Cell, Vol. 93, 111±123, April 3, 1998, Copyright 1998 by Cell Press
Crystal Structure of the Hemochromatosis Protein
HFE and Characterization of Its Interaction
with Transferrin Receptor
His-41 to aspartate increases the relative risk of devel-
oping hemochromatosis inindividuals whoareheterozy-
gous for the Cys260Tyr mutation (Feder et al., 1996;
Beutler, 1997). Unlike the Cys260Tyr substitution, the
or cell surface expression (Feder et al., 1997; Waheed
et al., 1997).(Ournumbering system begins withresidue
1 of the mature protein and differs from previous sys-
tems that began numbering at the intitial methionine
[see Experimental Procedures]. Thus, the residues we
refer to as Cys-260 and His-41 correspond to Cys-282
and His-63 in earlier publications.)
The homology between HFE and MHC molecules
does notsuggest an obvious role forHFE in ironhomeo-
stasis, since class I MHC molecules function in peptide
presentation to T cells. A connection between HFE and
iron absorption was recently made with the demonstra-
tion that HFE associates with transferrin receptor (TfR)
(Parkkila et al., 1997; Feder et al., 1998) and decreases
its affinity for iron-bound transferrin (diferric Tf [Fe-Tf])
by 5- to 10-fold (Feder et al., 1998). The Tf/TfR system
is a well-established pathway by which cells absorb
iron. Fe-Tf in the blood binds to cell-surface TfR and
triggers endocytosis of the Tf/TfR complex (reviewed in
Richardson and Ponka, 1997). Upon exposure to the
acidic pH of the endosome, iron is released from Tf and
enters a chelatable intracellular pool from which it is
utilized for the metabolic needs of the cell or incorpo-
rated into the storage protein ferritin. Apo-Tf remains
bound to TfR at the low pH of the acidic vesicle (?pH
6.0) and the apo-Tf/TfR complex is then recycled to the
cell surface where apo-Tf dissociates at the pH of blood
(?pH 7.4). The role of TfR in iron uptake has been well
characterized but the association with HFE was pre-
viously unnoticed, perhaps because HFE is not ex-
pressed in commonly used cultured cell systems (e.g.,
HeLa cells, 293 cells;J .N.F, unpublished data),possibly
an adaptive response to allow increased iron uptake.
The functional consequences of the primary mutation
in HFE, Cys260Tyr, were addressed by the demonstra-
tion that, unlike wild-type HFE, the Cys260Tyr mutant
does not interact with TfR and therefore does not de-
crease TfR's affinity for Fe-Tf (Feder et al., 1998). In
contrast, the His41Asp mutantform of HFE stillinteracts
with TfR but does not decrease the affinity for Fe-Tf to
the same extentas wild-typeHFE. Theresults described
above suggest that when HFE is expressed in the same
cell as TfR, less Tf-associated iron willbe taken into the
cell. In the absence of HFE, as in the hemochromatotic
individual carrying the Cys260Tyr mutation, more iron
would be brought into cells, ultimately resulting in in-
creased iron deposition in cells that normally use HFE
for modulation of iron intake.
In this study, we have determined the X-ray crystal
structure of a solubleform of humanHFE and character-
ized the interactions between HFE, Tf, and TfR. Com-
bined with the observation that HFE reduces the affinity
of TfR for Fe-Tf (Feder et al., 1998), our results support
a model in which HFE is involved in the canonical iron
absorption pathway and regulates iron intake by a
J ose ÂA. Lebro Ân,*?Melanie J . Bennett,*?
Daniel E. Vaughn,*# Arthur J . Chirino,*²
Peter M. Snow,² ³Gabriel A. Mintier,§
J ohn N. Feder,§and Pamela J . Bjorkman*²
*Division of Biology
²Howard Hughes Medical Institute
³Caltech Protein Expression Center
California Institute of Technology
Pasadena, California 91125
Menlo Park, California 94025
HFE is an MHC-related protein that is mutated in the
iron-overload disease hereditary hemochromatosis.
HFE binds to transferrin receptor (TfR) and reduces
its affinity for iron-loaded transferrin, implicating HFE
in iron metabolism. The 2.6 AÊcrystal structure of HFE
reveals the locations of hemochromatosis mutations
and a patch of histidines that could be involved in
pH-dependent interactions. We also demonstrate that
soluble TfR and HFE bind tightly at the basic pH of the
cell surface, but not at the acidic pH of intracellular
vesicles. TfR:HFE stoichiometry (2:1) differs from TfR:
transferrin stoichiometry (2:2), implying a different
mode ofbinding forHFE andtransferrinto TfR, consis-
tent with our demonstration that HFE, transferrin, and
TfR form a ternary complex.
Hereditary hemochromatosis (HH) is a disease charac-
terized by the excessive deposition of iron in different
organs of the body leading to multiorgan dysfunction.
HH is the most common autosomal recessive disorder
affecting individuals of Northern European descent. Ap-
proximately 1 in 200 to 1 in 400 Caucasian individuals
have HH, leading to an estimated carrier frequency of
between 1 in 8 and 1 in 10 (Merryweather-Clarke et al.,
1997, and references therein). A candidate gene for HH
was identifiednear,butnotin,the majorhistocompatibil-
ity complex (MHC)(Federet al., 1996). Thegene product
HFE is a 343 residue type Itransmembrane glycoprotein
that is homologous to class IMHC proteins and associ-
ates with the class I light chain ?2-microglobulin (?2m)
(Feder et al., 1996). Between 69% and 100% of HH
patients are homozygous for a mutation (845G→A) that
converts Cys-260 to a tyrosine (reviewed in Cuthbert,
1997), preventing formation of a disulfide bond in the
?3 domain and abrogating ?2m association as well as
cell-surface expressionof the protein(Federetal., 1997;
Waheed et al., 1997). A second mutation that converts
?These authors contributed equally to this work.
#Present address: Cold Spring Harbor Laboratory, 1 Bungtown
Road, Cold Spring Harbor, NY 11724.
asymmetry thatdeviatesfromthe pseudo-dyad symme-
try relating antibody constant domains. Specifically, the
HFE domains are related by a 161? rotation and a 13 AÊ
translation. This relative orientation is in the range seen
for the ?3-?2m relationship in class I molecules (146?-
160? rotation, 13±14 AÊtranslation), CD1 (170? rotation,
12 AÊtranslation) (reviewed in Zeng et al., 1997), and
FcRn (157? rotation, 13 AÊtranslation) (Burmeister et al.,
The HFE residue involved in the Cys260Tyr mutation
is in the ?3 domain, where it disulfide bonds with Cys-
203 (Figures 1A and 1C). His-41, the residue substituted
in the His41Asp mutation, is in a loop in the ?1 domain,
where it forms a salt bridge with Asp-73 (Figure 1B).
Substitution of an aspartate at position 41 would be
unlikely to affect the overall protein fold but would dis-
rupt the salt bridge with Asp-73, perhaps leading to a
local rearrangement of the loop to avoid juxtaposition
of two negative charges.
The surface of HFE includes a clusterof fourhistidine
residues accounting for one-third of the total histidines
inthe humanHFE heavy chain (Federetal., 1996)(Figure
1D). The clustering of these histidines with a nearby
tyrosine (HFE Tyr-118) bears some resemblance to the
composition of mononuclear iron±binding sites in sev-
eral proteins (Howard and Rees, 1991). Thus far, analy-
ses of crystals soaked at near-neutral or basic pH in
iron (10 mM Fe(III)-NTA or 10 mM Fe(II)(NH4)2SO4) or
chemically similarmetals (10 mM NiCl2or10 mM MnCl2)
have not yielded evidence of metalbinding in this region
(M. J . B., J . A. L., and P.J .B., unpublished data). How-
ever, the conditions under which the crystals were
soaked may not have been optimal for metal binding.
Table 1. Data Collection and Refinement Statistics for HFE
Unit cell dimensions
a, b, c (AÊ)
68.8, 100.1, 147.6
Reflections in working set
Reflections in test set
Rms deviations from ideality
Bond lengths (AÊ)
Bond angles (deg)
Number of nonhydrogen atoms
aRmerge(I) ? (?|(I(i) ? ?I(h)?|)/?I(i), summed over all reflections and
all observations, where I(i) is the ith observation of the intensity
of the hkl reflection and ?I(h)? is the mean intensity of the hkl
bRcryst(F) ??h|| Fobs(h)|?|Fc(h)||
calculated structure factor amplitudes for the hkl reflection. Rfreeis
calculated for a set of reflections that were not included in atomic
refinement (Bru Ènger, 1992b).
