REVIEW ARTICLE
The subtle side to hypoxia inducible factor (HIFa) regulation
Rebecca L. Bilton and Grant W. Booker
Department of Molecular Biosciences, The University of Adelaide, Australia
The transcription factor hypoxia inducible factor a-subunit
(HIFa) is pivotal in the cellular response to the stress of
hypoxia. Post-translational modification of HIFa by
hydroxylase enzymes has recently been identified as a key
Ôoxygen sensingÕ mechanism within the cell. The absence of
the substrate oxygen prevents the hydroxylases from modi-
fying HIFa during hypoxia and allows dramatic up-regula-
tion of both HIFa protein stability and transcriptional
activation capability. In addition to this oxygen-dependent
response, increased HIFa protein levels and/or enhanced
transcriptional activity during normoxic conditions can be
stimulated by various receptor-mediated factors such as
growth-factors and cytokines (insulin, insulin-like growth
factor 1 or 2, endothelial growth factor, tumour necrosis
factor a, angiotensin-2). Oncogenes are also capable of
HIFa activation. This induction is generally less intense than
that stimulated by hypoxia and although not fully elucida-
ted, appears to occur via hypoxia-independent, receptor-
mediated signal pathways involving either phosphatidyl
-inositol-3-kinase/Akt or mitogen activated protein kinase
(MAPK) pathways, depending on the cell-type. Activation
of Akt increases HIFa protein synthesis in the cell and results
in increased HIFa protein and transcriptional activity.
MAPK also activates HIFa protein synthesis and addi-
tionally may potentiate HIF1a transcriptional activity via a
separate mechanism that does not necessarily require protein
stabilization. Here we review the mechanisms and function

involved in their activation of HIFa upon stimulus by
hypoxia are reviewed elsewhere [7,8].
As elucidation of the hypoxic HIFa signalling pathway
continues, another side to HIFa biology has quietly emerged.
Zelzer and coworkers [9] were the first to demonstrate that
the growth-factors insulin and insulin-like growth factor-1
(IGF-1) activate HIF1 and that this has subsequently been
shown to occur through pathways separate to that employed
by the classical hypoxic pathway (Fig. 1). The list of
Correspondence to G. Booker, Department of Molecular Biosciences,
The University of Adelaide, North Terrace, Adelaide,
SA 5005, Australia.
Fax: + 61 88303 4348, Tel.: + 61 88303 3090,
E-mail:
Abbreviations: Akt, serine/threonine kinase (also known as protein
kinase B); ARNT, aryl-hydrocarbon receptor nuclear translocation;
bHLH-PAS, basic helix-loop-helix period-ARNT-single-minded;
CBP, CREB binding protein; CO, carbon monoxide; C-TAD,
C-terminal transcriptional activation domain; EGF, epidermal growth
factor; eIF-4E, eukaryotic initiation factor 4E; 4E-BP1, eIF-4E
binding protein 1; FGF-2, fibroblast growth factor-2; FIH-1, factor
inhibiting HIF; FRAP, FKBP(FK506 binding protein) rapamycin
associated-binding protein (also known as mTOR, mammalian target
of rapamycin); HER2
NEU
, heregulin-2 or EGF stimulated receptor
tyrosine kinase; HGF, hepatocyte growth factor; HIFa,hypoxia
inducible factor-1 or ) 2 a subunit; HRE, hypoxic response element;
IGF-1/IGF-2, insulin-like growth factor-1 or -2; IL-1b, interleukin-1b;
JNK, c-Jun amino-terminal kinase; MAPK, mitogen activated protein

