full-length
SMN2 pre-mRNA by restoring the splicing pattern
that gives rise to full-length SMN2 mRNA.
The clinical experience with nusinersen includes clinical trials in infants with genetically diagnosed and
presymptomatic SMA, infants with “infantile-onset
SMA” (most likely to develop type I SMA), and in
children and adolescents with “later-onset SMA” (most
likely to develop type II or type III SMA). The clinical
pharmacokinetics (PK) of nusinersen has been evaluated in a number of trials (see Methods), measuring
drug levels in both the cerebrospinal fluid (CSF) and
the plasma. Unlike in the plasma, however, where serial
sampling was performed, the sparse sampling in the
CSF did not allow for the calculation of basic PK
parameters in the CSF compartment such as volume of
distribution and clearance with reasonable confidence.
The objective of this work was to develop a population
PK model that simultaneously describes the PK of
nusinersen in the CSF and plasma, with the goal
of generating interpretable PK parameters—and their
interindividual variability—in both matrices.
Methods
Each of the above studies was approved by the respective local institution review board. Written informed
consent and assent (if applicable) were obtained before
any evaluations were conducted for eligibility. These
trials were conducted in compliance with the Declaration of Helsinki, the International Conference on
Harmonisation Good Clinical Practice guidelines, and
local regulatory requirements. The study designs for the
above studies are listed below (specific blood and CSF
PK sampling times for each study are shown in Table
S1 of the supplemental section).
Study Design
Data from 5 clinical studies were pooled for analysis
(Protocols ISIS 396443-CS1, ISIS 396443-CS2, ISIS
396443-CS3A, ISIS 396443-CS10, and ISIS 396443CS12). In these 5 studies limited PK samples were
obtained in the CSF, and serial PK samples were
obtained in the blood of SMA patients.
Study ISIS 396443-CS1 was an open-label singledose safety, tolerability, and dose-range-finding clinical
trial (n = 24) in patients aged 2 to 14 years with SMA. A
The Journal of Clinical Pharmacology / Vol 00 No 0 2017
brief PK summary from this study has been reported by
Chiriboga et al.8 Patients were randomized at baseline
to 1 of 4 independent treatment arms and received
a single intrathecal dose of 1, 3, 6, or 9 mg. CSF
samples were collected predose, at dose administration,
and 7 days after dosing in the 1-, 3-, and 6-mg dose
cohorts. In the 9-mg dose cohort, CSF samples were
collected postdose on day 8 or day 29 (5 patients at
each time point). Serial blood samples were collected on
day 1 with additional blood samples collected on days
2, 8, and 29. Patients randomized to cohort 1 (1 mg)
and completing study procedures were eligible to enroll
in study ISIS 396443-CS2, and patients randomized to
cohorts 2 to 4 (3, 6, and 9 mg) and completing study
procedures were eligible to enroll in study ISIS 396443CS10.
Study ISIS 396443-CS2 was an open-label multipledose safety, tolerability, and dose-range-finding clinical
trial (n = 32) in patients aged 2 to 14 years with
SMA. Patients were randomized at baseline to 1 of
4 independent treatment arms and received 2 (9-mg
cohort only) or 3 intrathecal doses of 3, 6, 9, or 12 mg
on days 1, 29 (3-, 6-, and 12-mg cohorts only), and
85. Six patients who participated in study ISIS 396443CS1 were eligible for and enrolled in study ISIS 396443CS2; 3 in cohort 1 (3 mg), and 3 in cohort 2 (6 mg).
CSF samples were collected predose before each dose
administration. Intensive blood profile samples were
collected on days 1, 29 (3-, 6-, and 12-mg cohorts only),
and 85 with additional blood samples collected on days
2, 8, 36, 86, and 92.
Study ISIS 396443-CS10 was an open-label singledose safety and tolerability clinical trial (n = 16) in
patients aged 2 to 14 years with SMA. Patients were
randomized at baseline to 1 of 2 independent treatment
arms and received a single intrathecal dose of 6 or
9 mg on day 1. All patients who enrolled in study ISIS
396443-CS10 participated in study ISIS 396443-CS1.
CSF samples were collected predose on day 1. Blood
samples were collected on days 1 (limited profile), 8, 85,
and 169.
Study ISIS 396443-CS3A is an ongoing open-label
multiple-dose safety, tolerability, and PK clinical trial
(n = 20) in patients aged >21 days to 7 months at the
time of screening with infantile-onset SMA. Patients
were randomized at baseline to 1 of 2 independent
treatment arms and received 3 intrathecal doses of 6 or
12 mg during a “loading” dosing phase on days 1, 15,
and 85. All patients received multiple 12-mg intrathecal
doses every 4 months during a “maintenance” dosing
phase beginning on day 253. CSF samples were collected predose before each dose administration. Blood
samples were collected on days 1 (limited profile),
2, 15, 85 (predose and 4 hours postdose), and 92,
followed by predose collections prior to each dose
3
Luu et al
during the “maintenance” dosing phase beginning on
day 253.