, where Fobs and Fcare the observed and
Groove Narrowing in HFE Prevents Peptide Binding
Although the overall structure of HFE resembles class I
MHC molecules, HFE lacks a functionalpeptide-binding
groove.Whereas class Imolecules bind short(8±10 resi-
dues) peptides (Rammensee et al., 1993), peptides are
not associated with HFE (Table 2). The crystal structure
reveals the reason for HFE's lack of peptide binding:
its counterpart of the MHC peptide-binding groove is
narrowed by a translation of the ?1 helix bringing it ?4
AÊcloser to the ?2 helix (Figure 2A). A striking feature of
the HFE groove is that the entire ?2 helix is almost
identically positioned tothe ?2helix inclass Imolecules,
whereas in the MHC-related molecules FcRn and CD1
both helices are repositioned.
The narrower groove in HFE results in burial of resi-
dues analogous to those forming pockets that interact
with peptides in class I binding clefts. Class I pockets
A and F accommodate the N and C termini of bound
peptides and are lined with mostly conserved residues,
while the intermediate pockets (B to E) interact with
peptide side chains and contain residues that vary (Fig-
ure2B; Saperetal., 1991;Matsumura et al., 1992). Over-
all, about half of the HFE residues located in positions
analogous to class I pockets A through F are buried,
preventing them from an interaction with peptide (Table
3). In class I pocket A, tyrosines 7, 59, 159, and 171
interact with the peptide N terminus, and Trp-167 is at
the groove rim in many class I alleles (Figure 2B). In
HFE, only two of the tyrosines are conserved (HFE Tyr-
10 and Tyr-160) and Tyr-10 is buried (Table 3). In addi-
tion, the side chain of HFE Gln-168 (class Iresidue 167)
mechanism that involves binding to TfR in a pH-depen-
Structure of HFE
Soluble HFE/?2m heterodimers were expressed in Chi-
nese hamster ovary (CHO) cells. The crystal structure
was determinedto 2.6 AÊby molecularreplacement using
a 2.0 AÊstructure of the human class I MHC molecule
HLA-A2 (Collins et al., 1994). The refined structure has
good stereochemistry (Table 1)and 94% of the residues
within allowed regions of the Ramachandran plot as
defined by Kleywegt and J ones (1996).
The overall structure of HFE resembles MHC class I
molecules (such as HLA-A2, with which it shares 37%
sequence identity). In both, the ?1and ?2domains form
a platform composed of eight antiparallel ? strands
topped by two antiparallel ? helices (Figure 1B) posi-
tioned on top of two immunoglobulin constant-like do-
mains: ?3 and the light chain ?2m (Figure 1A). The indi-
vidual domains in HFE can be superimposed upon the
corresponding domains in HLA-A2 with rms deviations
of less than 1.5 AÊfor most C? atoms, comparable to
superpositions of two other class I MHC-related pro-
teins: the neonatalFc receptor(FcRn)(Burmeisteret al.,
1994a) and CD1, a class Ib MHC protein (Zeng et al.,
1997)(see ExperimentalProcedures). InHFE, as in class
I and class I-related proteins, ?3 and ?2m interact with
Structure and Function of HFE
Figure 1. Crystal Structure of HFE and Comparison to a Class I MHC Molecule
(A) Ribbon diagram shows that HFE resembles class I molecules in the fold of the heavy chain (blue) and in its association with the ?2m light
chain (green). Cys-260, the residue substituted in the Cys260Tyr mutation, disulfide bonds with Cys-203.
(B) Ribbon drawing of a top view of the HFE ?1-?2 platform. His-41 (red), the site of the His41Asp mutation, interacts with Asp-73 (green).
(C) The HFE model in the region of the Cys-260Ð Cys-203 disulfide bond is shown superimposed on a 2Fo?Fcannealed omit electron density
map (Hodel et al., 1992) contoured at 1?. The average B factor for the residues shown is 48 AÊ2.
(D) Close-up of a histidine cluster and nearby tyrosine located underneath the right-hand side of the platform. His-94 is found in class I MHC
molecules (class I His-93); His-89 and His-123 are present only in human (Feder et al., 1996), rat (EMBL accession number AJ 001517), and
mouse (Hashimoto et al., 1997) HFE. His-87 is present in human, but not mouse or rat, HFE.
(A), (B), and (D) were made with Molscript (Kraulis, 1991) and rendered with Raster3D (Merritt and Murphy, 1994). (C) was prepared with O
(J ones and Kjeldgaard, 1997).
points into the groove to occlude pocket A (Figure 2C)
in a mannerreminiscent of Arg-164 in FcRn (Burmeister
et al., 1994a).
In addition to resulting in burial of pocket residues,
the groove narrowing causes several residues in HFE
to occupy positions that would clash witha bound pep-
tide. To identify only those HFE residues that would
clash with all bound peptides regardless of sequence,
the side chains in four defined nonameric peptides
bound to HLA-A2 (Madden et al., 1993) were truncated
to alanines and superimposed upon HFE. In total, eight
side chains from the HFE ?1 helix and two side chains
from the ?2 helix are incompatible with peptide binding
(Figure 2C). The clashes are caused both by the transla-
tion of the ?1 helix toward the ?2 helix and by the pres-
ence of largerside chains in HFE as compared to class
I (HFE Leu-69 versus class I Val-67, HFE Trp-72 versus
class I His-70, HFE Met-75 versus class I Thr-73, and
HFE Arg-153 versus class I Val-152). Additional HFE
residues probably also contribute to HFE's inability to
bind peptides with side chains largerthan alanine; e.g.,
Trp-114 on the HFE groove floor (Figure 2C).
The HFE counterpart of the peptide-binding groove
is distinct from the grooves in two other MHC class
I±related proteins of known structure (Figure 2A): FcRn,
which has an almost completely closed groove (Bur-
meisteret al., 1994a), and CD1(Zeng et al., 1997), which
contains a deep hydrophobic groove that binds lipids
(Beckman et al., 1994) and long (12±22 residues) hy-
drophobic peptides (Castan Äo et al., 1995). The totalsur-
face area of the groove in HFE (?415 AÊ2) is intermediate
between that in FcRn (?235 AÊ2) and class I molecules
(?760 AÊ2). CD1 has a narrower but deeper groove than
class I grooves, with the most extensive surface area
of all (?1440 AÊ2) (see Experimental Procedures).
The structural rearrangements resulting in grooves of
various sizes and shapes differ in HFE, FcRn, and CD1.
In contrast to HFE, in which the ?2 helix is positioned
similarly to its counterpart in class I molecules, the ?2
helix of FcRn or CD1 is kinked about a hinge point near
that the HFE-TfR interaction does not require the trans-
membrane domain of either protein (Feder et al., 1998),
soluble TfR is an appropriate reagent for analyses of
Inorder to measure the affinity betweenHFE and TfR,
HFE was covalently immobilized to a biosensor chip
using primaryaminechemistry (seeExperimentalProce-
dures). At pH 7.5, TfR binds immobilized HFE with high
affinity (KD?0.6 nM). The interaction was not affected
by the addition of the iron chelator pyrophosphate (PPi)
during the assay (data not shown), demonstrating that
HFE binding to TfR does not require iron. In contrast to
the high affinity binding between TfR and HFE, Tf does
not bind HFE (Table 4). Thus, the HFE-mediated de-
crease in the affinity of TfR for Fe-Tf (Feder et al.,
1998) does not involve a direct interaction between
HFE and Tf.
