Man Thai

Man Thai

Comparison between different models for CYP2D6 phenotype frequency

Marina Suzuki, PharmD

Liesl McCormick, PhD

Mariko Nanako, PhD

11/18/2016
 Man Thai

Introduction

Cytochrome P450 2D6 (CYP2D6) is one of the most common metabolic enzymes expressed
in the human liver that is responsible for the metabolism of 25% of clinical drugs including
tricyclic antidepressants, anticancer, opioids, antiarrhythmics, analgesics, antihistamines,
and antiviral agents [1,2,3]. CYP2D6 is highly polymorphic with more than 80 possible star
allelic variants that are described on Human Cytochrome P450 Allele Nomenclature
Database at http://www.cypalleles.ki.se/ [1,2,3]. Star allele nomenclature is the scientific
standard annotation for distinguishing genetic variants that might be due to single
nucleotide polymorphisms (SNPs), insertion/deletion, or copy number variants. Different
testing platforms can detect certain variants but not all, depending on the purpose of the
researcher. These platforms recognize the key difference in the nucleotide sequence and
match the SNPs to the referenced guideline. These variants differ in terms of enzymatic
activity that significantly affects patients ability to metabolize drugs. It has long been
confirmed that even though CYP2D6 genotypes vary across the world population, meaning
that some variants exist in all ethnic groups, some genotypes are specific to certain
subpopulations. Correlation of genotype to ethnicity, when taken into account, allows
phenotype predicting models to be more accurate and explain phenotype frequencies
better [4].

In the 1970s, CYP2D6 phenotype was first classified based on its metabolism level of probe
substrates such as dextromethorphan, sparteine, and debrisoquine [5, 6]. Metabolic activity
was measured by the ratio of probe drug to its metabolites, which could approximately
classify CYP2D6 phenotype into poor metabolizers (PMs), intermediate metabolizers (IMs),
extensive metabolizers (EMs), and ultrarapid metabolizers (UMs) [4, 7]. However, non-
consensus about which specific ratio depicted which phenotype continued to exist among
different laboratories [8-11]. Fortunately, along with advances in genotype sequencing,
researchers started to relate metabolizer phenotypes to DNA sequences instead of just
probe drugs metabolites quantification.

Gaedigk and colleagues originally developed a novel activity score (AS) allele quantification
model that assigns a continuous numeric value for corresponding CYP2D6 haplotypes
 Man Thai

based on approximate metabolic activity for individual star alleles variants [4, 12].
Individual star alleles receive an AS = 0 for non-functional enzymatic activity, AS = 0.5 for
reduced enzymatic activity, and AS = 1 for fully functional enzymatic activity. A total score
for each diplotype is required to determine phenotypes according to the proposed scale:
PM (AS = 0), IM (0 < AS < 1 ), EM (AS = 1-2), UM (AS > 2) [17]. This model is referred to as
AS Model A. This model gives rise to a more universal approach to classifying CYP2D6
phenotypes, but at the same time underestimates the distinct metabolic activity of over 81
SNPs by grouping haplotypes into nonfunctional (AS = 0), reduced function (AS = 0.5), and
functional (AS = 1). It is proposed that certain haplotypes in the reduced function group
such as CYP2D6*10, and CYP2D6*41 should receive different scores than CYP2D6*9 due to
pronounced different enzymatic activity [13]. A homozygous diplotype of CYP2D6*10 has a
score of 1.0, which corresponds to EM, or fully functional, but in reality has a much more
reduced enzymatic activity compared to a wild type CYP2D6*1.

Traditional AS Model A provides haplotype CYP2D6*10 with a score of 0.5 for its reduced
enzymatic activity. However, a group of researchers conducted a literature review of
previous researches from the Clinical Pharmacogenetics Implementation Consortium
(CPIC) and concluded that there exists more strong evidence rooting for a second model,
known as AS Model B that gives CYP2D6*10 a score of 0.25, rather than strong evidence
against this idea [13]. This new model promises a more specific quantification for CYP2D6
haplotypes [13]. However, there has been no comprehensive study that compares AS Model
A and AS Model B on a large population to see which model serves as a better clinical
diagnostic tool. This study will examine the phenotype translated from genotype data of a
large sample, and analyze the differences in phenotype frequency between these models.

