ABSTRACT
The MADS domain proteins APETALA1 (AP1), APETALA3 (AP3), PISTILLATA (PI),
and AGAMOUS (AG) specify the identity of
Arabidopsis
floral organs. AP1 and AG homocomplexes and AP3-PI heterocomplexes bind to CArG-box sequences. The DNA-binding properties of these complexes were investigated. We
find that AP1, AG and AP3-PI are all capable of recognizing the same DNA-binding sites, although with somewhat different affinities. In
addition, the three complexes induce similar conformational changes on a CArG-box sequence. Phasing analysis reveals that the induced distortion is DNA
bending, oriented toward the minor groove. The molecular dissection of AP1,
AP3, PI and AG indicates that the boundaries of the dimerization domains of
these proteins vary. The regions required to form a DNA-binding complex include, in addition to the MADS box, the entire L region
(which follows the MADS box) and the first putative amphipathic helix of the K
box in the case of AP3-PI, while for AP1 and AG only a part of the L region is needed. The
similarity of the DNA-binding properties of AP1, AP3-PI and AG is discussed with regard to the biological specificity
that these proteins exhibit.
According to a well-established genetic model (
1
-
3
), the identities of the organs of an
Arabidopsis
flower are specified by the action of at least five homeotic genes:
APETALA1
(
AP1
)
, APETALA2
(
AP2
)
, APETALA3
(
AP3
),
PISTILLATA
(
PI
)
and
AGAMOUS
(
AG
) (
4
-
8
). While these genes have been extensively characterized at the genetic level,
little is known about the molecular mechanisms by which the organ-identity proteins act. AP1, AP3, PI and AG are all MADS domain proteins (
4
,
6
-
8
). The MADS domain is a conserved DNA-binding/dimerization region present in a variety of transcription factors
from different organisms (SRF, serum response factor; MCM1; the MEF2 family) (
9
,
10
). Within the family of MADS domain proteins, a particular characteristic of the
plant proteins is that the vast majority of them contain another conserved
region, the K box (
11
,
12
). This region has similarity to the coiled-coil segment of keratin, suggesting that the K box may form amphipathic
alpha helices, perhaps involved in protein-protein interactions (
11
,
13
). SRF and MCM1 recognize CArG-box sequences (consensus CC(A/T)
6
GG) (
14
,
15
), and
in vitro
experiments have shown that AG (
16
-
20
), AP1 and AP3-PI (
20
) complexes bind to such sites. These four proteins exhibit partner specificity
for the formation of DNA-binding complexes: neither AP3 nor PI have been found to bind DNA by
themselves or in combination with either AP1 or AG (
20
).
Since these four related proteins act to specify the development of different
organ types in the
Arabidopsis
flower, we were interested in comparing the DNA binding properties of the AP1,
AP3-PI and AG complexes, which are dimers, in an effort to understand how the
biological specificity of these (presumed) transcription factors is achieved.
We show that the DNA-binding specificities of AP1, AP3-PI and AG dimers are very similar, since they recognize the same
DNA-binding sites, although differences in affinities were detected. The three
complexes are also similar in the distortion that they induce on the DNA, that
is (at least in part) DNA bending toward the minor groove. In addition, the
molecular dissection of AP1, AP3, PI and AG has revealed differences in the
regions that are required for dimerization among these four proteins, which
correlate with the partner specificity that they exhibit.
pSPUTK (Stratagene)-derived plasmids to produce AP1, AP3, PI and AG in
in vitro
transcription/translation reactions have been described previously (
20
). Several derivatives of
AP3
,
PI
and
AG
sequences were synthesized by PCR in order to make N- and C-terminal truncated proteins, as listed below, and cloned into
pSPUTK. Throughout this article, the N- (N-terminal extension that precedes the AG MADS-box), M- (MADS domain), L- (linker between the MADS domain and the K box), K- (K box) and C- (C-terminal) regions of AP1, AP3,
PI and AG, as well as the corresponding amino acid numbering, are as shown in
Figure 1 of ref.
20
. AP3[Delta]mlck: AP3 protein lacking the first 26 aa of the MADS box (Asn residue at
position 26 is changed into the initiation Met). AP3ml: truncated AP3 protein comprising only the MADS box and the L region (a stop
codon was introduced after the Gln residue at position 88). PIml: truncated PI protein comprising only the MADS box and the L region. AG[Delta]mlck: AG protein lacking the N region and the first half of the MADS box (Asn
residue is changed into the initiation Met). AGlkc: the AGlkc protein lacks the N region and the entire MADS box except the two last aa (it
therefore starts with the sequence Met-Glu-Tyr-Ser...). AGnm: truncated AG protein comprising only the N region and the
MADS box. AGnml: truncated AG protein comprising the N, MADS and L regions. AGnmlk: truncated AG protein lacking the C-terminal region.
