Structural basis for PAS domain heterodimerization in the basic helix--loop--helix-PAS transcription factor hypoxia-inducible factor
Biological responses to oxygen availability play important roles in development, physiological homeostasis, and many disease processes. In mammalian cells, this adaptation is mediated in part by a conserved pathway centered on the hypoxia-inducible factor (HIF). HIF is a heterodimeric protein complex composed of two members of the basic helix-loop-helix Per-ARNT-Sim (PAS) (ARNT, aryl hydrocarbon receptor nuclear translocator) domain family of transcriptional activators, HIFalpha and ARNT. Although this complex involves protein-protein interactions mediated by basic helix-loop-helix and PAS domains in both proteins, the role played by the PAS domains is poorly understood. To address this issue, we have studied the structure and interactions of the C-terminal PAS domain of human HIF-2alpha by NMR spectroscopy. We demonstrate that HIF-2alpha PAS-B binds the analogous ARNT domain in vitro, showing that residues involved in this interaction are located on the solvent-exposed side of the HIF-2alpha central beta-sheet. Mutating residues at this surface not only disrupts the interaction between isolated PAS domains in vitro but also interferes with the ability of full-length HIF to respond to hypoxia in living cells. Extending our findings to other PAS domains, we find that this beta-sheet interface is widely used for both intra- and intermolecular interactions, suggesting a basis of specificity and regulation of many types of PAS-containing signaling proteins.
Structural basis for PAS domain heterodimerization in
the basic helix–loop–helix-PAS transcription factor
Paul J. A. Erbel*
, Paul B. Card*
, Ozgur Karakuzu*, Richard K. Bruick*, and Kevin H. Gardner*
Departments of *Biochemistry and
Pharmacology, University of Texas Southwestern Medical Center, 5323 Harry Hines Boulevard, Dallas, TX 75390
Edited by Susan S. Taylor, University of California at San Diego, La Jolla, CA, and approved October 9, 2003 (received for review June 10, 2003)
Biological responses to oxygen availability play important roles in
development, physiological homeostasis, and many disease pro-
cesses. In mammalian cells, this adaptation is mediated in part by
a conserved pathway centered on the hypoxia-inducible factor
(HIF). HIF is a heterodimeric protein complex composed of two
members of the basic helix–loop–helix Per-ARNT-Sim (PAS) (ARNT,
aryl hydrocarbon receptor nuclear translocator) domain family of
transcriptional activators, HIF
and ARNT. Although this complex
involves protein–protein interactions mediated by basic helix–
loop–helix and PAS domains in both proteins, the role played by
the PAS domains is poorly understood. To address this issue, we
have studied the structure and interactions of the C-terminal PAS
domain of human HIF-2
by NMR spectroscopy. We demonstrate
PAS-B binds the analogous ARNT domain in vitro,
showing that residues involved in this interaction are located on
the solvent-exposed side of the HIF-2
residues at this surface not only disrupts the interaction between
isolated PAS domains in vitro but also interferes with the ability of
full-length HIF to respond to hypoxia in living cells. Extending our
ﬁndings to other PAS domains, we ﬁnd that this
is widely used for both intra- and intermolecular interactions,
suggesting a basis of speciﬁcity and regulation of many types of
PAS-containing signaling proteins.
ellular responses to oxygen availability are essential for the
development and homeostasis of mammalian cells, demon-
strated most critically by the link between the cellular adaptation
to reduced tissue oxygenation and disease progression (1, 2). In
mammalian cells, these responses are mediated in part by the
hypoxia-inducible factor (HIF), a heterodimeric transcription
factor composed of HIF
and aryl hydrocarbon receptor nuclear
translocator (ARNT, also known as HIF
) (3). HIF activity is
tightly controlled under normoxic conditions by multiple O
dependent hydroxylation events of the HIF
coordinately promote the ubiquitin-mediated destruction of this
protein (4) and impair its ability to interact with transcriptional
coactivators (5, 6) (Fig. 1a). These controls are relieved during
hypoxia, allowing HIF to activate the transcription of genes that
facilitate metabolic adaptation to low oxygen levels and increase
local oxygen supply by angiogenesis (7).
All three isoforms of HIF
(EPAS1), and -3
(8, 9) and ARNT belong to the basic helix–loop–helix (bHLH)–
Per-ARNT-Sim (PAS) family of eukaryotic transcription fac-
tors, which contain bHLH and PAS domains (Fig. 1). The bHLH
domains of these proteins serve as dimerization elements, help-
ing determine the specificity of complex formation while pro-
viding a DNA-binding interface composed of the basic regions
from each monomer (10). PAS domains are widespread com-
ponents of signal transduction proteins, currently identified in
⬎2,000 proteins from organisms in all three kingdoms of life.
