The Protein Kinase Complement of the Human Genome

Abstract

We have catalogued the protein kinase complement of the human genome (the "kinome") using public and proprietary genomic, complementary DNA, and expressed sequence tag (EST) sequences. This provides a starting point for comprehensive analysis of protein phosphorylation in normal and disease states, as well as a detailed view of the current state of human genome analysis through a focus on one large gene family. We identify 518 putative protein kinase genes, of which 71 have not previously been reported or described as kinases, and we extend or correct the protein sequences of 56 more kinases. New genes include members of well-studied families as well as previously unidentified families, some of which are conserved in model organisms. Classification and comparison with model organism kinomes identified orthologous groups and highlighted expansions specific to human and other lineages. We also identified 106 protein kinase pseudogenes. Chromosomal mapping revealed several small clusters of kinase genes and revealed that 244 kinases map to disease loci or cancer amplicons.

Full text

DOI: 10.1126/science.1075762
, 1912 (2002); 298Science et al.G. Manning,
Genome
The Protein Kinase Complement of the Human
www.sciencemag.org (this information is current as of November 15, 2007 ):
The following resources related to this article are available online at
http://www.sciencemag.org/cgi/content/full/298/5600/1912
version of this article at: including high-resolution figures, can be found in the onlineUpdated information and services,
http://www.sciencemag.org/cgi/content/full/298/5600/1912/DC1
can be found at: Supporting Online Material
found at: can berelated to this articleA list of selected additional articles on the Science Web sites
http://www.sciencemag.org/cgi/content/full/298/5600/1912#related-content
http://www.sciencemag.org/cgi/content/full/298/5600/1912#otherarticles
, 13 of which can be accessed for free: cites 37 articlesThis article
926 article(s) on the ISI Web of Science. cited byThis article has been
http://www.sciencemag.org/cgi/content/full/298/5600/1912#otherarticles
96 articles hosted by HighWire Press; see: cited byThis article has been
http://www.sciencemag.org/cgi/collection/cell_biol
Cell Biology : subject collectionsThis article appears in the following
http://www.sciencemag.org/about/permissions.dtl
in whole or in part can be found at: this article permission to reproduce of this article or about obtaining reprintsInformation about obtaining
registered trademark of AAAS. is aScience2002 by the American Association for the Advancement of Science; all rights reserved. The title CopyrightAmerican Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005.
(print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by theScience
on November 15, 2007 www.sciencemag.orgDownloaded from

ERK2 pathways, which contributes to the
increased proliferative rate of tumor cells. For
this reason, inhibitors of the ERK pathways
are entering clinical trials as potential anti-
cancer agents. In differentiated cells, ERKs
have different roles and are involved in re-
sponses such as learning and memory in the
central nervous system.
JNK1, JNK2, and JNK3
The JNKs were isolated and characterized as
stress-activated protein kinases on the basis of
their activation in response to inhibition of pro-
tein synthesis (8). The JNKs were then discov-
ered to bind and phosphorylate the DNA bind-
ing protein c-Jun and increase its transcriptional
activity. c-Jun is a component of the AP-1
transcription complex, which is an important
regulator of gene expression. AP-1 contributes
to the control of many cytokine genes and is
activated in response to environmental stress,
radiation, and growth factors — all stimuli that
activate JNKs. Regulation of the JNK pathway
is extremely complex and is influenced by
many MKKKs. As depicted in the STKE JNK
Pathway Connections Map, there are 13
MKKKs that regulate the JNKs. This diversity
of MKKKs allows a wide range of stimuli to
activate this MAPK pathway. JNKs are impor-
tant in controlling programmed cell death or
apoptosis (9). The inhibition of JNKs enhances
chemotherapy-induced inhibition of tumor cell
growth, suggesting that JNKs may provide a
molecular target for the treatment of cancer.
JNK inhibitors have also shown promise in
animal models for the treatment of rheumatoid
arthritis (10). The pharmaceutical industry is
bringing JNK inhibitors into clinical trials for
both diseases.
p38 Kinases
There are four p38 kinases: ␣,␤,␥, and ␦. The
p38␣enzyme is the most well characterized
and is expressed in most cell types. The p38
kinases were first defined in a screen for drugs
inhibiting tumor necrosis factor ␣–mediated in-
flammatory responses (11). The p38 MAPKs
regulate the expression of many cytokines. p38
is activated in immune cells by inflammatory
cytokines and has an important role in activa-
tion of the immune response. p38 MAPKs are
activated by many other stimuli, including hor-
mones, ligands for G protein–coupled recep-
tors, and stresses such as osmotic shock and
heat shock. Because the p38 MAPKs are key
regulators of inflammatory cytokine expres-
sion, they appear to be involved in human
diseases such as asthma and autoimmunity.
Recently, a major paradigm shift for
MAPK regulation was developed for p38␣.
The p38␣enzyme is activated by the protein
TAB1 (12), but TAB1 is not a MKK. Rather,
TAB1 appears to be an adaptor or scaffolding
protein and has no known catalytic activity.
This is the first demonstration that another
mechanism exists for the regulation of
MAPKs in addition to the MKKK-MKK-
MAPK regulatory module. This important
observation indicates that other adaptor pro-
teins should be scrutinized for potential roles
in regulating MAPK activity.
The importance of MAPKs in controlling
cellular responses to the environment and in
regulating gene expression, cell growth, and
apoptosis has made them a priority for re-
search related to many human diseases. The
ERK, JNK, and p38 pathways are all molec-
ular targets for drug development, and inhib-
itors of MAPKs will undoubtedly be one of
the next group of drugs developed for the
treatment of human disease (13).
References and Notes
1. C. Widmann et al., Physiol. Rev.79, 143 (1999).
2. L. B. Ray, T. W. Sturgill, Proc. Natl. Acad. Sci. U.S.A.
85, 3753 (1988).