Seven different probes (A-G) were used. Probes A and B are derived from the promoters of the
Arabidopsis
AP3
and
SUPERMAN
(
SUP
) genes, respectively, and have been described previously (
20
); probe D is derived from the
Arabidopsis
AGL5
promoter (
19
); sites C and E were obtained in sequence-selection experiments performed with AG (
17
; site clones #85 and #41, respectively); sites F and G were obtained in
sequence-selection experiments performed with AGL3 (
21
; site clones #3 and #103, respectively). All binding sites were cloned into
pGEM vectors. Probe A, 5'-
ggatcc
TCACTTAGTTTTCATCAACTTCTGAACTTACCTTTCATGGATTAGGCAATACTTT
Proteins were synthesized using the TnT coupled transcription/translation reticulocyte lysate system (Promega).
Labeled ([
35
S]methionine)
in vitro
translation reactions demonstrated that the proteins were produced in similar
amounts. Some of the C-terminal deletion derivatives of AP1, AP3, PI and AG proteins were
obtained by digesting the plasmids encoding the full-length proteins with internal restriction sites prior to the
in vitro
transcription reaction. AP1M-2, AP1M+3, AP1M+15, AP1M+29, AP1M+33, AP1ML+6, and AP1ML+34 proteins were obtained from RNAs synthesized from pSPUTK-AP1 linearized with
Bst
BI,
Hin
fI,
Rsa
I,
Ple
I,
Bsr
I,
Afl
II and
Alw
NI, respectively. To obtain AP3ML+12, AP3ML+31 and AP3ML+42, pSPUTK-AP3 was linearized with
Ple
I,
Alw
NI and
Fok
I, respectively. The PI open reading frame was linearized at
Ecl
136II and
Bpm
I sites to generate PIML+16 and PIML+20, respectively. AGNM+22 and AGNM+28 were obtained after digestion of pSPUTK-AG with
Ase
I and
Msp
I. The RNAs were purified by agarose gel electrophoresis prior to their use in
in vitro
translation reactions performed with standard reticulocyte lysate (Promega).
In vitro
translated proteins were tested for DNA-binding activity by electrophoretic mobility shift assay (EMSA). Binding
reactions were performed as described previously (
20
). Gels for resolving protein-DNA complexes were 5% (except when indicated otherwise)
polyacrylamide:bisacrylamide (60:1) in 1* TBE. Immunoprecipitation experiments were carried out as described
previously (
20
).
Saturation-binding assays to determine the dissociation constants (
K
d
) were carried out by incubating a fixed amount of
in vitro
translated proteins (2 [mu]l of the translation reaction) with increasing amounts of probes A or B
under the standard conditions (the incubation time after addition of the probe
was extended to 90 min to allow the binding reactions to reach equilibrium, as
determined in pilot experiments). Probes were used at concentrations between 1
and 80 nM, the concentration range depending on the protein/probe combination.
After gel electrophoresis, bound and free probe were quantitated with a
phosphorimager (Molecular Dynamics). The production of both the full-length and a truncated AP1 protein in the
in vitro
translation reactions resulted in the formation of three different AP1 DNA-binding complexes. The amount of probe bound by all of them was
quantitated, and the values obtained were used for the calculations as the
total amount of bound probe. DNA-binding reactions with AG also showed band shifts originated by truncated
AG proteins, but the amount of probe that was bound in the AG reactions is very
low. This ensured that the concentration of free probe at equilibrium was
approximately equal to the concentration of total probe, and therefore that the
values obtained for the probe bound only by full-length AG could be used to calculate the apparent
K
d
s.
K
d
s were estimated by the method of Scatchard and calculation of the least-square fit line of the primary data, wherein
K
d
= -1/slope (
22
).
For circular permutation analysis, two annealed complementary oligonucleotides
containing the site A CArG-box, 5'-CTAGAGCAATACTTT
For the phasing analysis, sequences containing the site A CArG-box separated by a linker of variable length from an A tract
(intrinsically bent toward the minor groove by approximately 54o;
25
) were cloned into the
Xba
I/
Sal
I sites of pBend2 (
23
). The distance between the center of the CArG-box and the center of the A tract was 21, 23, 26, 28 or 30 bases. Sequences were as follows: 5'-ctcagaTTT
The DNA-binding capabilities of AP1, AP3-PI and AG complexes were compared using several CArG-box containing sequences as binding sites.