These domains, shown to be protein–protein interaction ele-
ments in several systems (11), also appear to contribute to the
dimerization process and thus increase the specificity of bHLH-
PAS transcription factor formation (12, 13). In the case of the
兾ARNT complex, coimmunoprecipitation and gel mobil-
ity-shift experiments using truncated forms of HIF
suggest that although the bHLH domains alone are able to
dimerize, the PAS domains are required to build a stable
heterodimer capable of robust DNA binding (14, 15). These data
suggest a model of the complex where the bHLH, PAS-A, and
PAS-B domains of ARNT interact with their counterparts in
(Fig. 1a). However, most of this model remains speculative
in light of the sparse data describing how PAS domains bind to
each other, or more generally, to any protein partner.
To provide insight into this general topic of PAS domain
signaling, particularly its importance in the hypoxia response
pathway, we have studied the structure and interactions of the
C-terminal PAS domain of human HIF-2
NMR spectroscopy. We report that HIF-2
PAS-B adopts a
structure similar to other members of this family, with a central
-sheet flanked on one face by several
-helices. We further
show that HIF-2
PAS-B binds directly to the human ARNT
PAS-B domain in vitro, identifying the interface as a group of
residues located in the central strands of the
structure-based mutations of this interface in the PAS-B do-
mains of HIF-1
, we demonstrate that such changes
interfere with the binding of isolated PAS-B domains in vitro but
more importantly disrupt the ability of full-length HIF proteins
to respond to hypoxia in living cells. These observations led us
to compare PAS domains from multiple systems, showing that
-sheet interface participates in a wide range of inter- and
intramolecular interactions and suggesting a way that specificity
and regulation may be achieved among these versatile domains.
Materials and Methods
Protein Expression and Purification. DNA-encoding fragments of
PAS-B (residues 240–350) and ARNT PAS-B
(residues 356–470) were subcloned into the pG
pHis-parallel expression vectors, respectively (16, 17). Esche-
richia coli BL21(DE3) cells transformed with these plasmids
were grown in M9 media containing 1 g兾liter
Cl for U-
samples (supplemented with 3 g兾liter
glucose for U-
labeled samples). These cultures were grown at 37°C to an A
of 0.6–1.0, then induced overnight at 20°C by the addition of 0.5
The purification of HIF-2
PAS-B has been detailed (18).
NMR samples typically contained 0.9 mM protein in 50 mM Tris
buffer (pH 7.3), 15 mM NaCl, 5 mM DTT, 5 mM NaN
protease inhibitor mixture (Sigma) in 90% H
This paper was submitted directly (Track II) to the PNAS ofﬁce.
Abbreviations: HIF, hypoxia-inducible factor; PAS, Per-ARNT-Sim; HIF-2
PAS domain of human HIF-2
; ARNT, aryl hydrocarbon receptor nuclear translocator; bHLH,
basic helix–loop– helix; HSQC, heteronuclear sequential quantum correlation; CHO,
Chinese hamster ovary; HRE, hypoxia responsive element.
Data deposition: The atomic coordinates for the HIF-2
PAS-B domain have been deposited
in the Protein Data Bank (PDB ID 1P97).
To whom correspondence should be addressed. E-mail: kevin.gardner@utsouthwestern.
© 2003 by The National Academy of Sciences of the USA
December 23, 2003
no. 26 www.pnas.org兾cgi兾doi兾10.1073兾pnas.2533374100
unless otherwise noted. ARNT PAS-B was expressed and puri-
fied as described in Supporting Methods, which is published as
supporting information on the PNAS web site.
Parallel studies on human HIF-1
PAS-B used a construct
containing residues 238–349, chosen by homology with HIF-2
PAS-B. Expression and purification of HIF-1
PAS-B were done
as described for HIF-2
NMR Spectroscopy. All NMR data were recorded at 30°C with
Varian Inova 500 and 600 MHz spectrometers by using NMRPIPE
for data processing (19) and NMRVIEW for analysis (20). Chem-
ical-shift assignments were made by using standard methods (21)
as detailed in Supporting Methods.
Deuterium exchange reactions were started by resuspending
PAS-B in 99% D
rected pH 7.3). These samples were then placed into a pre-
warmed magnet (T ⫽ 30°C), and
H heteronuclear sequen-
tial quantum correlation (HSQC) spectra were sequentially
acquired approximately every 15 min. Observed
rates were converted into protection factors by using standard
Structure Determination. Interproton distance constraints were
obtained from 3D
N edited NOESY (
⫽ 150 ms),
edited NOESY (
⫽ 100 ms), and 2D NOESY (
⫽ 120 ms)
spectra. Hydrogen bond constraints (1.3 Å ⬍ d
⬍ 2.5 Å, 2.3
Å ⬍ d
⬍ 3.5 Å) were set for backbone amide protons
protected for ⬎30 min from exchange with D
O solvent (30°C,
pH 7.3). Constraints for the
dihedral angles were
generated by chemical-shift analyses by using TALOS (23), with
two times the standard deviation of TALOS predictions as the
bounds (minimum ⫾ 30°). For 19 residues without TALOS
dihedral angle constraints were obtained from an
analysis of a 3D HNHA spectrum. Finally, 78
dipolar coupling constraints were obtained from a sample
partially aligned in 5% (wt兾vol) DMPC兾 DHPC ratio of 3:1
(Avanti Polar Lipids) and 5 mM cetyltrimethylammonium bro-
mide at 35°C.