3. G. L. Johnson, R. Lapadat, ERK Pathway, Science’s
STKE (Connections Map, as seen November 2002),
http://stke.sciencemag.org/cgi/cm/stkecm;
CMP_10705.
4. 㛬㛬㛬㛬, JNK Pathway, Science’s STKE (Connections
Map, as seen November 2002), http://stke-
.sciencemag.org/cgi/cm/stkecm;CMP_10827.
5. 㛬㛬㛬㛬, p38 Pathway, Science’s STKE (Connections
Map, as seen November 2002), http://stke-
.sciencemag.org/cgi/cm/stkecm;CMP_10958.
6. G. Zhou et al., Biol. Chem.270, 12665 (1995).
7. M. K. Abe et al., Mol. Cell. Biol. 19, 1301 (1999).
8. J. M. Kyriakis et al., Nature 369, 156 (1994).
9. C. Tournier et al., Science 288, 870 (2000).
10. Z. Han et al.,J. Clin. Invest.108, 73 (2001).
11. J. C. Lee et al., Nature 372, 739 (1994).
12. B. Ge et al., Science 295, 1291 (2002).
13. J. M. English, M. H. Cobb, Trends Pharmacol. Sci. 23,
40 (2002).
REVIEW
The Protein Kinase Complement
of the Human Genome
G. Manning,
1
* D. B. Whyte,
1
R. Martinez,
1
T. Hunter,
2
S. Sudarsanam
1,3
We have catalogued the protein kinase complement of the human genome (the
“kinome”) using public and proprietary genomic, complementary DNA, and
expressed sequence tag (EST) sequences. This provides a starting point for
comprehensive analysis of protein phosphorylation in normal and disease
states, as well as a detailed view of the current state of human genome analysis
through a focus on one large gene family. We identify 518 putative protein
kinase genes, of which 71 have not previously been reported or described as
kinases, and we extend or correct the protein sequences of 56 more kinases.
New genes include members of well-studied families as well as previously
unidentified families, some of which are conserved in model organisms. Clas-
sification and comparison with model organism kinomes identified orthologous
groups and highlighted expansions specific to human and other lineages. We
also identified 106 protein kinase pseudogenes. Chromosomal mapping re-
vealed several small clusters of kinase genes and revealed that 244 kinases map
to disease loci or cancer amplicons.
Ever since the discovery nearly 50 years ago
that reversible phosphorylation regulates the ac-
tivity of glycogen phosphorylase, there has
been intense interest in the role of protein phos-
phorylation in regulating protein function. With
the advent of DNA cloning and sequencing in
the mid-1970s, it rapidly became clear that a
large family of eukaryotic protein kinases ex-
ists, and the burgeoning numbers of protein
kinases led to the speculation that a vertebrate
genome might encode as many as 1001 protein
kinases (1). The near-completion of the human
genome sequence now allows the identification
of almost all human protein kinases. The total
(518) is about half that predicted 15 years ago,
but it is still a strikingly large number, consti-
tuting about 1.7% of all human genes.
Protein kinases mediate most of the signal
transduction in eukaryotic cells; by modifica-
tion of substrate activity, protein kinases also
control many other cellular processes, includ-
ing metabolism, transcription, cell cycle pro-
gression, cytoskeletal rearrangement and cell
movement, apoptosis, and differentiation.
Protein phosphorylation also plays a critical
1
SUGEN Inc., 230 East Grand Avenue, South San
Francisco, CA 94080, USA.
2
Salk Institute, 10010
North Torrey Pines Road, La Jolla, CA 92037, USA.
3
Genomics and Biotechnology, Pharmacia Corpora-
tion, 230 East Grand Avenue, South San Francisco, CA
94080, USA.
*To whom correspondence should be addressed. E-
mail: gerard-manning@sugen.com
VIEWPOINT
6 DECEMBER 2002 VOL 298 SCIENCE www.sciencemag.org1912
on November 15, 2007 www.sciencemag.orgDownloaded from

role in intercellular communication during
development, in physiological responses and
in homeostasis, and in the functioning of the
nervous and immune systems. Protein ki-
nases are among the largest families of genes
in eukaryotes (2– 6) and have been intensive-
ly studied. As such, they made an attractive
target for an initial in-depth analysis of the
gene distribution in the draft human genome.
Mutations and dysregulation of protein ki-
nases play causal roles in human disease,
affording the possibility of developing ago-
nists and antagonists of these enzymes for use
in disease therapy (7–9). A complete catalog
of human protein kinases will aid in the
discovery of human disease genes and in the
development of therapeutics.
Comprehensive Discovery of Protein
Kinase Genes
Most protein kinases belong to a single su-
perfamily containing a eukaryotic protein ki-
nase (ePK) catalytic domain. We set out to
identify all sequenced human ePKs by
searching every available human sequence
source ( public and Celera genomic databas-
es, Incyte ESTs, in-house and GenBank
cDNAs and ESTs) with a hidden Markov
model (HMM) profile of the ePK domain
(10). This profile is sensitive enough to detect
short fragments of even very divergent ki-
nases that have little similarity to any single
known kinase. We extended these fragments
to full-length gene predictions using a com-
bination of EST and cDNA data, Genewise
homology modeling, and Genscan ab initio
gene prediction; more than 90% of the new
and extended sequences were verified by
cDNA cloning. We also identified 13 atypical
protein kinase (aPK) families. These contain
proteins reported to have biochemical kinase
activity, but which lack sequence similarity
to the ePK domain, and their close ho-
mologs (10). Some aPKs have structural
similarity to ePK domains (11). New aPKs
were identified with the use of additional
HMMs and Psi-Blast.
How Many Protein Kinases in the
Genome?