In vitro
translated AP1, AP3, PI and AG were incubated with probes A, B, C and D, and
the protein-DNA complexes analyzed by electrophoretic mobility shift assays (EMSA)
(Fig.
1
). Probes A, B and D contain CArG sequences that are found in the promoters of
three
Arabidopsis
genes (see Materials and Methods), while probe C is based on a synthetic AG-binding site identified in sequence-selection experiments (
17
). The probes were labeled to the same specific activity, allowing direct
comparison between the reactions containing the same protein. The shifted bands
present in the reactions with AP1 and AG correspond to protein-DNA complexes formed by the full-length proteins as well as by truncated proteins also produced in
the translation reactions. AP1 showed the strongest binding to probes A and D,
recognizing the probes in the order A~D>B>C (Fig.
1
). A similar behavior was observed for AP3-PI, while the affinities of AG for probes A, B and D were comparable and
higher than that for probe C (Fig.
1
). The binding of AG to probe C is revealed in a longer exposure of the
autoradiogram (Fig.
1
, lane 13).
Circular permutation analysis was used to determine whether AP1, AP3-PI and AG complexes induce conformational changes on the DNA upon binding
to a CArG-box sequence. This assay is based on the position-dependent effects of DNA distortion on the electrophoretic mobility
of DNA fragments of the same length (
26
). A series of probes were prepared in which the position of the site A CArG-box varies with respect to the ends of the fragments, that are otherwise
of identical sequence (Fig.
3
A). These circularly permutated probes were used in EMSAs with AP1, AP3-PI and AGnml. In all cases, protein-DNA complexes in which the CArG-box sequence is localized toward the center of the DNA
fragment (probes 3, 4 and 5) showed lower mobility than those in which the CArG-box is located near either end (probes 1, 2, 6 and 7) (Fig.
3
B), indicating that AP1, AP3-PI and AGnml induce DNA conformational changes. The unbound probes possessed similar
mobilities, regardless of the position of the CArG-box, suggesting that they do not contain significant intrinsic DNA bends
(data not shown). The distortion angles were calculated from the data obtained
in the circular permutation analysis (Fig.
3
C), and estimated to be 53o (AP1 and AP3-PI), and 73o (AGnml). The full-length AG protein was also used in EMSAs with the
circularly permutated probes, and its induced apparent bend angle was estimated
to be 70o (data not shown).
To investigate the role of the K box and other regions that are C-terminal to the MADS box in DNA-binding complex formation, a series of C-terminal deletion derivatives of AP1, AP3, PI and AG were
produced by
in vitro
transcription/translation. Regions C-terminal to the AP1 MADS box are required to form a DNA-binding complex, since neither AP1M-2 nor AP1M+3 derivatives have such activity (Fig.
5
A; AP1M-2 and AP1M+3 truncated proteins lack the last two amino acids of the MADS box or
contain the first three amino acids of the L region, respectively). The K box
is not required for DNA binding, as AP1M+15, AP1M+29 and AP1M+33 truncated proteins were capable of DNA binding (Fig.
5
A). Therefore, the `core' AP1 protein (minimal DNA-binding domain) consists of the MADS box and part of the L region. AP1M+15 binds to DNA, but at much reduced levels compared with AP1M+29 (Fig.
5
A; comparable amounts of the truncated proteins were produced in the
in vitro
translation reactions). AP1 truncated proteins were also used to show that the
DNA-binding complex is a protein dimer. The presence of both AP1M+29 and AP1ML+34 in the DNA-binding reaction leads to the formation of a single additional
complex of intermediate mobility, corresponding to a heterodimer of both
protein forms (Fig.
5
A, lanes 10-12).
AP1, AP3-PI and AG dimers were tested for DNA-binding with seven different CArG-box containing sequences, and of the resulting 21 different
protein-DNA combinations only one failed to show DNA-binding, that between AP1 and probe E. Some of the probes used were
synthetic binding sites identified in random sequence-selection experiments performed with either AG or AGL3 (
17
,
21
), but were nevertheless also bound by AP1 and AP3-PI. These results indicate that the sets of sequences recognized by AP1,
AP3-PI and AG dimers are largely overlapping. Moreover,
AGL5
has been proposed to be regulated by AG (
19
); however, the CArG-box (probe D in this study) that might mediate such regulation is also
very efficiently bound by AP1 and AP3-PI. Similarly, a CArG-box present in the
AP3
promoter (probe A), that might be involved in the autoregulation of
AP3
expression by AP3-PI (
7
,
31
,
32
), is also bound by AP1 and AG; and the three complexes recognized the probe
derived from the
SUP
promoter (probe B). It is noteworthy that the three probes that are derived
from the
Arabidopsis
genome were bound with much higher affinities than those obtained from sequence-selection experiments, showing that the sequence-selection experiments did not unequivocally identify the highest
affinity binding sites, and questioning the biological significance of the
consensus sequences that are defined in those experiments. The similarity (or
identity) of the sequences recognized by AP1, AP3-PI and AG implies that it would not be feasible to try to identify
downstream genes of each particular MADS box protein complex by scanning
Arabidopsis
genomic sequences for CArG motifs. In addition, and most importantly, this
similarity raises a question about the
Arabidopsis
MADS domain homeotic proteins that has been asked previously for other
transcription factors: how do proteins that recognize the same or very similar
sets of binding sites regulate the expression of different groups of downstream
genes?