Study ISIS 396443-CS12 is an ongoing open-label
multiple-dose safety, tolerability, and PK clinical trial
(n = 43) in patients aged 2 to 14 years with SMA.
Patients received multiple intrathecal doses of 12 mg
on days 1, 169, 351, and 533. All patients who enrolled
in and completed study ISIS 396443-CS12 participated
in either study ISIS 396443-CS2 or ISIS 396443-CS10.
CSF samples were collected predose on all dosing days.
Blood samples were collected on days 1 (limited profile),
85, 169, 351, and 533.
Dose Scheme
The intended age-based dose scheme applied in the
later nusinersen trials were determined based on the
age-to-CSF volume relationship equation reported by
Matsuzawa et al.9 In this reference, Matsuzawa et al
performed 3-dimensional magnetic resonance imaging
in healthy children aged 1 month to 10 years and
examined the CSF along with other brain regions. The
CSF volume data from this study were best fitted to a
model in which CSF volume (mL) = 149.88 + 13.90 ×
[ln(age [months]/100) + 0.7596]. From this equation,
the applied age-based dose schemes were determined,
with each dose intending to provide consistent exposure
across the age groups: 9.6 mg (0-3 months), 10.3 mg
(3-6 months), 10.8 mg (6-12 months), 11.3 mg (1224 months), and 12 mg (>24 months). Based on this
scheme, infants less than 2 years received an age-based
dosing, as listed, and children >2 years received a fixed
12-mg dose. Of note is that this dosing scheme was not
made based on PK or PK/PD results and, thus, was
considered purely experimental. Dose escalation during
the development of this program was based primarily
on safety monitoring starting at 1 mg in the first trial.
However, the escalations were made with the desire to
achieve the preclinical efficacious CSF concentrations
seen in monkeys.
Bioanalytical Assays
A noncompetitive hybridization nuclease-based
enzyme-linked immunosorbent assay (ELISA) method
or an electrochemiluminescence (ECL) method was
used to quantify intact nusinersen in human CSF and
plasma. The ELISA method was used to quantify
intact nusinersen in CSF and plasma for study ISIS
396443-CS1 only. The ECL method was used to
quantify intact nusinersen in CSF and plasma for
studies ISIS 396443-CS2, ISIS 396443-CS10, ISIS
396443-CS3A, and ISIS 396443-CS12. The sensitivities
of the ELISA (1.5 ng/mL) and ECL (0.05 ng/mL)
assays were sufficient to characterize the disposition
of nusinersen well below levels associated with
pharmacologic activity. For the ELISA method,
the quantitation range for a sample is 1.5 ng/mL
to 150 ng/mL from lower limit of quantification to
upper limit of quantification, respectively. For the
ECL method, the quantitation range for a sample is
0.05000 ng/mL to 10.00 ng/mL (from lower to upper
limit of quantification). Samples with concentrations
previously observed or expected to be above the
upper limit of quantification were diluted within the
range using a suitable volume of blank matrix prior
to analysis. Calibration standards were within 25%
of the theoretical concentration at the lower limit
of quantification and within 20% of the theoretical
concentration for all other levels, except for the anchor
point, which had no precision criteria. The criteria for
quality controls were within 20% of the theoretical
value if the precision of each replicate was within 20%.
During validation of either assay, the intra-assay and
interassay precisions were 20%, and accuracies were
within ±20% of the theoretical value for each QC
level. Cross-validation of the 2 methods confirmed
suitability of transition from ELISA to ECL. Analyses
for CSF and plasma concentrations using ELISA and
ECL were conducted with fully validated methods,
and samples were analyzed under United States Food
and Drug Administration Good Laboratory Practice
guidelines.
Modeling Methodology
NONMEM 7.2 (Icon, Ellicott City, Maryland) was
used for all model estimation. First-order conditional
estimation interaction was used for the base model as
well as the final model. Perl speaks NONMEM (PsN;
Uppsala Universitet, Uppsala, Sweden) 4.2.0, Xpose
4.0.4, R 3.3.1 (with the installation and implementation of the following library packages: ggplot2, lattice,
gridExtra, GGally, and plyr) were used for postprocessing and plotting results. The prediction-corrected visual
predictive check (pcVPC) with parameter uncertainty
was conducted using PsN.10
The intersubject variability in the structural PK parameters was modeled using multiplicative exponential
random effects of the form:
θ i = θ · eηi
(1)
where θ is the typical individual (population mean)
value of the PK parameter and ηi denotes the interindividual variability (IIV) accounting for the i-th
individual’s deviation from the typical value having 0
mean and variance ω2 .