When the HFE-TfR interaction was monitored at pH
7.5 in the reverse orientation by injecting HFE over pri-
mary amine±coupled TfR, the affinity is significantly
lower (KD?240 nM) (Table 4). In order to investigate
whether this decreased affinity was a result of the
method of TfR immobilization (see Experimental Proce-
dures), we injected HFE overTfR that was noncovalently
coupled via its 6xHis-tag to a biosensorchip derivatized
with nickel nitrilotriacetic acid (Ni-NTA). However, since
HFE interacts directly with the Ni-NTA chip at high HFE
concentrations (?50 nM; Figure 3C) (perhaps due to
nickel binding to the His cluster on HFE; Figure 1D),
we could not obtain a precise value for the KDof this
interaction. Nonetheless, the approximate affinity and
dissociation rate derived from the interaction of HFE
with His-tagged TfR captured on the Ni-NTA chip are
consistent with those derived using covalently immobi-
lized TfR and not with those derived using soluble TfR
and immobilizedHFE (Figure 3; Table4). Thus, TfR binds
to immobilized HFE more tightly than HFE binds to im-
The biosensor studies demonstrate that TfR and HFE
interact strongly with an affinity either comparable to
the affinity of TfR for its Tf ligand or no more than 100-
fold weaker(Figure 3; Table 4). We cannot ascertain the
reason for the coupling-dependent affinity difference of
the HFE-TfR interaction, but note that coupling-depen-
dent differences have beenobserved in other biosensor
assays (Kuziemko et al., 1996; Vaughn and Bjorkman,
1997). In those cases, the higher affinity values corre-
sponded more closely to values derived from cell bind-
ing assays. By analogy, we expect the higheraffinity (KD
?0.6 nM; Table 4) to be more relevantforthe physiologi-
cal interaction of TfR and HFE so that TfR binds HFE at
least as tightly as it binds its Tf ligand. In vivo, complex
formationbetweenTfR and HFE is likely to be even more
favored than predicted from the affinities determined
in these biosensor studies due to tethering of the two
proteins on the same membrane.
At pH 6.0, HFE shows either only very weak binding
(KD? 10 mM) or no detectable binding to TfR (Table 4;
Figures 3A and 3B). Thus, the TfR-HFE binding affinity
drops from nanomolar to essentially undetectable over
a change in pH of less than two units (Table 4). Histidine
residues, with their near-neutral pKa's, are likely candi-
dates to mediate pH-dependent binding over this pH
Table 2. Amino Acids Recovered from Acid Elutionsa
Cycle NumberHFE FcRnUL18
aTotalyield (pmols)ofaminoacids fromeachN-terminalsequencing
cycle for acid eluates derived from equivalent amounts of HFE,
FcRn, and UL18.Previous studies established that UL18 and classi-
cal class I molecules, but not FcRn, associate with endogenous
peptides when expressed as soluble proteins in CHO cells (Chap-
man and Bjorkman, 1998, and references therein). Only those amino
acid residues that showedanincrease inthe absolute amountrecov-
ered compared to the previous cycle were considered significant.
Analysis of the HFE and FcRn eluates by matrix-assisted, laser
desorption, time-of-flight mass spectrometry using a PerSeptive
Biosystems ELITE mass spectrometer did not reveal the presence
of N-terminally blocked peptides (data not shown).
a proline. HFE is the only class I homolog that has a
proline at this position(HFE Pro-166, FcRnPro-162, and
CD1 Pro-169) in which a kink is not seen. Instead, HFE
Pro-166 is accommodated without significant structural
rearrangements relative to class Imolecules (Figure 2A),
which contain a valine at this position (Val-165). The
reason appears to be that a local distortion, perhaps
related to a neighboring disulfide bridge involving class
I Cys-164, occurs at this point in class I molecules, and
the -NH group of the subsequent class I residue (Val-
165) is not hydrogen bonded in the helix. Thus, this
position in the ?2 helix can accommodate a proline as
occurs in HFE without altering the structure.
HFE Binds Tightly to TfR at pH 7.5
but Not at pH 6
In order to further characterize the interaction of HFE
with TfR, we expressed a soluble form of TfR (residues
121±760) corresponding to a previously characterized
proteolytic fragment purified from human placenta (Tur-
kewitz et al., 1988a). Since TfR is normally a disulfide-
linked homodimeric type II membrane glycoprotein
(Schneider et al., 1984), we firstverified that the proper-
ties of soluble recombinant TfR are similar to those of
its membrane-bound counterpart. Analytical ultracen-
trifugation demonstrated that soluble TfR is dimeric
(data not shown).Ina surface plasmon resonance (SPR)
based assay, soluble TfR binds Fe-Tf with a KD?3 nM
at pH 7.5 (Figure 3; Table 4; KD's cited in the text are
the medianof the rangeof values derived underdifferent
experimental conditions), consistent with the affinity of
Fe-Tf for membrane-bound TfR (5 nM; Richardson and
Ponka, 1997, and references therein). Soluble TfR also
retains membrane-bound TfR's pH-dependent affinity
for apo-Tf, binding at pH 6.0 (KD?8 nM) but not at pH
7.5 (Figure 3; Table 4). Since soluble TfR retains the
binding properties and physicalcharacteristics of mem-
brane-bound TfR and because previous studies implied
Structure and Function of HFE
range (Fersht, 1985). Specifically, neutral histidines
could be involved in binding of TfR and HFE at the
pH of the cell surface (?pH 7.4) with protonation of
histidines in acidic vesicles (?pH 6.0) then mediating
the dissociation of TfR and HFE. The prominent patch
of histidines in HFE (Figure 1D) and/or His-41, the site
of one of the HH mutations, could be involved in these
sorts of pH-dependent interactions (see Discussion).
Insight into the molecular mechanism of HFE in iron
uptake regulation is provided by the recent discovery
that HFE binds to TfR (Parkkila et al., 1997; Feder et al.,
1998) and thereby reduces its affinity for Fe-Tf (Feder
et al., 1998).The HH mutations eithereliminate the bind-
ing of HFE to TfR (Cys260Tyr) or alter HFE's ability to
reduce the affinity between TfR and Fe-Tf (His41Asp)
(Feder et al., 1998). How perturbation of the HFE-TfR
association might result in increased iron absorption in
the smallintestine,as seenin patients withHH (McLaren
et al., 1991), will require additional studies (for further
discussion see Feder et al., 1998). However, a higher
affinity between TfR and Fe-Tf due to the absence of
fully functional HFE could lead to increased iron uptake
by cells in some tissues, ultimately causing excess iron
deposition in the major organs, a primary defect in HH
(Bacon and Tavill, 1996).
To further define the HFE-TfR association, we have
begun to characterize the interaction between soluble
forms of HFE and TfR. Considering that HFE was only
recently identified as a component in the TfR pathway,
HFE and TfR form a surprisingly high affinity complex
at the slightly basic pH found atthe cellsurface (KD?0.6
nM), whileat the acidic pHcorresponding to intracellular
vesicles there is little or no binding of HFE to TfR. The
pH-dependent affinityofthe HFE-TfR interactionisremi-
niscent of the interaction between TfR and apo-Tf and
of the interactionbetween FcRn and its immunoglobulin
G (IgG) ligand, interactions that each show a sharp pH
dependence in the range between pH 6 and pH 7.5
(reviewed in Richardson and Ponka, 1997; J unghans,
1997). In the TfR-apo-Tf and FcRn-IgG systems, the
receptors' functions involve trafficking through acidic
compartments as a complex with ligand and release of
ligand at the slightly basic pH of blood. The sharp pH
dependence of the HFE-TfR interaction, an unusual fea-
ture of a protein±protein interaction unless it is required
for pH-regulated binding during trafficking, implies that
HFE enters the cell along with TfR-Tf complexes, then
dissociates from TfR in acidic vesicles. Thus, studies of
HFE trafficking and/or recycling might be pertinent to
HFE's role in the regulation of iron homeostasis.
Histidine residues are likely candidates for mediating
pH-dependent protein interactions at pH values near
neutral. Histidines haveapKaof 6.6inmodelcompounds
(Fersht, 1985) and are therefore likely to be neutral at
the basic pH of blood and to carry a positive charge at
the pH of acidic intracellular vesicles. The FcRn-IgG
system is a well-characterized example of a pH-depen-
dent affinity difference mediated throughtitration of his-
tidines (Vaughn and Bjorkman, 1998). By analogy, the
pH dependence of the HFE-TfR interaction could be due
to a favorable interaction at neutral or slightly basic pH
involving uncharged histidines on one or both proteins,
which becomes unfavorable upon acquiring positive
charge(s) by protonation at acidic pH. The distribution
of histidines in TfR is unknown, but the HFE structure
includes an intriguing patchof histidines (Figure 1D)that
could act as a pH-dependent switch to modulate the
interaction. In addition to the potential involvement of
the clustered histidines inintermolecular pH-dependent
interactions, HFE His-41 is involved in an intramolecular
pH-dependent interaction in that it forms a salt bridge
HFE and Tf Bind to TfR with Different
Stoichiometries and Can Bind Simultaneously
to Form a Ternary Complex
We used a gel filtration assay to determine the stoichi-
ometry of the interaction of TfR with Tf and with HFE.