smooth muscle cells, where several stimulatory factors
increased the amount of protein observed to levels signifi-
cantly greater than those induced by hypoxia [13]. Whilst
apparently minor in comparison to the in vitro induction by
hypoxia, the gene expression changes resulting from the
receptor-mediated pathways are nonetheless important.
These stimuli often elicit small changes in housekeeping
functions that accumulate over extended periods of time [25].
Receptor-mediated HIFa regulation has been shown to
occur via two well characterized signalling pathways, the
Ras/MEK/MAPK and PI3K/Akt/FRAP kinase cascades
[24,26,27] (Fig. 3). Although the end result is enhanced
HIFa protein levels and/or transcriptional activation even
under normoxic conditions, the molecular mechanisms
involved must differ from those of hypoxia, as low oxygen
tension and activation of the MAPK and Akt pathways can
co-operate to enhance the induction of HIFa activity [20,
27–29]. There is some evidence to suggest that co-operation
of these kinase pathways with hypoxia may occur via the
hypoxic generation of reactive oxygen species (ROS) as an
intermediate signalling step [30] (reviewed in [31]). This may
also be the pathway by which some of the stimulatory
factors such as thrombin, angiotensin and IL-1b influence
HIFa, as their ability to activate HIF1a was blocked by
ROS inhibitors and antioxidants [13,14,32]. In this way,
Fig. 2. Fold induction of pTK-HRE luciferase reporter construct in
stably transfected 3T3L1 adipocyte cells. Treatments include dipyridyl
(100 n
M
), serum free, insulin (100 n

different pathways occurs in the hepatoma cell-line HepG2
where both TNFa and IL-1b were shown to increase DNA
binding by HIF, but only IL-1b was able to increase the
observed HIF1a protein levels [12]. This suggests a different
mechanism of action for each cytokine in this cell type. In
contrast to these findings, the two cytokines were reported
to act in the same manner to increase transcription of
HIF1a mRNA by twofold to threefold resulting in increased
protein levels in synovial fibroblasts [16]. Furthermore, the
differences in stimuli-induced signalling are highlighted by
the example of IGF-1 stimulation in different cell types.
IGF-1 stimulation allows visualization of HIF1a protein
and increases transcriptional activity of HIF1a via activa-
tion of MAPK in mouse embryo fibroblasts [35], whereas in
the U373 glioblastoma cell line these effects require Akt [20].
Both kinase pathways are reported to additively increase
HIF1a translation and thus protein level in HCT116 colon
carcinoma cells [26]. Finally, the family members HIF1a
and HIF2a are both responsive to receptor-mediated
stimuli, but not necessarily the same stimuli within a single
cell type, even when both homologues are coexpressed [29].
Whilst confusing, this complexity and cross-talk between
signalling pathways is not uncommon for growth factor
stimuli. Dependant on cell type and signal intensity,
stimulation of receptors by insulin can alternatively activate
either MAPK or Akt, and this can result in the completely
disparate outcomes of proliferation (mitogenic) or glucose
uptake (metabolic) [34].
Phosphorylation
Upon polyacrylamide gel electrophoresis, HIFa protein

reported by several groups. These researchers showed that
activated recombinant or endogenous MAPK was able to
phosphorylate either full-length HIFa or a C-TAD-fusion
product when supplied as a substrate [27,41,42]. In those
studies employing HIF1a-fusion proteins expressed in
COS-7 cells, the region targeted by MAPK was shown to
lie within residues 786–826 of the C-TAD [27] or residues
531–826 spanning both the inhibitory domain and C-TAD
[42]. Although Sodhi et al. [42] identified up to eight serine
residues within this inhibitory region that contain adjacent
proline residues that may serve as putative consensus target
sites for the MAPK family, the specific HIFa residue(s) that
are phosphorylated by MAPK have yet to be identified. A
proposed function for phosphorylation leading to increased
HIF transcriptional activity is through the derepression of
the inhibitory domain that lies between the two transcrip-
tional activation domains of the HIF a-subunit [42].
Regions within this inhibitory domain have been shown
to be important for the interaction of HIFa with factor
inhibiting HIF-1 (FIH-1) [43,44] identified as the asparagine
hydroxylase [45]. In the three-dimensional structure, these
regions may form part of the FIH-1 recognition site. One
possible explanation for the observed derepression of the
HIFa inhibitory domain, is that phosphorylation of
residues within this domain may prevent docking of FIH,
and thus prevent the subsequent asparagine hydroxylation.
This would result in a derepression of transcriptional
activity, as CBP/p300 would be able to associate with HIFa.
As mentioned previously, hypoxia is able to activate
MAPK in some cell lines [27–29]. However, activation of

