1
Introduction
The Notch receptor is a single pass trans-membrane protein which, during maturation,
may be cleaved by a furin-like convertase (at S1) in the trans-Golgi to generate a
non-covalently associated heterodimer at the cell surface. Canonical Notch signalling
is initiated when a cell-surface expressed Delta/Serrate/LAG-2 (DSL) ligand binds
to the Notch receptor expressed on an opposing cell surface (Fig. 1). Endocytosis
of the Notch-ligand complex by the ligand-expressing cell leads to ADAM metalloprotease
mediated cleavage at S2 and removes the extracellular fragment of the heterodimer.
The membrane tethered fragment is then cleaved by γ-secretase complex at S3 to release
the intracellular fragment of Notch (NICD). This translocates to the nucleus and assembles
into a transcriptional activation complex which includes a DNA binding protein of
the CSL family and its co-activator Mastermind-like. This new assembly relieves repression
of Notch target genes such as Hes-1. In addition to trans-activating Notch–ligand
complexes, the receptor can also form cis-inhibitory complexes (Fig. 1) when Notch
and ligand are expressed on the same cell surface. Cis-inhibition serves to limit
the zone of Notch activity and is particularly important in developmental programs
in Drosophila such as the wing disc and eye.
2
Domain architecture of Notch receptor
The extracellular domain (ECD) of the Notch receptor varies from species to species;
Drosophila and mammalian receptors are much larger than their counterparts from Caenorabditis
elegans, although each invariably maintains the same molecular architecture (see review
[1]). Drosophila has a single Notch receptor, C. elegans has two (LIN-12, GLP-1) while
mammals have four paralogues. At the N-terminal end, Drosophila Notch and human Notch-1
(Fig. 2) both contain 36 EGF-like domains, a subset of which contain calcium-binding
sites (cbEGF). Following the EGF-like domains are three Lin-12-Notch (LNR) repeats,
and a hydrophobic region which has been shown to mediate heterodimerisation (HD).
Together, the LNR repeats and the heterodimerisation domain form the negative regulatory
region (NRR), adjacent to the cell membrane. This region prevents ligand-independent
activation of the Notch receptor by concealing and protecting the S2 cleavage site
from metalloproteases [2]. The S3 cleavage site lies within the transmembrane segment
and is cleaved by the γ-secretase complex to liberate NICD. NICD contains a RAM domain,
seven ankyrin repeats (ANK), a transcription activation domain (TAD) and PEST domain.
Both the RAM domain and ANK repeats have been identified as regions involved in the
interaction with CSL transcription factors [3]. The TAD region is found in Notch-1
and -2 but not in -3 and -4 in mammals. The C-terminal PEST domain is involved in
NICD degradation by proteolysis. Mutations which lead to deletions of this region
are associated with T-cell leukaemias, emphasising the important functional role of
regulated NICD degradation [4].
2.1
Structure of the ligand binding domain (LBD) of human Notch1
Deletion analyses, in combination with cell aggregation assays, identified EGF domains
11 and 12 of the Drosophila Notch receptor as the major ligand-binding site. This
region was found to be sufficient to bind in a calcium-dependent manner to Notch ligands,
but it did not show full functionality in vivo, indicating that additional sites were
involved in Notch activation and regulation [5,6]. The large size and disulphide-bonded
complexity of the full length Notch receptor precluded structural studies on the intact
native molecule, however analysis of other unrelated EGF domain-rich proteins such
as fibrillin-1 demonstrated that a molecular dissection approach employing in vitro
redox refolding could be used to determine the structure of short fragments containing
two or more EGF domains [7,8]. A fragment of human Notch-1 EGF11–13, encompassing
the ligand-binding region, was subsequently expressed in bacteria, refolded in vitro
and shown to be capable of binding to ligand in a Ca2+ dependant manner in FACS assay
when biotinylated and anchored to Streptavidin beads and also in Surface Plasmon Resonance
(SPR) studies [9,10]. A study of calcium-binding mutations introduced into a slightly
larger fragment hNotch-1 EGF11–14 showed that the calcium-dependent structure of EGF12
but not EGF11 or EGF13 was key to hDll-1 binding, suggesting that the ligand-binding
site resides in Notch-1 EGF12 and/or the EGF11–12 junction. The solution structure
of hNotch-1 EGF11–13 showed that EGF11 and 12 packed against each other, with Ca2+
dependent pairwise domain interactions stabilised by a conserved aromatic residue
(Y/F/W) in the N-terminal domain packing against a hydrophobic residue (I/L/V/P) and
a conserved glycine in the C-terminal domain. In contrast, there was a poorly defined
orientation for EGF13. The tilt angle of ∼20° for hNotch-1 EGF11–12 was similar to
that seen for calcium binding (cb)EGF domain pairs of fibrillin-1, but the twist angle
of 120° was very different and at least in part due to the two amino acid linker between
EGF11 and 12, rather than the one amino acid linker seen in fibrillin-1. This NMR
structure allowed modelling of other contiguous regions of Notch cbEGF domains as
rod-shaped structures, since linker length and residues important for pairwise interactions
are highly conserved.