AtpH 7.5, TfR andFe-Tf form acomplex with2:2 stoichi-
ometry (Figure 4), consistent with earlier results (Enns
and Sussman, 1981) and with the hypothesis that each
polypeptide chain in the TfR homodimer binds to one
Tf molecule. At the same pH, TfR complexes with HFE
with 2:1 stoichiometry (Figure 4), corresponding to one
TfR homodimer binding only one HFE. At pH 6.0, HFE
and TfR do not form a complex that is observed on the
sizing column (data not shown), as expected from the
biosensor studies (Table 4).
The observation that HFE and Tf bind to TfR with
differentstoichiometries implies thatTfR uses adifferent
mode of binding to interact with each protein, which in
turn suggested that a ternary complex of TfR, Tf, and
HFE could form. We used an anti-HFE monoclonal anti-
body (1C3) that does not interfere with HFE binding
to TfR to investigate whether Tf coimmunoprecipitates
along with HFE and TfR. HFE, TfR, and Fe-Tf were incu-
bated at a 1:2:2 molar ratio (corresponding to one HFE,
one TfR homodimer, and two molecules of Fe-Tf), fol-
lowed by immunoprecipitation. SDS-PAGE analysis
demonstrates that bands corresponding to HFE, TfR,
and Fe-Tf are present regardless of the orderof addition
of the proteins (Figure 5).
HFE was initially implicated in iron metabolism by the
discovery that it is mutated in patients afflicted with
HH, an iron-overload disorder (Feder et al., 1996). The
majority of HH patients are homozygous fora single site
mutation leading to the Cys260Tyr substitution (re-
viewed in Cuthbert, 1997) that disrupts the interaction
between HFE and its ?2m light chain and prevents cell-
surface expression (Feder et al., 1997; Waheed et al.,
1997). The 2.6 AÊcrystal structure of HFE confirms that,
as predicted from its sequence (Feder et al., 1996),
HFE closely resembles class IMHC molecules and that
Cys-260 is involved in a disulfide bridge analogous to
those found in class I MHC ?3 domains. The residue
(His41Asp)is locatedinaloop withinthe ?1domain.This
mutation may alter the structure of the protein locally
by a loop rearrangement to avoid juxtaposition of the
substituted aspartate with Asp-73, a residue with which
His-41 normally interacts. The existence of these muta-
tions in HH patients implies that properly folded HFE is
required for prevention of iron overload.
Figure 2. The Counterpart of the Class I Peptide-Binding Groove is Narrowed in HFE by Translation of the ?1 Helix
(A) C? stereo superpositions based on C? atoms in the platform ? strands of HFE with class I and class I±related proteins. Top, HLA-A2
(green, including a ball-and-stick representation of bound peptide; PDB code 2CLR) and HFE (magenta). Heavy chains share 37% amino acid
sequence identity. Middle, Rat FcRn (green; PDB code 1FRU) and HFE (magenta). Heavy chains share 28% sequence identity. HFE Pro-166
(labeled) is analogous to FcRn Pro-162, located at a kink in the FcRn ?2 helix. Bottom, Mouse CD1 (green) (Zeng et al., 1997) and HFE
Structure and Function of HFE
Table 3. Comparison of Residues in Peptide-Binding Grooves of HLA-A2 and HFE
Clash with Peptide?b
Not in a pocket
A, B, C, D
C, D, E
Not in a pocket
C, D, E
aPocket residues in the peptide binding groove of HLA-A2 are defined as having ?5.0 AÊ2of solvent accessible surface area with a 1.4 AÊ
probe radius but ?5.0 AÊ2with a 5.0 AÊprobe radius. Analogous residues in HFE are listed for comparison. ªBuriedº indicates an HFE residue
that is inaccessible to a 1.4 AÊprobe. Surface areas were calculated excluding water molecules and bound peptide using the coordinates of
HLA-A2 (PDB code 2CLR) and HFE.
bSteric clashes with polyalanine peptides defined as described (Figure 2C).
cResidues that have accessible surface in the binding cleft of HLA-A2 and hence constitute pockets (Saper et al., 1991), but are accessible
to a 5.0 AÊprobe and do not meet the criterion for pocket residues described above.a
with Asp-73 (Figure 1B) that would be more stable at
acidic compared to basic pH. Interestingly, this residue
is mutated to aspartate in some HH patients (Feder et
al., 1996; Beutler, 1997), destroying its potential to form
a pH-dependent salt bridge.
BecauseHFE is structurally similarto MHC molecules,
another candidate for its ligand recognition site is the
HFE counterpart of the MHC peptide-binding groove, a
characteristic feature of classical class Imolecules and
CD1. Biochemical and structural analyses demonstrate
that HFE does not bind peptides or other small mole-
cules in this region. Although the region of HFE involved
in ligand binding remains to be established, the nar-
rowing of the HFE groove comparedto class Imolecules
suggests that HFE does not use what remains of its
groove for ligand recognition in a manner analogous to
peptide binding in MHC molecules. That is, although
TfR could bind to this general area on HFE, the HFE
groove is not large enough to accept a loop from TfR
in the position where a peptide would bind to a class I
molecule. Alternatively, HFE may use an entirely differ-
entsurfaceforbinding ligands.A precedentforanMHC-
related molecule that binds ligands using a molecular
surface other than the groove is found in the example
of FcRn. Like HFE, FcRn functions ina recognitionevent
(binding and transporting IgG) that bears no resem-
blance to the antigen presentation functions of classical
class IMHC molecules and CD1, and its counterpart of
the MHC peptide-binding groove is closed and is not
the binding site for IgG (Burmeister et al., 1994b).
The interaction between HFE and TfR will ultimately
require a detailed characterization of both proteins,
which we have initiated by an analysis of their complex.
Our observation that the stoichiometry of TfR's interac-
tion with Fe-Tf (2:2) differs from the stoichiometry of its
interaction with HFE (2:1) implies that HFE and Tf bind
(magenta). Heavy chains share 22% sequence identity. HFE Pro-166 (labeled) is analogous to CD1 Pro-162, located at a kink in the CD1 ?2
(B) Peptide-binding groove of HLA-A2 with labeled binding pockets.
(C) HFE counterpart of the class I binding groove showing residues that would clash with a bound peptide. HFE side chains shown in white
(same in HFE and HLA-A2) and green (different in HFE and HLA-A2) have at least two steric clashes with polyalanine versions of four different
nonameric peptides (from HLA-A2 structures, PDB codes 1HHG, 1HHI, 1HHJ , and 1HHK; Madden et al., 1993) that were superimposed upon
HFE after alignment of HLA-A2 and HFE. Steric clashes are defined as occurring when atoms are closer than the sum of their van der Waals
radii minus 0.4 AÊ. The blue side chain would clash with peptides containing side chains larger than alanine.
All panels were made with Molscript (Kraulis, 1991) and rendered with Raster3D (Merritt and Murphy, 1994).