Interoccasion variability (IOV) was deemed to be
important, especially because of the intrathecal route
of administration, sampling procedure, and infrequent
dosing intervals (4-6 months), and was also evaluated as an additional level of random effect on
4
The Journal of Clinical Pharmacology / Vol 00 No 0 2017
CSF clearance. The relationship was expressed as
follows:
Pi = TVPex p (I O V )
(2)
where
I O V = κ1i ·OCC1 +κ2i ·OCC2 + . . . + κni ·OCCn
(3)
Pi is the parameter of interest, TVP is the typical
value of the parameter, OCCn (occasion) has the value
of 1 for the nth occasion, otherwise 0, for each individual. κ 1 . . . n -values are random variables assumed
to be normally distributed around 0 with identical
variance denoted by π 2 P .11 Four occasions were defined
in this analysis: each dose event was assigned a unique
occasion starting with 1 for the first dose, 2 for the
second dose, and 3 for the third dose. For the fourth
and additional doses, the occasions were assigned to 4.
Goodness of fit was determined based on diagnostic
plots and precision of the parameter estimates. Hypothesis testing to discriminate among alternative hierarchic
structural models was performed using the likelihood
ratio test. When alternative models were compared,
the difference in the NONMEM objective function
was approximately χ 2 distributed with n degrees of
freedom, where n was the difference in number of
parameters between the hierarchic models. A decrease
of 6.64 in the value of the objective function was
considered significant under the likelihood ratio test
(n = 1, P < .01).
The full covariate model approach was used.12 As
opposed to stepwise covariate modeling, the full covariate model approach emphasized parameter estimation
rather than stepwise hypothesis testing.12 This method
initially built a stable and structurally sound base model
using the standard goodness-of-fit diagnostics. Then,
extensive graphical evaluation was utilized to assess
the correlation between each covariate and each η
as well as evaluating potential relationship between
each covariate and each model parameter. The final
model was then constructed, and point estimates and
parameter precision were evaluated for all base model
parameters and covariate model parameters.
The performance of the final model was evaluated
by conducting a visual predictive check (VPC). Simulations were performed using the subjects’ characteristics and the dosing and sampling history from the
original data set using PsN. From these simulations,
concentration-time data were summarized using median and 2.5th and 97.5th percentiles. Because the
diagnostic value of a VPC can be hampered by binning
across a large variability in dose and/or influential
covariates, the pcVPC was used. The pcVPC offers a
solution to these problems while retaining the visual
interpretation of the traditional VPC.10 In a pcVPC, the
variability originating from binning across independent
variables is eliminated by normalizing the observed and
simulated dependent variable based on the typical population prediction for the median independent variable
in the bin.
Because the observed data from the CSF were
very sparse, limiting the ability to calculate the CSF
terminal half-life for the individual patients in the
dataset, the approach was taken to calculate this parameter based on the model-derived individual post
hoc estimates of the microconstants. The CSF portion of the structural model was set up to be identical to a traditional 2-compartment model, and the
CSF β half-life was calculated based on the following
formula13 :
Beta H L =
ln2
0.5(k12 +k21 +k10 − (k12 +k21 +k10 )2 − 4k21 k10 )
(4)
Simulations
Population simulations were performed using the final
model to test if the age-based dose scheme (see Dose
Scheme section) implemented in the trials was valid. In
these simulations a data set of 1000 virtual subjects with
normal distribution in age ranging from 0.1 month to
17 years with a mean age of 5 years (these statistics
were based on the modeled data set) was created;
each subject was randomly assigned to an age value
based on this distribution. SMA patients tend to have
low body mass index despite increased fat mass and
overweight.14 Thus, SMA patients are not expected to
have normal age-weight correlations, and the Center
for Disease Control age-weight chart was not used.
The body weight was then assigned to the subjects
according to their assigned age using the age–body
weight relationship determined from the original data
set. This age–body weight relationship was modeled
by a piecewise relationship (pivoted among 0, 1, and
6 years of age). The fitting was implemented using the
R function: lm(formula=BWTbs(AGE, df=NULL,
knots=c(0,1,6)), data=d1) under the “splines” library
package. Once the age–body weight relationship was
determined, the body weight was calculated based
on the subject’s assigned age. The simulation outputs were exported to WINNONLIN for determining area under the concentration-time curve (AUCinf )
and peak concentration (Cmax ). Box plots of AUCinf
and Cmax were presented by age (with intersubject
variability and residual variability included but not
IOV).
