Abstract

Craniosynostosis, Boston type is an autosomal dominant disorder that results in the premature fusion of cal-varial bones and ensuing abnormalities in skull shape. We showed previously that this disorder is tightly linked to the Msx2 homeobox gene on the long arm of chromosome 5, and that affected individuals bear a mutated copy of Msx2. In addition, transgenic mice in which either mutant or wild-type mouse Msx2 is overexpressed in the developing skull also exhibit craniosynostosis. That both mutant and wild-type Msx2 elicit craniosynos-tosis in transgenic mice and that the Boston type mutation is dominant led us to hypothesize that the mutation might enhance the normal function of Msx2. The mutation is located in position 7 of the N-terminal arm of the homeodomain, a region implicated in both target sequence recognition and protein-protein interactions. Here we test the hypothesis that the Pro148→ His mutation alters the DNA binding properties of Msx2. Using gel shift and binding site selection analyses, we show that the mutation enhances the affinity of Msx2 for a set of known Msx2 target sequences but has little or no effect on the site specificity ofMsx2 binding. The enhancement of Msx2 binding is due largely if not entirely to an increased stability of the mutant Msx2-DNA complex. These data provide a molecular-level explanation of how the Pro148→His mutation enhances Msx2 function and thus leads to the dominant craniosynostosis phenotype.

Introduction

Craniosynostosis, the premature fusion of calvarial bones with consequent abnormalities in skull shape, is a relatively common birth defect (1/3000 live births). It has both environmental and genetic causes and is a feature in over 100 genetic syndromes (1,2). Among these is craniosynostosis, Boston-type, a highly penetrant, autosomal dominant disorder (3,4). We showed previously that individuals affected with craniosynostosis, Boston-type bear a mutated copy of Msx2, a highly conserved homeobox gene that functions in the regulation of inductive tissue interactions during embryogenesis (5). The mutation is a C to A transversion resulting in the substitution of a histidine for a proline in position 7 of the homeodomain (P148H). We demonstrated further that transgenic mice expressing the mutant Msx2 gene exhibit craniosynostosis, strongly supporting the proposition that the Pro148→His mutation is the cause of the disorder in humans (6). Overexpression/misexpression of the wild-type Msx2 gene in murine embryos can also produce craniosynostosis (6), a finding which when combined with the dominant nature of the mutation in humans argues that the mutation enhances the normal activity of Msx2. The molecular mechanism by which the Pro148→His substitution might cause this enhancement is the focus of this paper.

Extensive physical studies and domain swaps performed on several homeodomain proteins have demonstrated the importance of the N-terminal arm in minor groove DNA contacts (7–10) and in interactions with auxiliary factors (11). It is thus plausible that the Pro148→His mutation could influence Msx2 target sequence recognition, the ability of Msx2 to interact with partner proteins, or both. A preliminary analysis suggested that there are no gross differences in the DNA binding properties of mutant and wild-type Msx2 proteins (12). Here, in a more detailed study, we demonstrate that bacterially expressed Pro148→His mutant Msx2 does in fact bind with higher affinity to several Msx2 DNA target sequences than does its wild-type counterpart. The basis of this increased affinity is a reduction in the dissociation rate of the Msx2-DNA complex. These data suggest a straightforward mechanism by which the Pro148→His mutation could lead to enhanced Msx2 activity cand thus to mutant phenotypes in humans and mice.

Results

We first asked whether there is a difference between the mutant and wild-type Msx2 proteins in affinity for a consensus Msx-class binding site. We designed a double stranded oligonucleotide containing the consensus Msx binding site, TAATTG (13,14). Using gel shift analysis, we measured the DNA binding affinity of both the wild-type and P148H Msx2 for this optimal Msx2 binding site. A binding site titration (15) was performed in which a varying amount of radiolabeled oligonucleotide was added to a series of binding reactions. The Msx2-DNA complexes were assayed by EMSA and quantitated in a phosphoimager. We estimated relative binding affinities of the mutant and wild-type Msx2 proteins by comparing the midpoints of the binding curves (15), an example of which is shown in Figure 1.