We identified 478 human ePKs and 40 aPK
genes (Table 1 and Fig. 1) (table S1). Of
these 518 protein kinases, 24 are absent from
the public Genpept database, and 47 more are
published only as hypothetical proteins or are
not described as kinases. Many more are
annotated only by automatic methods, or are
fragmentary sequences and have not been
individually studied. Most new kinases come
from new and little-studied families, as tar-
geted cloning has previously identified most
members of well-known families. However,
new members were found even in some of the
best studied kinase families. One new mem-
ber of the cyclin-dependent kinase (CDK)
family was found: CDK11 is a close paralog
of CDK8 (91% protein sequence identity for
Fig. 1. Dendrogram of 491 ePK domains from 478 genes. Major groups (Table 1) are labeled and
colored. For group-specific and comparative genomic trees, see www.kinase.com/human/kinome.
Table 1. Kinase distribution by major groups in human and model systems. A detailed classification is available in tables S1 and S6.
Group Families Subfamilies Yeast
kinases
Worm
kinases
Fly
kinases
Human
kinases
Human
pseudogenes
Novel
human
kinases
AGC 14 21 17 30 30 63 6 7
CAMK 17 33 21 46 32 74 39 10
CK1 3 5 4851012 5 2
CMGC 8 24 21 49 33 61 12 3
Other 37 39 38 67 45 83 21 23
STE 3 13 14251847 6 4
Tyrosine kinase 30 30 0 90 32 90 5 5
Tyrosine
kinase–like
7 13 0 15 17 43 6 5
RGC 1 1 0 27 6 5 3 0
Atypical-PDHK 1 1 2 1 1 5 0 0
Atypical-Alpha 1 2 0 4 1 6 0 0
Atypical-RIO 1 3 2 3 3 3 1 2
Atypical-A6 1 1 1 2 1 2 2 0
Atypical-Other 7 7 2 1 2 9 0 4
Atypical-ABC1 1 1 3 3 3 5 0 5
Atypical-BRD 1 1 0 1 1 4 0 1
Atypical-PIKK 1 6 5 5 5 6 0 0
Total 134 201 130 454 240 518 106 71
REVIEW
www.sciencemag.org SCIENCE VOL 298 6 DECEMBER 2002 1913
on November 15, 2007 www.sciencemag.orgDownloaded from

most of their length), a kinase that interacts
with cyclin C and RNA polymerase II (12). A
CDK11 ortholog exists in mouse, but fly
(Drosophila melanogaster), worm (Caeno-
rhabditis elegans), and yeast (Saccharomyces
cerevisiae) have only a single member of this
CDK8/CDK11 family. The Nek (NimA-
related kinase) family is also thought to have
a role in the cell cycle; we discovered four
new Neks to bring the human total to 11 Nek
kinases. Within the mitogen-activated protein
kinase (MAPK) cascade, we found two new
Ste11/MAP3K (MAP kinase kinase kinase)
and two new Ste20/MAP4K (MAP kinase
kinase kinase kinase) genes, all of which have
restricted expression that may explain their
failure to be previously cloned. For instance,
only 14 ESTs are known from MAP3K8, and
all but one derive from testis, lung, or brain
libraries, indicating that these new genes may
have evolved to mediate specialized roles in
selected tissues.
Classification and Phylogeny of the
Human Kinome
To compare related kinases in human and
model organisms and to gain insights into
kinase function and evolution, we classified
all kinases into a hierarchy of groups, fami-
lies, and subfamilies. This extends the Hanks
and Hunter (13) human kinase classification
of five broad groups, 44 families, and 51
subfamilies by adding four new groups, 90
families, and 145 subfamilies (Table 1 and
Fig. 1) (table S1). Kinases were classified
primarily by sequence comparison of their
catalytic domains (10), aided by knowledge
of sequence similarity and domain structure
outside of the catalytic domains, known bio-
logical functions, and a similar classification
of the yeast, worm, and fly kinomes (4).
Of the four new groups, STE consists of
MAPK cascade families (Ste7/MAP2K, Ste11/
MAP3K, and Ste20/MAP4K). The CK1 group
contains CK1, TTBK (tau tubulin kinase), and
VRK (vaccinia-related kinase) families. TKL
(tyrosine kinase–like) is a diverse group of fam-
ilies that resemble both tyrosine and serine-
threonine kinases. It consists of the MLK
(mixed-lineage kinase), LISK (LIMK/TESK),
IRAK [interleukin-1 (IL-1) receptor–associated
kinase], Raf, RIPK [receptor-interacting protein
kinase (RIP)], and STRK (activin and TGF-␤
receptors) families. Members of the RGC (re-
ceptor guanylate cyclase) group are also similar
in domain sequence to tyrosine kinases.
Phylogenetic comparison of the human ki-
nome with those of yeast, worm, and fly (4)
confirms that most kinase families are shared
among metazoans and defines classes that are
expanded in each lineage. Of 189 subfamilies
present in human, 51 are found in all four
eukaryotic kinomes, and these presumably
serve functions essential for the existence of a
eukaryotic cell. An additional 93 subfamilies
are present in human, fly, and worm, implying
that these evolved to fulfill distinct functions in
early metazoan evolution. Comparison with the
draft mouse genome indicates that more than
95% of human kinases have direct orthologs in
mouse; additional orthologs may emerge as that
genome sequence is completed.
The functions of human kinases can be
inferred from family members in model
organisms. For instance, the BRSK (brain-
selective kinase) family has two uncharacter-
ized human members that are selectively ex-
pressed in brain. They are orthologous to
worm SAD-1, which has a role in presynaptic
vesicle clustering (14), suggesting a con-
served function. A highly conserved ascidian
(chordate) homolog is also expressed in neu-
ral tissue and is asymmetrically localized to
the posterior end of the embryo, suggesting a
second role in embryonic axis determination
(15). Conversely, we identified four families
with orthologs in human, fly, and worm
where no functional data are available for any
member. Their phylogenetic distribution
hints at roles fundamental to metazoan biol-
ogy of which we are still ignorant.