5
Luu et al
Table 1. Subject Demographic Based on the Pooled Data Set Used in the Analysis
Number of
Subjects (N)
Age, y, Median
(Range)
Male
Female
Weight, kg, Median
(Range)
All
72a
5 (0.10 to 17)
37
35
15.2 (5.1 to 83)
CS1
24
6 (2 to 12)
9
15
18.6 (10.3 to 52.1)
CS2
32
5 (2 to 15)
19
13
21.5 (10.8 to 83.1)
CS10
16
6 (2 to 11)
5
11
18.4 (10.3 to 52.1)
CS3A
20
0.425 (0.10 to 0.57)
12
8
6.58 (5.07 to 9.25)
CS12
43
7 (3 to 17)
22
21
16.9 (10.3 to 83.1)
Study
Race
White 62, Black 4,
Asian 3, Other 3
White 19, Black 1,
Asian 2, Other 2
White 28, Black 2,
Asian 1, Other 1
White 13, Black 1,
Asian 1, Other 1
White 16, Black 1,
Asian 1, Other 2
White 39, Black 1,
Asian 2, Other 1
CSF Data
Points
Plasma
Data Points
279
1181
15
105
53
458
16
118
89
183
100
293
a
Note that many subjects continued on from one study to another; thus, the total number of unique subjects in all of the studies is less than the sum of the
numbers of subjects in each study.
Figure 1. Diagram of the structural model. CLp represents plasma clearance; CLCSF , CSF clearance; QCSF and Qp , the intercompartmental clearances
within the CSF and plasma, respectively; VCSF , VCNS_tissue , Vp , and Vsystemic_tissue , apparent volumes in the CSF, CNS tissue, plasma, and systemic tissue
compartment, respectively.
Results
Table 1 lists the demographics and the data points
used in the model. In brief, the model included
72 subjects (37 males and 35 females) having 279
CSF and 1181 plasma concentration data points. The
base and full structural model are diagrammed in
Figure 1. The model consisted of a total of 4 compartments, with 2 compartments representing the CNS
(a CSF compartment and a CNS tissue compartment)
and 2 compartments representing the plasma and peripheral tissue. The model was defined in terms of
the following parameters: CLp represents the plasma
clearance, CLCSF represents the CSF clearance, QCSF
and Qp represent the intercompartmental clearances
within the CSF and plasma, respectively, and VCSF ,
Vp , Vsystemic_tissue and VCNS_tissue represent the modelestimated apparent volume of distribution in the CSF,
plasma, systemic tissue, and CNS tissue, respectively.
All clearance rate constants in the model were assumed
to be linear.
Parameters of the base model and final model are
shown in Table 2. Overall, the base model estimated
the parameters with reasonable precision as shown.
Diagnostic plots of the base model are shown in
Figure 2, indicating reasonable goodness of fit. For
the base model, the random effects associated with
the IIVs were assigned to CLp , CLCSF , VCSF , and Vp ;
their estimated IIVs were 58.5, 24.8, 104, and 63.2%,
respectively. Interoccasion variability was estimated to
be 38.1%.
Baseline body weight (BWT), baseline age, baseline
body surface area, and baseline height were found to be
highly collinear to each other. Thus, covariate testing
was not performed on all of these variables. Because
BWT was considered most physiologically relevant, it
was selected to be tested for covariate effect. In the final
model, a power model was used to relate BWT to VCSF
6
The Journal of Clinical Pharmacology / Vol 00 No 0 2017
Table 2. Parameters of the Base Model and Final Modela
Base Model
Parameter
Definition
Typical Value
(%RSE)
Unit
%IIV
η Shrinkage
(%)
Typical Value
(%RSE)
Unit
%IIV
η Shrinkage
(%)
0.572 (11.3)
L/h
NE
NA
0.568 (11.7)
L/h
NE
NA
0.069 (18.1)
L/h
NE
NA
0.0712 (19.5)
L/h
NE
NA
2.50 (11.2)
0.133 (8.87)
0.441 (21.7)
L/h
L/h
L
58.5
24.8
104
1.78
19.4
15.2
2.36 (5.04)
0.136 (9.12)
0.433 (21.7)
L/h
L/h
L
29.5
24.4
88.1
7.25
20.7
18.0
32.0 (21.7)
L
63.2
3.66
29.0 (11.4)
L
39.0
10.0
429 (19.2)
L
NE
NA
418 (20.9)
L
NE
NA
258 (17.1)
L
NE
NA
263 (18.9)
L
NE
NA
NA
NA
NA
None
38.1
NA
4.51
NA
NA
0.689 (12.2)
NA
None
38.1
NA
5.21
NA
NA
None
NA
NA
0.596 (52.3)
None
NA
NA
NA
None
NA
NA
0.047 (21.7)
None
NA
NA
0.493 (2.39)
None
NA
6.94
( shrinkage)
0.494 (2.45)
None
NA
6.81
( shrinkage)
Qp
Intercompartmental
clearance between the
plasma and the systemic
tissue compartments
Intercompartmental
QCSF
clearance between the
CSF fluid and the CNS
tissue compartments
Plasma clearance
CLp
CSF clearance
CLCSF
Volume of distribution in
VCSF
the CSF
Volume of distribution in
Vp
the plasma
Volume of distribution of
Vsystemic_tissue
the systemic tissue
compartment
VCNS_tissue
Volume of distribution of
the CNS tissue
compartment
IOV
Interoccasion variability
BWT on CLp
Covariate effect of the
baseline body weight
(BWT) on CLp
BWT on VCSF
Covariate effect of the
baseline body weight
(BWT) onVCSF
BWT on Vp
Covariate effect of the
baseline body weight
(BWT) on Vp
(Proportional) Residual error
(proportional structure)
a
Final Model
%IIV indicates intersubject variability (%); IOV, interoccasion variability; NA, not applicable; NE, not estimated or fixed to 0; %RSE, relative standard error (%).