Star allele CYP2D6 phenotype prediction is useful for personalized pharmacogenomics as

providers can adjust medication dosage properly for individuals based on their genotype.
Therefore, there is a great interest in translating CYP2D6 genotype to phenotype for clinical
applications. Since 2009, CPIC has constantly updated its guidelines for adjusting dosage or
complete replacement of drugs that are primarily metabolized by CYP2D6 for
corresponding CYP2D6 phenotypes, such as codeine, tricyclic antidepressants, and
selective serotonin reuptake inhibitors [14-16]. The most important application is for
 Man Thai

patients with PM and UM phenotypes, since they are vulnerable to more adverse events
during treatment due to low drug efficacy and high risk to toxicity.

Methods

Subjects. Study protocols were reviewed and granted an exempt status by the Institutional
Review Board at Pacific University. Study participants (n=85822) were patients who
previously submitted DNA samples to Molecular Testing Laboratory (MTL, Vancouver, WA)
between 2014 and 2016. Patients signed contract forms and understood that MTL reserves
the right to provide their de-identified information for the purpose of research studies.

CYP2D6 genotyping. Buccal samples were sent to MTL laboratory for purification and
analysis. The purification follows a fully automated protocol, which clears off contaminants
and proteins, collecting only highly pure DNA that is ready for immediate downstream
analysis. DNA samples and master mix were prepared with probe and primers for the
appropriate assay. Patient samples and 2 positive and negative controls were tested for
better accuracy. Gene amplification and real time PCR were performed on Taqman real-
time PCR platform with full automation. Genotyping of single polymorphisms for CYP2D6
was based on the fluorescent detection. Activity score was assigned to corresponding
CYP2D6 star nomenclature based on Table 1.

Analysis. De-identified data, CYP2D6 genotype, star alleles, gender, ethnicity, and
prescribed medications were extracted from the MTL data bank and analyzed. Activity
scores and phenotypes were assigned to CYP2D6 genotypes based on previously published
evidence [4]. All analyses were performed in R version 3.3.1, an open source program
developed by the R Foundation for Statistical Computing (Vienna, Austria). Student t-test
and chi-squared tests were done to examine differences in CYP2D6 metabolizer
phenotypes among reported ethnicity and gender. Difference in activity scores and
phenotypes between Model A and Model B was analyzed by t-test.

Activity score assignment. Two primary models Model A and Model B - were used to
assign activity score. Assigned activity scores followed the Human Cytochrome P450 Allele
Nomenclature Database at http://www.cypalleles.ki.se/, summarized in Table 1. Increasing
 Man Thai

values correlate with increasing enzyme activity, for instance, a 0 scored allele is
nonfunctional, and a 2 scored allele encodes an enzyme with higher activity compared to
a 1 scored allele. Sums of activity score for both alleles were used for phenotype
categorization, where phenotypes are PM (AS = 0), IM( 0 < AS <1), EM (1 AS 2), and UM
(AS >2).

Since EM is the most pervasive phenotype with the largest AS range (1 AS 2) in a

population, differences in phenotype frequencies between Model A and Model B may be
underestimated by EMs. Hence, we added variant models that categorizes EM group on a
finer level to look for marginal differences. Variant models are mostly similar to primary
models, except the total score range for EM (in primary models) is now split into EM and
Slow EM. Specifically, diploids whose AS = 2 is considered EM, where diploids whose 1 AS
<2 is Slow EM.

Results

This is the first study that compares the two models of activity score with CYP2D6
genotypes from an enormous dataset (n=82855). The sorted data (n=44187) used for
statistical analysis only includes the following variables: gender, genotypes, and ethnicity.
The exclusion criteria for sorting data are unknown CYP2D6 genotype, genotypes with
undefined activity scores, and incorrectly formatted data (Figure 1).

