Báo cáo khoa học: Site-directed mutagenesis and footprinting analysis of the interaction of the sunflower KNOX protein HAKN1 with DNA ppt

Binding of HAKN1 to different oligonucleotides indicated that HAKN1 prefers the sequence TGACA TGTCA, with changes within the GAC core more pro-foundly affecting the interaction.. Concer

of the interaction of the sunflower KNOX protein HAKN1 with DNA

Mariana F Tioni, Ivana L Viola, Raquel L Chan and Daniel H Gonzalez

Ca´tedra de Biologı´a Celular y Molecular, Facultad de Bioquı´mica y Ciencias Biolo´gicas, Universidad Nacional del Litoral, Santa Fe, Argentina

Homeobox genes encode a group of eukaryotic

tran-scription factors generally involved in the regulation of

developmental processes [1] These genes contain a

region coding for the homeodomain, a 60 amino acid

protein motif that interacts specifically with DNA [2]

The homeodomain folds into a characteristic

three-helix structure Helices I and II are connected by a

loop, while helices II and III are separated by a turn,

resembling prokaryotic helix-turn-helix transcription

factors However, unlike helix-turn-helix-containing

proteins, most homeodomains are able to bind DNA

as monomers with high affinity, through interactions made by helix III (the so-called recognition helix) and

a disordered N-terminal arm located beyond helix I [3–6]

In plants, the first homeobox was identified in the maize gene Knotted1 (kn1; [7]) Dominant mutations in kn1, which is normally active only in meristematic cells, affect leaf development due to its aberrant expression in these organs [8] Additional kn1-like genes (also termed knox genes) have been isolated from maize and other monocot and dicot species

Keywords

DNA-binding specificity; footprinting;

homeodomain; KNOX protein; recognition

code

Correspondence

D H Gonzalez, Ca´tedra de Biologı´a Celular

y Molecular, Facultad de Bioquı´mica y

Ciencias Biolo´gicas (UNL), CC 242 Paraje El

Pozo, 3000 Santa Fe, Argentina

Fax ⁄ Tel: +54 342 4575219

E-mail: dhgonza@fbcb.unl.edu.ar

(Received 13 July 2004, revised 31 August

2004, accepted 21 September 2004)

doi:10.1111/j.1432-1033.2004.04402.x

The interaction of the homeodomain of the sunflower KNOX protein HAKN1 with DNA was studied by site-directed mutagenesis, hydroxyl radical footprinting and missing nucleoside experiments Binding of HAKN1 to different oligonucleotides indicated that HAKN1 prefers the sequence TGACA (TGTCA), with changes within the GAC core more pro-foundly affecting the interaction Footprinting and missing nucleoside experiments using hydroxyl radical cleavage of DNA showed that HAKN1 interacts with a 6-bp region of the strand carrying the GAC core, covering the core and nucleotides towards the 3¢ end On the other strand, protec-tion was observed along an 8-bp region, comprising two addiprotec-tional nucleo-tides complementary to those preceding the core Changes in the residue present at position 50 produced proteins with different specificities An I50S mutant showed a preference for TGACT, while the presence of lysine shifted the preference to TGACC, suggesting that residue 50 interacts with nucleotide(s) 3¢ to GAC Mutation of Lys54 fi Val produced a protein with reduced affinity and relaxed specificity, able to recognize the sequence TGAAA, while the conservative change of Arg55fi Lys completely abol-ished binding to DNA Based on these results, we propose a model for the interaction of HAKN1 with DNA in which helix III of the homeodomain accommodates along the major groove with Arg55, Asn51, Lys54 and Ile50, establishing specific contacts with bases of the GACA sequence or their complements This model can be extended to other KNOX proteins given the conservation of these amino acids in all members of the family

Abbreviations

TALE, three-amino-acid loop extension.

(reviewed in [9]), indicating that this class of genes

constitutes a family present throughout the plant

king-dom The knox family of genes can be subdivided into

two classes, I and II, by sequence relatedness and

expression patterns [10] Based on the expression

pat-terns [11–13], analysis of mutants [14–17] and

over-expression studies [18–21] it was proposed that class

I knox genes are involved in the maintenance of

meris-tematic cells in an undifferentiated state Indeed,

over-expression of some class I genes in Arabidopsis and

tobacco produces the proliferation of meristems on the

surface of leaves

The proteins encoded by knox genes belong to the

three-amino-acid loop extension (TALE) superclass

Members of this superclass contain three extra amino

acids within the loop connecting helices I and II [22]

and are present in several eukaryotic kingdoms,

sug-gesting that they represent an early evolutionary

acqui-sition

Concerning their interaction with DNA, studies with

proteins from barley [23], tobacco [24], rice [25] and

maize [26] indicate that they bind related sequences

containing a TGAC core (GTCA in the

complement-ary strand), considerably different from the sequence

TAAT recognized by most homeodomains [27]