The human genome has approximately twice
as many kinases as those of fly or worm, after
idiosyncratic worm-specific expansions are
trimmed (4). Accordingly, most kinase families
have twice as many human members as they
have in worm or fly. However, the expansion is
not uniform: 25 subfamilies—including CDK5,
CDK9, and Erk7—have just one member in
each organism, indicating critical unduplicated
functions. Conversely, substantial human ex-
pansions occurred in several families, with the
most striking example being Eph family recep-
tor tyrosine kinases (RTKs), where there are 14
genes in human and only 1 in fly and worm
(Table 2). These expanded families function
predominantly in processes that are more ad-
vanced in human, such as the nervous and im-
mune systems, angiogenesis, and hemopoiesis,
as well as functions that are less obviously
enhanced, such as apoptosis, MAPK signaling,
calmodulin-dependent signaling, and epidermal
growth factor (EGF) signaling.
Fourteen families are found only in hu-
man. The Tie family of RTKs are expressed
in endothelial cells and function in angiogen-
esis, and the Axl RTKs (Axl, Mer, and Ty-
ro3) function in both hemopoietic and neural
tissues. The Trio and RIPK families have
invertebrate homologs that lack kinase do-
mains. They are involved in muscle function
and apoptotic signaling via tumor necrosis
factor (TNF), Fas, and NF-␬B, respectively.
Lmr, NKF3, NKF4, NKF5, and HUNK are
novel families whose functions are largely
unknown, and BCR, FAST, G11, H11, and
DNAPK are atypical kinases.
The human expansions of many of these
families can be traced both to large duplica-
tions of multigene loci (“paralogons”) and to
local tandem duplications of smaller loci of-
ten containing just one gene. This supports
recent findings that vertebrate genome com-
plexity may derive from ancient large-scale
duplications as well as a continuing series
of smaller scale duplications (16 –18). For
instance, each of the four human epidermal
growth factor receptors (EGFRs) maps
close to one of the four HOX clusters,
implying that the proposed double duplica-
tion of that cluster early in vertebrate evo-
lution created the EGFR family from a
single ancestral EGFR gene (19). Similarly,
the eight genes of the VEGFR and PDGFR
(vascular endothelial growth factor and
platelet-derived growth factor receptors)
families map to three of the four paraHOX
clusters, and they probably derive from
duplications of the single ancestral paraHOX
locus as well as local duplications within the
paraHOX loci (table S3). The common an-
cestry of PDGFR and VEGFR families is
supported by the Drosophila kinome, which
contains two genes whose sequences are in-
termediate between those two families (4).
We mapped all kinase genes to chromo-
somal loci to look for origins of kinase ex-
pansions and to link kinases with known
disease loci. The map was created using the
Celera and public genome assemblies and
literature references (table S2). Although the
overall kinase distribution is similar in den-
sity to that of other genes, many pairs of
closely related genes from the same families
map closer to each other than expected by
chance, indicating that they may have arisen
through local chromosomal duplications (ta-
ble S3). Seven pairs are within 30 kb of each
other, all in tandem orientation. Another six
pairs are within 1 Mb of each other, and 15
more within 10 Mb. In all, 66 genes map
unusually near to close paralogs, indicating
that at least 6% of kinases may have arisen by
local duplications. Most of these genes are
from families that are highly expanded in
human compared with worm and fly, further
supporting a recent origin. The multigene
duplications are thought to have arisen most-
ly during early vertebrate evolution, but some
local duplications may also have happened at
this time. For instance, the clustering of
PDGFR␤and CSF-1 receptor (c-fms) genes
is conserved in pufferfish (20).
Chromosomal Mapping and Disease
The knowledge of the exact chromosomal lo-
cations of genes afforded by the complete hu-
man genome assemblies is increasingly valu-
able in pinpointing candidate disease genes
within loci that are associated with specific
diseases. Comparison of the kinase chromo-
somal map with known disease loci indicates
that 164 kinases map to amplicons seen fre-
quently in tumors (21) and 80 kinases map to
loci implicated in other major diseases (table
REVIEW
6 DECEMBER 2002 VOL 298 SCIENCE www.sciencemag.org1914
on November 15, 2007 www.sciencemag.orgDownloaded from

S2). Although each locus covers many genes,
these data provide entry points for studying
both the function of these kinases and their
potential as the causative principle of these
diseases. The role of kinases as biological
control points and their tractability as drug
targets make them attractive targets for dis-
ease therapy.
Catalytically Inactive Kinases
Several ePK domains are known to lack kinase
activity experimentally, and these have been
postulated to act as kinase substrates and scaf-
folds for assembly of signaling complexes (22–
24). Our sequence analysis shows that 50 hu-
man kinase domains lack at least one of the
conserved catalytic residues (Lys
30
, Asp
125
,
and Asp
143
) (table S5) and are predicted to be
enzymatically inactive. Twenty-eight inactive
kinases belong to families where all members
are inactive in human, fly, and worm, and even
in yeast. Thus, surprisingly, nearly 10% of all
kinase domains appear to lack catalytic activity.
However, these domains are otherwise well
conserved and are likely to maintain the typical
kinase domain fold. This suggests that this do-
main can have generalized noncatalytic func-
tions; it is also possible that they use a modified
catalytic mechanism that does not require these
residues. This has been shown for the Wnk
family, where Lys
13
is thought to replace Lys
30
in adenosine triphosphate (ATP) binding (25).
