569709

ryngologyHead and Neck SurgeryHeineman et al

2015 The Author(s) 2010

OTOXXX10.1177/0194599815569709Otola

Reprints and permission:

sagepub.com/journalsPermissions.nav

Original ResearchEndocrine Surgery

In Silico Analysis of RET Variants in

Medullary Thyroid Cancer: From the
Computer to the Bedside

Otolaryngology
Head and Neck Surgery
2015, V
ol. 152(4) 650654
American Academy of
OtolaryngologyHead and Neck
Surgery Foundation 2015
Reprints and permission:
sagepub.com/journalsPermissions.nav
DOI: 10.1177/0194599815569709
http://otojournal.org

Thomas E. Heineman1, Rohan Joshi, MD2, Marc A. Cohen, MD2,

William I. Kuhel, MD2, and David I. Kutler, MD2
No sponsorships or competing interests have been disclosed for this article.

Abstract
Objective. The American Thyroid Association (ATA) medullary thyroid cancer (MTC) guidelines group RET variants, in
the setting of familial medullary thyroid cancer and multiple
endocrine neoplasia type 2, into 4 classes of severity based on
epidemiological data. The aim of this study was to determine if
genotype correlates with phenotype in RET missense mutations.
Study Design. In silico mutational tolerance prediction.
Setting. Academic research hospital.
Subjects and Methods. We analyzed all RET variants currently
listed in the ATA guidelines for the management of MTC using
2 computer-based (in silico) mutation tolerance prediction
approaches: PolyPhen-2 HumVar and PolyPhen-2 HumDiv.
Our analysis also included 27 different RET single-nucleotide
polymorphisms resulting in missense variants.
Results. There was a statistically significant difference in the
overall HumDiv score between ATA groups A and B (P = .025)
and a statistically significant different HumVar score between
benign polymorphisms and ATA group A (P = .023). Overall,
RET variants associated with a less aggressive clinical phenotype generally had a lower Hum Div/Var score.
Conclusions. Polyphen-2 Hum Div/Var may provide additional
clinical data to help distinguish benign from MEN2/familial
medullary thyroid carcinomacausing RET variants as well
as less aggressive phenotypes (ATA A) from more aggressive
ones (ATA B-C). In silico genetic analyses, with proper validation, may predict the phenotypic severity of RET variants, providing clinicians with a tool to aid clinical decision making in
cases in which the RET variant is currently unknown or little
epidemiological data are available.
Keywords
medullary thyroid cancer, multiple endocrine neoplasia, thyroidectomy, in silico analysis, RET gene

Received September 13, 2014; revised December 4, 2014; accepted

January 7, 2015.

ultiple endocrine neoplasia (MEN) type 2 is a rare

genetic disorder, in which nearly all patients develop
medullary thyroid cancer (MTC).1 Variants in the
RET gene, a cell-surface tyrosine kinase, can result in oncogenic activation of cell growth and differentiation.2 MEN2
can be classified into 3 distinct subclasses: MEN2A, MEN2B,
and familial medullary thyroid carcinoma (FMTC). Over the
past 2 decades, analysis of genotype-phenotype relationships
has shown a distinct relationship between a given variant in
the RET gene and MEN2 disease presentation. While the significance of certain variants are well documented, there
remain some variants whose effect on RET activation is still
unknown.
MEN2 is an autosomal dominant condition that may affect
several members of a family over several generations. Once a
proband is diagnosed with MEN2, the current recommendation is to screen family members and perform a prophylactic
thyroidectomy in the kindred at risk of developing MTC.3 The
known variant data combined with the need to better understand the proper age for prophylactic thyroidectomy culminated in the American Thyroid Association (ATA) guidelines
for MEN2A/FMTC based on 3 classes of severity: surgery
before 5 years of age (Level C), before 5 years of age but possibly delayed (Level B), and before 10 years of age (Level A).4
1

Weill Cornell Medical College, New York, New York, USA

Department of OtolaryngologyHead and Neck Surgery, Weill Cornell
Medical College/New York Presbyterian, New York, New York, USA
2