improve HIF transcriptional activity by derepression of
the HIFa inhibitory domain or simply favouring a confor-
mation that supports the active domain. Given the varia-
tions encountered so far within the characterization of the
receptor-mediated signalling pathway, it is not surpri-
sing that, in a few cell types, the up-regulation of HIF
transcriptional activity via MAPK activity has been attri-
buted to an increase in observable HIF1a protein [22,24]. In
this way, MAPK may act via a similar mechanism to Akt to
improve HIFa protein synthesis. The effects of MAPK on
protein stability or transcriptional activation need not
necessarily be mutually exclusive.
Akt
The serine/threonine kinase Akt has also been identified as a
signalling intermediate downstream of the receptor-medi-
ated factors that alter HIFa regulation. Unlike stimulation
by MAPK, Akt activity increases HIF transcriptional
activation by increasing the pool of available HIFa protein
within the cell. The use of chemical inhibitors such as
wortmannin and LY 294002 that block the phosphatidyl-
inositol 3-kinase (PI3K) family of enzymes, or dominant
negative mutants of the PI3K/Akt pathway were shown to
inhibit factor- or hypoxia-stimulated HIF1a protein accu-
mulation as detected by Western blot [10,46]. A reduction in
the levels of observable HIFa protein resulted in loss of
DNA binding ability of HIF and failure to up-regulate the
transcription of reporter constructs or endogenous down-
stream target genes [10,46]. Similarly over-expression of
members of the PI3K/Akt pathway or inhibition of PTEN,
a negative regulator of Akt, resulted in increased levels of

ofMAPKinthesecellswasshowntoleadtoatransient
activation of eIF-4E and its effects on HIF1a protein
synthesis were additive to those of PI3K/Akt/FRAP [26]
(Fig. 3). Enhanced levels of HIFa synthesis may explain the
previous reports for which activation of MAPK resulted in
an increase in observed HIFa protein levels [22,24]. The
increase in HIFa protein synthesis appears to be relatively
gene specific since the translation of control luciferase-
reporter or ARNT mRNA was not altered [17]. In addition,
over-expression of both FRAP and eIF-4E have been
previously shown to disproportionately increase the trans-
lation of specific target genes [25,47]. This mechanism of
increased translation is in contrast to that employed by
hypoxic stimuli for which it has been repeatedly shown, for
most cell types, that there is no alteration of either HIFa
mRNA levels or the rate of de novo protein synthesis when
oxygen levels are limiting [17,48].
Increased translation of HIFa mRNA ultimately leads to
an increase in the HIFa protein pool, thus explaining initial
reports that observed increased HIFa protein in response to
stimulatory factors. Given that the prolyl (and presumably
the asparaginyl) hydroxylase enzymes are believed not to be
at high concentrations within the cell [49], increasing the
availability of their HIFa substrate may easily titrate them
out. As well as overwhelming the HIFa degradation
mechanisms, substrate saturation also relieves the transcrip-
tional repression due to the asparagine hydroxylase, FIH-1.
Thus it is plausible that even small increases in total HIFa
protein via up-regulated translation could saturate one or
both of these enzymes. Overwhelming the hydroxylase

RasG12V onco-proteins, as well as constitutively active
Akt, however, there were variations in the amount of HIFa
proline-hydroxylation detected for each stimulus [51]. The
antibody generated during this work, which targets the
hydroxylated proline 564 of HIF1a (531 of HIF2a) [51], will
prove a valuable tool in delineating the exact hydroxylation
status of HIFa protein during all types of stimuli.
Nitric oxide, carbon monoxide and cell
confluence
Nitric oxide (NO) has been shown to increase HIFa protein
levels, DNA binding and transcriptional activity in endo-
thelial, smooth muscle, Hep3B and LLC-PK
1
cells during
normoxia [21,52]. Paradoxically, it has also been reported
that NO can also have the completely opposite effect
of inhibition of both basal- and hypoxia-induced expression
of HIF target genes in endothelial cells [53]. Inhibition of
hypoxia-induced DNA-binding activity by carbon monox-
ide (CO) or NO exposure was also seen in several other cell
types [54,55], although reduced HIFa protein expression
was only observed in one case [55]. These inhibitory effects
may be stimuli specific as CO did not prevent the
stabilization of HIF1a protein and transcriptional activity
induced by either cobalt chloride or the iron-chelator
desferrioxamine [55]. The differences in these findings
indicate that cell type, concentration of NO or CO stimuli
and cellular oxygen status are important experimental
considerations and suggest that CO and NO may mediate
their effects through multiple targets within the HIF