and CLP , and a linear model was used to relate BWT to
Vp as defined in the following equations:
Vcs f (i) =Vcs f e
eta(i)
BW Ti
M BW T
θ V cs f
,
(5)
V p(i) =V p eeta(i) [1+ V p(i) (BW T − M BW T )],
C L p(i) = C L p eeta(i)
BW Ti
M BW T
θC L p
,
(6)
(7)
where VCSF(i) is the individual specific value of CSF
volume, VCSF is the population value of CSV volume,
θ VCSF is the estimated power coefficient scaling VCSF(i)
and VCSF based on body weight (BWTi ), Vp(i) is the
individual specific value of plasma volume, θ Vp(i) is
the linear parameter scaling BWT to Vp(i) , MBWT is
the median body weight based on the data. CLp(i) is
the individual plasma volume, CLp is the population
volume, and θ CLp is the estimated power coefficient
scaling CLp(i) and CLp based on BWT.
A summary of model performance of the base model
relative to the full model is presented in Table 2. The
final model resulted in a –89-point change in OFV
compared to the base model. Incorporating BWT into
the final model reduced the IIVs by –29, –0.4, –16,
and –24 for CLp , CLCSF , VCSF , and Vp , respectively,
confirming that BWT was a meaningful covariate.
Diagnostic plots of the base model are shown in
Figure 2. The diagnostic plots of the final model are
shown in Figure 3, indicating reasonable goodness of
fit. The pcVPC plots of the final model are shown
in the supplementary Figure S1. Separate diagnostic
plots for plasma and CSF are shown in supplementary
Figures S2 and S3, respectively. There appeared to be
a clear trend in the population model-predicted–vs–
dependent variable plot for the CSF data, indicating
that additional sources of variability had not been
accounted for by the model.
7
Luu et al
Figure 2. Diagnostic plots of the base model fit. DV indicates dependent variable (concentration); PRED, population model prediction; IPRED, individual
prediction; red line, line of identity or line through 0; blue line, trend line; gray shade, 95% confidence interval of the trend line.
Figure 3. Diagnostic plots of the final model fit. DV indicates dependent variable (concentration); PRED, population model prediction; IPRED, individual
prediction; red line, line of identity or line through 0; blue line, trend line; gray shade, 95% confidence interval of the trend line.
The median β half-life of nusinersen in the CSF was
163 days as calculated from the post-hoc estimates of
each individual’s PK parameters using equation 4. The
CSF half-life did not change with age (Table 3).
Results of the population simulations are shown
in Figure 4. For the fixed-dose scenario (all patients
received 12-mg dose), neonates and infants up to 3
months tended to be overexposed in terms of AUCinf
and Cmax . However, the protocol-based dosing (agebased up to 2 years and fixed dose >2 years) ensured
a more consistent AUCinf and Cmax .
The simulated median profiles of nusinersen
following a single dose in all 4 compartments are
shown in Figure 5. Nusinersen had a steep distribution
phase followed by a prolonged elimination phase in the
CSF. CSF and CNS tissues were quickly equilibrated,
followed by plasma uptake and then systemic tissue
uptake.
Discussion
SMA is one of the most common genetic causes
of infant death.15 Clinical trial results of nusinersen
administered intrathecally showed good safety and
tolerability profiles and promising clinical outcomes.8
In a phase 2 open-label study, patients with type I
8
The Journal of Clinical Pharmacology / Vol 00 No 0 2017
Table 3. Model-Derived CSF Terminal Half-Life Values by Age Groups
Terminal Elimination Half-Life
Age Group
N
Mean (Days)
SD
Median (Days)
Range (Days)
All (0.1 to 15 years)
0 to 3 months
3 to 6 months
6 to 12 months
1 to 2 years
2 to 6 years
>6 years
72
5
8
7
9
22
21
172
169
166
180
180
169
173
32.7
18.7
22.0
28.3
40.0
31.0
40.1
163
159
165
171
172
163
160
123 to 310
155 to 199
135 to 196
145 to 224
147 to 289
123 to 247
139 to 310
SD, standard deviation.