Initial structures were determined without manual assign-
ments by using
ARIA1.2 (24, 25) and subsequently refined with a
mix of automated and manual assignment of NOESY spectra. Of
1,000 structures, the 20 lowest-energy structures were analyzed
MOLMOL (26) and PROCHECK-NMR (27).
From this ensemble, the structure closest to the mean was
superimposed against other PAS domains with the
Swiss Protein Data Bank program (28) with the automatic fit
option. The calculated rms deviations ranged between 1.4 and
1.65 Å for HERG (Research Collaboratory for Structural Bio-
informatics Protein Data Bank ID 1BYW), hPASK (1LL8),
RmFixL (1D06), and Phy3 (1G28). The HIF-2
was also used to generate a model of the HIF-1
structure (74% sequence identity) by using
and ARNT PAS-B Titration. Titrations were conducted by the
stepwise addition of natural abundance ARNT PAS-B (up to 800
M) to a sample of 200-
at 35°C. The peak heights
PAS-B signals that do not show ARNT-dependent
chemical shift changes (38 residues) were fit to Eq. 1 to obtain
the corresponding K
⌬I ⫽ 1 ⫺ 兵⌬I
⫻ 关共A ⫹ P
⫺ 共共A ⫹ P
⫺ 共4 ⫻ A ⫻ P
兴兾关2 ⫻ P
where ⌬I is the observed change in peak height at ARNT
concentration A, ⌬I
is the change in peak height at saturation,
is the total HIF
concentration. Eq. 1 is similar to the
equation used to extract K
from chemical-shift changes ob-
served in titrations of complexes undergoing fast exchange (30),
and we apply it here only to sites without chemical-shift changes
(fast exchange) to ensure that the observed peak line widths are
a population-weighted average of the free- and bound-state line
widths (31). The binding of HIF-2
PAS-B mutants to ARNT
PAS-B was assessed by adding 900
M natural abundant ARNT
PAS-B to 250
Mutagenesis. Point mutants of full-length HIF-1
created from wild-type DNA and primers including the desired
mutation(s). PAS-B domains containing these mutations were
obtained by PCR amplification of the corresponding full-length
sequence and subcloned into the pG
1-parallel vector. Trans-
formation, protein induction, and purification were performed
as described above.
Transfections. Cells were plated onto 48-well plates (3.5 ⫻ 10
cells per well) in 200
l of HyQ DME兾F-12 1:1 media (HyClone)
supplemented with 5% FBS 24 h before transfection. Cells were
transfected with 10 ng of each HIF
construct and 20 ng of the
3HRE-tk-luc (HRE, hypoxia-responsive element) luciferase re-
porter construct (8) by using the Lipofectamine PLUS reagent
(Invitrogen). After 3 h, the media were changed and, after an
additional 2 h, cells were incubated for 15 h under normoxic or
hypoxic (1.0% O
) conditions. Luciferase activity was measured
as described (32).
Solution Structure of HIF-2
PAS-B. We determined the solution
structure of HIF-2
PAS-B by using standard double- and
triple-resonance NMR experiments conducted on uniformly
C labeled protein samples (Fig. 2). This structure is
Fig. 1. Oxygen-dependent regulation and domain architecture of HIF pro-
teins. (a) HIF regulation is tightly linked to intracellular oxygen levels. Under
normoxic conditions, HIF
is posttranslationally hydroxylated, promoting its
degradation [modiﬁcation of the oxygen-dependent degradation domain
(ODD)] and interfering with its ability to interact with CBP兾p300 coactivators
(modiﬁcation of the transcriptional activation domains NTAD and CTAD).
These modiﬁcations are not made under hypoxic conditions, allowing HIF
accumulate and enter the nucleus where it associates with ARNT and binds to
HREs upstream of hypoxia-activated genes. The red box highlights the HIF
and ARNT PAS-B domains. (b) Domain topology of HIF
subunits, including a
bHLH domain, two PAS domains, and C-terminal regulatory domains. A
sequence alignment of the HIF
PAS-B orthologs is shown, with bold letters
indicating the mutated residues described in the text. HIF-2
structure elements are indicated with a gray background.