Although AP1, AP3-PI and AG recognize similar sets of target sites, their intrinsic DNA-binding specificities are not identical: differences in the
in vitro
DNA-binding affinities are detected. It is possible that these differences
contribute to the biological specificity of these proteins. However, if subtle
differences in DNA-binding affinities are, by themselves, the main determinants of the
functional specificity of these four homeotic proteins, their concentrations in
the cell should be critical and thus finely regulated. The available data, on
the contrary, have not revealed a tight link between protein concentration and
developmental outcome. First, none of the
ap1
,
ap3
,
pi
or
ag
alleles studied to date has been shown to be a haplo-insufficient mutation with respect to organ identity. In addition,
AG
,
AP3
and
PI
have been ectopically expressed under the control of the constitutive 35S
promoter and shown to produce the expected organ identity changes (
31
-
33
). These data indicate that the levels of expression of AP1, AP3, PI and AG can
be varied within a certain range without affecting their control of organ
identity (it remains an open possibility that the level of protein of each gene
is also regulated posttranscriptionally). Certain thresholds of homeotic
protein concentration or function likely exist: the phenotype that is conferred
by the ectopic expression of
AG
or
AP3
can vary in its severity between different transgenic lines, presumably owing
to different levels of transgene expression (
31
,
33
). Nonetheless, the only functions identified in the ectopic expression
experiments are those that are particular to the wild-type expression of each of those genes, and no new or different functions
are shown by these proteins in the different transgenic lines. Therefore, the
thresholds of protein concentration or function could in part be related to the
DNA-binding activity of each of these proteins, but they do not indicate that
the specific functions of each protein can be changed by under- or overexpression, as would be expected if subtle DNA affinity
differences were responsible for specific functions.
AP1, AP3-PI and AG dimers were found to induce similar degrees of DNA bending
toward the minor groove. It is noteworthy that a truncated core AG protein, AGnml, induced the same DNA distortion as the full-length AG protein, suggesting that the results obtained in the circular
permutation and phasing analysis experiments were not affected by a possible
extended shape of the proteins (as has been described in other cases;
34
). In addition, the crystal structure of core SRF bound to DNA has recently been
determined and showed the DNA bent around the protein (
9
). The similarity of the conformational changes induced by AP1, AP3-PI and AG dimers suggests that the different regulatory specificities of
these three complexes do not arise through the generation of different DNA
structures that could direct the formation of transcripton complexes with
distinct functional properties. It therefore seems at least possible that the
biological specificity of AP1, AP3, PI and AG cannot be explained on the basis
of their intrinsic DNA-binding properties alone. Consistent with this interpretation,
in vivo
analyses of the activity of chimeric genes formed by swapping regions between
AP1, AP3, PI and AG have shown that, at least in some cases, the MADS domains
can be interchanged without them determining the specific functions of the
resulting chimeric proteins (
35
). In addition, we have recently found that the DNA-binding specificity of AP1, AP3, PI and AG can be altered without
affecting their functions
in vivo
(J. L. Riechmann and E. M. Meyerowitz, unpublished results). Another possible
mechanism by which the MADS domain homeotic proteins could direct the
development of different organs is that they may act in conjunction with
cofactors that modulate their ability to regulate the transcription of
downstream genes. This could be a process in which DNA bending by the plant
MADS-domain proteins might be involved, through determining DNA topology in
nucleoprotein complexes, allowing interactions with other proteins that may
bind to adjacent DNA sites, or facilitating the recognition by accessory
proteins of their respective target sites, as has been suggested for SRF (
9
).