From Figure 1, the midpoints of the binding titration curves differed significantly between the wild-type Msx2 versus its mutant congener. In the case of the wild-type Msx2, the half maximal point was ∼20 nM, while for the mutant Msx2 it was 125 nM, a six-fold difference. In three repetitions of this experiment, with two different preparations of Msx2 -type and Msx2 p148h proteins, the half maximal point for the Msx2 p148h ranged from six- to ninefold greater than that of the wild-type Msx2. These data demonstrate that the p148h mutation substantially enhances the binding of Msx2 to its target sequence.

To determine more generally whether the P148H mutation influences the DNA binding specificity of Msx2, we carried out a binding site selection analysis with the mutant and wild-type Msx2 proteins. Our approach was similar to that described by Abate and colleagues (13). We prepared a set of oligonucleotides consisting of 14 randomized bases flanked by a common sequence of 15. Opposite strands were synthesized and the resultant double stranded oligonucleotides were end-labeled and allowed to bind purified wild-type GST-Msx2 or GST-Msx2 P148H. Msx2-DNA complexes were gel-purified and the oligonucleotides amplified by PCR. Three rounds of binding and purification were carried out, and, after the final round, the selected oligonucleotides were cloned and sequenced.

The results of this analysis are shown in Figure 2. For both mutant and wild-type Msx2 proteins, the majority of the selected sequences were AT-rich and resembled the TAAT core characteristic of homeodomain proteins. There were, in addition, a few GC rich sequences which did not fit into a consensus (data not shown). Gel shift analysis revealed that they did not bind Msx2; hence, they were the residual unselected sequences in the protein-bound oligonucleotide fraction. Given that bases on the 5′ side of the TAAT core are known to interact with N-terminal arm of the homeodomain (7), and that mutations in the N-terminal arm can alter homeodomain DNA binding specificity (10,16) we expected that the mutant and wild-type forms of Msx2 might select different sequences in this region. Our data suggest, however, that this is not the case. Mutant and wild-type Msx2 exhibited indistinguishable base preferences on the 5′ side of the TAAT. Both proteins showed slight selection against A in positions −2 and −3 relative to the T of the TAAT core. Furthermore, both proteins had virtually identical base preferences within and 3′ of the TAAT core. Hence, both mutant and wild-type Msx2 essentially identical sets of sequences, which were also largely the same as those selected by the Msx1 homeodomain (13). Our data thus do not support the thesis that the P148H mutation influences the sequence-specificity of Msx2-DNA binding, but suggest, rather, that the major influence of the mutation, at least at the level of DNA binding, is to enhance generally the affinity of Msx2 for a range of different target sequences.

Figure 1

Binding of Msx2 wild-type and Msx2 P148H to an optimum Msx2 binding site. Msx2 wild-type and Msx2 P148H were expressed in bacteria as glutathione-S-transferase (GST) fusion proteins and affinity purified by chromatography on a GST column. The pure proteins were incubated with varying amounts of radiolabeled, Msx-class optimum binding site. Binding reactions were subjected to electrophoresis on TBE-acrylamide gels. Protein-DNA complexes were visualized (A) quantitated (B) in a phosphoimager. The midpoints of the titration curves provide a measure of the association constants. Note that the midpoint of the WT Msx2 titration curve is at a DNA concentration of approximately 20 nM, while the midpoint of the P148HMsx2 curve is at approximately 125 nM, a six-fold difference.

Figure 1

Binding of Msx2 wild-type and Msx2 P148H to an optimum Msx2 binding site. Msx2 wild-type and Msx2 P148H were expressed in bacteria as glutathione-S-transferase (GST) fusion proteins and affinity purified by chromatography on a GST column. The pure proteins were incubated with varying amounts of radiolabeled, Msx-class optimum binding site. Binding reactions were subjected to electrophoresis on TBE-acrylamide gels. Protein-DNA complexes were visualized (A) quantitated (B) in a phosphoimager. The midpoints of the titration curves provide a measure of the association constants. Note that the midpoint of the WT Msx2 titration curve is at a DNA concentration of approximately 20 nM, while the midpoint of the P148HMsx2 curve is at approximately 125 nM, a six-fold difference.

Figure 2

Influence of P148H mutation on Msx2 DNA binding specificity. We prepared a set of double stranded, radiolabeled 29-mer oligonucleotides whose central 14 base pairs were randomized. These oligonucleotides were incubated with either GST-Msx2 wild-type or GST-Msx2P148H and electrophoresed on a TBE acrylamide gel. Protein-DNA complexes were eluted and the DNA amplified by PCR. Selected DNA was then subjected to two additional rounds of Msx2 binding and gel purification. After the final round, selected DNA binding sites were cloned into a plasmid vector and sequenced. Shown above are the raw sequence data (A) and a graphical representation of the base preferences (B) for Msx2 wild type and Msx2P148H.