Eluci-dating the structural basis for this difference would

help to understand at the molecular level how KNOX

transcription factors recognize their DNA target site

In this study, we analysed the interaction of the

homeodomain of HAKN1, a sunflower class I KNOX

protein [28], with DNA Based on studies of wild-type and mutant forms of the homeodomain, we propose a model for the complex between HAKN1 and its target site This model must be applicable to all KNOX homeodomains, as important amino acids are con-served within this family

Results

Expression and DNA binding analysis of the HAKN1 homeodomain

The homeodomain of the KNOX transcription factor HAKN1 was expressed in Escherichia coli as a fusion with the maltose binding protein using vector pMALc2 The fusion protein was purified by affinity chromatography in amylose resin and used for DNA– protein interaction studies A 24-bp oligonucleotide (HAKN1 binding site; BS1) containing the sequence TGT(G⁄ C)ACA was used as DNA target This seq-uence was designed against a compilation of seqseq-uences bound by KNOX transcription factors from different species, and contains the TGAC (GTCA) core that is present in all of them

Figure 1A shows an electrophoretic mobility shift assay performed with HAKN1 and oligonucleotide BS1 or variants containing changes at single positions (sequences shown in the right panel) We have arbi-trarily numbered from 1 to 7 those positions present in the strand that contains the central G Two shifted

B

A

C

Fig 1 Binding of HAKN1 to different

oligo-nucleotides (A) Electrophoretic mobility

shift assay performed with 30 ng of HAKN1

and oligonucleotides containing different

variants of the sequence TGT(G ⁄ C)ACA

(numbers indicated above each lane) (B)

Competition assay of HAKN1 binding to BS1

using a 15-fold molar excess of different

oligonucleotides (numbers indicated above

each lane) as competitors The sequence of

the 7-bp core present in each

oligonucleo-tide is shown in (C) for reference

Modifica-tions with respect to BS1 are shown within

black boxes.

bands of similar intensity were observed in this

experi-ment The relative intensity of the low mobility

com-plex varied when different protein preparations were

used We speculate that this behavior may arise from

aggregation of the protein Nevertheless, different

pro-tein preparations showed the same specificity and

affin-ity when considering the amount of bound protein as

the sum of both shifted bands These bands displayed

similar footprinting patterns (see below), suggesting

that a single HAKN1 homeodomain is bound to DNA

in both complexes This is strengthened by the fact

that only monomer–DNA complexes were observed in

crosslinking experiments (data not shown)