This situation of diverse and highly specific
in vivo
functions by related proteins with similar DNA-binding properties is reminiscent of that encountered for the
Drosophila
homeotic selector proteins. Homeodomain proteins also show very similar
intrinsic DNA-binding specificities
in vitro
(with affinities on the order of
K
d
= 10
-8
-10
-9
M) (
36
). Some differences in the DNA-binding specificities are also detected, which might contribute in part to
the functional specificity of the proteins (
37
). However, the analyses of different mutant and chimeric proteins in ectopic
expression experiments have shown that the specificity of action of the
homeodomain proteins
in vivo
also depends on protein-protein interactions (
38
). Examples of direct interactions between the MADS box proteins of animals and
fungi and additional cofactors are already abundant. Some of these interactions
result in modulation of the MADS box protein activity and a concomitant cell-specific differential gene expression, eventually leading to cell
specialization or to different developmental pathways. The yeast MADS domain
protein MCM1 is required for transcription in the three yeast cell types, but
through interactions with different cofactors ([alpha]1 protein, [alpha]2 homeodomain protein) it regulates the transcription of cell-type specific genes. Thus the regulatory activities of MCM1
are determined by the availability of accessory proteins in conjunction with
the sequence context of the MCM1 binding sites (
10
,
39
). The MADS domain protein MEF2A physically interacts with muscle bHLH
transcription factors to control the cascade of myogenic development through
cooperative activation of muscle gene expression (
40
,
41
).
As expected from the high degree of sequence similarity, the organization of the
AG, AP3 and PI MADS boxes is similar to that of SRF: the basic N-terminal half is essential for DNA-binding and the C-terminal half is required for dimerization (
9
,
42
). Since the MADS box proteins bind to DNA as dimers, the minimal DNA-binding domain includes the conserved 56 aa MADS box, and an additional C-terminal extension, whose sequence is not conserved throughout the
family but is necessary for dimerization. This extension is of ~24 aa in SRF and MCM1 (
9
,
16
,
42
). In addition to the MADS box, the minimal DNA-binding domains of AP1 and AG include extensions of ~20 amino acids (part of the L region), and similar results have been
obtained recently with the
Arabidopsis
AGL2 protein, whose core includes the MADS domain and the first 21 aa of the L
region (
43
). On the other hand, core AP3 and PI proteins comprise the entire L region and
part of the K box (a total C-terminal extension to the MADS domain of ~50 aa). The involvement of the first amino acids of the K-box in dimerization has also been recently shown for the
Antirrhinum
homologous proteins of AP3 and PI: DEF and GLO, respectively (
44
). The difference in the size of the core proteins, AG and AP1 on one hand, and
AP3 and PI on the other, correlates with the partner specificity that these
proteins possess: AG and AP1 form DNA-binding homodimers but not DNA-binding heterodimers with AP3 or PI, which form a DNA-binding AP3-PI heterodimer (
20
).
Based on the presumptive coiled-coil structure of the K box, and by analogy to leucine zipper proteins, it
has been suggested that this region could be involved in promoting dimerization
(
12
). The analysis of C-terminal deletion mutants described here shows that the entire K box (in
the case of AP1 and AG), or a substantial part of it (in the case of AP3 and
PI), is dispensable for the formation of DNA-binding dimers. It is possible that the K box plays a role in dimer
stabilization, but might not be required in the mild conditions used in the DNA-binding experiments. Consistent with this notion, it has been shown that
deletion of part of the K box of an epitope-tagged PI protein (a deletion that did not include the region shown here
as forming part of the core protein) reduced, but did not abolish, the
immunoprecipitation of labeled AP3 protein (
7
). Alternatively, the K-box could be involved in interactions with additional (unknown) cofactors
of the plant MADS box proteins.
In summary, the finding of differences in the organization of the AP1, AG and
AP3 and PI proteins, and its correlation with the partner specificity that
these proteins exhibit for the formation of DNA-binding dimers (
20
), support the idea that selective dimerization is part of the mechanism by
which these proteins achieve their functional specificity. On the other hand,
the DNA-binding activities of these dimers (AP1, AP3-PI and AG) are very similar, suggesting that the biological
specificity that these proteins possess may not be explained on the basis of
their intrinsic DNA-binding specificity alone. It is likely that at least part of their
biological specificity is achieved through selective interactions with
additional transcription factors, a mechanism that appears to be a common theme
for the MADS box proteins of animals and fungi.
We are grateful to Hong Ma for providing binding site probes, to Sankar Adhya
for pBend2, and to members of the laboratory for valuable comments on the
manuscript. This work was supported by US National Science Foundation grant MCB-9204839 to E. M. M.; J. L. R. was supported by a fellowship from
Ministerio de Educación y Ciencia (Spain) and was formerly an EMBO Postdoctoral Fellow.
REFERENCES
Return