Erbel et al. PNAS
December 23, 2003
based on ⬎2,500 geometric constraints obtained from measure-
ments of interproton distances, dihedral angles, and
residual dipolar couplings of a partially oriented sample (Table
1). All of these data are well satisfied by the high-precision
ensemble of the 20 lowest-energy structures subsequently used
for further analysis.
PAS-B adopts a typical
PAS domain fold, char-
acterized by several
-helices flanking a five-stranded antipar-
sheet. The similarity of this structure to other PAS
domains is demonstrated by the low-backbone rms deviation
values (1.4–1.65 Å) of pairwise comparisons between represen-
tative PAS structures and HIF-2
PAS-B. Although several
other PAS domains bind cofactors within their hydrophobic
cores to regulate protein–protein interactions in response to
various physical stimuli (11), a combination of NMR, mass
spectrometry, and visible spectroscopy shows that HIF-2
PAS-B does not copurify with any such compound (data not
shown). Further, no preformed cavities are present in the protein
core, even at sites occupied by ligands in some other PAS
domains (17, 33–36) (Fig. 2b).
Identification of ARNT PAS-B-Binding Surface on HIF-2
PAS domains in bHLH-PAS transcription factors are thought to
cooperate with the bHLH domains to facilitate dimerization (12,
13), which implies that the HIF
and ARNT PAS domains bind
to each other (Fig. 1a). To experimentally demonstrate this
interaction, we titrated unlabeled ARNT PAS-B into
PAS-B and monitored changes in the HIF-2
H HSQC spectrum (Fig. 3a). Peaks in these spectra showed
both chemical-shift changes and line broadening on addition of
ARNT PAS-B, consistent with binding on the intermediate and
fast exchange time scales. In contrast, we found that HIF-2
PAS-B signals were not affected by the addition of a PAS domain
from PAS kinase, a protein not involved in the hypoxia response
(17) (data not shown), suggesting that the changes observed on
addition of ARNT PAS-B reflect a specific HIF-2
The ARNT-induced changes in the HIF-2
line widths dem-
onstrate two important effects. First, we observed a general
increase in line width for HIF-2
peaks during the titration,
which we attribute to the slower tumbling of the larger 27-kDa
heterodimeric complex compared with an isolated HIF-2
PAS-B domain. By monitoring this broadening via the decrease
in peak heights as ARNT PAS-B was added, we observed a
titration consistent with a 1:1 binding event with a 30
3b). This effect saturated at a 1:3 (HIF兾ARNT) ratio, establish-
ing that it is not caused by nonspecific increases in sample
viscosity or aggregation. Second, we observed that a subset of
residues preferentially broadened on the addition of substoi-
chiometric amounts of ARNT PAS-B. Such differential effects
have been observed in several complexes (17, 37, 38) and arise
from exchange broadening at sites experiencing significant
chemical-shift changes on complex formation. Mapping sites
that exhibit either this differential broadening or significant
ARNT-induced chemical shift changes onto the HIF-2
structure shows that they cluster on the face of the central
-sheet (Fig. 4). This provides a chiefly hydrophobic surface for
ARNT binding that is conserved among the HIF isoforms
(Fig. 1), suggesting that the PAS-B domains of all three interact
similarly with ARNT.
Evidence for the importance of this interface in the HIF兾ARNT
PAS-B dimer was obtained from studies of PAS-B domains con-
taining point mutations. Based on our structure, we altered three
Fig. 2. Solution structure of HIF-2
PAS-B. (a) Superimposition of 20 lowest-
energy structures for HIF-2
PAS-B, calculated as indicated in the text. (b)
Ribbon diagram of the structure closest to the mean of the ensemble shown
in a. Circles indicate the approximate locations of the ligand-binding sites of
several PAS domains (17, 33–36).
Table 1. Statistics for HIF-2
PAS-B solution structure
List of constraints
NOE distance restraints
Hydrogen bond restraints 60
Dihedral angle restraints 96
H residual dipolar couplings 78
Mean rms deviation from
NOE, Å 0.022 ⫹兾⫺ 0.002
Dihedral angles, deg 1.04 ⫹兾⫺ 0.16
Average number of:
NOE violations ⬎0.5 Å 0
NOE violations ⬎0.3 Å 1.9 ⫹兾⫺ 1.2
Dihedral violations ⬎5° 1.6 ⫹兾⫺ 1.1
Mean rms from idealized covalent
Bonds, Å 0.0045
Angles, deg. 0.65
Impropers, deg. 1.69
Geometric analysis of residues
6–91 and 98–112
rms deviation to mean 0.53 ⫹兾⫺ 0.07 Å (backbone)
1.08 ⫹兾⫺ 0.10 Å (all heavy)
Ramachandran analysis (
PROCHECK) 81.0% most-favored
16.4% additionally allowed
1.6% generously allowed
www.pnas.org兾cgi兾doi兾10.1073兾pnas.2533374100 Erbel et al.