Figure 3. Biosensor Assays of TfR-Tf and TfR-HFE Binding
In each panel, the injected protein is indicated in front of an arrow pointing to the immobilized protein (coupled covalently via primary amines
[A and B] or coupled noncovalently via a 6xHis-tag to a Ni-NTA chip [C]). The model used to fit the data is listed along with a derived affinity
constant(s). ªHeterogeneousº refers to two classes of noninteracting binding sites on the coupled protein with different KD's (KD,1and KD,2),
Structure and Function of HFE
of the processed HFE chain as identified by N-terminal sequencing
is residue 1;previous numbering systems starting atthe initialmethi-
onine of the signal peptide refer to it as 23 (Cuthbert, 1997, and
Purified HFE or the control proteins FcRn and UL18 (0.25 mg of
each) were treated with acetic acid and analyzed for the presence
of bound peptides using established methods (Ro Ètzschke et al.,
1990) as previously described for UL18 and FcRn (Chapman and
Bjorkman, 1998). Low±molecular weight filtrates of the acid eluates
were lyophilized, and halfofeacheluate was analyzed by automated
Edman degradation using an Applied Biosystems model 477A pro-
tein sequencer for pool sequencing (Table 2).
differently to TfR, most likely to distinct regions of the
receptor (Figure 4). Regardless of where these proteins
bind to TfR, the finding that HFE and Fe-Tf can bind
simultaneously to TfR to form aternary complex demon-
strates that HFE does not occlude both Tf-binding sites
onTfR.SinceTfR is homodimeric (Schneideretal., 1984;
Turkewitz etal., 1988a), itseems reasonableto postulate
a 2-fold symmetric structure for a Tf-TfR complex in
whicheachpolypeptide chainbinds toone Tfto produce
2:2stoichiometry (Enns and Sussman,1981).Thefinding
that only one HFE binds to a TfR homodimer raises the
possibility that HFE binding induces asymmetry in the
TfR homodimer, resulting in only a single optimal HFE-
binding site. The resulting asymmetric receptor might
then be expected to bind Fe-Tf with lower affinity.
The crystal structure of HFE and the characterization
of its interactionwithTfR reported here provide a frame-
work for studies to elucidate the role of HFE in iron
homeostasis under normal circumstances and in the
disease state caused by iron overload. In addition, the
structure of HFE provides the first example of the use
of the MHC fold for a recognition event outside of the
immune system. Despite differences in function and li-
gand specificity, the three-dimensional structures of
classical class I molecules CD1, FcRn, and HFE are
remarkably similar, raising the intriguing questions of
why and how evolution has selected the MHC fold for
such diverse biological roles.
Crystallization and Data Collection
Crystals (space group P212121; a ? 68.8 AÊ, b ? 100.1 AÊ, c ? 147.6
AÊ; two molecules per asymmetric unit) of HFE were grown in 1:1
hanging drops containing Vibrio cholerae neuraminidase±treated
HFE (14mg/ml) and 16% (w/v) PEG 4000, 0.4 M ammonium acetate,
and 0.1 M sodium citrate (pH 5.9), then improved by microseeding
and macroseeding.Before datacollection, crystals were transferred
to a cryoprotectant solution (22% PEG 4000, 0.4 M ammonium ace-
tate, 0.1 M sodium citrate [pH 5.9] and 7.5% glycerol). Initial data
were collected at ?150?C from a single crystal to 2.9 AÊusing a MAR
Research detector at the Stanford Synchrotron Radiation Labora-
tory beamline 7±1. A second dataset was collected to 2.6 AÊusing
Fuji image plates and anoff-line scannerat the BrookhavenNational
Laboratory beamline X4A. The diffraction was anisotropic, ex-
tending beyond 2.0 AÊalong c*and to 2.6 AÊalong a*and b*. Data
were processed and scaled with DENZO and SCALEPACK (Otwi-
nowski and Minor, 1996) (Table 1).
Structure Determination and Refinement
The structure was determined by molecular replacement using
AMoRe (Navaza, 1994). A self-rotation function(15.0±4.0AÊ) revealed
a noncrystallographic 2-fold axis positioned at 45? in the x-y plane.
Cross-rotation and -translation functions (15.0±4.0 AÊ) using the 2.0
AÊstructure of HLA-A2 (PDB code 2CLR with the peptide omitted,
nonconserved side chains truncated to alanine, residues 124±156
deleted) as a search model yielded a solution for the first molecule
(correlation coefficient: 26.6%;R factorof55.0%). Thesecond mole-
cule was found inapartialtranslation functionwith the firstmolecule
fixed in which the coordinates for the first molecule were rotated
according to the noncrystallographic 2-fold and used as a search
model(correlation coefficient: 29.7%; R factorof 53.2%).Rigid body
refinement (6.0±3.5AÊ) of bothmolecules resulted inanRcrystof50.9%
(Rfree? 50.5%). Averaged and solvent-flattened maps calculated to
2.9 AÊwith DM (Cowtan, 1994) showed density for residues 124±156
and the ?1 helix (shifted ?4AÊrelative to HLA-A2).Residues 124±156
were modeled using O (J ones and Kjeldgaard, 1997), and the ?1
helix was positioned by rigid body refinement. Further rebuilding
was done using averaged simulated annealing Fo?aveomit maps
(Hodel et al., 1992) (throughout the model in ?10% increments) and
conventional (2Fo?Fc)?calcand (Fo?Fc)?calcmaps. Anisotropy and
bulk solvent corrections were applied, and the model was refined
against the Brookhaven dataset (15±2.6 AÊ) with tight NCS restraints
(300 kcal/mol´ AÊ2) and individual temperature (B) factors using
XPLOR (Bru Ènger, 1992a). Despite the relatively high mean B factor
Expression, Purification, and Characterization
of Soluble HFE
A construct encoding soluble HFE (residues 1±275 of the mature
protein) was subcloned after sequencing into the expression vector
PBJ 5-GS that carries the glutamine synthetase gene as a selectable
marker and means of gene amplification in the presence of methio-
nine sulfoximine (Bebbington and Hentschel, 1987). HFE and human
?2m expression vectors were cotransfected into CHO cells. Selec-
tion, amplification, and maintenance of methionine sulfoximine±
resistant cells and identification of HFE-expressing cells were done
as described (Chapman and Bjorkman, 1998). HFE/?2m heterodim-
ers were isolated from supernatants of cells grown in a hollow fiber
bioreactor device (Unisyn Fibertec) at yields up to 35 mg/liter using
an immunoaffinity column made with an anti-HFE monoclonal anti-
body (1C3) (J . A. L., H. Shen, P. J . B., and S. Ou, unpublished data).
HFE was eluted from the 1C3 column using 50 mM diethylamine
(pH 11.0) into tubes containing 1M monobasic sodium phosphate,
then further purified by anion exchange chromatography using an
FPLC mono Q column (Pharmacia Biotech). N-terminal sequence
analysis of purified protein yielded the sequences RLLRSHSLHYLF
and IQRTPKIQVYSR corresponding to correctly processed mature
HFE and human ?2m. In ournumbering system, the first amino acid
each representing the indicated percent of the total binding sites.
(A) Plots of equilibrium binding response (Req) versus the log of the concentration of injected protein derived from biosensor experiments in
which the binding response closely approached or reached equilibrium. Best-fit binding curves to the experimental data points are shown as
continuous lines. KD's derived from independent experiments performed on chips coupled to different densities agreed to within a factor of
(B) Sensorgrams (thick colored lines) from kinetics-based experiments overlaid with the calculated response (thin black lines) derived using
the model indicated on each panel. One representative set of injections from experiments performed in triplicate is shown for each interaction
(analyses from triplicate experiments reported in Table 4).
(C) Sensorgrams from kinetics-based experiments using TfR noncovalently coupled to a Ni-NTA chip. The HFE→TfR analysis is complicated
by a significant interaction of HFE with the Ni-NTA chip itself (blank responses and their corresponding binding responses after blank
subtraction are shown in the same color). Because the blanks represent a high proportion (more than half) of the total binding, the resulting
subtracted curves do not yield a precise value for the KD.
Figure 5. TfR, Tf, and HFE Can Form a Ternary Complex
Individual proteins ormixtures of proteins were immunoprecipitated
with 1C3 (anti-HFE) and analyzed on a 10% reducing SDS-PAGE
gel (conditions chosen to maximize separation between TfR and
Tf). Proteins listed in parentheses were incubated together first,
followed by addition of the third protein. The ?2mlight chain of HFE
and the antibody light chain are present on gels composed of a
higher percentage of acrylamide (data not shown).