A high resolution crystal structure was later solved for an EGF11–13 fragment (Fig.
2a), which showed a calcium-stabilised rod-shape for all three domains with dimensions
100 × 24 × 20 Å [11]. As seen in the NMR structure the interdomain packing interactions
were key to maintaining the rigidity of the domain pair (Fig. 2b). The stability of
EGF13 in the crystal structure was attributed to the addition of a recombinant C-terminal
tag which was absent in the earlier construct used for NMR. The overall elongated
shape of this region provides an extensive protein–protein interaction surface which
is likely to be of functional significance. Although EGF12 appears to be the core
ligand binding domain, it is possible that additional contacts from neighbouring domains
may contribute to the interaction. Further work on defining the ligand binding site
using site-directed mutagenesis, chemical shift mapping and other interaction studies
will help to clarify the binding interface.
O-glycosylation of the extracellular domain is known to play a key role in regulating
Notch signalling. Multiple sites occur across the receptor, including one O-fucose
site and one O-glucose site in EGF12. Over-expression of Ofut1 (O-fucosyl transferase)
in Notch-expressing cells increases Notch-Serrate binding, whereas it decreases the
Notch-Delta binding. These results are in contrast to the effects of extending the
fucose moiety with N-acetylglucosamine (GlnNAc) by the glycosyltransferase Fringe.
This additional modification decreases Notch-Serrate binding and increases Notch-Delta
binding [12]. Mutation of the O-fucose site in EGF12 (Ser to Ala) of Drosophila Notch
eliminates the inhibitory effect of Serrate on Notch receptor because fucosylation
and subsequent Fringe modifications do not occur. This mutant was found to be functional
during embryonic neurogenesis but not at the dorsoventral boundary. An O-fucose site
substitution in EGF12 (Thr to Ala) of mouse Notch1 receptor (Notch112f) demonstrated
that homozygous (Notch112f/12f) mice were viable and fertile, though they showed decreased
Notch signalling leading to defective T-cell development [13,14]. These data show
that post-translational modifications of Notch can modulate Notch ligand interactions,
but the relative importance of each site glycosylated and the molecular basis for
this regulation is unknown. Recently, controlled in vitro glycosylation of synthetic
mouse Notch-1 EGF12 was performed and the structure solved by NMR [15]. Only minor
structural changes in the domain were observed with the addition of a single fucose
residue but addition of a GlcNac residue onto fucose resulted in a significant conformational
change in the beta-hairpin of EGF12. Although interesting, the functional significance
of these data is not clear. The NMR experiments were performed in the absence of Ca2+
on an isolated domain. Since N-terminal linkage of EGF11 and Ca2+ are both required
for EGF12 to adopt its native ligand-binding structure, further comparative experiments
in the presence of Ca2+ are required to confirm the structural effects of O-linked
modifications to EGF12, and their impact on ligand binding.
2.2
Structure of negative regulatory region (NRR) of Notch1
The membrane-proximal NRR, which contains the LNR repeats, the dimerisation domain
and both S1 and S2 cleavage sites, is well characterised at an atomic level (Fig.
2c) compared to the EGF repeat region, and has been reviewed in detail recently [16].