residues with solvent-exposed side chains within the H
strands (Q322E, M338E, and Y342T) (Fig. 4b).
spectra of this triple mutant (trHIF-2
PAS-B) retain the chemical-
shift dispersion and general pattern of the wild-type protein,
confirming that the protein structure remains intact (Fig. 5a). The
ARNT-binding capability of this mutant was assessed by comparing
H HSQC spectra before and after addition of unlabeled
ARNT PAS-B. As demonstrated by the minimal ARNT-induced
changes in peak locations and intensities, the interaction of the
triple mutant HIF-2
PAS-B with ARNT has been very signifi-
cantly weakened. These data establish that subtle changes on the
surface of the H
strands of HIF-2
PAS-B can disrupt the
HIF–ARNT PAS-B interaction.
Comparison of HIF-1
PAS-B. Sequence alignments of
indicate that the PAS-B domains of these
proteins are extremely similar (74% identity; Fig. 1b). Never-
theless, the HIF-1
homolog of our HIF-2
Fig. 3. Characterization of the HIF-2
兾ARNT PAS-B-binding interaction. (a)
Titration of unlabeled ARNT PAS-B (black, 0
M; light blue, 200
M; blue, 600
M; red, 800
M ARNT) into a 200
Arrow shows direction of peak shifts with increasing amounts of ARNT.
Residues with peak broadening beyond detection during the titration are
.(b) Normalized peak heights of HIF-2
PAS-B (38 resonances)
plotted against increasing amounts of ARNT PAS-B. The concentration depen-
dence of the observed reduction in peak heights can be ﬁt to a 1:1 binding
event with a K
M (dotted line).
Fig. 4. Identiﬁcation of the ARNT PAS-B-binding site on the surface of HIF-2
PAS-B. (a) Chemical-shift changes from
H HSQC spectra of HIF-2
the presence of 800
M ARNT are plotted as a function of residue number. The red line indicates chemical-shift changes ⬎0.16 ppm. Residues with peak
broadening beyond detection are shown as an arbitrary chemical-shift change of 0.5 ppm. Secondary structure elements are indicated with a gray background.
(b) Surface representations of HIF-2
PAS-B showing the location of the ARNT PAS-B-binding site. Colors indicate residues with large chemical-shift changes (⬎0.4
ppm) or broadening beyond detection (red), residues with a signiﬁcant chemical-shift changes (⬎0.16 ppm) (orange), and the site of the mutated residues that
disrupt the HIF兾ARNT PAS-B interactions (yellow). Figs. 4 and 6 were made with
Fig. 5. Point mutations in the HIF
-sheet disrupt the binding
of ARNT PAS-B. (a) Superimposed
H HSQC spectra of 250
PAS-B (Left) or triple mutant (Q322E兾M338E兾Y342T) (Right). Spectra in
the presence of 900
M unlabeled ARNT PAS-B are shown with red contours;
those without ARNT are shown in black contours. Similar data for HIF-1
are provided in Supporting Methods.(b) PAS-B domain interaction is impor-
tant to form a biologically active HIF兾ARNT complex. A construct expressing a
luciferase reporter under the control of an HRE promoter was transfected into
Ka-13 (columns 1–5) or CHO (column 6) cells along with various HIF
structs. Values represent the average luciferase activity of three samples, with
bars indicating standard error. Luciferase expression was induced by cotrans-
fection of HIF-1
(column 2) or HIF-2
(column 4), particularly under hypoxic
conditions. Cotransfection of trHIF-1
(column 3) or trHIF-2
proteins containing the three PAS-B mutations, shows a
signiﬁcant drop in luciferase activity compared with wild-type HIF
Erbel et al. PNAS
December 23, 2003
was too poorly behaved in solution to be amenable to structure
determination. However, by combining the high sequence iden-
tity between these PAS-B domains and the HIF-2
solution structure, we generated a homology model of HIF-1
PAS-B (29). This guided our mutation of three solvent-exposed
residues in the HIF-1
at sites analogous to those changed in HIF-2
ingly, this HIF-1
PAS-B triple mutant was significantly better
behaved in solution than the wild-type domain. To determine
PAS-B bound ARNT PAS-B in a similar
fashion as HIF-2
PAS-B, we recorded
H HSQC of
wild-type and triple-mutant HIF-1
PAS-B in the presence and
absence of ARNT PAS-B (Fig. 7, which is published as support-
ing information on the PNAS web site). In the case of wild-type
PAS-B, the combination of specific peak shifting and
line broadening on the addition of ARNT PAS-B indicated
binding. In contrast, this interaction was disrupted in the triple
mutant (Fig. 7), demonstrating that the interface identified in
is as crucial for the ARNT-binding function of HIF-1
Functional Importance of PAS-B in Full-Length HIF
. Our demonstra-
tion of specific interactions between the isolated HIF
ARNT PAS-B domains suggested that they function as dimer-
ization elements in the HIF heterodimer. To address this ques-
tion in living cells with full-length HIF
, we used a luciferase-
based assay that uses Chinese hamster ovary (CHO) cells lacking
(CHO-Ka13) (39). CHO-Ka13 cells were tran-
siently transfected with full-length HIF
constructs under the
control of the HIF-1 promoter (40) and encoding either wild-
type or a mutant sequence containing the three PAS-B domain
mutations. Low levels of DNA were transfected to roughly match
the HIF activity in wild-type CHO cells. The abilities of these
constructs to activate transcription were determined by
measuring the expression of luciferase under the control of a
minimal HRE promoter (8).