Characterization of the population. The dataset population (n=44187) consists of

several ethnicity groups, including but not limited to White, Black, Hispanic, Asian,
Native American/Alaskan, Hawaiian or Pacific Islander and Mixed race or Other.
Ethnicity distribution varies widely, but is mostly skewed towards White (n=28749),
Hispanic or Latino (n=7629), and Black (n=5368). There are several observed star alleles
from the data, yet the most prevalent star-alleles across the whole population are *1, or
wildtype (f=0.39), *4 (f=0.156), and *2 (f= 0.186) (Figure 2).

Statistical analysis between Model A and Model B and variant models. Descriptive
analyses of variable phenotype of both models were performed with R. Phenotype
 Man Thai

distributions by Model A and Model B are not significantly different (chi-squared, p>0.05)
(Table 2, 3 & 4). In the variant models, phenotype distributions are not significantly
different (chi-squared, p > 0.05) (Table 5). This result confirmed that either model is
equally appropriate to determine CYP2D6 enzyme activity for the population. In fact, Model
A (and variant model) produced fewer UMs and more EMs (and more Slow EMs) than
Model B (and variant model) by only 3 subjects, which is reasonable since Model B is more
specific to minimal spacing between activity score (0.25 stepwise), and thus gives the same
allele a higher score than Model A, contributing to the different ratio of UM and EM in the
results. Specifically, these three subjects are all Black females with *29/*29XN genotypes,
hence they received AS = 1.5 (Model A and variant model), and AS = 2.25 (Model B and
variant model).The most prominent phenotype observed with both Model A and Model B
from this population is EM with fully functional CYP2D6 enzyme, whereas the least
common is UM (Figure 3). In the variant models, the most prominent phenotype in both
models is Slow EM, and the least common is still UM (Figure 4).

Ethnicity and phenotype are significantly related to one another (chi squared test,
p<0.001). This strong relationship between ethnicity and phenotype should be taken in
future phenotype predicting models for better accuracy. Contingency tables in R offered a
generalized observation for phenotype frequency within each ethnic group (Tables 2 and
3). There are significant differences between the phenotype frequency of certain ethnic
groups and the whole population (chi-squared, x2 d.f = 1 > 5.02). Hispanic/Latino subjects are
more likely to be UMs than the whole population. Ethnic groups Asian, Hawaiian/Pacific
Islander, Black, and White have higher tendency to be EMs. While Asian and Black
subjects are more prone to be IMs, Hispanic/Latino subjects are least prone to be IMs.
When it comes to PMs, White, Hispanic/Latino, and Mixed race subjects have the most
likelihood, whereas Asian, Black, and Hawaiian/Pacific Islander have the least likelihood
(Figures 2 and 3).

Discussion
Models A and B assigned similar phenotypes to the population (unpaired t-test, p > 0.05).
In particular, Model B is more fine-tuned to 0.25 spacing between score levels, whereas
 Man Thai

model As scores are 0.50 apart. This can be explained since the only 4 rare star alleles
variants subjected to different scores are not common throughout the population, making
little significant difference in the final score, thus the phenotype assignments. Furthermore,
there were about 30 star alleles observed in the data set, which might have underestimated
the difference if there is any. In such a way, it may be safe to use either model to assign
phenotypes to any subject with a known genotype in a population. If researchers are
interested in looking at marginal differences between these models in future studies, a
sample with more rare star allele variants that receive different scores from Model A and
Model B, namely *9, *29, *45, *46 may yield more meaningful results

We can draw fairly generalized presumptions about each ethnic group, and the whole
population. The most common CYP2D6 phenotype for each ethnic group and the whole
population is EM, or slower EM (for variant models). On the contrary, the least common
CYP2D6 phenotype is UMs. This phenomenon translates very well into reality, as EMs and
slow EMs have fully functional enzyme activity, which are the 'standard', whereas UMs are
at the other extreme, or the 'anomaly' that should be much less typical. At first, one may
infer that PMs should lie on the opposite extreme and should account for a somewhat
similar percentage as UMs, but in fact, PMs are even more common than IMs, making it the
second most common phenotype in the population. Professionals that utilize the results of
this study should be aware of the unusually great distribution of PMs. One hypothesis is
that PMs are one of the primary phenotypes in the population, and IMs are just a
subsequent hybridization of PMs and EMs. Certain ethnic groups are notably less prone to
be PMs compared to the whole population, as well as other groups: 'Asian',
'Hawaiian/Pacific Islander', and 'Mixed Race', which also agrees with a report across the
world population [18]. These ethnic groups have fewer PMs than IMs, and fewer PMs
compared to the whole population. On the contrary, 'White' and 'American Indian/Native
Alaskan' subjects are more likely to be PMs across the whole population. The composition
of IMs and PMs, surprisingly, is widely varied depending on ethnicity. For instance,
'American Indian/ Native Alaskan' and 'Hispanic/Latino' subjects are much less likely to be
IMs compared to other ethnic groups and the whole population. On the contrary, Black
and Mixed Race are most likely to be IMs out of all ethnic groups. Ethnic group 'Other'
 Man Thai