Figure 2

Influence of P148H mutation on Msx2 DNA binding specificity. We prepared a set of double stranded, radiolabeled 29-mer oligonucleotides whose central 14 base pairs were randomized. These oligonucleotides were incubated with either GST-Msx2 wild-type or GST-Msx2P148H and electrophoresed on a TBE acrylamide gel. Protein-DNA complexes were eluted and the DNA amplified by PCR. Selected DNA was then subjected to two additional rounds of Msx2 binding and gel purification. After the final round, selected DNA binding sites were cloned into a plasmid vector and sequenced. Shown above are the raw sequence data (A) and a graphical representation of the base preferences (B) for Msx2 wild type and Msx2P148H.

The basis of the enhanced DNA binding activity of the P148H Msx2 could be an increased association rate, a reduced dissociation rate, or both. To assess the influence of the P148H mutation on these rate processes, we used a gel shift assay, comparing the dissociation rates of complexes formed between the consensus Msx2 binding site and the P148H and wild-type Msx2 proteins. We allowed complexes to form, and, at time zero, added non-labeled consensus Msx2 oligonucleotide. We sampled the reaction at various times and loaded aliquots on a running acrylamide gel. A representative experiment is shown in Figure 3. The time-dependent decay of Msx2-DNA complexes showed clearly that the dissociation rate of the wild-type Msx2-DNA complex is significantly higher than that of the mutant protein-DNA complex. After 0.5 min, 66% of the wild-type complex dissociated, while only 27% of the mutant complex dissociated. We estimate the half-life of the wild-type complex to be 0.5 min or less, while that of the mutant complex is nearly 5 min. These half-lives should be considered minimal as our time course data do not include the time required for the protein-DNA complexes to enter the gel (∼2–5 min). Nevertheless, our data show that the increased affinity of the P148H mutant Msx2 protein for an Msx2 consensus binding site is caused largely if not entirely by an enhancement in the stability of the Msx2-DNA complex.

Discussion

We have addressed the underlying molecular cause of Boston-type craniosynostosis. We show that a mutation found exclusively in individuals affected with this disorder alters the DNA binding properties of the Msx2 homeoprotein. Although a preliminary study revealed no gross differences in DNA binding activity of mutant and wild-type Msx2 (12), our more extensive analysis has shown that the mutated (P148H) form of Msx2 interacts more avidly with consensus Msx2 binding sites than does its wild-type counterpart. This enhanced DNA binding affinity is caused largely if not entirely by enhanced stability of the Msx2-DNA complex. A binding site selection analysis revealed that the P148H mutation has no discernible effect on the sequence-specificity of Msx2 binding. This finding, together with our demonstration that the P148H mutation enhances the affinity of Msx2 for a consensus binding site, suggests that the P148H mutation has a generalized effect on Msx2 binding to different classes of binding sites.

Binding site selection studies, as well as physical analyses performed on several homeodomain proteins, have shown that the N-terminal arm has an important part in DNA binding (7–9, 16). Residues 3, 5 and 7 make base-specific contacts in the minor groove on the 5′ side of the TAAT core sequence (7,8). Residue 6 contacts the sugar-phosphate backbone (7,9). Such contacts are at least in part responsible for some of the differences between homeodomain proteins in the DNA sequences to which they bind. Most significantly, mutations in residues 6 and 7 affect base preferences throughout the region contacted by the N-ter-minal arm, perhaps by influencing the overall conformation of the arm within the minor groove (10). Our results show that although the P148H mutation (corresponding to residue 7 of the homeodo-main) does not influence Msx2 base preferences, it does increase the affinity of Msx2for several target sequences, consistent with the well-documented involvement of the N-terminal arm in DNA binding. It is not obvious, however, why this mutation should have a positive rather than a negative effect on Msx2-target sequence interactions. The P148H substitution might, for example, reduce the flexibility of the arm and thus stabilize its minor groove contacts. It is also possible that the mutation affects charge density in the N-terminal arm. Protonation the histidine in the local environment of the Msx2 protein-DNA complex would produce a positive charge in close proximity to the phosphate backbone and would thus presumably enhance DNA binding. We also point out that as residues in the arm do not function autonomously in sequence recognition, but act together with residues in helices I, II and III (16), the influence of the P148H mutation may extend beyond the N-terminal arm into other portions of the homeodomain.