Analysis of the interaction of HAKN1 with different

oligonucleotides indicates that modifications in the

outermost positions (1 and⁄ or 7) do not significantly

affect binding (Fig 1A, lanes BS1, 1,7, 7T, 1 and 7C),

while certain inner nucleotides, notably those located

at positions 4–6, are critical for binding (Fig 1A, lanes

4, 6A and 5) Regarding position 7, the change of A

for T does not seem to affect binding, while the

intro-duction of C partially reduces the amount of complex

formed Mutations at positions 2 (not shown) and 3

(lane 3) have only a moderate effect Similar

obser-vations could be made in experiments in which the

binding to oligonucleotide BS1 was competed with a

15-fold molar excess of different oligonucleotides

(Fig 1B) These results indicate that HAKN1 mainly

recognizes the GAC (GTC) trinucleotide and displays

lower specificity at outer positions The GAC triplet is

contained within the TGAC sequence, found to be

part of the binding sites of the barley KNOX protein

Hooded [23] and of maize Knotted1 [26] This element

is also present in the sequence GTNAC, postulated to

be important for the binding of the tobacco protein

NTH15 to DNA [24], provided that N is G or C

Analysis of DNA binding by hydroxyl radical

footprinting and interference assays

A more detailed picture of the binding of HAKN1 to

its target site was obtained by the analysis of

footprint-ing patterns after cleavage of free and protein-bound

DNA with hydroxyl radicals generated by Fe–EDTA

complexes For this purpose, a dimer of the

corres-ponding oligonucleotide ligated through its EcoRI

cohesive site was cloned into the BamHI site of

pBlue-script SK– Cleavage with HindIII and XbaI produces

a 94-bp fragment that contains two HAKN1 binding

sites in opposite orientations After HAKN1 binding

to the 94-bp oligonucleotide, labeled specifically at one

of its 3¢ ends by filling-in the HindIII site, the complex

was subjected to hydroxyl radical attack, and free and

bound DNA were separated, recovered from the gel and analysed by denaturing polyacrylamide gel electro-phoresis (Fig 2A) Because the oligonucleotide con-tains two sites in opposite orientation, both strands of the binding site can be observed in a single footprint-ing assay Analysis of the cleavage patterns indicates that HAKN1 protects six nucleotides from the strand carrying the sequence TGTGACA (hereafter named the top strand) The protected area includes GACA and two adjacent nucleotides (GA) towards the 3¢ end (Fig 2A) On the bottom strand, the protected region covers two additional nucleotides, AC complementary

to GT in TGTGACA (Fig 2A) For both strands, the highest protection is observed within the GAC core, suggesting that the protein makes closer contacts in this region This agrees with the important role of these nucleotide positions in determining the binding strength of HAKN1 to DNA shown by electrophoretic mobility shift assays When the oligonucleotide labeled

at its XbaI site (at the opposite 3¢ end) was used, foot-printing patterns were identical to those described above, indicating that HAKN1 makes equivalent contacts with both binding sites present in the 94-bp fragment

Footprinting analysis was also performed with a similar oligonucleotide containing two mutated sites [BS(mut1,7); AGTGACT instead of TGTGACA, mutations underlined) The results obtained were essentially the same (not shown), suggesting that HAKN1 contacts the nucleotide adjacent to the GAC core and its complement on the other strand whether they are A or T

Information about the nucleotide positions that influence binding of HAKN1 to DNA was obtained from missing nucleoside (interference) experiments Here, DNA is treated with hydroxyl radical-generating agents before protein binding, thus producing a popu-lation of molecules with single cleavages along the phosphodiester backbone This population is incubated with the protein of interest and subjected to an elec-trophoretic mobility shift assay from which the free and bound fractions are recovered Molecules with cleavages at positions important for binding are then under-represented in the bound fraction and, depend-ing on the binddepend-ing conditions, over-represented in the free fraction Figure 2B shows a missing nucleoside experiment using HAKN1 and the 94-bp DNA frag-ment containing two binding sites previously labeled in one of its 3¢ ends (HindIII or XbaI sites) and treated with Fe–EDTA It is noteworthy that there is a good correlation between the region protected by HAKN1 and the nucleotide positions important for binding This means that all nucleotides in the protected area

establish contacts that contribute to binding efficiency.

Again, the GAC core seems to be particularly

import-ant, but outside positions are also required (Fig 2B)

Within the core, modifications to G and A or their

complements influence binding more markedly These

results agree with the fact that mutations of these two

nucleotides abolish binding of HAKN1 to DNA On

the other hand, because nucleotides at outside posi-tions can be mutated without significant loss in bind-ing efficiency, it can be assumed that they mainly participate in nonspecific contacts, such as those estab-lished with the sugar–phosphate backbone

The results of footprinting and missing nucleoside experiments also indicate that HAKN1 does not make

Fig 2 Hydroxyl radical footprinting and interference analysis of HAKN1 binding to DNA An oligonucleotide containing two HAKN1 binding sites (BS1) in opposite orientations was labeled in the 3¢ end of either strand (HindIII or XbaI sites) and subjected to hydroxyl radical attack either after (A) or before (B) HAKN1 binding Free (F) and bound (B) DNA were separated and analysed A portion of the same fragment digested with defined restriction enzymes was used as a standard (S) to calculate the position of the footprint Letters to the right of each panel indicate the DNA sequence (5¢ end in the upper part) of the corresponding strand in this region In the lower part, the sequence of the binding site is shown and the protected positions are indicated in bold and underlined The GAC (GTC) core that shows the highest protec-tion is shaded.