should be studied further, especially the composition, for example what ethnic group
identify themselves as 'Other' in order to make meaningful assumptions.

So far it is not possible to approximate CYP2D6 phenotype solely based on ethnicity, since
genotypes within an ethnic group can also vary. The first suggestion from this study is to
test for UMs and PMs, since these phenotypes, though much less common, experience more
serious effects, either toxicity or low efficacy with a regular dose. IMs and EMs will not be
as much disadvantaged in a case of non-diagnosis as UMs and PMs.

The strength of this study is the large nationwide sample with various ethnic groups, hence
it can approximately reflect the whole population of America. This study opens up several
directions for future research: comparison between phenotype frequencies across the
literature, across ethnic groups, world population, and our study population. However,
there are still certain limitations with the study. First, the sorted data only contain half of
the raw data due to extensive exclusion criteria that ensured proper input for the model.
Second, because our study consists of statistical analysis with computer software, we
cannot conduct confirmation tests with probe drugs. Furthermore, any generalized model
tends to simplify input to approximate an output that is guaranteed to be relevant to a
certain degree of confidence. Thus, AS Models may not be useful for marginal differences
between phenotype frequencies.
 Man Thai

Figure 1. Schematic of sorting methodology. Flowchart describes the workflow from

sorting raw data, assigning appropriate activity score and phenotypes. Statistical tests
examine the relationship between ethnicity, gender, and CYP2D6 enzyme activity level. Any
subject that does not meet the inclusion criteria is excluded. The listed star alleles are well-
characterized and thus are included in the final dataset and continue to statistical analysis.
 Man Thai

Figure 2. CYP2D6 phenotype distribution within each ethnic group. The color-coded
phenotype groups are ultra-metabolizer (UM), intermediate metabolizer (IM), extensive
metabolizer (EM), and poor metabolizer (PM). These phenotypes are plotted with
increasing enzyme activity level. The population graph generated by model A and model B
are not statistically different on a 95% confidence interval; hence, only one graph from
model A needs to be included as a standard by which each ethnic group can be classified.
 Man Thai

Figure 3. Mosaic plots of ethnicity and phenotype by Model A and Model B. The color-coded
phenotype groups are ultra-metabolizer (UM), intermediate metabolizer (IM), poor
metabolizer (PM), and extensive metabolizer (EM). The far left vertical axis of each graph
describes the frequency of specific CYP2D6 phenotype groups. The population spread is
skewed to EM, followed by IM and PM, whereas UM is far less common.
 Man Thai

Figure 4. Mosaic plots of ethnicity and phenotype by Model A and Model B. The color-coded
phenotype groups are ultra-metabolizer (UM), intermediate metabolizer (IM), poor
metabolizer (PM), extensive metabolizer (EM) and slow extensive metabolizer (Slow EM).
The far left vertical axis of each graph describes the frequency of specific CYP2D6
phenotype groups. The population spread is skewed to Slow EM, followed by EM, IM and
PM, whereas UM is far less common.
 Man Thai

Table 1. Activity score assignment with model A and model B for observed alleles. The left
column consists of the activity score for associated alleles. These combinations followed
the Human Cytochrome P450 Allele Nomenclature Database at
http://www.cypalleles.ki.se/. The right column includes the alleles that match the activity
score. Different numbers denote different single polymorphic star alleles but do not
necessarily define activity level. Any star allele with XN can be interpreted as allele
duplication whose score is doubled of that of a single allele.
 Man Thai