Figure 4. Population simulations (n = 1000 virtual subjects) showing AUCinf (A and B) and Cmax (C and D) relative to age following either a fixed
dose (A and C, 12 mg single dose for all virtual subjects, n = 1000) or a protocol-based (single dose, dose level based on age) dose adjustment (C
and D, age-adjusted for subjects 2 years, 12 mg for subjects >2 years as implemented in the protocols). For each box in the plot, thick middle line is
the median, lower and upper edges are first quartile and third quartile, respectively, and whiskers are minimum (lower whisker) and maximum (upper
whisker).
SMA who were treated with nusinersen demonstrated
significant (P = .01) improvements in motor function
scores, incremental achievement of motor milestones
such as head control, rolling, sitting, and improvements in neuromuscular electrophysiology compared
to baseline and published natural history data.16 In
this article we report a population PK model for
intrathecally dosed nusinersen describing the pooled
time-concentration data from 5 clinical trials in the
CSF and plasma of pediatric patients with SMA. With
sparse data in the CSF, the population PK model
enabled the estimation of PK parameters in the CSF by
integrating the CSF PK data with rich plasma PK data.
The PK of nusinersen in the CSF and plasma were best
described, simultaneously, by a 4-compartment model
with 2 compartments representing the CSF and CNS
9
Luu et al
Figure 5. Simulated median PK profiles of nusinersen in the CSF, CNS tissue, plasma, and systemic tissue following a single 12-mg fixed dose.
tissue and 2 compartments representing the plasma and
peripheral tissue.
CSF formation in adults has been traditionally reported to be 500-600 mL/d with a turnover rate of
about 4 times per day.17 Recent imaging data, however,
indicated that CSF turnover rate could be much faster
than traditionally reported: 140 times a day, with fluxes
of more than 22 mL/min.18 CSF turnover in infants
is thought to be higher than that in children and
adults.19 Nevertheless, nusinersen has a CSF half-life
much beyond the human CSF turnover rate and was
predicted by our model to be 163 days. This halflife value was in agreement with the values calculated
from monkeys, where more CSF PK sampling took
place, which were reported to be >100 days.20 The
extended half-life was most likely due to distribution
from CSF to the CNS tissue compartment and slow
equilibration back to the CSF before it was cleared
into the systemic circulation. The inclusion of the CNS
tissue compartment in the model was also supported
by measurable brain concentration data obtained from
monkeys following an intrathecal dosing at clinically
relevant doses (unpublished data). Post hoc CSF halflife values for all 72 subjects included in the population
PK modeling indicated that the CSF half-life of nusinersen was not age (or body-weight) dependent.
Covariate testing resulted in the BWT being a clinically relevant covariate on the VCSF . In the evaluation
of the covariate effect of BWT on CSF exposures,
population simulations were performed to investigate
whether covariate-adjusted dosing was warranted. In
these simulations we tested whether (1) fixed dosing
was feasible with nusinersen across all age groups and
(2) the age-adjusted dosing for the youngest patients
(2 years) according to the dose scheme implemented
in the protocols was justified. These results indicated
that fixed dosing led to higher exposure in young
patients (0-3 months) in terms of Cmax and AUCinf .
The protocol age-based dosing scheme ensured a more
comparable exposure across all age groups (Figure
4). Based on these covariate-exposure relationships,
however, a more simplified dosing scheme than the
protocol scheme might be feasible. For example, agebased dosing may be applied only for patients 0-3
months rather than 0-24 months. Alternatively, because
no dose-limiting toxicity has been reported in any of
the trials, a fixed-dose scheme could be recommended
despite the inconsistent exposure across the age groups.
The PK profiles of nusinersen in the plasma following intrathecal dosing was biphasic, which was commonly described by a 2-compartment model following a
systemic (subcutaneous) administration.21,22 Although
the influence on plasma Vp and CLp appeared to
be clinically relevant (parameter-covariate plots not
shown), the covariate-adjusted dosing for nusinersen
was considered based only on CSF PK rather than
plasma PK because the CNS is the site of action. In
addition, the SMN target for nusinersen is located in
the CNS, not in the circulation or peripheral tissue.
The sparseness of the CSF data limited our ability
to test for more refined mechanism(s) of distribution
of nusinersen in the spinal cord and in the brain. The
aim of our top-down approach was to develop a parsimonious model using the available data, not intending
10
to capture the complex physiology of the CNS space.