with the finding that stimulation of HIF1a by insulin was
only possible when cells were cultured at low density [11]
that suggests the capacity for induced up-regulation of
HIFa is prevented at high density. Given these completely
disparate results, there can be no consensus currently made
as to a mechanism for cell-density mediated effects on HIFa
and clearly this area requires more research. However, the
phenomenon of confluence is an important consideration
during in vitro cell-based assays, particularly because the
effects of confluence may be due in part to localized
hypoxia. Confluence should be carefully monitored during
analysis of HIFa activation by each nonhypoxic stimulus so
the mechanism by which that stimulus contributes to HIFa
can be clearly defined. Finally, the contribution that density
makes to HIFa activity in vivo within tissues is unknown.
Possibly it forms part of the basal level of HIF activity and
this may be different in each tissue type, depending on how
tightly packed the cells are. Although HIFa is ubiquitously
expressed within all cells, the level of normoxic HIFa
protein observable and also the capacity for inducible
up-regulation varies in different cell types [61].
A role for receptor-mediated HIFa
in vivo
?
HIFa is a transcription factor with a complex set of multiple
regulatory mechanisms. Activation through various recep-
tor-mediated pathways, to influence only a subset of these
regulatory mechanisms, allows for a moderate induction of
HIFa and consequently a small increase in the transcription
of downstream target genes. Given the subtle effect upon

whose expression is up-regulated during adipogenesis [65]
Ó FEBS 2003 Kinase pathways influence HIFa (Eur. J. Biochem. 270) 795
and IGF-2 levels are extremely high in embryogenesis [66].
Interestingly, HIF1a has been shown to be stabilized and
activated by the cytokine TNFa during inflammation in
normoxic wounds, allowing increased expression of the HIF
target-gene vascular endothelial growth factor in order to
promote wound healing [67]. It is possible that these receptor-
mediated effects, particularly through HIFa protein synthe-
sis, are only able to occur in active tissues, especially because
the translation initiation factor eIF-4E is only abundant in
nonquiescent cells [25]. Stimulation of HIFa though oxygen-
independent mechanisms could increase expression of genes
that promote angiogenesis, vasodilation, glucose uptake or
glycolysis to provide increased nutrient supply to those
tissues requiring it. Many of these HIF target genes have
other regulatory elements within their promoters and their
expression is a balance between converging signals. This may
be the case with many of the glycolytic genes such as
hexokinase that have glucose or carbohydrate responsive
elements nearby to the hypoxic response elements that may
combine synergistically to regulate gene expression [68,69].
Further work is required to clarify the molecular details of
the receptor-mediated signalling pathways and their different
effects on HIFa activity. However the activation of HIFa by
receptor-mediated signals has been established. It will also be
important to define the role(s) for these signalling pathways
and to investigate the possibility that this type of receptor-
mediated induction of HIFa may regulate the transcription
of only a subset of HIF responsive genes for specific functions

domain a hypoxic switch. Science. 295, 858–861.
7. Semenza, G.L. (2001) HIF-1, O(
2
), and the 3 phds: how animal
cells signal hypoxia to the nucleus. Cell. 107, 1–3.
8. Lando, D., Gorman, J.J., Whitelaw, M.L. & Peet, D.J. (2003)
Oxygen-dependent regulation of the hypoxia-inducible factors
by prolyl and asparaginyl hydroxylation. Eur. J. Biochem. 270,
781–790.
9. Zelzer, E., Levy, Y., Kahana, C., Shilo, B.Z., Rubinstein, M. &
Cohen, B. (1998) Insulin induces transcription of target genes
through the hypoxia-inducible factor HIF-1a/ARNT. EMBO J.
17, 5085–5094.
10. Zhong, H., Chiles, K., Feldser, D., Laughner, E., Hanrahan, C.,
Georgescu, M.M., Simons, J.W. & Semenza, G.L. (2000)
Modulation of hypoxia-inducible factor 1alpha expression by the
epidermal growth factor/phosphatidylinositol 3-kinase/PTEN/
AKT/FRAP pathway in human prostate cancer cells:
implications for tumor angiogenesis and therapeutics. Cancer
Res. 60, 1541–1545.
11. Feldser, D., Agani, F., Iyer, N.V., Pak, B., Ferreira, G. &
Semenza, G.L. (1999) Reciprocal positive regulation of hypoxia-
inducible factor 1a and insulin-like growth factor 2. Cancer Res.
59, 3915–3918.
12. Hellwig-Burgel, T., Rutkowski, K., Metzen, E., Fandrey, J. &
Jelkmann, W. (1999) Interleukin-1b and tumor necrosis factor-a
stimulate DNA binding of hypoxia-inducible factor-1. Blood. 94,
1561–1567.
13. Richard, D.E., Berra, E. & Pouyssegur, J. (2000) Nonhypoxic
pathway mediates the induction of hypoxia-inducible factor