Most Notch receptors are cleaved at the S1 site to form a heterodimer molecule, but
the requirement of such processing for activity is debatable. A recent report of a
S1-cleaved human Notch-1 NRR region shows that there is little change in the overall
structure whether or not the S1 loop is present [17]. Several disease-causing mutations
associated with T-cell leukaemias have been mapped to the NRR region and have been
found to lead to a gain of Notch signalling suggesting that the NRR acts as an activation
switch in the receptor. The first NMR structure of the LNR module showed that it has
little secondary structure, but is stabilised by three disulphide bonds and a single
calcium ion [18]. In an early study calcium was shown to play a key role in stabilising
the inactive state of the NRR since EDTA treatment resulted in receptor activation
and NICD production in a ligand independent manner [19]. A crystal structure of the
NRR region of human Notch-2 showed that the S2 cleavage site is buried because of
an extensive interaction surface between the LNR repeats and the heterodimerisation
domain (HD), which constitutes two subunits tightly entwined in an α-β-sandwich. As
a consequence of the cap-like covering of LNR over the HD region, the protein is locked
in an autoinhibited mode. This finding immediately poses the question of how the autoinhibition
is relieved and the S2 cleavage site exposed? The current favoured mechanism is that
ligand binding and subsequent initiation of ligand endocytosis generates a mechanical
pull on the NRR region which leads to LNR repeats being pulled away unmasking the
S2 cleavage site (Fig. 1). This is consistent with experimental data which show a
requirement for ligand endocytosis in Notch signalling (see review by Weinmaster in
this series). However an alternative allosteric mechanism cannot be excluded whereby
ligand binding causes a conformational change in the NRR leading to S2 exposure. Irrespective
of which of these models is correct, mutations found in T-ALL patients appear to disrupt
the hydrophobic core of the HD region leading to exposure of the S2 cleavage site,
even in the absence of ligand [20].
2.3
Structure of Notch intracellular domain (NICD)
The NICD, encompassing the RAM, ANK and PEST domains is by far the best characterised
region in the Notch receptor at the atomic level (for a detailed review see [21])
The ANK domain (Fig. 2d), which comprises seven ANK repeats, was the first region
to be studied, probably because of the known classical ANK fold comprising a pair
of antiparallel helices, and its observed role in protein–protein interactions [22].
Subsequent biochemical and structural analysis of the Notch transcription complex
partners has revealed that the interaction between NICD and CSL is dependant predominantly
on the RAM domain and less on the ANK domain [23]. In contrast, the binding of the
co-activator MAML1 is independent of RAM domain, but instead requires the ANK domain.
It was found that neither NICD nor CSL can bind MAML1 alone, but bind cooperatively
to it in a complex (Fig. 2e) [24]. A recent report shows that NICD, CSL and MAML1
cooperatively form a dimer complex on a paired recognition site of a promoter Hes5
[25]. This finding demonstrates how promoter organisation controls responsiveness
to Notch signalling.
2.4
Structure of Notch receptor: a rod or a rope
Despite advances in structure determination of limited fragments of Notch comprising
the key domain types, the quaternary structure of the ectodomain remains unsolved.
Since much of the extracellular region of Notch comprises calcium binding EGF-like
domains, it is reasonable to base working models of receptor shape on structural features
of this domain type. Most of the receptor is predicted to have a rigid near-linear
architecture (Fig. 3), but potential sites of flexibility may occur at the interfaces
of cbEGF/EGF and EGF/EGF domains which appear much less conserved [9,26]. An alternative
bent structure has been proposed (Fig. 3) based on non-linear pairwise domain interactions,
as shown for the EGF pair from a merozoite protein from Plasmodium falciparum, which
predicts a shorter, more compact, receptor [27]. This model is intriguing in a way
that it provides a potential explanation for genetic and biochemical data which implicate
domains distant from EGF11–12, such as the Abruptex region, in modulating ligand dependent
activation. The quaternary structure of the complete extracellular domain (ECD) of
human Notch1 receptor and Drosophila Notch receptor has recently been investigated
by single particle electron microscopy and antibody labelling [28]. These are necessarily
challenging experiments as the inherent resolution of the technique is poorly matched
to the small size of the domains involved. Baculovirus-expressed proteins were directly
captured on grids and dimensions of ∼25 nm estimated for the long axis of each protein.