symmetrical contacts with its target site The protein

establishes contacts with both strands at the right side

of the GAC core, while only one strand seems to be

contacted at the left side This lack of symmetry and

the extension of the contacts most probably indicate

that only one molecule of HAKN1 is bound at each

target site

Binding of HAKN1 single-site mutants to DNA

The picture that emerges from our results is that

HAKN1 binds an 8-bp region of DNA with a tGACa

(tGTCa) specificity core An interesting question is

how the HAKN1 homeodomain interacts with this

sequence and which amino acids are involved in

sequence-specific contacts To answer this, we have

analysed the effect of single-site mutations on HAKN1

binding to TGACA and variants of this sequence It is

logical to assume that changes in amino acids involved

in the interaction must influence binding efficiency In

addition, some substitutions may alter binding

specific-ity, indicating the existence of contacts between a given

residue and defined positions within the DNA

Residue 50 (53 in TALE homeodomains) is usually

involved in determining the different specificities

among related homeodomains [27,29–31] In

homeo-domains that bind the canonical TAAT sequence,

residue 50 interacts with nucleotides located 3¢ to this site [27,31] We reasoned, then, that changing Ile50, present in HAKN1 and all KNOX proteins, may influ-ence sequinflu-ence preferinflu-ences at external positions of the core As a first approach, we mutated Ile50 to Ser, pre-sent in the yeast TALE protein MATa2 [32] The ana-lysis of binding of I50S–HAKN1 to variants of the HAKN1 binding site indicates a preference for an oligonucleotide containing the sequence TGACT, while the wild-type HAKN1 homeodomain binds TGACA and TGACT with similar efficiency (Fig 3A) This suggests that residue 50 interacts with the 3¢ region of the top strand (and⁄ or the 5¢ region of the bottom strand), outside the GAC core This is also evident in competition experiments (Fig 3B), where oligonucleo-tides BS(mut1,7) and BS(mut7T) compete more effi-ciently than variants with A [BS1 and BS(mut1)] or C [BS(mut7C)] at this position Changes at other posi-tions within the target DNA sequence produced sim-ilar effects on binding than with the wild-type protein (Fig 3)

To further explore the hypothesis that residue 50

is oriented towards the 3¢ end of the top strand, we also mutated Ile50 to Lys, present in Drosophila bicoid [33] I50K–HAKN1 shows a net preference for an oligonucleotide containing the sequence TGACCC [BS(mut7C)] over the original TGACAG, present in

Fig 3 DNA binding preferences of HAKN1 mutants at position 50 (A) Electrophoretic mobility shift assay of I50S–HAKN1 (30 ng) binding to BS1 and BS(mut1,7) (B) Binding

of I50S–HAKN1 to BS(mut1,7) was com-peted with a 100-fold molar excess of oligo-nucleotides with different sequences (depicted in Fig 1) (C) Binding of I50K– HAKN1 (30 ng) to different oligonucleotides was analysed by an electrophoretic mobility shift assay In (D), the binding of different amounts (50, 100 and 250 ng) of either HAKN1 or I50K-HAKN1 to oligonucleotides BS1 and BS(mut7C) is shown Oligonucleo-tide sequences are shown in Fig 1.

BS1 and BS(mut1) (Fig 3C) This result confirms that

residue 50 interacts with nucleotides adjacent to the

TGAC core Binding analysis with different

oligonuc-leotides indicated that I50K–HAKN1 is also able to

interact with oligonucleotide BS(mut6G), that contains

a TGAG core (Fig 3C) In fact, when higher protein

concentrations were used in the assays, binding to

TGAGAG was considerably better than to TGACAG

(not shown), suggesting that Lys50 may also be able to

contact the fourth position of the core, thus changing

the preference for G The inclusion of Lys at position

50, in addition to promoting a change in specificity,

resulted in a protein with increased affinity towards its

preferred binding site (Fig 3D) An additional,

fast-migrating band observed in this experiment is present

in free DNA and may represent noncovalent

oligo-nucleotide dimers interacting through their cohesive

ends We have observed that the presence of this

spe-cies does not affect the intensity of the shifted band

The increased affinity dispalyed by I50K–HAKN1

may arise from the fact that lysine is able to establish

hydrogen bonds with DNA, which are more stable

than the van der Waals contacts established by Ile

The interaction of mutants at position 50 with their

preferred binding sites was also analysed by

footprint-ing experiments I50S–HAKN1 protects a region

cov-ering five nucleotides of the top strand and six

nucleotides of the bottom strand (Fig 4A) This region

is coincident with the one more strongly protected by wild-type HAKN1, but is shorter towards the 3¢ end