Chen, E., Gottschalk, A.R., Ryan, H.E., Johnson, R.S., Jefferson,
A.B., Stokoe, D. & Giaccia, A.J. (2000) Loss of PTEN facilitates
HIF-1-mediated gene expression. Genes Dev. 14, 391–396.
21.Sandau,K.B.,Faus,H.G.&Brune,B.(2000)Inductionof
hypoxia-inducible-factor 1 by nitric oxide is mediated via the PI
3K pathway. Biochem. Biophys. Res. Commun. 278, 263–267.
22. Sheta, E.A., Trout, H., Gildea, J.J., Harding, M.A. & Theodorescu, D.
(2001) Cell density mediated pericellular hypoxia leads to
induction of HIF-1a via nitric oxide and Ras/MAP kinase
mediated signaling pathways. Oncogene. 20, 7624–7634.
796 R. L. Bilton and G. W. Booker (Eur. J. Biochem. 270) Ó FEBS 2003
23. Mukhopadhyay, D., Tsiokas, L. & Sukhatme, V.P. (1998) High
cell density induces vascular endothelial growth factor expression
via protein tyrosine phosphorylation. Gene Expr. 7, 53–60.
24. Treins, C., Giorgetti-Peraldi, S., Murdaca, J., Semenza, G.L. &
Van Obberghen, E. (2002) Insulin stimulates hypoxia-inducible
factor 1 through a phosphatidylinositol 3-Kinase/Target of
rapamycin-dependent signaling pathway. J. Biol. Chem. 277,
27975–27981.
25. Gingras, A.C., Raught, B. & Sonenberg, N. (2001) Regulation of
translation initiation by FRAP/mTOR. Genes Dev. 15, 807–826.
26. Fukuda, R., Hirota, K., Fan, F., Jung, Y.D., Ellis, L.M. &
Semenza, G.L. (2002) IGF-1 induces HIF-1-mediated VEGF
expression that is dependent on MAP kinase and PI-3-kinase
signaling in colon cancer cells. J. Biol. Chem. 277, 38205–38211.
27. Minet, E., Arnould, T., Michel, G., Roland, I., Mottet, D., Raes,
M., Remacle, J. & Michiels, C. (2000) ERK activation
upon hypoxia: involvement in HIF-1 activation. FEBS Lett.
468, 53–58.
28. Mazure, N.M., Chen, E.Y., Laderoute, K.R. & Giaccia, A.J.

37. Wang, G.L., Jiang, B.H. & Semenza, G.L. (1995) Effect of
protein kinase and phosphatase inhibitors on expression of
hypoxia-inducible factor 1. Biochem. Biophys. Res. Commun.
216, 669–675.
38. Pugh, C.W., O’Rourke, J.F., Nagao, M., Gleadle, J.M. &
Ratcliffe, P.J. (1997) Activation of hypoxia-inducible factor-1;
definition of regulatory domains within the a subunit. J. Biol.
Chem. 272, 11205–11214.
39. Gradin, K., Takasaki, C., Fujii-Kuriyama, Y. & Sogawa, K.
(2002) The transcriptional activation function of the HIF-like
factor requires phosphorylation at a conserved threonine. J.Biol.
Chem. 277, 23508–23514.
40. Jewell,U.R.,Kvietikova,I.,Scheid,A.,Bauer,C.,Wenger,R.H.
& Gassmann, M. (2001) Induction of HIF-1a in response to
hypoxia is instantaneous. FASEB J. 15, 1312–1314.
41. Richard,D.E.,Berra,E.,Gothie,E.,Roux,D.&Pouyssegur,J.
(1999) p42/p44 mitogen-activated protein kinases phosphorylate
hypoxia-inducible factor 1a (HIF-1a) and enhance the
transcriptional activity of HIF-1. J. Biol. Chem. 274, 32631–32637.
42. Sodhi,A.,Montaner,S.,Patel,V.,Zohar,M.,Bais,C.,Mesri,
E.A. & Gutkind, J.S. (2000) The Kaposi’s sarcoma-associated
herpes virus G protein-coupled receptor up-regulates vascular
endothelial growth factor expression and secretion through
mitogen-activated protein kinase and p38 pathways acting on
hypoxia-inducible factor 1a. Cancer Res. 60, 4873–4880.
43. O’Rourke, J.F., Tian, Y.M., Ratcliffe, P.J. & Pugh, C.W. (1999)
Oxygen-regulated and transactivating domains in endothelial PAS
protein 1: comparison with hypoxia-inducible factor-1a. J. Biol.
Chem. 274, 2060–2071.
44. Mahon, P.C., Hirota, K. & Semenza, G.L. (2001) FIH-1: a novel