After expressing either wild-type HIF-1
, the hypoxia
response of the transfected CHO-Ka13 cells mimicked that of
wild-type CHO cells, confirming that HIF
activity can be
rescued by expression from these plasmids (Fig. 5b). Cells
transfected with HIF-2
constructs either lacking the entire
PAS-B domain (⌬240–350) or containing a mutation known to
unfold PAS-B (C339P; unfolded as established by NMR) showed
no significant luciferase activity under any conditions (data not
shown). Although these results suggest that HIF
a crucial role in forming a biologically active HIF
heterodimer, the question remained whether the subtler PAS-B
modifications that disrupted in vitro binding would have a similar
effect. Therefore, we transfected CHO-Ka13 cells with full-
constructs containing the three PAS-B
). For both mutant proteins, the HRE-driven
luciferase response is decreased compared with that observed in
cells containing vectors expressing wild-type HIF
. In the case
, the response is virtually eliminated (Fig. 5b, column
3), whereas a less pronounced but significant reduction is
observed for trHIF-2
(Fig. 5b, column 5). These results are
surprising in light of the three apparently redundant sets of
protein–protein interactions implied by the model of the HIF
heterodimer (Fig. 1a). Our data show that the subtle disruption
of the PAS-B interaction within the full length HIF
sufficient to prevent the formation of the HIF transcription
factor, thereby abolishing the hypoxia response.
Despite the intense interest in hypoxia signaling, previous un-
derstanding of the roles of various protein–protein interactions
in the HIF heterodimer remained unclear. Much of the data
underlying prior work in this area were based on deletions
generated at a time when the lack of structural data hampered
the design of constructs cleanly corresponding to PAS domain
boundaries. Unfortunately, this approach has complicated the
interpretation of these findings with respect to the functional
importance of individual PAS domains. In this context, our
results provide the structural insight to clarify how PAS-B
domains contribute to the stability of HIF and other bHLH-PAS
transcription factor complexes. We demonstrate that the PAS-B
domains of HIF
and ARNT directly associate in vitro, using a
group of predominantly hydrophobic residues located on the
solvent-exposed face of the HIF
-sheet. Critically, mutations
that disrupt the interaction between isolated PAS-B domains in
vitro also interfere with the ability of the full-length heterodimer
to activate transcription in living cells. These results establish the
importance of the
-sheet interaction surface of HIF
domains within the complexes that are central to the hypoxia
More broadly, the
-sheet interface we identified appears to
be important for a wide array of inter- and intramolecular
interactions within several PAS domains (Fig. 6). Evidence of
this commonality is provided by PAS domains from two blue-
light photoreceptors, phototropin1 (AsLOV2) (36) and photo-
active yellow protein (33). Both of these domains are flanked by
-helical elements that fold back on the PAS domain
itself, associating with the same
-sheet region used by HIF
bind ARNT (Fig. 6). An additional example is provided by the
human ether-a-go-go-related gene (HERG) potassium channel,
which contains a PAS domain that controls the poststimulation
deactivation kinetics of this channel, possibly by interacting with
Fig. 6. Versatility of protein interactions involving PAS domain
is shown in the same orientation as Fig. 4b and colored by residues experiencing
H chemical shifts on complex formation (red) and those used to generate the complex-disrupting trHIF-2
(blue). Phototropin (AsLOV2) (36) and
photoactive yellow protein (33) highlight the
-helices external to the PAS core (magenta) and any atoms located within 5 Å of those helices (pink). HERG (42)
shows functionally important, solvent-exposed residues (dark blue) and residues present in a surface hydrophobic patch suggested to be important for channel
function (light blue) (42).
www.pnas.org兾cgi兾doi兾10.1073兾pnas.2533374100 Erbel et al.
the S4-S5 linker (41). Point mutations in the PAS domain that
impair channel function have been identified by both alanine
scanning mutagenesis (42) and sequencing of HERG genes from
individuals affected by long QT syndrome (41), an inherited
cardiac disorder associated with defects in HERG and other ion
channels. Biochemical studies of channels containing one of
several conservative mutations in solvent-exposed residues
(F29L, F29A, N33T, Y43A, and R56Q; Fig. 6) demonstrate that
the region analogous to the HIF兾ARNT interface plays an
important role in regulating HERG channel kinetics. In sum-
mary, these comparisons indicate that the exposed face of the
-sheet in PAS domains is well suited for making many
functionally important types of associations.