This article was presented at the International Federation of Head and

Neck Oncologic Societies 5th World Congress and American Head and
Neck Society 2014 meeting; July 26-30, 2014; New York, New York (abstract
ID 55282).
Corresponding Author:
David I. Kutler, MD, Department of OtolaryngologyHead and Neck
Surgery, New York Presbyterian Hospital, Weill Cornell Medical College,
1305 York Ave, 5th Floor, New York, NY 10021, USA.
Email: dik2002@med.cornell.edu

Downloaded from oto.sagepub.com at Airlangga University on September 4, 2015

651

Heineman et al
Level D represents MEN2B and has the earliest need for prophylactic thyroidectomy.
Recently, several computer-modeling algorithms have
been developed to analyze the disruptive effect of a given
variant on protein function, termed in silico tolerance prediction, as they calculate whether a protein could tolerate a particular missense mutation.5 As any amino acid change can be
either a benign polymorphism or a disease-causing mutation,
in silico tolerance prediction attempts to predict which amino
acid substitutions will more severely disrupt protein function
by a multifaceted analysis. The programs assess pathogenicity
through 2 major criteria. The first is the evolutionary conservation of the mutated protein region, which assumes that
highly conserved regions between species are of more functional importance than lesser conserved regions. The second is
the change in amino acid side-chain encoded by the missense
mutation, since amino acids have a wide range of physical
properties, which may affect protein folding and function.
These properties include variability in amino acid side chain
size and polarity. This methodology is often used to rank
mutations observed in a genome by predicted effect size but
had not been systematically applied to the specific study of
genotype-phenotype associations for the RET gene.
This is in contrast to in vitro studies that required introducing
a particular variant into cell lines and observing the change in
cellular phenotype.6 In silico analyses of a subset of RET variants
were correlated with in vitro studies by Cosci et al.7 They analyzed 6 known and 7 unknown variants and found a direct relationship between the transforming ability of a particular variant
in cell lines with the Align-GVGD score. Furthermore, this work
proved the validity of in silico analysis as a valid alternative to in
vitro studies. Nakao et al8 used an in silico method to analyze two
MEN 2B variants, p.(Q781R) and p.(V804M). The p.(V804M)
variant was also examined in the context of 6 different tandem
variants. To our knowledge, these 2 studies are the only in silico
analyses of the RET proto-oncogene, but they both examine only
a fraction of the known MTC-causing RET variants. This is the
first attempt to our knowledge to analyze of all the ATA-classified
MEN-causing variants and compare them with known benign
polymorphisms of the RET gene.

msa/clustalw2/).10 This software calculates the best match for

the input sequences of interest and identifies similarities and
differences between the species. The NF2 alignment generated by Clustal W was analyzed using 2 Web-based softwares:
PolyPhen-2 HumDiv and HumVar.11 HumDiv and HumVar
are similar algorithms that were trained on 2 different mutational data sets to allow for 2 different analytical approaches
to any given genetic variant. The software inputs include the
interspecies RET gene alignment as well as the mutation in
question. The various RET mutations we analyzed were taken
from the ATA guidelines. Of note, ATA class D mutations were
excluded from analysis because this subset of RET mutations
is responsible for MEN2B, a distinct clinical entity with a very
different clinical phenotype from MEN2A/FMTC. The RET
benign polymorphisms analyzed as a control were taken from
the dbSNP database and were chosen only if they were validated though multiple, independent submissions as well as the
1000 Genome project. Both the RET mutations and polymorphisms are listed in Table 1.
The output scoring system for both PolyPhen-2 programs
are on a scale from 0 (benign) to 1 (probably damaging).
Because all mutations analyzed were previously published
(ATA guidelines) and no individual patient data were used, a
Weill Cornell Medical College Institutional Review Board
committee exemption was granted for this study.
Lastly, we used 2 other available in silico variant analysis
softwares in addition to PolyPhen-2: SIFT (Sorting Intolerant
From Tolerant) and Align GVGD.12,13 These programs had
worse variant discrimination within the RET gene than
PolyPhen-2 and were not included in this report. Because
every algorithm is different in design, one program may be
superior to the others in determining the degree of protein disruption for a given genetic disease.