The 50 “inactive”kinase domains fall into
three main categories. First are domains that
may act as modulators of other catalytic do-
mains. GCN2 and JAK ( Janus kinase) family
kinases have dual ePK domains, one of which
is inactive and may regulate the active do-
main (26). Similarly, the inactive ePK do-
main of receptor guanylate cyclases (RGCs)
is thought to regulate the activity of the
neighboring guanylate cyclase domain, in a
manner that is modulated by ATP binding
and phosphorylation (27).
Second are other kinases with high similar-
ity to the canonical ePK domain profile. These
include the Ras pathway scaffold proteins KSR
(kinase suppressor of Ras) (23) and the previ-
ously undescribed KSR2, titin, ILK (integrin-
linked kinase), PSKH2 ( protein serine kinase
H2), and unpublished kinases from the STLK
and Trbl families. The scaffold protein CASK
(calcium/calmodulin-dependent serine kinase)
contains an inactive protein kinase domain and
an inactive guanylate kinase domain, both of
which act as protein-protein interaction do-
mains (28, 29). This group also contains several
RTKs where an inactive kinase may dimerize
with and act as a substrate of another RTK:
Ryk, CCK4, the ephrin receptors EphA10 and
EphB6, and ErbB3 (24).
Third is a group whose members have
very weak similarities to the kinase domain
profile, and may have quite divergent func-
tions. Of 37 “weak”kinase domains (whose
kinase HMM E-value score is greater than
1e-30), 26 lack one or more catalytic resi-
dues. Note, however, that other weakly scor-
ing kinases have been shown experimentally
to have catalytic activity, including Bub1 (e-
11 E value), VRK1 (e-10), PRPK (e-5), and
haspin (e-3) (30 –33).
Other Functional Domains in Protein
Kinases
Most protein kinases act in a network of
kinases and other signaling effectors, and are
modulated by autophosphorylation and phos-
phorylation by other kinases. Other domains
within these proteins regulate kinase activity,
link to other signaling modules, or subcellu-
Table 2. Kinase families expanded in human relative to those in fly and worm. See table S6 for more details.
Function Family Human Fly Worm Notes
Immunology, hemopoiesis, JAK 4 1 0 Couple cytokine receptors to transcription
angiogenesis PDGFR/VEGFR 8 2 0 Angiogenesis, vascular growth factor receptors
Tec 5 1 0 Nonreceptor tyrosine kinase
Src 11 2 3 Nonreceptor tyrosine kinase
IRAK 4 1 1 IL-1 receptor–associated kinase
Tie 2 0 0 Tie and Tek RTKs
IKK 4 2 0 I␬B kinase, NF-␬B signaling
RIPK 5 0 0 Receptor-interacting protein kinase, NF-␬B signaling
Axl 3 0 0 Immune system homeostasis
Neurobiology Eph 14 1 1 Ephrin receptors
Trk 3 0 0–1 Neurotrophin receptors
MAPK cascades Ste11 9 2 2 (MAP3K)
Ste20 31 13 12 (MAP4K)
Ste7 8 4 10 (MAP2K) Has distinct worm-specific expansion
Apoptosis DAPK 5 1 1 Death-associated protein kinase family
RIPK 5 0 0 Transduces death signal from TNF-␣receptor
Lmr 3 0 0 Lmr1, aka apoptosis-associated tyrosine kinase (AATYK)
Calcium signaling CaMK1 5 1 1 Calmodulin (CaM)–regulated kinases
CaMK2 4 1 1 Calmodulin (CaM)–regulated kinases
EGF signaling EGFR 4 1 1 Epidermal growth factor receptor family
RSK/RSK 4 1 1 Ribosomal protein S6 kinases; RSK1-3 activated by
MAPK in response to EGF
Tao 3 1 1 Tao3 activated by EGFR
Src 11 2 3 Src implicated in EGF signaling
HUNK 1 0 0 Hormonally up-regulated Neu-associated kinase
Other Trio 3 0 0 Fly and worm orthologs lack the kinase domain
Trbl 3 1 0 Unpublished homologs of Drosophila trbl
PDK 5 1 1 Mitochondrial pyruvate dehydrogenase kinases
HIPK 4 1 1 Homeodomain-interacting protein kinases
STKR 12 5 3 TGF-␤, Activin receptors
BRD 4 1 1 Bromodomain-containing atypical kinases
Wnk 4 1 1 Implicated in hypertension
NKF3 2 0 0 Uncharacterized (new kinase family 3)
NKF4 2 0 0 Uncharacterized (new kinase family 4)
NKF5 2 0 0 Uncharacterized (new kinase family 5)
CDKL 5 1 1 Cyclin-dependent kinase-like
REVIEW
www.sciencemag.org SCIENCE VOL 298 6 DECEMBER 2002 1915
on November 15, 2007 www.sciencemag.orgDownloaded from

larly localize the protein. We identified 83
additional types of domain present in 258 of
the 518 kinases, using profiles from the Pfam
HMM collection (Table 3). In general, mem-
bers of the same kinase family have the same
domain structure, but some domain shuffling
is seen, where individual members of fami-
lies have gained or lost a domain and so may
have altered function. For instance, the death
domain is found in all four IRAK kinases as
well as in single members of the DAPK and
RIPK families.