This comparison also suggests a route that might be exploited
to regulate the formation of the HIF
–ARNT complex by using
small organic compounds. Although neither a natural ligand nor
a ligand-binding cavity has been identified for the HIF
domains, this should not be interpreted to mean this domain
cannot bind any small compounds. Indeed, it has been demon-
strated that certain artificial chemicals can specifically bind into
the well packed hydrophobic core of a PAS domain from PAS
kinase (Fig. 2b) (17), likely producing significant conformational
changes therein. Parallel studies of two PAS domains with
natural cofactors, AsLOV2 and photoactive yellow protein, have
shown that relatively small structural changes in these internally
bound compounds are sufficient to displace
onto the same
-sheet interface we identified in HIF
Combining these observations and an earlier model proposed for
PAS domain signaling (44) raises the possibility that binding
compounds in the HIF
PAS-B core could generate structural
changes that disrupt the interactions needed to form the HIF
heterodimer, analogous to the HIF
here. Given the central role of HIF in the hypoxia response
and the importance of this pathway in cancer progression,
this approach may serve as the basis for a novel therapeutic
strategy (1, 2).
We thank M. Rhima for technical assistance; P. Ratcliffe (Oxford
University, Oxford) for the gift of the Ka13 cell line; and D. Russell, J.
Garcia, and S. McKnight for comments on the manuscript. This work was
supported by grants from the National Institutes of Health (CA90601 to
K.H.G. and CA95471 to K.H.G. and R.K.B.), the University of Texas
Southwestern Endowed Scholars Program (to K.H.G. and R.K.B.), and
the Robert A. Welch Foundation (I-1424 to K.H.G.). P.B.C. was
supported by a National Institutes of Health Training Grant GM08297
to the University of Texas Southwestern Graduate Program in Molecular
Biophysics. R.K.B. is supported by a Career Award in the Biomedical
Sciences from the Burroughs Wellcome Fund.
1. Semenza, G. L. (2000) Genes Dev. 14, 1983–1991.
2. Ratcliffe, P. J., Pugh, C. W. & Maxwell, P. H. (2000) Nature 6, 1315–1340.
3. Wang, G., Jiang, B., Rue, E. & Semenza, G. (1995) Proc. Natl. Acad. Sci. USA
4. Min, J. H., Yang, H., Ivan, M., Gertler, F., Kaelin, W. G., Jr., & Pavletich, N. P.
(2002) Science 296, 1886–1889.
5. Dames, S. A., Martinez-Yamout, M., De Guzman, R. N., Dyson, H. J. &
Wright, P. E. (2002) Proc. Natl. Acad. Sci. USA 99, 5271–5276.
6. Freedman, S. J., Sun, Z. Y., Poy, F., Kung, A. L., Livingston, D. M., Wagner,
G. & Eck, M. J. (2002) Proc. Natl. Acad. Sci. USA 99, 5367–5372.
7. Semenza, G. L. (1999) Annu. Rev. Cell Dev. Biol. 15, 551–578.
8. Tian, H., McKnight, S. & Russell, D. (1997) Genes Dev. 11, 72–82.
9. Gu, Y.-Z., Moran, S. M., Hogenesch, J. B., Wartman, L. & Bradfield, C. A.
(1998) Genes Dev. 7, 205–213.
10. Massari, M. E. & Murre, C. (2000) Mol. Cell. Biol. 20, 429–440.
11. Taylor, B. L. & Zhulin, I. B. (1999) Microbiol. Mol. Biol. Rev. 63, 479–506.
12. Pongratz, I., Antonsson, C., Whitelaw, M. L. & Poellinger, L. (1998) Mol. Cell.
Biol. 18, 4079–4088.
13. Hogenesch, J. B., Chan, W. K., Jackiw, V. H., Brown, R. C., Gu, Y.-Z.,
Pray-Grant, M., Perdew, G. H. & Bradfield, C. A. (1997) J. Biol. Chem. 272,
14. Jiang, B.-H., Rue, E., Wang, G. L., Roe, R. & Semenza, G. L. (1996) J. Biol.
Chem. 271, 17771–17778.
15. Gradin, K., McGuire, J., Wenger, R., Kvietikova, I., Whitelaw, M., Toftgard,
R., Tora, L., Gassmann, M. & Poellinger, L. (1996) Mol. Cell. Biol. 16,
16. Sheffield, P., Garrard, S. & Derewenda, Z. (1999) Protein Expr. Purif. 15, 34–39.
17. Amezcua, C. A., Harper, S. M., Rutter, J. & Gardner, K. H. (2002) Structure
(Cambridge, U.K.) 10, 1349–1361.