Comparisons to Class I-Related Proteins
and Analyses of Groove Surfaces
Alignments were performed with LSQMAN (Kleywegt, 1996) (3.0 AÊ
maximum matching distance for C? pairs). Rms deviations for HFE
superimposed with HLA-A2 (PDB code 2CLR): 1.5 AÊ(?1; 71 C?'s),
1.2 AÊ(?2; 83 C?'s), and 0.8 AÊ(?3; 76 C?'s); FcRn (PDB code 1FRU)
superimposed with HLA-A2: 1.1 AÊ(?1; 62 C?'s), 1.6 AÊ(?2; 75 C?'s),
and 1.3 AÊ(?3; 77 C?'s); CD1 superimposed with HLA-A2: 1.2 AÊ(?1;
47 C?'s), 1.5 AÊ(?2; 71 C?'s), and 1.3 AÊ(?3; 81 C?'s).
Groove surface areas were calculated as follows. First, groove
residues were identifiedas thosehaving ?5.0 AÊ2of solvent-accessi-
ble surface area using a 1.4 AÊprobe radius but ?5.0 AÊ2of solvent-
accessible surface area calculated using a 5.0 AÊprobe radius in
XPLOR (Bru Ènger, 1992a). Using GRASP (Nicholls et al., 1991), we
then (i) built a molecular surface of only the groove residues, (ii)
selected the contiguous groove surface between the ?1 and ?2
helices by scribing its perimeter, and (iii) calculated its surface area.
This residue-based method of calculating groove surface areas dif-
fers from the atom-based method described by Zeng et al. (1997).
For comparison, we also used the atom-based method, which
yielded estimates for the groove surface areas in HFE (?410 AÊ2),
class I (?690 AÊ2), and CD1 (?1390 AÊ2) that were similar both to the
values calculated with the residue-based method (see text) and to
values previously reported (Zeng et al., 1997). However, we were
unable to define a contiguous groove surface for FcRn using the
atom-based method. The residue-based method yields a value of
?235 AÊ2, approximately half of the value reported by Zeng et al.
(1997), presumably because the grooves were defined differently.
Figure 4. Gel Filtration Chromatographic Demonstration that TfR
Binds to Tf and to HFE with Different Stoichiometries
TfR and Fe-Tf(leftpanel)orTfR and HFE (rightpanel)were incubated
at pH 7.4 at the indicated molar ratios, then passed over a sizing
columnto separate TfR:TforTfR:HFE complexes fromuncomplexed
proteins. At a 2:2 molar ratio of TfR to Tf and a 2:1 molar ratio
of TfR to HFE, virtually all of the protein chromatographed as the
complex. When the input ratio of TfR to Tf was greater than 2:2
there was excess TfR, and when it was less than 2:2 there was
excess Tf (verified by SDS-PAGE analysis; data not shown). Like-
wise, when the input ratio of TfR to HFE was greater than 2:1 there
was excessTfR, and whenitwas less than2:1therewas excessHFE.
Schematic representations of the TfR:Tf and TfR:HFE complexes
shown beside the chromatograms are consistent with the data,
but do not represent the only possible models accounting for the
Expression, Purification, and Characterization of TfR
A soluble version of TfR, normally a type II membrane glycoprotein
(Schneideret al., 1984), was expressed in a lytic baculovirus/insect
cellexpressionsystem.The portion ofthe humanTfR geneencoding
residues 121±760 (the C-terminal amino acid of wild-type TfR) was
fused 3? to a gene segment encoding the hydrophobic leader pep-
tide from the baculovirus protein gp67, a 6xHis-tag, and a factorXa
cleavage site in a modified version of the pAcGP67A expression
vector (Pharmingen). The N-terminal start site for soluble TfR was
chosen based on studies of a previously characterized soluble pro-
teolytic fragment of TfR beginning at residue 121, which has been
crystallized (Borhani and Harrison, 1991) and forms a stable dimer
that binds Tf, although it lacks two interchain disulfides involving
Cys-89and Cys-98(Turkewitz etal., 1988a). Recombinant virus was
generated by cotransfection of the transfer vector with linearized
viralDNA (Baculogold; Pharmingen). TfR was purified fromsuperna-
tants of baculovirus-infected High 5 cells using Ni-NTA chromatog-
raphy (Ni-NTA superflow; Qiagen) followed by gel filtration chroma-
tography using a Superdex-200 FPLC column (Pharmacia). A farUV
of 62AÊ2(WilsonB factor?67 AÊ2),the modelis generally welldefined
in the electron density (Figure 1C). The model (Rcryst ? 23.3%;
Rfree? 27.7%) includes 272 out of 275 residues in the recombinant
HFE heavy chain and all 99 of the residues in ?2m. No ordered
density was observed for carbohydrate at the three potential
N-linked glycosylation sites at positions 88, 108, and 212. A large
difference electron density peak near Phe-76 in an apolar pocket
was not conclusively identified but was modeled as water in the
absence of other chemical information. Residues 1±3 are not seen
in the electron density and 14 side chains are disordered and were
modeled as alanines (HFE residues 18, 42, 53±57, 63, 66±67, 106,
and 177; ?2m residues 48 and 75). Several regions in loops include
residues with real space correlation values (J ones and Kjeldgaard,
1997) below one standard deviation from the mean (residues 17±23,
53±61, 90±91, 106±108, 174±176, 225±230).
Structure and Function of HFE
Table 4. Biosensor Analyses of TfR Binding to Tf and HFE
Fe-Tf (pH 7.5)c
Fe-Tf (pH 7.5)d
apo-Tf (pH 6.0)e
apo-Tf ? PPi(pH 7.5)e
HFE (pH 7.5)f
HFE (pH 6.0)f
0.81 ? 0.1
1.3 ? 0.2
130 ? 10
3.1 ? 105
(7.3 ? 0.7) ? 105
(8.1 ? 0.9) ? 105
1.8 ? 10?3
(1.3 ? 0.2) ? 10?3
(9.4 ? 2) ? 10?4
(1.1 ? 0.1) ? 10?1
TfR (pH 7.5)g
TfR (pH 6.0)g
Fe-Tf (pH 7.5)g
apo-Tf (pH 6.0)g
0.33 ? 0.02
(3.8 ? 0.2) ? 106
(1.2 ? 0.1) ? 10?3
*Not determined because the experiment could not be performed (see Experimental Procedures).
N.B. No significant binding at concentrations up to 1 ?M.
aDetermined from equilibrium binding data. Only the higher affinity of two noninteracting binding sites is reported when binding curves were
fit to a model assuming two independent classes of binding sites (see Figure 3A).
bDetermined from the ratio of the kinetic constants (kd/ka) from experiments performed in triplicate when a standard deviation is given. Only
the higher affinity of two noninteracting binding sites is reported when sensorgrams were fit to a model assuming two independent classes
of binding sites (see Figure 3B).
c6xHis-tagged TfR was noncovalently immobilized to a density of 220 RU on an Ni-NTA sensor chip (Figure 3C).
dTfR was covalently immobilized to a density of 2310 RU for the equilibrium experiments and 420 RU for the kinetic experiments.
eTfR was covalently immobilized to a density of 1460 RU for the equilibrium experiments and 420 RU for the kinetic experiments. The
equilibrium measurements at pH 6 did not fully equilibrate during the injection time, thus 15 nM is the upper limit for the KD. 10 mM PPiwas
added to the buffer at pH 7.5 to prevent loading of apo-Tf with trace amounts of iron in the buffers.
fTfR was covalently immobilized to a density of 1600 RU for the equilibrium experiments and 420 RU for the kinetic experiments.
gHFE was covalently immobilized to a density of 3800 RU for the equilibrium experiments and 418 RU for the kinetic experiments.
CD spectrumof the purified protein(data not shown)verified folding
and resembled the spectrum of the proteolytic fragment of TfR
(Turkewitz et al., 1988b).
solutions to the activated flowcell. For the kinetic experiments in-
volving TfR, the 6xHis-tag was removed by factorXa(New England
Biolabs) treatment according to the manufacturer's instructions,
followed by purificationon a Biospincolumn (Biorad). Proteins were
injected at room temperature in 50 mM PIPES (pH 6.0 or pH 7.5),
150 mM NaCl, 0.005% BIAcore surfactant P20. All injections onto a
TfR- or HFE-coupled flowcellwere followed by an identical injection
onto a mock-coupledflowcelloraflowcellcoupled withan irrelevant
protein in order to subtract out significant nonspecific responses.
To achieve a defined orientation and to avoid exposing TfR to
the low pH conditions required for primary amine coupling (which
produce a conformational change resulting in self-association at
pH ?6 [Turkewitz et al., 1988b]; J . A. L., P. J . B., and P. Poon,
unpublished data), we injected soluble HFE over 6xHis-tagged TfR
noncovalently coupled to a biosensor chip derivatized with Ni-NTA.