The most common domains mediate inter-
actions with other signaling proteins: 24 kinases
contain Src homology 2 (SH2) domains that
bind to phosphotyrosine residues; other domains
link to small guanosine triphosphatase (GTPase)
signaling (38 kinases with RhoGEF, RhoGAP,
RBD, PBD, RGS, CNH, HR1, or TBC do-
mains), lipid signaling (42 kinases with
DAG_PE, C2, PX, or PH domains), and calci-
um signaling (28 kinases with CaM, IQ, or
OPR/PB1 domains); target the protein to the
cytoskeleton (seven kinases with spectrin, cofi-
lin, myosin head, or FCH domains); or mediate
interactions with other proteins (46 kinases:
Death, SH3, SAM, LIM, or ankyrin domains) or
RNA (three kinases with RRM, DSRM, and
putative RNA binding Tudor domains). Most of
the domains found in new or extended sequenc-
es are the same as those already seen in other
family members, but some unpredicted domains
are found, such as the previously unpublished
leucine-rich repeat kinase (LRRK) family, con-
taining arrays of leucine-rich repeats, as well as
armadillo and ankyrin repeats.
Most of the 58 RTKs, 12 receptor serine-
threonine kinases, and five receptor guanylate
cyclases also have recognizable ligand-binding
and other extracellular domains, along with
clear signal peptides and transmembrane re-
gions. Several nonreceptor tyrosine kinases are
also targeted to the membrane by lipidation or
protein-protein interactions. Three kinases are
targeted to the endoplasmic reticulum, five or
six are likely to be mitochondrial, and most of
the rest are thought to be cytoplasmic, nuclear,
or both.
Two hundred and sixty kinases contain no
additional Pfam domains. Many are small
proteins containing little more than an ePK
domain and may be controlled by additional
regulatory subunits, such as cyclins, which
control CDK activity. Others contain con-
served sequences that have not yet been clas-
sified as domains and whose functions are
unknown.
Thirteen kinases have dual ePK domains,
in which both domains appear to be active
[six ribosomal S6 kinase (RSK) family ki-
nases and two Trio family kinases] or the
second domain is inactive (the four JAK fam-
ily kinases and GCN2). The two RSK do-
mains are involved in a kinase relay: Erk
phosphorylates and activates the CAMK-
group domain of RSK2, leading to autophos-
phorylation on a linker region that then al-
lows PDK1 to phosphorylate and activate the
second AGC-group kinase domain (34).
Kinase Pseudogenes
The genome also contains many nonfunctional
copies of kinase genes that are not expressed or
encode degenerate, truncated proteins. These ki-
nase pseudogenes are derived mostly from ret-
roviral transposition and genomic duplications.
Pseudogenes can confuse gene predictions,
cross-hybridize with probes for functional
genes, and contribute to disease by homologous
recombination with their parental genes (35,
36). We identified 106 pseudogenes containing
similarity to the ePK domain or to an aPK (table
S4); several other pseudogene fragments that
lack a kinase domain were found but are not
included here. All but two pseudogenes
have open reading frames (ORFs) interrupt-
ed by stop codons or frameshifts, which
were verified by multiple independent se-
quence sources. These ORFs typically have
high protein sequence similarity to a func-
tional (“parent”) kinase; most are partial
gene fragments. The two putative pseudo-
genes with complete ORFs (CK2a-rs and
STLK6-rs) lack introns and obvious pro-
moters, are absent from EST databases,
have ⬎98.5% DNA sequence identity to
their parents, and contain remnants of
polyA tails in their genomic sequences.
They are probably young processed pseudo-
genes whose sequences have not yet
diverged.
Seventy-five kinase pseudogenes lack in-
trons. Some are duplications of intronless genes
Table 3. Most common Pfam domains in protein kinases. See table S7 for a fuller listing.
Domain name Number
of genes
Number
of
domains
Function class
Protein kinase C terminal domain 44 44 Accessory domain
Immunoglobulin domain (Ig) 30 254 Extracellular, protein interactions
Fibronectin type III domain (FnIII) 28 194 Extracellular, protein interactions
SH2 domain 25 27 Adaptor: Binds phosphotyrosine
SH3 domain 27 28 Adaptor: Binds proline-rich
motifs
PH domain 23 22 Signaling; phospholipid binding
Diacylglycerol binding (C1, DAG_PE) 23 33 Phospholipid binding
Calmodulin binding motif 23 25 Not in Pfam. From literature and
sequence alignment
SAM domain (Sterile alpha motif) 15 16 Dimerization domain
Ephrin receptor ligand binding
domain
14 14 Ligand binding
CNH domain 12 12 Cytoskeletal?
HEAT, armadillo/␤-catenin repeats 10 27 Protein interaction
Activin receptor 11 11 Ligand binding
Ankyrin repeat (ANK) 9 59 Protein interaction
Regulator of G protein signaling
(RGS)
7 7 GTPase interaction
PDZ/DHR/GLGF domain 7 7 Membrane targeting
Ubiquitin-associated domain A
(UBA)
7 8 Protein degradation
Receptor L domain 7 14 Ligand binding
Furin-like cysteine rich region 7 21 Receptor dimerization?
p21-Rho-binding domain (PBD,
CRIB)
9 9 GTPase interaction
Phosphatidylinositol 3⬘-kinase (PI3K) 6 6 Catalytic: Protein kinase
FAT 6 6 Accessory domain for PI3K
FATC 6 6 Accessory domain for PI3K
Alpha kinase 6 6 Catalytic: Atypical kinase
C2 domain 6 6 Ca
2⫹
, phospholipid binding
Guanylate cyclase catalytic domain 5 5 Catalytic: cGMP production
HSP90-like ATPase 5 5 Catalytic: Atypical kinase
ANF receptor 5 5 Ligand binding
Kinase-associated domain 1 (KA1) 5 5 Unknown
Bromodomain 8 13 Acetyl-lysine (chromatin) binding
domain
HR1 repeat 5 13 GTPase interaction
Leucine-rich repeat 5 30 Ligand binding, protein
interaction
ABC1 family 5 5 Catalytic: Atypical kinase
Death domain 6 6 Dimerization domain
BTK motif 4 4 Signaling
RhoGEF domain 4 5 GTPase interaction (guanine
exchange factor)
REVIEW
6 DECEMBER 2002 VOL 298 SCIENCE www.sciencemag.org1916
on November 15, 2007 www.sciencemag.orgDownloaded from

or of single exons of larger genes, but most
appear to derive from viral retrotransposition of
a processed transcript. Additionally, some in-
tron-containing pseudogenes such as AurAps2
contain some parental introns but lack others,
and may result from retrotransposition of a
partially spliced transcript.