Figure 3

Effect of the P148H mutation on the stability of Msx2 DNA binding. Purified GST-Msx2 wild type and GST-Msx2 P148H were incubated with a radiolabeled consensus Msx binding site. Formation of Msx2-DNA complexes was as described in Figure 1. After 15 min, a 50-fold molar excess of non-labeled consensus Msx binding site was added to each reaction. At the time intervals indicated, aliquots of the reactions were withdrawn and immediately loaded on a running TBE acrylamide gel. Approximately 2–5 min elapsed between withdrawal of the sample from the binding reaction and entry into the gel. Complexes were visualized (A) and quantitated (B) with a phosphoimager.

Figure 3

Effect of the P148H mutation on the stability of Msx2 DNA binding. Purified GST-Msx2 wild type and GST-Msx2 P148H were incubated with a radiolabeled consensus Msx binding site. Formation of Msx2-DNA complexes was as described in Figure 1. After 15 min, a 50-fold molar excess of non-labeled consensus Msx binding site was added to each reaction. At the time intervals indicated, aliquots of the reactions were withdrawn and immediately loaded on a running TBE acrylamide gel. Approximately 2–5 min elapsed between withdrawal of the sample from the binding reaction and entry into the gel. Complexes were visualized (A) and quantitated (B) with a phosphoimager.

The Boston craniosynostosis mutation behaves as an autosomal dominant in humans (3,4). Our finding that transgenic mice in which either the mutant or wild-type Msx2 gene is overexpressed also exhibit craniosynostosis is consistent with the possibility that the mutation acts via a dominant positive mechanism, i.e. that the mutation augments the normal function of Msx2. The data presented here on the enhancement of the DNA binding affinity of Msx2 are entirely consistent with this scenario. Although the magnitude of the enhancement in DNA binding is less than ten-fold, it may nevertheless be sufficient to explain the observed phenotypes. Measurements of the degree to which Msx2 is overexpressed in the developing cranial bones of Msx2 transgenic mice indicate a subtle effect, that two-fold or less overexpression is capable of causing sutural fusion (Liu, Y-H., Kundu, R., and Maxson, R., unpublished observations). Hence, it appears that the developing skull is highly sensitive to the level of Msx2 gene expression. A slight increase in the DNA binding affinity of the Msx2 protein may thus be sufficient to elicit craniosynostosis.

We stress, however, that a simple, linear model relating the abundance of Msx2 or its affinity for target sites to the fractional occupancy of such sites and thus to phenotype is likely to be an oversimplification. One complicating factor is the finding that the biological specificity of homeodomain proteins depends in part on auxiliary proteins with which they interact. For example, the exd-Ubx interaction enhances the DNA binding specificity of Ubx (17–20), similarly, the interaction between the yeast α2 and MCM1 proteins modifies the activity and DNA binding properties of α2 (11). Such interactions depend on the N-terminal arms of the Ubx and α2 homeoproteins (18,19). If Msx2 function similarly involves protein-protein interactions mediated by the N-terminal arm, then the P148H mutation may, in addition to affecting DNA binding, perturb an interaction between Msx2 and a partner protein. Such an effect could influence Msx2 target sequence recognition indirectly, or might alter the ability of Msx2 to interact with the transcriptional machinery and thus regulate downstream genes.

In spite of these potential complexities, our data provide a framework in which to understand how a mutation in Msx2 can lead to the craniosynostosis phenotype. Further analysis of the molecular and developmental consequences of the P148H mutation is likely to illuminate not only the role of Msx2 in cranial development, but also general relationships between homeodo-main structure, DNA target sequence recognition and interacting partner proteins.

Materials and Methods

Bacterial expression of wild-type and P148H Msx2

Wild-type and P148H Msx2 cDNA were excised as a BamHI-EcoRI fragment from pBSK-Msx2 wild-type or pBSK-Msx2C→A (P148H) (6). They were directionally cloned between the BamHI and EcoRI sites of pGEX-3X, which results in an in-frame fusion of the Msx2 proteins with glutathione-S-transferase (GST). Plasmids encoding the fusion proteins were transformed into the protease deficient bacterial strain BL21(DE3). GST-Msx2 proteins were expressed and purified as described (21).