Statistical Analysis
Statistical significance was calculated using SPSS (version
18, SPSS Inc, Chicago, Illinois). A 2-tailed Fisher exact test
was used with a significance threshold of P < .05, with the in
silico results binned into lower risk (<0.9) and higher risk
(>0.9).

Results

Materials and Methods

In Silico Analysis
Before an in silico tolerance prediction of known RET gene
mutations could be performed, it was necessary to create an
alignment of the RET gene. An alignment between other eukaryotic genomes allows regions of the RET gene to be identified as
being conserved and, therefore, predicted to be of importance to
the protein function or nonconserved, which can be predicted to
be of lesser importance. To generate an alignment, we used the
HomoloGene system (http://www.ncbi.nlm.nih.gov/homologene), part of the National Center for Biotechnology Information
(NCBI).9 The sequence of the Homo sapiens NF2 gene along
with the sequences of other eukaryotic species were collected.
After collecting the various sequences, we calculated the
gene alignment using Clustal W (http://www.ebi.ac.uk/Tools/

The HumDiv and HumVar scores were generated for each

ATA RET variant and are given in Table 1, with variants
predicted as damaging indicated by a gray shade. The median
PolyPhen-2 HumDiv scores of benign variants and ATA
groups A to C were 0.76 (n = 27), 1.0 (n = 23), 1.0 (n = 26),
and 1.0 (n = 6), respectively. The median PolyPhen-2 HumVar
scores of benign variants and ATA groups A to C were 0.21
(n = 27), 0.99 (n = 23), 0.98 (n = 26), and 0.99 (n = 6), respectively. All variants in ATA group C had the highest PolyPhen-2
score. Medians are reported because the data do not fall on a
normal distribution.
We found a statistically significant difference when a
binary analysis was performed, examining those variants with
a classification of probably damaging by the in silico analysis (score >0.9) between benign variants (26%) and ATA group

Downloaded from oto.sagepub.com at Airlangga University on September 4, 2015

652

OtolaryngologyHead and Neck Surgery 152(4)

A (65%) using HumVar (P = .025). There was also a statistically significant difference using HumDiv between ATA
groups A (78%) and B (96%; P = .025). The data used for this
analysis are given in Table 2. Overall, RET variants associated with a less aggressive clinical phenotype generally had a
lower Hum Div/Var score.
In addition to MTC risk levels, we mined the ATA genotype-phenotype data for other presentations of a RET variant,
among the most studied of which is primary hyperparathyroidism. In variants known to have a higher predisposition to
primary hyperparathyroidism, 24 of 26 (92%) had the highest
disruption score. This is in comparison to 9 of 24 (38%) variants having the highest score with lesser predispositions to
primary hyperparathyroidism. There are insufficient data to
analyze the risk of pheochromocytoma and other manifestations of RET variants.
When proposing a diagnostic test, a clinician needs to
know the sensitivity and specificity of the test and positive
(PPV) and negative predictive value (NPV) of the test result if
presented with a patient having an unknown RET variant. The
sensitivity and specificity of HumVar in distinguishing
disease-causing (MEN2/FMTC) from nondisease-causing
(benign) variants were 74.0% and 78.6%, respectively (PPV =
86.1%, NPV = 62.9%). The sensitivity and specificity of
HumDiv in the same analysis were 88.0% and 57.1%, respectively (PPV = 78.6%, NPV = 72.7%). When analyzing the
tests ability to distinguish less severe phenotypes (ATA A)
from more severe (ATA B, C), HumVar had a sensitivity and
specificity of 81.8% and 41.2% (PPV = 73.0%, NPV = 53.9%)
compared with HumDiv with a sensitivity and specificity of
97.0% and 28.4% (PPV = 72.7%, NPV = 83.3%).