Twenty-nine kinase pseudogenes contain
clear introns and probably arose by genomic
duplication. In some cases, these are part of a
large duplicon (2, 5) containing multiple du-
plicated genes. Such cases include two p70
ribosomal protein S6 kinase ( p70S6K) pseu-
dogenes, which appear to arise from intrach-
romosomal duplications of the p70S6K locus.
These duplications are 20 kb and 70 kb in
length, and are 90 to 95% identical in DNA
sequence to the original locus.
A few pseudogenes have no obvious hu-
man parent but have functional orthologs in
rodents and probably indicate the decay of
previously functional genes. They include the
polo-like kinase SGK384ps, whose mouse
ortholog is intact, and the human orthologs of
rat guanylate cyclases CGD and KSGC.
Although pseudogenes appear to be evo-
lutionary relicts, some may have some resid-
ual or cryptic function. Many pseudogenes
are transcribed: 26 kinase pseudogenes are
seen in cDNA and EST databases (table S4),
some represented by as many as 50 ESTs.
The prevalence of pseudogenes varies great-
ly between kinase families (Table 1) (table S4).
The MARK (microtubule affinity-regulating ki-
nase) family kinases displays the largest ratio of
pseudogenes to functional genes (28/4), fol-
lowed by p70S6K (4/1), Erk3 (4/1), phospho-
rylase kinase ␥1 (3/1), and casein kinase 1␣
(3/1). Frequent copying of a gene by retroviral
insertion might indicate a functional role for the
gene in retroviral function, but no viral function
or source for MARK genes is yet known.
Comparison with Sequence Databases
We compared our nonredundant set of cloned
and predicted kinase protein sequences with the
published predictions from Celera and public
genome projects (2, 5) and with a recent release
of the public GenPept database (10). Figure 2
shows the extent to which the best match in each
database agrees with our sequences. All three
databases contain at least fragments of most
kinases, but far fewer genes are in perfect agree-
ment. In many cases the public sequences come
from partial clones that lack the NH
2
-or
COOH-termini (43 and 15 genes, respectively),
often from large-scale sequencing projects that
do not individually annotate sequences. In other
cases, the public sequence has overextended the
true start site where upstream stop codons are
absent. We used similarity to rodent orthologs to
trim sequences to a strongly predicted transla-
tional start site in nine cases. Other discrepan-
cies come from sequencing errors, alternative
splicing, and sequencing of partially spliced
cDNAs. In all cases, our unique sequence is
supported by strong sequence similarity to ho-
mologs or by cDNA cloning.
In some cases, our additional sequence
greatly changes the predicted function of a
gene, such as the addition of a predicted
signal peptide to the Lmr1 tyrosine kinase;
the previously published form of this gene
(AATYK) was based on a cDNA lacking this
domain, which created a cytoplasmic protein
(37). We also identified full-length forms of
two related new genes, Lmr2 and Lmr3,
which together form a new family of predict-
ed receptor tyrosine kinases with vestigial
extracellular regions. Their biological roles
are currently under investigation.
Gene predictions from the public genome
project (Ensembl) and Celera differ from those
we obtained largely as a result of misprediction
of exon boundaries and splitting of single genes
into multiple predicted genes. Ensembl incor-
porates public sequence data from RefSeq and
Swiss-Prot, giving perfect agreement with our
sequences for many genes. The distance be-
tween the GenPept and Ensembl traces in Fig. 2
indicates the extent of recent new sequence
publication from large-scale cDNA sequencing
projects and individual cloning driven by
genomic data. The Celera predictions were en-
tirely computational, and so have very few per-
fect predictions. However, for genes not present
in public databases, many Celera predictions
agree better with our sequences than those from
Ensembl (not shown).
A comparison with “known” protein ki-
nases encounters several problems with over-
and under-classification of genes as kinases, as
well as with partial sequences. GenPept con-
tains multiple sequences for most kinases,
many of which are partial fragments or contain
multiple sequencing errors. It also contains chi-
meric genes such as the nonexistent zona pel-
lucida kinase (38). The proliferation of different
names for the same kinase adds to the problem
of creating an accurate nonredundant list of
kinases. Ensembl and Celera predictions in-
clude several pseudogenes (36 and 29, respec-
tively), and also annotate as kinases a number
of genes that are homologous to noncatalytic
regulatory subunits of protein kinase complexes
or to kinases other than protein kinases.
All 518 kinases are found in at least one of
the expressed sequence databases (dbEST,
Incyte, and GenBank cDNAs), indicating that
all are genuine, transcribed genes. Many ki-
nases are expressed in low amounts in a
restricted distribution, so the presence of all
kinases in EST or cDNA databases implies
that these databases contain fragments of
most human genes.
Summary
The sequencing of the human genome has pro-
vided a starting point for the identification of
most, if not all, human members of the eukary-
otic protein kinase superfamily, and many atyp-
ical kinases. We used the published human
genome sequences, combined with other se-
quence databases and directed cloning and se-
quencing of individual genes to discover, ex-
tend, or correct 125 kinase gene sequences, and
define a nonredundant set of 518 human protein
kinase genes. This set accounts for almost all
human protein phosphorylation and collective-
ly mediates most cellular signal transduction
and many other processes. Comparative se-
Fig. 2. Comparison of our kinase protein sequences with the best matches in Celera, Ensembl, and
GenPept databases. Each point shows the number of genes for which the percentage difference
between our sequence and the database is greater than the value indicated. Insert table indicates
number of sequences where differences between our sequence and closest database match is ⬎2%,
⬎50%, or ⬎95%.