Overall, these dimensions support a more compact rather than extended architecture
for the receptor; measurements >100 nm would have been expected for an extended rod-shaped
molecule since each EGF domain is ∼3 nm long. Furthermore, the density maps suggested
that the ECD existed as a homodimer and could adopt at least three conformations which
the authors proposed may be of functional significance to the cis- and trans-regulatory
interactions mediated by Notch. However it should be noted that due to low yields
of protein, both Drosophila and human proteins were directly captured on grids rather
than undergoing extensive purification. This could result in inclusion of non-native
receptor in the data set. Ideally future single-particle studies should include comparative
analyses of purified and characterised proteins, in the presence and absence of Ca2+,
since Ca2+ is known to have substantial effects on the conformation of tandem cbEGF
domains. These could be complimented with antibody labelling studies using a panel
of monoclonal antibodies specific for individual Notch EGF domains to provide distance
restraints in conjunction with structure and dynamics studies of fragments containing
cbEGF/EGF linkages.
3
Domain architecture of canonical ligands
The Notch receptor is activated by Delta/Delta-like and Serrate/Jagged ligand families
(Fig. 4), both of which contain a Notch-binding site within a DSL domain. Drosophila
contains one member of each family (Delta and Serrate), while mammalian ligands are
more complex with three members of the Delta family (Dll-1, -3 and -4) and two members
of the Jagged family (J-1 and -2). Preceding the DSL domain is a disulphide-bond stabilised
module at the N-terminus of Notch ligands (MNNL) which is of unknown structure, but
is functionally important since many missense mutations affecting this region of hJ-1
give rise to Alagille syndrome. Following the DSL domain is a series of EGF-like domains
which range in number from 16 (Jagged family), 5–9 (Delta family) and one in DSL-1
(secreted ligand) of C. elegans
[29]. The first two EGF-like domains in Serrate/Jagged1/2 and Delta/Delta-like-1 are
unusual in that, although they contain 6 conserved cysteine residues, they have very
short loop sequences and resemble a motif seen in the OSM-11 protein in C. elegans
known as the Delta/OSM-11 domain (DOS domain) [30]. The remaining EGF domains are
more orthodox in sequence. A juxtamembrane cysteine rich domain (CRD) distinguishes
Serrate/Jagged from the Delta family. On the cytosolic side, a PDZL domain was identified
in some vertebrate ligands (Jagged-1, Delta-like-1, 4). This domain facilitates the
interaction with proteins at the adherens junction in promoting cell–cell adhesion
and inhibiting cell motility [31].
The crystal structure of a human Jagged-1 fragment containing the DSL domain and the
first three EGF-like domains have given the first high resolution insights into ligand
domain organisation. The four domains form an elongated structure (Fig. 4a), similar
to the ligand binding EGF11–13 region of Notch, with dimensions of 120 × 20 × 20 Å
[11]. The DSL domain revealed a unique fold with no known structural homologues, but
showed similarity to an EGF domain, since it contained a double-stranded antiparallel
β-sheet. The disulphide bond pattern in the DSL domain was found to occur between
consecutive cysteine residues instead of the C1–C3, C2–C4, C5–C6 arrangement usually
associated with the EGF fold. However, the loop between C1 and C2 superimposed on
the loop from C5 and C6 suggesting that the DSL domain might have evolved from a truncation
of tandem EGF domains. Sequence alignment of DSL domains from both ligand families
identified a group of highly conserved residues (human Jagged1 F199, R201, R203, F207)
which mapped to one face of the DSL domain forming a putative Notch binding site (Fig.
4b). Functional analysis of site-directed DSL mutants confirmed the importance of
these residues in both cis- and trans-regulatory interactions with Notch.
Analysis of the three EGF domains revealed interesting differences. EGF3 showed a
classical EGF fold whereas EGF1 and 2 showed a trimmed EGF fold with no canonical
secondary structure. These data confirm the observation of Lissemore and Starmer based
on sequence alignments that EGF1 and 2 are different from EGF3 [32]. A solution NMR
structure of a 44 aminoacid synthetic peptide corresponding to exon6 of human Jagged-1
consisting of the C-terminal region of EGF1 and EGF2 showed that there were strong
hydrophobic interactions between Y255, W257 of EGF1 and I266, P279, W280 of EGF2 indicating
the rigidity of these tandem repeats even in solution [33]. Non-canonical EGF domains
have been observed in the Delta and OSM-11 (DOS) motif identified in the OSM-11 and
related proteins of C. elegans, which may act as Notch coligands. The DOS motif comprises
of a non-canonical EGF domain and additional sequences including two conserved cysteine
residues and a tryptophan residue which may suggest it evolved from two non-canonical
EGF domains. However it remains to be confirmed whether it represents a completely
novel domain structure with a non-EGF disulphide pairing (as observed in the DSL domain)
or comprises a non-canonical EGF domain with an additional disulphide bond.