DNA mobility shift assays

Oligonucleotides were end-labeled with [γ-32P]ATP using T4 polynucleotide kinase and opposite strands were allowed to anneal (22). Binding reactions were carried out by incubating 50 ng of purified GST fusion protein with 1 ng of probe in a buffer containing 10 mM Tris-HCl (pH 7.5), 50 mM potassium chloride, 0.05% Nonidet P-40, 0.1 µg of bovine serum albumin per µl, 5% (vol/vol) glycerol and 5 mM Dithiothreitol in 30 µl reaction volume. Binding reactions were incubated for 15 min at room temperature, loaded without dye on 5% polyacrylamide, TBE gels, and electrophoresed at 8 V/cm for 2 h. Gels were dried and protein-DNA complexes were visualized and quantitated with a phosphoimager. For measurements of the dissociation rates of the Msx2-DNA complexes, complexes were formed as described above, and, after 15 min of incubation at room temperature, a 50-fold excess nonlabeled self-competitor was added. Aliquots of the binding reaction were withdrawn at various times and immediately loaded on a running polyacryl-amide gel. Approximately 4 min elapsed between withdrawal of the sample from the binding reaction and entry into the gel.

DNA binding site selection

Oligonucleotide selection was performed essentially as described by Abate and coworkers (13). Briefly, a DNA fragment containing 14 bp random sequence flanked on either side by 15 bases of nonrandom sequence [5′-AGACGGATCCATTGCA(N14)CTG-TAGGAATTCGGA-3′] was 32P-end-labeled and used in the gel shift assay as described above. Msx2-bound DNA was identified by autoradiography and extracted from the gel. The DNA was PCR-amplified, gel purified, end-labeled and used as a probe in a second round of gel retardation assays. A total of three rounds of selection were carried out. The resulting Msx2-bound DNA was cut with BamHI and EcoRI and cloned into BamHI-EcoRI digested pBluescript SK vector (Stratagene). Individual clones were se-quenced by the dideoxy chain termination method (22).

Acknowledgments

We thank Dr Richard Maas and members of the Maxson laboratory for critical comments on this manuscript. This work was supported by NIH grants HD22416 and DE09165.