Role of prolyl hydroxylation in oncogenically stabilized hypoxia-
inducible factor-1alpha. J. Biol. Chem 277, 40112–40117.
52. Palmer, L.A., Gaston, B. & Johns, R.A. (2000) Normoxic
stabilization of hypoxia-inducible factor-1 expression and
activity: redox-dependent effect of nitrogen oxides. Mol.
Pharmacol. 58, 1197–1203.
53. Kourembanas, S., McQuillan, L.P., Leung, G.K. & Faller, D.V.
(1993) Nitric oxide regulates the expression of vasoconstrictors
and growth factors by vascular endothelium under both normoxia
and hypoxia. J. Clin. Invest. 92, 99–104.
54. Liu, Y., Christou, H., Morita, T., Laughner, E., Semenza, G.L.
& Kourembanas, S. (1998) Carbon monoxide and nitric
oxide suppress the hypoxic induction of vascular endothelial
growth factor gene via the 5¢ enhancer. J. Biol. Chem. 273, 15257–
15262.
55. Huang, L.E., Willmore, W.G., Gu, J., Goldberg, M.A. & Bunn,
H.F. (1999) Inhibition of hypoxia-inducible factor 1 activation by
carbon monoxide and nitric oxide. Implications for oxygen
sensing and signaling. J. Biol. Chem. 274, 9038–9044.
56. Melillo, G., Musso, T., Sica, A., Taylor, L.S., Cox, G.W. &
Varesio, L. (1995) A hypoxia-responsive element mediates a novel
pathway of activation of the inducible nitric oxide synthase
promoter. J.ExpMed.182, 1683–1693.
Ó FEBS 2003 Kinase pathways influence HIFa (Eur. J. Biochem. 270) 797
57. Lee, P.J., Jiang, B.H., Chin, B.Y., Iyer, N.V., Alam, J., Semenza,
G.L. & Choi, A.M. (1997) Hypoxia-inducible factor-1 mediates
transcriptional activation of the heme oxygenase-1 gene in
response to hypoxia. J. Biol. Chem. 272, 5375–5381.
58. Zeng, G. & Quon, M.J. (1996) Insulin-stimulated production of
nitric oxide is inhibited by wortmannin. Direct measurement in

Smith, A., Fundele, R., Stewart, F., Kelsey, G., Fowden, A., Sibley,
C. & Reik, W. (2002) Placental-specific IGF-II is a major
modulator of placental and fetal growth. Nature. 417, 945–948.
67. Albina, J.E., Mastrofrancesco, B., Vessella, J.A., Louis, C.A.,
Henry, W.L. Jr & Reichner, J.S. (2001) HIF-1 expression in
healing wounds: HIF-1a induction in primary inflammatory cells
by TNF-a. Am. J. Physiol. Cell Physiol. 281, C1971–C1977.
68. Mathupala, S.P., Rempel, A. & Pedersen, P.L. (2001) Glucose
catabolism in cancer cells: identification and characterization of a
marked activation response of the type II hexokinase gene to
hypoxic conditions. J. Biol. Chem. 276, 43407–43412.
69. Dang,C.V.,Lewis,B.C.,Dolde,C.,Dang,G.&Shim,H.(1997)
Oncogenes in tumor metabolism, tumorigenesis, and apoptosis.
J. Bioenerg Biomembr. 29, 345–354.
798 R. L. Bilton and G. W. Booker (Eur. J. Biochem. 270) Ó FEBS 2003