Further, unresolved questions about ligand structure include the role of the MNNL
domain. This region is not conserved between the two ligand families but the number
of missense mutations associated with Alagille syndrome which affect amino acids within
this domain from Jagged-1 indicate an important functional role which has yet to be
elucidated. Similar to the Notch receptor, the cell surface organisation of the ligands
remain unsolved. Do the ligands extend away from the cell surface, or adopt a more
compact, membrane proximal, architecture? What is the molecular basis for the differential
response of each ligand family to O-fucosylation of Notch.
4
Notch–ligand interactions
There are two different modes of canonical Notch–ligand interactions. Ligand expressed
on the surface of a signal-sending cell can bind in trans to a receptor on the receiving
cell initiating Notch signalling [34]. Alternatively ligand can be expressed on the
same cell surface as the receptor resulting in a cis inhibitory interaction that limits
Notch activity [35]. Although these two modes of interaction are well established
from a variety of experimental data, the molecular basis for activating and inhibitory
complexes is poorly understood. It has been unambiguously established that Notch EGF11–12
and the DSL domain residues interact to promote Notch trans-activation. In contrast,
the molecular requirements for cis interactions are less clear cut, although the loss
of both cis-inhibition and trans-activation, observed when a panel of Serrate DSL
mutants were over expressed in the imaginal disc, suggest that the DSL region is involved
in both [11]. Early reports proposed that the Abruptex region of Notch receptor (EGF24–29)
might be involved in cis interactions [36]. A recent study showed that the Notch EGF10–12
region is required for cis-inhibition [37]. Although similar regions of receptor and
ligand are implicated, it is impossible to predict whether or not a single molecular
complex involving DSL and EGF11–12 underlies both types of interaction. NMR analysis
and in silico docking of isolated receptor ligand fragments have suggested that two
distinct complexes involving the same regions could form, however in the absence of
more native-like flanking sequences, these data should be regarded as preliminary.
More convincing data may come from the expression of full length Notch mutants and
observation of their effects in an experimental system which can distinguish cis-inhibition
and trans-activation.
4.1
Insights into Notch–ligand interaction: affinity vs. avidity
Although new insights into the Notch–ligand recognition event have been gained from
the application of structural biology, the establishment of a quantitative assay is
required to identify the affinity of different Notch ligand complexes. This is required
to fine tune the design of fragments for co-crystallisation, to determine the contribution,
if any, of flanking regions and post translational-modifications to the interaction,
and to rationalise the contribution of affinity/avidity to the activation mechanism.
A relatively weak K
D of 130 μM was measured for the calcium-dependent interaction of EGF11–14 Notch1
and DSL EGF-3 of Dll-1 in an SPR study [10]. In solid phase assays, mouse soluble
Dll-1 produced from eukaryotic cells showed saturable binding at nM concentrations
to larger fragments of Notch 1 and Notch 3 (N-terminus to EGF15) and a soluble Jagged1
construct gave a K
D(app) value of 0.7 nM suggesting that regions outside the EGF11–12 region may contribute
to ligand binding [38,39]. However the ligand fragments were expressed as Fc fusions,
so avidity effects may lie behind the very different K
D(app) values reported, compared to the studies performed with monomeric reagents.
Similar ELISA based assays of fragments of Drosophila Notch and Delta suggested that
the ligand-binding region of Notch also bound to the Abruptex region and a role of
Abruptex in maintaining the inactive state of the receptor was proposed [40]. Comparative
analysis of Fc vs. non-Fc-fused ligand constructs would establish the effect of avidity
on the measured K
D(app), and pave the way for systematic analysis of the contribution of regions outside
the recognition domains, and/or post-translational modifications to binding using
ELISA and SPR solid phase assays. These data would greatly facilitate the design of
receptor and ligand fragments with optimal binding affinity, which may prove suitable
for co-crystallisation studies.