For example, temporal and spatial distribution within
the spinal cord, brain, and other CNS tissues were not
accounted for. Others have developed multicompartment distribution models and computational models to
describe drug distribution in the CSF for intrathecally
dosed drugs.23,24 Thus, the physiologically relevant kinetics of nusinersen could be better accounted for with
more relevant data. In addition, the model assumed the
transport from the CSF to the plasma compartment to
be the predominant mechanism of CSF clearance and
that possible elimination due to other mechanisms (ie,
potential metabolism within the CNS tissues, including brain) were assumed to be relatively insignificant.
Attempts at including additional elimination rate constants were unsuccessful due to the sparseness of the
CSF data.
SMA disease severity or type has not been shown
to affect CSF volume or turnover rate. In addition,
impairment of organs involved in the uptake and
clearance of ASOs (eg, liver and kidney) has not been
associated with the disease state in humans. Thus, we
did not consider the disease state to be a mechanistically
relevant covariate and did not include it in the covariate
testing.
The CNS tissue, particularly the brain tissue, is
the intended site of action, where SMN neurons are
degenerated in SMA patients. Although we separated
the CSF from the CNS tissue as distinct compartments
in our model, there are, as structured by the model,
transfer kinetics between the 2 compartments. At steady
state, we expect the CSF concentration to be proportional to the CNS tissue concentration. The assumption
of the structural model, however, can be supported with
additional data, such as brain tissue data, which are
difficult to collect in the clinical setting but could be
performed in rare cases by biopsy or with postmortem
sampling.
Because the model was meant to be descriptive
rather than fully mechanistic, it has inherent limitations. Thus, the parameters of this model should
be carefully interpreted. First, for example, the CSF
volume estimated from the model should be considered
an apparent volume of distribution rather than an
actual physiological volume. The final model estimated
the typical CSF volume to be 433 mL, whereas the
physiological volume of the CSF is reported to be about
150 mL in humans.25 Second, due to the sparse sampling in the CSF (with no distinct distribution profile),
we limited the model to include a single transport rate
constant from CSF to plasma but no transport rate constant from the plasma to the CSF. Thus, CSF reabsorption of drug is not accounted for by the model. Third,
the CNS compartment is a lumped compartment consisting possibly of the spinal cord tissue, subarachnoid
The Journal of Clinical Pharmacology / Vol 00 No 0 2017
space, and brain tissue. Thus, this compartment should
not be interpreted solely to mean the brain compartment. Last, we did not explore between-trial variability,
which can be important, especially because, in our
trials, many patients transition from 1 trial to the next.
We, however, did test for IOV and found it to be
important, as shown in the results. Although betweentrial variability and IOV are different, we feel that IOV,
in some regards, indirectly accounts for some of the
variability associated with the between-trial variability.
Conclusions
The PK of nusinersen in the cerebral spinal fluid and
plasma of 72 infants and children with SMA across
5 clinical trials was analyzed via population-based
modeling. The CSF and plasma PK of nusinersen were
simultaneously described by a 4-compartment model.
The only relevant covariate explaining variability in
the PK of nusinersen was body weight (and thus, age,
because the 2 variables were highly collinear in this
population), which influences the VCSF , Vp , and CLp
but not CLCSF . Simulations indicated that our agebased dosing scheme ensured more comparable median
exposure in terms of Cmax and AUC relative to a fixed
dosing regimen for all patients. However, because no
dose-limiting toxicity has been reported in any of the
trials, a fixed-dose scheme (12 mg across all age groups)
was recommended. Sex and race did not likely explain
variability in the PK of nusinersen. The long terminal
half-life of nusinersen supported infrequent dosing, eg,
once every 4 or 6 months in pediatric patients with
SMA.
Disclosure
All authors are employees and stockholders of Ionis
Pharmaceuticals, Inc.
References
1. D’Amico A, Mercuri E, Tiziano FD, Bertini E. Spinal muscular
atrophy. Orphanet J Rare Dis. 2011;6:71.
2. van der Steege G, Cobben JM, Osinga J, Scheffer H, van Ommen
GJ, Buys CH. A sublocus of the multicopy microsatellite marker
CMS1 maps proximal to spinal muscular atrophy (SMA) as
shown by recombinant analysis. Hum Genet. 1995;96(5):589–591.
3. Ahmad S, Bhatia K, Kannan A, Gangwani L. Molecular mechanisms of neurodegeneration in spinal muscular atrophy. J Exp
Neurosci. 2016;10:39–49.
4. Dubowitz V. Very severe spinal muscular atrophy (SMA type
0): an expanding clinical phenotype. Eur J Paediatr Neurol.
1999;3(2):49–51.