Phenotype Model A (subjects) Model B (subjects)

UM 295 298

EM 30469 30466

IM 5805 5805

PM 7618 7618

Table 2. Comparison of the resulting phenotypes from model A and model B. The
phenotypes assigned by these different models are not significantly different on 95%
confidence interval. However, Model A produced fewer UM and more EM than Model B,
which is reasonable since Model B is more specific to minimal spacing between activity
score (0.25 stepwise), and thus gives the same allele higher score than model A,
contributing to the different ratio of UM and EM in the results.
 Man Thai

American Asian Black Hawaiian or Hispanic or Mixed Race Other White

Indian or Pacific Latino
Native Islander
Alaskan

EM 0.6973 0.8006 0.6804 0.8289 0.7747 0.6930 0.6442 0.6639

IM 0.0917 0.1630 0.2075 0.1497 0.0789 0.1814 0.1366 0.1292

PM 0.2018 0.0332 0.1071 0.0214 0.1368 0.1163 0.2132 0.2005

UM 0.0092 0.0032 0.0050 0.0000 0.0096 0.0093 0.0060 0.0064

Table 3 Contingency table of ethnicity and CYP2D6 phenotypes in the population by Model
A. Numeric values describe the frequency of specific CYP2D6 phenotypes among each
ethnic group. On the far left, abbreviated phenotype groups are ultra-metabolizer (UM),
intermediate metabolizer (IM), extensive metabolizer (EM) and poor metabolizer (PM).
 Man Thai

American Indian or Asian Black Hawaiian or Hispanic or Mixed Other White

Native Alaskan Pacific Islander Latino Race

EM 0.6973 0.8006 0.6804 0.8289 0.7747 0.6930 0.6442 0.6639

IM 0.0917 0.1630 0.2075 0.1497 0.0789 0.1814 0.1366 0.1292

PM 0.2018 0.0332 0.1071 0.0214 0.1368 0.1163 0.2132 0.2005

UM 0.0092 0.0032 0.0050 0.0000 0.0096 0.0093 0.0060 0.0064

Table 4 Contingency table of ethnicity and CYP2D6 phenotypes in the population by model
B. Numeric values describe the frequency of specific CYP2D6 phenotypes among each
ethnic group. On the far left, abbreviated phenotype groups are ultra-metabolizer (UM),
intermediate metabolizer (IM), extensive metabolizer (EM), and poor metabolizer (PM).
 Man Thai

Phenotype Model A (subjects) Model B (subjects)

UM 295 298

EM 8169 8169

Slow EM 22300 22297

IM 5805 5805

PM 7618 7618

Table 5. Comparison of the resulting phenotypes from variant Model A and variant Model
B. The phenotypes assigned by these different models are not significantly different on
95% confidence interval. However, variant Model A produced fewer UM and more Slow EM
than variant Model B, which is reasonable since variant Model B is more specific to minimal
spacing between activity score (0.25 stepwise), and thus gives the same allele higher score
than variant model A, contributing to the different ratio of UM and Slow EM in the results.

Literature cited:
 Man Thai

[1] Zhou S.F. Polymorphism of human cytochrome P450 2D6 and its clinical significance:
part II. Clin. Pharmacokinet., 2009; 48 (7), 761804.

[2] Zhou S.F. Polymorphism of human cytochrome P450 2D6 and its clinical significance:
Part I. Clin. Pharmacokinet., 2009; 48 (6), 689723.

[3] He Z.X.;Chen X.W.; Zhou Z.W.; Zhou S.F. Impact of physiological, pathological and
environmental factors on the expression and activity of human cytochrome P450 2D6 and
implications in precision medicine. Drug Metab Rev., 2015, 150.

[4] Gaedigk, A.; Simon, S. D.; Pearce, R. E.; Bradford, L. D.; Kennedy, M. J.; Leeder, J. S. The
CYP2D6 activity score: translating genotype information into a qualitative measure of
phenotype. Clin. Pharmacol. Ther., 2008, 83 (2), 234-42.