Discussion
A variant in the RET proto-oncogene may result in a spectrum
of disease, from no symptoms, to any combination of MTC,
primary hyperparathyroidism, pheochromocytoma, Hirschsprung
disease, and cutaneous lichen amyloidosis.4 The ATA classification of genotype-phenotype correlations is the result of
several hundred studies of families worldwide with a spectrum of variants. While the better understood variants may
involve many kindred, a significant percentage of variants
involve only a small number of families, making a consensus
on the pathogenicity of the variant extremely challenging.
The traditional method in analyzing the oncogenic potential of a RET variant was the in vitro focus formation assay, in
which the variant was introduced into cell lines to study
growth curves. This method is time-consuming, expensive
and does not always correlate with disease severity.14 An alternative to in vitro analysis is in silico analysis.
Overall, our computer-based analysis of RET variants generally reproduced the ATA classification. The in silico data
appeared to be binary, in that the scores were largely either
very low (<0.5) or very large (>0.95).
There was no statistically significant difference between
the ATA levels B and C; however, there was a large population
of high RET disruption in group B. Similarly, the ATA

Table 1. Benign Polymorphism: ATA Risk Level Variants.a

RET Variant
Benign
p.(D102N)
p.(T278N)
p.(V347A)
p.(R368H)
p.(A386V)
p.(V397M)
p.(R417H)
p.(C430G)
p.(G446C)
p.(L452I)
p.(V467L)
p.(D489N)
p.(L501V)
p.(N554S)
p.(S561R)
p.(Y606H)
p.(A680T)
p.(R694Q)
p.(K710E)
p.(T754M)
p.(G830R)
p.(E867D)
p.(G885R)
p.(V906M)
p.(D1031N)
p.(T1038I)
p.(R1050Q)
ATA A
p.(C515S)
p.(G533C)
p.(R600Q)
p.(K603E)
p.(Y606C)
p.(S649L)
p.(K666E)
p.(E768D)
p.(N777S)
p.(L790F)
p.(Y791F)
p.(V804L)
p.(V804M)
p.(R833C)
p.(R844Q)
p.(S891A)
p.(R912P)
p.(R844Q)
p.(S891A)
p.(R912P)
p.(R844Q)
p.(S891A)
p.(R912P)

Downloaded from oto.sagepub.com at Airlangga University on September 4, 2015

HumDiv

HumVar

0.01
0.00
0.86
0.15
0.21
0.96
0.00
0.03
0.99
0.04
1.00
0.00
0.00
0.99
0.76
0.00
0.40
0.51
0.99
1.00
0.01
1.00
1.00
1.00
1.00
0.98
1.00

0.004
0.004
0.207
0.035
0.023
0.736
0.002
0.015
0.503
0.015
0.878
0.003
0.002
0.81
0.319
0.002
0.063
0.067
0.888
1
0.026
0.973
1
0.998
0.998
0.662
0.997

0.61
1.00
0.01
0.49
0.99
1.00
0.99
0.99
0.33
1.00
0.56
1.00
1.00
1.00
1.00
0.97
1.00
1.00
0.97
1.00
1.00
0.97
1.00

0.13
1.00
0.00
0.12
0.60
1.00
0.80
0.86
0.15
0.99
0.47
1.00
1.00
1.00
1.00
0.94
1.00
0.99
0.94
1.00
0.99
0.94
1.00
(continued)