REVIEW
www.sciencemag.org SCIENCE VOL 298 6 DECEMBER 2002 1933
on November 15, 2007 www.sciencemag.orgDownloaded from

quence analysis and mapping predict function
and possible disease association for many
kinases, and give clues to their evolutionary
origin. Comprehensive kinome-scale ap-
proaches are now feasible, including RNA
and protein expression profiling, and high-
throughput functional assays using constitu-
tively active and dominant-negative kinase
constructs. These will facilitate the study of
the role of kinases in a wide range of biolog-
ical processes, and the development of selec-
tive inhibitors and activators for research and
therapeutic purposes.
This large and well-curated sequence set
also casts a light on the current state of
human genome analysis. All 518 genes are
covered by some EST sequence, and ⬃90%
are present in gene predictions from the
Celera and public genome databases, al-
though those predictions are often fragmen-
tary or inaccurate and are frequently misan-
notated (39).
References and Notes
1. T. Hunter, Cell 50, 823 (1987).
2. E. S. Lander et al., Nature 409, 860 (2001).
3. G. M. Rubin et al., Science 287, 2204 (2000).
4. G. Manning, G. Plowman, T. Hunter, S. Sudarsanam,
Trends Biochem. Sci. 27, 514 (2002).
5. J. C. Venter et al., Science 291, 1304 (2001).
6. T. Hunter, G. D. Plowman, Trends Biochem Sci. 22,18
(1997).
7. P. Blume-Jensen, T. Hunter, Nature 411, 355 (2001).
8. T. Hunter, Cell 100, 113 (2000).
9. P. Cohen, Nature Rev. Drug Discovery 1, 309 (2002).
10. See supporting data on Science Online and at www.
kinase.com/human/kinome.
11. H. Yamaguchi, M. Matsushita, A. C. Nairn, J. Kuriyan,
Mol. Cell 7, 1047 (2001).
12. P. Rickert et al., Oncogene 12, 2631 (1996).
13. S. K. Hanks, T. Hunter, FASEB J. 9, 576 (1995).
14. J. G. Crump, M. Zhen, Y. Jin, C. I. Bargmann, Neuron
29, 115 (2001).
15. Y. Sasakura, M. Ogasawara, K. W. Makabe, Mech. Dev.
76, 161 (1998).
16. A. McLysaght, K. Hokamp, K. H. Wolfe, Nature Genet.
31, 200 (2002).
17. X. Gu, Y. Wang, J. Gu, Nature Genet. 31, 205 (2002).
18. L. Abi-Rached, A. Gilles, T. Shiina, P. Pontarotti, H.
Inoko, Nature Genet. 31, 100 (2002).
19. J. Spring, Nature Genet. 31, 128 (2002).
20. G. F. How, B. Venkatesh, S. Brenner, Genome Res. 6,
1185 (1996).
21. S. Knuutila et al., Am. J. Pathol. 152, 1107 (1998).
22. C. G. Zervas, N. H. Brown, Curr. Biol. 12, R350 (2002).
23. D. K. Morrison, J. Cell Sci. 114, 1609 (2001).
24. M. Kroiher, M. A. Miller, R. E. Steele, Bioessays 23,69
(2001).
25. B. Xu et al., J. Biol. Chem. 275, 16795 (2000).
26. M. Chen et al., Mol. Cell. Biol. 20, 947 (2000).
27. M. Chinkers, D. L. Garbers, Science 245, 1392 (1989).
28. Y. Li, O. Spangenberg, I. Paarmann, M. Konrad, A.
Lavie, J. Biol. Chem. 277, 4159 (2002).
29. K. Tabuchi, T. Biederer, S. Butz, T. C. Sudhof, J. Neu-
rosci. 22, 4264 (2002).
30. H. Tanaka et al., J. Biol. Chem. 274, 17049 (1999).
31. S. Lopez-Borges, P. A. Lazo, Oncogene 19, 3656
(2000).
32. T. W. Seeley, L. Wang, J. Y. Zhen, Biochem. Biophys.
Res. Commun. 257, 589 (1999).
33. Y. Abe et al., J. Biol. Chem. 276, 44003 (2001).
34. M. Frodin, C. J. Jensen, K. Merienne, S. Gammeltoft,
EMBO J. 19, 2924 (2000).
35. B. S. Emanuel, T. H. Shaikh, Nature Rev. Genet. 2, 791
(2001).
36. B. Cormand, A. Diaz, D. Grinberg, A. Chabas, L. Vilage-
liu, Blood Cells Mol. Dis. 26, 409 (2000).
37. E. Gaozza, S. J. Baker, R. K. Vora, E. P. Reddy, Onco-
gene 15, 3127 (1997).
38. P. Bork, Science 271, 1431 (1996).
39. We wish to thank the dozens of kinase researchers at
SUGEN for their contributions to understanding the
kinome at many levels. We particularly thank G.
Plowman who guided the initial stages of the project,
S. Caenepeel for extensive sequence analysis of ki-
nases, and G. Charydczak for the computational sup-
port that made the genome mining possible. The
SUGEN sequencing group provided cDNA confirma-
tion of most predicted sequences. T.H. is a Frank and
Else Schilling American Cancer Society Research Pro-
fessor and serves on the Scientific Advisory Board of
SUGEN.
Supporting Online Material
www.sciencemag.org/cgi/content/full/298/5600/1912/
DC1
Materials and Methods
SOM Text
Tables S1 to S7
REVIEW
6 DECEMBER 2002 VOL 298 SCIENCE www.sciencemag.org1934
on November 15, 2007 www.sciencemag.orgDownloaded from