Only kinetic experiments were performed using this chip, since the
amount of His-tagged TfR released from the Ni-NTA chip became
significant during the long injections required for the equilibrium-
based measurements. The Ni-NTA chip could not be used for bind-
ing studies at pH 6.0, as the 6xHis interaction with nickel is not
stable at this pH.
Biosensor-Based Affinity Measurements
A BIAcore 1000 biosensor system (Pharmacia LKB Biotechnology)
was used to assay interactions between HFE, TfR, and human Tf
(Sigma; Fe-Tfwas furtherpurified by gel filtration chromatography).
Binding between a molecule coupled to a biosensor chip and a
second molecule injected over the chip results in changes in the
SPR signal that are read out in real time as resonance units (RU)
(Malmqvistand Granzow,1994).We derivedequilibriumdissociation
constants (KD's) whenever possible using two methods. In the first
(KD,eqcolumn in Table 4), binding reactions were allowed to closely
approach or to reach equilibrium by using long injections (50 min)
with slow flow rates (5 ?l/min) over biosensor chips coupled to
high densities (?1000 RU). KD,eq values were derived by nonlinear
regressionanalysis of plots of Req(the equilibriumbinding response)
versus the log of the concentration of the injected protein (Figure
3A). The fit of data to binding models assuming one ormore classes
of interacting or noninteracting binding sites was then examined,
and the appropriate model was chosen as described (Vaughn and
Bjorkman, 1997). In the second method (KD,calc in Table 4; KD,calc ?
kd/ka; kaand kdare the association and dissociation rate constants,
respectively), kinetic constants were derived from binding experi-
ments conducted for shortertimes (2±4 min) using fasterflow rates
(50 ?l/min) over chips coupled at lower densities (?400 RU). These
conditions were chosen to minimize mass transport effects upon
the kinetics of binding reactions (Karlsson and Fa Èlt, 1997), which
are not a concern for the equilibrium measurements. Kinetic con-
stants werederived fromsensorgramdatausing simultaneous fitting
to the association and dissociation phases of the interaction and
global fitting to all curves in the working set (Figures 3B and 3C) as
implemented in BIAevaluation version 3.0.
TfR (20 ?g/ml in 5 mM maleate [pH 6.0]) or HFE (55 ?g/ml in 20
mMsodiumacetate [pH 5.0])was immobilizedusing standard amine
coupling chemistry on a CM5 chip (Pharmacia LKB Biotechnology).
Higher coupling densities for the equilibrium-based experiments
were achieved by increasing the time of exposure of the protein
Gel Filtration Analyses of TfR-Tf and TfR-HFE Stoichiometries
Protein concentrations were determined spectrophotometrically at
280 nm using the following extinction coefficients: HFE, 96570
M?1cm?1; TfR monomer, 93790M?1cm?1;Tf, 83360 M?1cm?1. Extinc-
tion coefficients were first calculated from the protein sequences,
then A280 measurements for a fixed amount of each protein were
compared in 6 M GuHCl and aqueous solutions and the coefficient
was adjusted if necessary. For the TfR:Tf experiments, molar ratios
from 3:2 to 1:2 of TfR and Fe-Tf were incubated for 20 min at room
temperature in 20 mM Tris (pH 7.4), 150 mM NaCl, 0.02% NaN3,
keeping the amount of TfR fixed at 200 pmol in a total volume of
25 ?l. For the TfR:HFE experiments, molar ratios from 3:1 to 1:1 of
TfR and HFE were incubated as described above, keeping the
amount of HFE fixed at 360 pmol ina total volume of 25 ?l.Samples
(25 ?l) were injected onto a Superose 6B FPLC column (Pharmacia),
eluted with the same bufferat0.5 ml/min,and the fractions analyzed
by SDS-PAGE (data not shown).
Coimmunoprecipitation of HFE, TfR,
and Tf in a Ternary Complex
TfR (138 pmol), Fe-Tf(138 pmol), and HFE (69 pmol)were incubated
for 15 min at room temperature in 20 mM Tris (pH 7.4), 150 mM
NaCl, 0.025% NaN3(incubation buffer). Allthree proteins were incu-
bated for 15 min or two of the proteins were preincubated for 15
min, followed by addition of the third protein and a second 15 min
incubation. The reaction mixture was then diluted from 130 ?l to 1
ml using the same buffer, and 5 ?l of ascites containing the 1C3
anti-HFE monoclonal antibody were added. After incubation for 60
min at room temperature, 30 ?l of protein G beads (Pharmacia; 2
mg of proteinG/ml ofbeads)were added, and incubated with mixing
for 60min. Afterpelleting, the beads were washed once withincuba-
tion buffer, twice with phosphate buffered saline, 0.05% Tween 20
(Sigma), and once again with incubation buffer, then boiled in 20 ?l
of SDS-PAGE loading buffer containing 10% (v/v) 2-mercaptoetha-
nol and loaded onto a 10% SDS-PAGE gel (Figure 5). Bands corre-
sponding to the individual proteins were identified by comparison
with the migration of samples of purified proteins (data not shown).
Cowtan, K. (1994). J oint CCP4 and ESF-EACBM Newsletter onPro-
tein Crystallography 31, 34±38.
Cuthbert, J .A. (1997). Iron, HFE, and hemochromatosis update. J .
Invest. Med. 45, 518±529.
Enns, C.A., and Sussman, H.H. (1981). Physical characterization of
the transferrin receptor in human placentae. J . Biol. Chem. 256,
Feder, J .N., Gnirke, A., Thomas, W., Tsuchihashi, Z., Ruddy, D.A.,
Basava, A., Dormishian, F., Domingo, R., Ellis, M.C., Fullan, A., et
al. (1996). A novel MHC class I-like gene is mutated in patients with
hereditary hemochromatosis. Nature Genet. 13, 399±408.
Feder, J .N., Tsuchihashi, Z., Irrinki, A., Lee, V.K., Mapa, F.A., Mori-
kang, E., Prass, C.E., Starnes, S.M., Wolff, R.K., Parkkila, S., et al.
(1997). The hemochromatosis founder mutation in HLA-H disrupts
?2-microglobulin interaction and cell surface expression. J . Biol.
Chem. 272, 14025±14028.
Feder, J .N., Penny, D.M., Irrinke, A., Lee, V.K., Lebro Â n, J .A., Watson,
N., Tsuchihashi, Z., Sigal, E., Bjorkman, P.J ., and Schatzman, R.C.
(1998). The hemochromatosis gene product complexes with the
transferrin receptor and lowers its affinity for ligand binding. Proc.
Natl. Acad. Sci. USA 95, 1472±1477.
Fersht, A. (1985). Enzyme Structure and Mechanism, 2nd Edition.
(New York: W.H. Freeman and Company).
Hashimoto, K., Hirai, M., and Kurosawa, Y. (1997). Identification
of a mouse homolog for the human hereditary haemochromatosis
candidate gene. Biochem. Biophys. Res. Comm. 230, 35±39.
Hodel, A., Kim, S.-H., and Brunger, A.T. (1992). Modelbias inmacro-
molecular structures. Acta Crystallogr. A 48, 851±858.
Howard, J .B., and Rees, D.C. (1991). Perspectives on non-heme
iron protein chemistry. Adv. Protein Chem. 42, 199±280.
J ones, T.A., and Kjeldgaard, M. (1997). Electron density map inter-
pretation. Meth. Enzymol. 277, 173±208.
J unghans, R.P.(1997).Finally! The Brambellreceptor(FcRB).Media-
tor of transmission of immunity and protection from catabolism for
IgG. Immunol. Res. 16, 29±57.
Karlsson, R., and Fa Èlt, A. (1997). Experimental design for kinetic
analysis of protein±protein interactions with surface plasmon reso-
nance biosensors. J . Immunol. Methods 200, 121±133.
Kleywegt, G.J . (1996). Use of non-crystallographic symmetry inpro-
tein structure refinement. Acta Crystallogr. D 52, 842±857.
Kleywegt, G.J ., and J ones, T.A. (1996). Phi/Psi-chology: Rama-
chandran revisited. Structure 4, 1395±1400.