[5] Mahgoub, A.; Idle, J. R.; Dring, L. G.; Lancaster, R.; Smith, R. L., Polymorphic
hydroxylation of Debrisoquine in man. Lancet, 1977, 2 (8038), 584-6.

[6] Eichelbaum, M.; Spannbrucker, N.; Steincke, B.; Dengler, H. J., Defective N-oxidation of
sparteine in man: a new pharmacogenetic defect. Eur. J. Clin. Pharmacol., 1979, 16 (3), 183-
7.

[7] Zanger, U. M.; Raimundo, S.; Eichelbaum, M. Cytochrome P450 2D6: overview and
update on pharmacology, genetics, biochemistry. Naunyn Schmiedebergs Arch. Pharmacol.,
2004, 369 (1), 23-37.

[8] Gaedigk, A.; Bradford, L. D.; Marcucci, K. A.; Leeder, J. S. Unique CYP2D6 activity
distribution and genotype-phenotype discordance in black Americans. Clin. Pharmacol.
Ther., 2002, 72 (1), 76-89.

[9] Schmid, B.; Bircher, J.; Preisig, R.; Kupfer, A. Polymorphic dextromethorphan
metabolism: co-segregation of oxidative O-demethylation with debrisoquin hydroxylation.
Clin. Pharmacol. Ther., 1985, 38 (6), 618-24.

[10] Wojtczak, A.; Rychlik-Sych, M.; Krochmalska-Ulacha, E.; Skretkowicz, J. CYP2D6

phenotyping with dextromethorphan. Pharmacol. Rep., 2007, 59 (6), 734-8.
 Man Thai

[11] Frank, D.; Jaehde, U.; Fuhr, U. Evaluation of probe drugs and pharmacokinetic metrics
for CYP2D6 phenotyping. Eur. J. Clin. Pharmacol., 2007, 63 (4), 321-33.

[12] Kirchheiner, J. CYP2D6 phenotype prediction from genotype: which system is the best?
Clin. Pharmacol. Ther., 2008, 83 (2), 225-7.

[13] Hicks J.K.; Swen J.J.; Gaedigk A. Challenges in CYP2D6 phenotype assignment from
genotype data: a critical assessment and call for standardization. Curr. Drug. Metab., 2014;
15: 21832.

[14] Crews, K. R.; Gaedigk, A.; Dunnenberger, H. M.; Klein, T. E.; Shen, D. D.; Callaghan, J. T.;
Kharasch, E. D.; Skaar, T. C. Clinical Pharmacogenetics Implementation Consortium (CPIC)
guidelines for codeine therapy in the context of cytochrome P450 2D6 (CYP2D6) genotype.
Clin. Pharmacol. Ther., 2012, 91 (2), 321-6.

[15] Hicks, J. K.; Bishop, J. R.; Sangkuhl, K.; Mller, D. J.; Ji, Y.; Leckband, S. G.; ... & Skaar, T. C.
Clinical Pharmacogenetics Implementation Consortium (CPIC) guideline for CYP2D6 and
CYP2C19 genotypes and dosing of selective serotonin reuptake inhibitors. Clin. Pharmacol.
Ther., 2015, 98 (2), 127-134.

[16] Hicks, J. K.; Swen, J. J.; Thorn, C. F.; Sangkuhl, K.; Kharasch, E. D.; Ellingrod, V. L.; ... &
Stingl, J. C. Clinical Pharmacogenetics Implementation Consortium guideline for CYP2D6
and CYP2C19 genotypes and dosing of tricyclic antidepressants. Clin. Pharmacol. Ther.,
2013, 93(5), 402-408.

[17] Relling, M. V.; Klein, T. E. CPIC: Clinical Pharmacogenetics Implementation Consortium

of the Pharmacogenomics Research Network. Clin. Pharmacol. Ther., 2011, 89 (3), 464-7.

[18] Gaedigk, A.; Sangkuhl, K.; Whirl-Carrillo M.; Klein, T. & Leeder, J.S.Prediction of CYP2D6
phenotype from genotype across world populations. Genetics in Medicine, 2016.