5. Crawford TO, Pardo CA. The neurobiology of childhood spinal
muscular atrophy. Neurobiol Dis. 1996;3(2):97–110.
6. Bogdanik LP, Osborne MA, Davis C, et al. Systemic, postsymptomatic antisense oligonucleotide rescues motor unit maturation
delay in a new mouse model for type II/III spinal muscular
atrophy. Proc Natl Acad Sci USA. 2015;112(43):E5863–E5872.
11
Luu et al
7. Hua Y, Sahashi K, Hung G, et al. Antisense correction of SMN2
splicing in the CNS rescues necrosis in a type III SMA mouse
model. Genes Dev. 2010;24(15):1634–1644.
8. Chiriboga CA, Swoboda KJ, Darras BT, et al. Results from
a phase 1 study of nusinersen (ISIS-SMN(Rx)) in children
with spinal muscular atrophy. Neurology. 2016;86(10):890–
897.
9. Matsuzawa J, Matsui M, Konishi T, et al. Age-related volumetric
changes of brain gray and white matter in healthy infants and
children. Cereb Cortex. 2001;11(4):335–342.
10. Bergstrand M, Hooker AC, Wallin JE, Karlsson MO. Predictioncorrected visual predictive checks for diagnosing nonlinear
mixed-effects models. AAPS J. 2011;13(2):143–151.
11. Karlsson MO, Sheiner LB. The importance of modeling interoccasion variability in population pharmacokinetic analyses. J
Pharmacokinet Biopharm. 1993;21(6):735–750.
12. Gastonguay M. A full model estimation approach for covariate
effects: inference based on clinical importance and estimation
precision [abstract]. AAPS J. 2004;6.
13. Gabrielsson JW, Weiner D. Pharmacokinetic and Pharmacodynamic Data Analysis, 3rd ed. Stockholm: Swedish Pharmaceutical Press; 2000: 82.
14. Sproule DM, Montes J, Montgomery M, et al. Increased fat mass
and high incidence of overweight despite low body mass index
in patients with spinal muscular atrophy. Neuromusc Disord.
2009;19(6):391–396.
15. Moultrie RR, Kish-Doto J, Peay H, Lewis MA. A review on
spinal muscular atrophy: awareness, knowledge, and attitudes. J
Genet Couns. 2016.
16. Finkel R, Chiriboga, CA, Vjsar, J, et al. Interim results of a
phase 2 clinical study of nusinersen (ISIS-SMNRx) in patients
with infantile-onset spinal muscular atrophy (AAN 2016, Poster
P5.004). Poster presented at: 68th Annual American Academy of
Neurology.
17. Artru AA. Spinal Cerebral Fluid Chemistry and Physiology. New
York: Elsevier; 1999: 216.
18. Brinker T, Stopa E, Morrison J, Klinge P. A new look at
cerebrospinal fluid circulation. Fluids Barriers CNS. 2014;11:10.
19. Bissonnette B. Pediatric Anesthesia. Shelton, CT: PMPH-USA;
2014: 455.
20. Rigo F, Chun SJ, Norris DA, et al. Pharmacology of a central
nervous system delivered 2’-O-methoxyethyl-modified survival
of motor neuron splicing oligonucleotide in mice and nonhuman
primates. J Pharmacol Exp Ther. 2014;350(1):46–55.
21. Shimizu R, Kitade M, Kobayashi T, Hori S, Watanabe A.
Pharmacokinetic-pharmacodynamic modeling for reduction of
hepatic apolipoprotein B mRNA and plasma total cholesterol
after administration of antisense oligonucleotide in mice. J
Pharmacokinet Pharmacodyn. 2015;42(1):67–77.
22. Geary RS, Norris D, Yu R, Bennett CF. Pharmacokinetics,
biodistribution and cell uptake of antisense oligonucleotides.
Adv Drug Deliv Rev. 2015;87:46–51.
23. Kuttler A, Dimke T, Kern S, Helmlinger G, Stanski D, Finelli
LA. Understanding pharmacokinetics using realistic computational models of fluid dynamics: biosimulation of drug distribution within the CSF space for intrathecal drugs. J Pharmacokinet
Pharmacodyn. 2010;37(6):629–644.
24. Ummenhofer WC, Arends RH, Shen DD, Bernards CM. Comparative spinal distribution and clearance kinetics of intrathecally administered morphine, fentanyl, alfentanil, and sufentanil.
Anesthesiology. 2000;92(3):739–753.
25. Johanson CE, Duncan JA 3rd, Klinge PM, Brinker T, Stopa
EG, Silverberg GD. Multiplicity of cerebrospinal fluid functions:
new challenges in health and disease. Cerebrospinal Fluid Res.
2008;5:10.
Supporting Information
Additional Supporting Information may be found in
the online version of this article at the publisher’s website.