Kraulis, P. (1991). MOLSCRIPT: a program to produce both detailed
and schematic plots of protein structures. J . Appl. Crystallogr. 24,
Kuziemko, G.M., Stroh, M., and Stevens, R.C. (1996). Cholera toxin
binding affinity and specificity for gangliosides determined by sur-
face plasmon resonance. Biochemistry 35, 6375±6384.
Madden, D.R., Garboczi, D.N., and Wiley, D.C. (1993). The antigenic
identity of peptide-MHC complexes: a comparisonof five viral pep-
tides presented by HLA-A2. Cell 75, 693±708.
Malmqvist, M., and Granzow, R. (1994). Biomolecular interaction
analysis. Methods 6, 95±98.
Matsumura, M., Fremont, D.H., Peterson, P.A., and Wilson, I.A.
(1992). Emerging principles for the recognition of peptide antigens
by MHC class I molecules. Science 257, 927±934.
McLaren, G.D., Nathanson, M.H., J acobs, A., Trevett, D., and Thom-
son, W. (1991). Regulationof intestinal iron absorption and mucosal
iron kinetics in hereditary hemochromatosis. J .Lab. Clin. Med. 117,
Merritt, E.A., and Murphy, M.E.P. (1994). Raster3D Version 2.0Ð a
program for photorealistic molecular graphics. Acta Crystallogr. D
Merryweather-Clarke, A.T., Pointon, J .J ., Shearman, J .D., and Rob-
son, K.J . (1997). Global prevalence of putative haemochromatosis
mutations. J . Med. Genet. 34, 275±278.
Navaza, J . (1994). AMoRe: an automated package for molecular
replacement. Acta Crystallogr. A 50, 157±163.
Nicholls, A., Sharp, K.A., and Honig, B.H. (1991). Protein folding
We thankT. L.Chapmanforperforming the acid elutions, I.Nangiana
and D. Penny for assistance with protein expression, the Caltech
PPMAL for peptide and protein analyses, Drs. C. Ogata and M.
Soltis for assistance with synchrotron data collection, Drs. I. A.
Wilson and B. Segelke for providing CD1 coordinates prior to PDB
release and helpful discussions regarding groove surface area cal-
culations, Dr. P. Poon for performing the analytical ultracentrifuga-
tion studies of TfR, and Dr. L. Sa Ânchez for critical reading of the
manuscript. M. J . B. was supported by a grant from the Cancer
Research Institute, andJ .A. L. was supported by aFORD fellowship.
Atomic coordinates for HFE have been deposited with the Protein
Received J anuary 16, 1998; revised February 26, 1998.
Bacon, B.R., and Tavill, A.S. (1996). Hepatology: A Textbook of Liver
Disease. (Philadelphia: Saunders).
Bebbington, C.R., and Hentschel, C.C.G. (1987). DNA Cloning. (Ox-
Beckman, E.M., Porcelli, S.A., Morita, C.T., Behar, S.M., Furlong,
S.T., and Brenner, M.B. (1994). Recognition of a lipid antigen by
CD1-restricted (???) T cells. Nature 372, 691±694.
Beutler, E. (1997). The significance of the 187G (H63D) mutation in
hemochromatosis. Am. J . Hum. Genet. 61, 762±764.
Borhani, D.W., and Harrison, S.C. (1991). Crystallization and X-ray
diffraction studies of a soluble form of the human transferrin recep-
tor. J . Mol. Biol. 218, 685±689.
Bru Ènger,A.T. (1992a). XPLOR Version 3.1ASystemforX-ray Crystal-
lography and NMR. (New Haven: Yale University Press).
Bru Ènger, A.T. (1992b). Free R value: a novel statistical quantity for
assessing the accuracy of crystal structures. Nature 355, 472±475.
Burmeister, W.P., Gastinel, L.N., Simister, N.E., Blum, M.L., and
Bjorkman, P.J . (1994a). Crystal structure at 2.2 AÊresolution of the
MHC-related neonatal Fc receptor. Nature 372, 336±343.
Burmeister, W.P., Huber, A.H., and Bjorkman, P.J . (1994b). Crystal
structure of the complex of rat neonatal Fc receptor with Fc. Nature
Castan Äo, A.R., Tangri, S., Miller, J .E.W., Holcombe, H.R., J ackson,
M.R., Huse, W.D., Kronenberg, M., and Peterson, P.A. (1995). Pep-
tide binding and presentationby mouse CD1. Science 269, 223±226.
Chapman, T.L., and Bjorkman, P.J . (1998). Characterization of a
murine cytomegalovirus class I major histocompatibility complex
(MHC) homolog: comparison to MHC molecules and to the human
cytomegalovirus MHC homolog. J . Virol. 72, 460±466.
Collins, E.J ., Garboczi, D.N., and Wiley, D.C. (1994). Three-dimen-
sional structure of a peptide extending from one end of a class I
MHC binding site. Nature 371, 626±629.
Structure and Function of HFE
and association: insights from the interfacial and thermodynamic
properties of hydrocarbons. Proteins 11, 281±296.
Otwinowski, Z., and Minor,W. (1996). Processing ofX-ray diffraction
data collected in oscillation mode. Meth. Enzymol. 276, 307±326.
Parkkila, S., Waheed, A., Britton, R.S., Bacon, B.R., Zhou, X.Y., To-
matsu, S., Fleming, R.E., and Sly, W.S. (1997). Association of the
transferrin receptor in human placenta with HFE, the protein defec-
tive in hereditary hemochromatosis. Proc. Natl. Acad. Sci. USA 94,
Rammensee, H.-G., Falk, K., and Ro Ètzschke, O. (1993). Peptides
naturally presented by MHC class Imolecules. Annu. Rev. Immunol.
Richardson, D.R., and Ponka, P. (1997). The molecularmechanisms
of the metabolism and transport of iron in normal and neoplastic
cells. Biochim. Biophys. Acta 1331, 1±40.
Ro Ètzschke, O., Falk, K., Deres, K., Schild, H., Norda, M., Metzger,
J ., J ung, G., and Rammensee, H.-G. (1990). Isolation and analysis
of naturally processed viral peptides as recognized by cytotoxic T
cells. Nature 348, 252±257.
Saper, M.A., Bjorkman, P.J ., and Wiley, D.C. (1991). Refined struc-
ture of the human histocompatibility antigen HLA-A2 at2.6 AÊresolu-
tion. J . Mol. Biol. 219, 277±319.
Schneider, C., Owen, M.J ., Banville, D., and Williams, J .G. (1984).
Primary structure of human transferrin receptor deduced from the
mRNA sequence. Nature 311, 675±678.
Turkewitz, A.P., Amatruda, J .F., Borhani, D., Harrison, S.C., and
Schwartz, A.L. (1988a). A high yield purification of the human trans-
ferrin receptor and properties of its major extracellular fragment. J .
Biol. Chem. 263, 8318±8325.
Turkewitz, A.P., Schwartz, A.L., and Harrison, S.C. (1988b). A pH-
dependent reversible conformational transition of the human trans-
ferrin receptor leads to self-association. J . Biol. Chem. 263, 16309±
Vaughn, D.E., and Bjorkman, P.J . (1997). High affinity binding of the
neonatal Fc receptor to its IgG ligand requires receptor immobiliza-
tion. Biochemistry 36, 9374±9380.
Vaughn, D.E., and Bjorkman, P.J . (1998). Structural basis of pH
dependent antibody binding by the neonatal Fc receptor. Structure
Waheed, A., Parkkila, S., Zhou, X.Y., Tomatsu, S., Tsuchihashi, Z.,
Feder, J .N., Schatzman, R.C., Britton, R.S., Bacon, B.R., and Sly,
W.S. (1997). Hereditary hemochromatosis: effects of C282Y and
H63D mutations on association with ?2-microglobulin, intracellular
processing, and cell surfaceexpression of the HFE protein inCOS-7
cells. Proc. Natl. Acad. Sci. USA 94, 12384±12389.
Zeng, Z.-H., Castan Ä o, A.R., Segelke, B.W., Stura, E.A., Peterson,
P.A., and Wilson, I.A. (1997). Crystal structure of mouse CD1: an
MHC-like fold with a large hydrophobic binding groove. Science