18. Erbel, P. J. A., Barr, K., Gao, N., Gerwig, G. J., Rick, P. D. & Gardner, K. H.
(2003) J. Bacteriol. 185, 1995–2004.
19. Delaglio, F., Grzesiek, S., Vuister, G. W., Zhu, G., Pfeifer, J. & Bax, A. (1995)
J. Biomol. NMR 6, 277–293.
20. Johnson, B. A. & Blevins, R. A. (1994) J. Biomol. NMR 4, 603–614.
21. Sattler, M., Schleucher, J. & Griesinger, C. (1999) Prog. NMR Spectrosc. 34,
22. Bai, Y., Milne, J. S., Mayne, L. & Englander, S. W. (1993) Proteins 17, 75–86.
23. Cornilescu, G., Delaglio, F. & Bax, A. (1999) J. Biomol. NMR 13, 289–302.
24. Bru¨nger, A. T., Adams, P. D., Clore, G. M., Delano, W. L., Gros, P.,
Grosse-Kunstleve, R. W., Jiang, J. S., Kuszewski, J., Nilges, M. & Pannu, N. S.
(1998) Acta Crystallogr. D 54, 905–921.
25. Nilges, M. & O’Donoghue, S. I. (1998) Prog. NMR Spectrosc. 32, 107–139.
26. Koradi, R., Billeter, M. & Wu¨thrich, K. (1996) J. Mol. Graphics 14, 51–55.
27. Laskowski, R. A., Rullman, J. A. C., MacArthur, M. W., Kaptein, R. &
Thornton, J. M. (1996) J. Biomol. NMR 8, 477–486.
28. Guex, N., and Peitsch, M. C. (1997) Electrophoresis 18, 2714–2723.
29. Marti-Renom, M. A., Stuart, A., Fiser, A., Sanchez, R., Melo, F. & Sali, A.
(2000) Annu. Rev. Biophys. Biomol. Struct. 29, 291–325.
30. Dalvit, C., Flocco, M., Knapp, S., Mostardini, M., Perego, R., Stockman, B. J.,
Veronesi, M. & Varasi, M. (2002) J. Am. Chem. Soc. 124, 7702–7709.
31. Palmer, A. G., III, Kroenke, C. D. & Loria, J. P. (2001) Methods Enzymol. 339,
32. Bruick, R. K. (2000) Proc. Natl. Acad. Sci. USA 97, 9082–9087.
33. Borgstahl, G. E. O., Williams, D. R. & Getzoff, E. D. (1995) Biochemistry 34,
34. Crosson, S. & Moffat, K. (2001) Proc. Natl. Acad. Sci. USA 98, 2995–3000.
35. Gong, W., Hao, B., Mansy, S. S., Gonzalez, G., Gilles-Gonzalez, M. A. & Chan,
M. K. (1998) Proc. Natl. Acad. Sci. USA 95, 15177–15182.
36. Harper, S. M., Neil, L. C. & Gardner, K. H. (2003) Science 301, 1541–1544.
37. Sette, M., van Tilborg, P., Spurio, R., Kaptein, R., Paci, M., Gualerzi, C. O. &
Boelens, R. (1997) EMBO J. 16, 1436–1443.
38. Matsuo, H., Walters, K. J., Teruya, K., Tanaka, T., Gassner, G. T., Lippard,
S. J., Kyogoku, Y. & Wagner, G. (1999) J. Am. Chem. Soc. 121, 9903–9904.
39. Wood, S. M., Wiesener, M. S., Yeates, K. M., Okada, N., Pugh, C. W., Maxwell,
P. H. & Ratcliffe, P. J. (1998) J. Biol. Chem. 273, 8360–8368.
40. Minet, E., Ernest, I., Michel, G., Roland, I., Remacle, J., Raes, M. & Michiels,
C. (1999) Biochem. Biophys. Res. Commun. 261, 534–540.
41. Chen, J., Zou, A., Splawski, I., Keating, M. T. & Sanguinetti, M. C. (1999)
J. Biol. Chem. 274, 10113–10118.
42. Morais Cabral, J. H., Lee, A., Cohen, S. L., Chait, B. T., Li, M. & Mackinnon,
R. (1998) Cell 95, 649–655.
43. Hoff, W. D., Xie, A., Van Stokkum, I. H. M., Tang, X. J., Gural, J., Kroon, A. R.
& Hellingwerf, K. J. (1999) Biochemistry 38, 1009–1017.
44. Cusanovich, M. A. & Meyer, T. E. (2003) Biochemistry 42, 4759–4770.
Erbel et al. PNAS
December 23, 2003