653

Heineman et al
distinction between risk levels B and C is not entirely clear.
The current recommendation for age of prophylactic surgery
for level C is before age 5 years, while the recommendation
for level is to consider surgery before age 5 but may be delayed
if the patient has a normal annual basal serum calcitonin, normal annual neck ultrasound, less aggressive MTC family history, and family preference.4 This highlights the clinical
heterogeneity within the ATA risk levels and may indicate the
need for a more advanced classification scheme. Perhaps,
instead of comparing the in silico analysis to the current gold
standard, the ATA risk levels, both systems could be reconciled into a unified classification.
Notably, tolerance predictors were able to successfully
determine pathogenic MEN2/FMTC-causing variants from
benign polymorphisms. This may provide important evidence
to determine whether polymorphisms of undetermined significance are potentially pathologic. With the rapidly decreasing
cost and increasing availability of DNA sequencing, the ability to interpret and make meaningful predictions of sequencing data is becoming ever more important.
In silico tolerance predictors have the potential to fill this void
but are not yet ready to make perfect RET variant predictions. We
postulate that the inability of these methods to detect perfect differences within our analyzed variants was several fold.
Firstly, genotype-phenotype variation may be influenced
by epigenetic factors. The ability to regulate the expression of
various genes through the covalent modification of DNA and
histones, known as epigenetics, has been shown to play an
important role in the activation of proto-oncogenes, thereby
driving the development of cancer.15 These changes are not
detectable with standard DNA sequencing.
Another potential contributor to the discordance between
in silico tolerance predictions is the complex genetic makeup
of every patient as well as the theoretical nature of the in
silico analyses, both tolerance predictions and structural
analysis. The correlation between clinical presentation and
the in silico analyses were not perfect. Clearly, each case of
MEN2/FMTC will have a unique genetic milieu with a combination of local, nonpathologic polymorphisms as well as
other genes that may retard or accelerate tumorigenesis. The
mutation analysis programs may also miss important regions
of the protein that are not conserved or would have a larger
biological role than would be expected, resulting in erroneous predictions. In subsequent studies, we would like to
examine more RET missense variants to further elucidate
which mutation prediction methodologies have the best
prognostic value.
We do not advocate that computer algorithms replace
patient preference, clinical acumen, and past epidemiological
and in vitro studies; rather, we propose it be added to the discourse and decision making on an individual patient basis.
Patients and families with diseases on the MEN spectrum
require care at tertiary care hospitals that can deliver integrated care with surgeons, endocrinologists, and geneticists.
Lastly, in silico mutation predictors have the capability of
analyzing any genetic variant and may be applied to many different genetic problems facing otolaryngologists.

Table 1. (continued)
RET Variant
ATA B
p.(C609F)
p.(C609G)
p.(C609R)
p.(C609S)
p.(C609Y)
p.(C611F)
p.(C611G)
p.(C611R)
p.(C611S)
p.(C611W)
p.(C611Y)
p.(C618F)
p.(C618G)
p.(C618R)
p.(C618S)
p.(C618Y)
p.(C620F)
p.(C620G)
p.(C620R)
p.(C620S)
p.(C620W)
p.(C620Y)
p.(C630F)
p.(C630R)
p.(C630S)
p.(C630Y)
ATA C
p.(C634F)
p.(C634G)
p.(C634R)
p.(C634S)
p.(C634W)
p.(C634Y)

HumDiv

HumVar

1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.96
0.96
0.96
0.02
0.96
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.97
1.00

1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.82
0.69
0.69
0.02
0.69
0.95
0.97
0.98
0.95
0.96
1.00
0.96
0.98
0.78
0.98

1.00
1.00
1.00
1.00
1.00
1.00

1.00
1.00
1.00
1.00
1.00
1.00

Abbreviation: ATA, American Thyroid Association.

a
HumDiv and HumVar scores each ATA RET variant as well as benign variants.Variants predicted as damaging by the in silico are indicated by a gray
shade.

Table 2. Polymorphism-ATA A/B Group Comparison.a

HumDiv Number <0.9

HumVar Number <0.9c
Total number

Benign

ATA A

ATA B

16
22
28

5
7
17

1
6
27

Abbreviation: ATA, American Thyroid Association.

a
Cumulative HumDiv and HumVar scores for ATA class A and B RET
variant as well as benign variants. A Fisher test of significance P value
between the 2 groups of variants is indicated with a line under the stated
P value.
b
Bolded values have a Fisher test significance at HumVar P = .02.
c
Bolded values have a Fisher test significance at HumDiv P = .03.