As described earlier the early steps leading to activation upon ligand binding are
undefined. Is the receptor activated solely by mechanotransduction initiated by ligand
endocytosis and/or do a series of conformational changes lead to an allosteric effect
on S2 cleavage? Cell aggregation assays performed on Serrate DSL mutants demonstrated
that F249A and R253A retained their ability to bind to Notch despite being defective
in transactivation, suggesting that binding alone is not sufficient to initiate the
signalling process [11]. These mutants would be prime candidates to test in quantitative
binding assays, since native-like binding would suggest the involvement of these conserved
residues in stabilising a conformation required for activation, rather than contributing
to the stability of the complex. Interestingly two separate studies showed that N-terminal
linkage of EGF10 reduced the ligand binding affinity of a receptor fragment encompassing
EGF11–12 [10,27]. In the latter study, ligand binding could be rescued by the introduction
of a calcium-binding mutation into EGF11 which uncouples the EGF10–11 interface. These
data suggest that conformational changes in regions adjacent to EGF12 have the capacity
to modulate ligand binding, but until these experiments are repeated using larger
fragments of Notch, it is unclear if they have physiological relevance. Ligand-dependent
clustering of Notch has been proposed to allow trans-regulatory complexes to withstand
the mechanical pulling force generated by ligand endocytosis during the activation
process suggesting that avidity effects may contribute to the mechanism of activation
[41]. Atomic force microscopy on live cells showed that Notch signalling is linked
to the adhesion force between cells expressing receptor and ligand [42]. The surface
of S2 cells expressing Notch and Delta revealed marked differences with Notch-expressing
cells displaying a topology of fibres whereas Delta expressing cells showed bumps.
The authors predicted that Notch might exist as a monomer or oligomer, but proposed
that Delta formed a multimer. On the basis of this it was postulated that Delta and
Notch-expressing cells resemble ‘Velcro’. O-fut1 down-regulation by RNA interference
produced a marked difference of surface topology and resulted in almost zero detachment
force to Delta expressing cells. It is possible therefore that changes in glycosylation
state could alter the conformation of the receptor/ligand and their ability to cluster
which in turn would modulate binding and activation.
5
Notch–ligand therapeutics: where lies the specificity?
Consistent with its central role in metazoan cell–cell communication, dysregulation
of Notch signalling leads to several genetic and acquired diseases. Several mutations
were identified in receptors and ligands causing developmental disorders Alagille
syndrome, Spondylocostal dysostosis, Tetrology of Fallot, aortic valve disease and
adult-onset conditions such as CADASIL [43]. Notch signalling has also been found
to be dysregulated in several cancers which make it a potential target for anti-cancer
therapeutics [44–47]. The current strategies target key events in Notch signalling
in the form of decoy soluble ligands and receptors, TACE inhibitors, γ-secretase inhibitors
(GSIs), and transcription complex inhibitors. The GSIs have reached clinical trials,
but they are not selective for different Notch receptors and are associated with significant
levels of intestinal toxicity. With the availability of high resolution structures
for regions involved in the activation process such as NRR, paralogue-specific antibodies
are under development which are designed to block activation [48]. Wu et al., have
recently shown that dual inhibition of Notch-1 and Notch-2 cause intestinal toxicity,
whereas specifically targeting Notch-1 or Notch-2 with anti-NRR antibodies avoids
toxicity [49]. These data suggest that therapeutic antibodies targeted against other
regions of the receptor and ligand involved in binding and activation may also allow
specific targeting of Notch/ligand paralogues.
6
Conclusions and perspective
It is perhaps surprising, given its known role in development and disease, that the
Notch signalling pathway remains relatively poorly understood at a molecular level.
However, recent years have seen the publication of high resolution structures for
individual parts of the receptor, ligand, and transcriptional complex (Table 1), suggesting
that a more complete molecular description is a realistic prospect. In future, combining
biochemical and biophysical skills in the areas of protein purification, O-glycosylation,
receptor endocytosis, and structural biology should allow a molecular dissection of
the activation mechanism and its regulation by post-translational modifications and,
as a consequence, offer new targets for cancer therapies.