References

1
Cohen
M.M.
Jr.
Craniosynostosis update 1987
Am. J. Med. Genet. Suppl.
 , 
1988
, vol. 
4
 (pg. 
99
-
148
)
2
Cohen
M.M.
Jr.
Sutural biology and the correlates of craniosynostosis
Am. J. Med. Genet.
 , 
1993
, vol. 
47
 (pg. 
581
-
616
)
3
Warman
M.L.
Mulliken
J.B.
Muller
U.
Hayward
P.G.
Newly recognized autosomal dominant craniosynostotic syndrome
Am. J. Med. Genet.
 , 
1993
, vol. 
46
 (pg. 
444
-
449
)
4
Muller
U.
Warman
M.L.
Mulliken
J.B.
Weber
J.L.
Assignment of a gene locus involved in craniosynostosis to chromosome 5qter.Hum
Mol. Genet.
 , 
1993
, vol. 
2
 (pg. 
119
-
122
)
5
Jabs
E.W.
, et al.  . 
A mutation in the homeodomain of the humanMsx2 gene in a family affected with autosomal dominant craniosynostosis
Cell
 , 
1993
, vol. 
75
 (pg. 
443
-
450
)
6
Liu
Y. H.
, et al.  . 
Premature suture closure and ectopic cranial bone in mice expressing Msx2 transgenes in the developing skull
Proc. Natl Acad. Sci. USA
 , 
1995
, vol. 
92
 (pg. 
6137
-
6141
)
7
Kissinger
C.R.
Liu
B.S.
Martin-Blanco
E.
Kornberg
T.
Pabo
C.O.
Crystal structure of an engrailed homeodomain-DNA complex at 2.8A resolution: a framework for understanding homeodomain-DNA interactions
Cell
 , 
1990
, vol. 
63
 (pg. 
579
-
590
)
8
Wolberger
C.
Vershon
A.K.
Liu
B.
Johnson
A.D.
Pabo
C.O.
Crystal structure of a MAT alpha 2 homeodomain-operator complex suggests a general model for homeodomain-DNA interactions
Cell
 , 
1991
, vol. 
67
 (pg. 
517
-
528
)
9
Klemm
J.D.
Rould
M.A.
Aurora
R.
Herr
W.
Pabo
C.O.
Crystal structure of the Oct-1 POU domain bound to an octamer site: DNA recognition with tethered DNA-binding modules
Cell
 , 
1994
, vol. 
77
 (pg. 
21
-
32
)
10
Ekker
S.C.
Jackson
D.G.
von-Kessler
D.P.
Sun
B.I.
Young
K.E.
Beachy
P.A.
The degree of variation in DNA sequence recognition among four Drosophila homeotic proteins
EMBO J.
 , 
1994
, vol. 
13
 (pg. 
3551
-
3560
)
11
Vershon
A.K.
Johnson
A.D.
A short, disordered protein region mediates interactions between the homeodomain of the yeast a2 protein and the MCM1 protein
Cell
 , 
1993
, vol. 
72
 (pg. 
105
-
112
)
12
Semenza
G.L.
Wang
G. L.
Kundu
R.
DNA binding and transcriptional properties of wild-type and mutant forms of the homeodomain proteinMsx2
Biochem. Biophys. Res. Commun.
 , 
1995
, vol. 
209
 (pg. 
257
-
262
)
13
Catron
K.M.
Iler
N.
Abate
C.
Nucleotides flanking a conserved TAAT core dictate the DNA binding specificity of three murine homeodo-main proteins
Mol. Cell. Biol.
 , 
1993
, vol. 
13
 (pg. 
2354
-
2365
)
14
Shang
Z.
, et al.  . 
Design of a ‘minimAl’ homeodomain: the N-terminal arm modulates DNA binding affinity and stabilizes homeodomain structure
Proc. Natl Acad. Sci. USA
 , 
1994
, vol. 
91
 (pg. 
8373
-
8377
)
15
Emerson
B.M.
Lewis
C.D.
Felsenfeld
G.
Interaction of specific nuclear factors with the nuclease-hypersensitive region of the chicken adult beta-globin gene: nature of the binding domain
Cell
 , 
1985
, vol. 
41
 (pg. 
21
-
30
)
16
Isaac
V.E.
Sciavolino
P.
Abate
C.
Multiple amino acids determine the DNA binding specificity of the Msx-1 homeodomain
Biochemistry
 , 
1995
, vol. 
347
 (pg. 
127
-
134
)
17
van Dijk
M.A.
Murre
C.
Extradenticle raises the DNA binding specificity of homeotic selector gene products
Cell
 , 
1994
, vol. 
78
 (pg. 
617
-
624
)
18
Johnson
F.B.
Parker
E.
Krasnow
M.A.
Extradenticle protein is a selective cofactor for the Drosophila homeotics: role of the homeodomain and YPWM amino acid motif in the interaction
Proc. Natl Acad. Sci. USA
 , 
1995
, vol. 
92
 (pg. 
739
-
743
)
19
Chang
C.P.
Shen
W.F.
Rosenfeld
S.
Lawrence
H.J.
Largman
C.
Cleary
M.L.
Pbx proteins display hexapeptide-dependent cooperative DNA binding with a subset of Hox proteins
Genes Dev.
 , 
1995
, vol. 
9
 (pg. 
663
-
674
)
20
Chan
S.K.
Jaffe
L.
Capovilla
M.
Botas
J.
Mann
R.S.
The DNA binding specificity of Ultrabithorax is modulated by cooperative interactions with extradenticle, another homeoprotein
Cell
 , 
1994
, vol. 
78
 (pg. 
603
-
615
)
21
Smith
D.B.
Purification of glutathione-S-transferase fusion proteins
Methods Mol. Cell. Biol.
 , 
1993
, vol. 
4
 (pg. 
220
-
229
)
22
Ausubel
F.M.
Brent
R.
Kingston
R.E.
Moore
D.D.
Siedman
J.G.
Smith
J.A.
Struhl
K.
Current Protocols in Molecular Biology
 , 
1989
New York
Wiley

Author notes

+
Present address: Howard Hughes Medical Institute, Brigham & Women's Hospital, 20 Shattuck St Boston, MA 02115, USA