Downloaded from oto.sagepub.com at Airlangga University on September 4, 2015

654

OtolaryngologyHead and Neck Surgery 152(4)

Conclusions
We have shown that PolyPhen-2 has the potential to distinguish benign from disease causing RET variants as well as
ATA group A and B from one another on the basis of the variant alone with a moderate degree of sensitivity and specificity.
We acknowledge the current limitations of this analysis; however, we propose that that the accuracy of in silico genetic
analytics will only improve with increasing computing power
in the setting of the widespread availability of sequencing.
The use of computer-based protein modeling to predict clinical outcomes for rare, genetic diseases, such as the timing of
prophylactic thyroidectomy in MEN2, may add crucial evidence to aid in decision making. Furthermore, if a previously
unknown variant (non-ATA classified) is discovered in a
patient, in silico analysis may help determine the potential
aggressiveness and predict the ideal age for prophylactic thyroidectomy for that patients kindred.
Author Contributions
Thomas E. Heineman, acquisition, analysis, drafting, approval,
accountable; Rohan Joshi, analysis, drafting, approval, accountable;
Marc A. Cohen, data interpretation, drafting, approval, accountable;
William I. Kuhel, data interpretation, drafting, approval, accountable; David I. Kutler, analysis, drafting, approval, accountable.

Disclosures
Competing interests: None.
Sponsorships: None.
Funding source: None.

References
1. Mulligan LM, Kwok JB, Healey CS, et al. Germ-line variants
of the RET proto-oncogene in multiple endocrine neoplasia type
2A. Nature. 1993;363:458-460.
2. Takahashi M, Buma Y, Iwamoto T, et al. Cloning and expression
of the ret protoconcogene encoding a tyrosine kinase with two
potential transmembrane domains. Oncogene. 1988;3:571-578.
3. Waguespack SG, Rich TA, Perrier ND, Jimenez C, Cote GJ.
Management of medullary thyroid carcinoma and MEN2 syndromes in childhood. Nat Rev Endocrinol. 2011;7:596-607.

4. American Thyroid Association Guidelines Task Force. Medullary thyroid cancer: management guidelines of the American
Thyroid Association. Thyroid. 2009;19:565-582.
5. Tavtigian SV, Deffenbaugh AM, Yin L, et al. Comprehensive statistical study of 452 BRCA1 missense substitutions with classification of eight recurrent substitutions as neutral. J Med Genet.
2006;43:295-305.
6. Mise N, Drosten M, Racek T, Tannapfel A, Putzer BM. Evaluation of potential mechanisms underlying genotype-phenotype
correlations in multiple endocrine neoplasia type 2. Oncogene.
2006;25:6637-6647.
7. Cosci B, Vivaldi A, Romei C, et al. In silico and in vitro analysis
of rare germline allelic variants of RET oncogene associated with
medullary thyroid cancer. Endocr Relat Cancer. 2011;18:603-612.
8. Nakao KT, Usui T, Ikeda M, et al. Novel tandem germline RET
proto-oncogene variants in a patient with multiple endocrine neoplasia type 2B: report of a case and a literature review of tandem
RET variant with in silico analysis. Head Neck. 2013;35:363-368.
9. Geer LY, Marchler-Bauer A, Geer RC, et al. The NCBI BioSystems database. Nucleic Acids Res. 2010;38:D492-D496.
10. Larkin MA, Blackshields G, Brown NP, et al. Clustal W and
Clustal X version 2.0. Bioinformatics. 2007;23:2947-2948.
11. Adzhubei IA, Schmidt S, Peshkin L, et al. A method and server
for predicting damaging missense mutations. Nat Methods.
2010;7:248-249.
12. Tavtigian SV, Deffenbaugh AM, Yin L, et al. Comprehensive statistical study of 452 BRCA1 missense substitutions with classification of eight recurrent substitutions as neutral. J Med Genet.
2006;43:295-305.
13. Kumar P, Henikoff S, Ng PC. Predicting the effects of coding
non-synonymous variants on protein function using the SIFT
algorithm. Nat Protoc. 2009;4:1073-1082.
14. Fugazzola L, Muzza M, Mian C, et al. 2008 RET genotypes in
sporadic medullary thyroid carcinoma and R770Q-in a patient
with mixed medullar/follicular thyroid tumour. Exp Clin Endocrinol Diabetes. 2010;118:550-553.
15. Greger V, Passarge E, Hopping W, Messmer E, Horsthemke B.
Epigenetic changes may contribute to the formation and spontaneous regression of retinoblastoma. Hum Genet. 1989;83:155-158.

Downloaded from oto.sagepub.com at Airlangga University on September 4, 2015