of the top strand This result further suggests that Ile50 contacts the nucleotides located 3¢ to the TGAC core, as its replacement by a smaller residue such as Ser allows better access of this region to the modifying agent Conversely, I50K–HAKN1 shows an extended footprinting pattern towards the 3¢ end of the top strand and the 5¢ end of the bottom strand (Fig 4B) This agrees with the presence of a larger residue that makes stable contacts with this region of DNA The interaction of mutants at position 50, and par-ticularly of I50K–HAKN1 with DNA, provides a framework to build a model of HAKN1–DNA inter-actions, taking into account experiments performed with other homeodomains The protein bicoid, for example,

is able to bind the sequence TAATCC that contains the canonical TAAT box [31] Lys50 of bicoid puta-tively interacts with the CC dinucleotide, as its muta-tion to Gln changes its preference to TAATTG [29] A reciprocal change, Gln50 to Lys, in engrailed or fushi tarazu shifts sequence preferences from TAATTA or TAATTG to TAATCC [30,34] We postulate, then, that positioning of the HAKN1 homeodomain along the TGAC core in DNA must be equivalent to that adopted by other homeodomains along the TAAT sequence The third position of both sequences con-tains an adenine, known to interact with Asn51,

Fig 4 Hydroxyl radical footprinting of I50S–HAKN1 (A) and I50K–HAKN1 (B) bound to their preferred binding sites After binding and hydro-xyl radical attack, free (F) and bound (B) DNA were separated and analysed The left and right panels in (A) and (B) represent the top and bottom strands of the binding site, respectively A portion of the same fragment digested with defined restriction enzymes was used as a standard to calculate the position of the footprint Letters to the right of each panel indicate the DNA sequence (5¢ end in the upper part) of the corresponding strand in this region Below the footprints, the sequence of the corresponding binding site is shown and the protected positions are indicated in bold and underlined.

universally conserved among homeodomains [2,31].

The importance of this interaction is reflected by the

fact that this nucleotide cannot be mutated without a

complete loss of HAKN1 binding The fourth base in

TAAT is usually recognized by a nonpolar amino acid

(mostly Ile or Val) present at position 47 [2,31]

HAKN1 contains Asn at this position, which may be

too small to establish specific contacts with bases

Asn47 does not make specific contacts in the

homeo-domain–DNA complexes of MATa2 and extradenticle

[5,35] Here, we favour the hypothesis that the fourth

position of the core is contacted by Lys54, because the

nucleotide next to that contacted by Asn51 is

recog-nized by residue 54 in other homeodomains (see

below) In support of a prominent role of Lys54, its

mutation to Val produces a significant decrease in

DNA binding (not shown) In addition, K54V–

HAKN1 binds with similar efficiency to sequence

vari-ants containing either A [BS(mut6A)] or C (BS1) at

the fourth position of TGAC, suggesting that it has a

decreased discrimination capacity with respect to

wild-type HAKN1 (Fig 5) An oligonucleotide containing

TGAG [BS(mut6G)], however, is bound with reduced

efficiency, suggesting that the mutant homeodomain

retains partial specificity Discrimination at other

posi-tions of the bound region is similar to those displayed

by the wild-type protein Although the results obtained

do not necessarily indicate a direct role of Lys54 in

establishing contacts with DNA, a plausible

explan-ation is that this residue interacts with at least one of

the members of the CÆG pair at the fourth position of

TGAC in the HAKN1–DNA complex

The two leftmost positions of the core interact

through the minor groove with the N-terminal arm in

most homeodomains [2,31] In yeast MATa2, for

example, the N-terminal arm makes base-specific

con-tacts with the first two nucleotides of a TTAC core [5]

Hence, we replaced the N-terminal arm (residues 1–9)

of HAKN1 with the same portion of MATa2, to

determine if a change in specificity was observed The

resulting protein, Na–HAKN1, showed an overall

reduced affinity but the same sequence preferences as

HAKN1 (Fig 6) It is noteworthy that it did not bind

oligonucleotide BS(mut4), which contains a TTAC

core on the complementary strand This indicates that

the N-terminal arm of MATa2 is not able to interact

with DNA within the context of the HAKN1

homeo-domain as it does within MATa2 Poor binding may

arise from incorrect folding of the chimeric protein or

from the fact that important contacts with DNA are

lost upon replacement of the HAKN1 N-terminal arm

In addition to a role of the N-terminal arm in

contact-ing the first two amino acids of the core, examination

Fig 6 Effect of changes within the N-terminal arm and position 55

on the binding of HAKN1 to oligonucleotides BS1 and BS(mut4) Binding to oligonucleotides containing the sequences TGACA (BS1)

or TTACA [BS(mut4)] was analysed using 30 ng of proteins HAKN1, R55K–HAKN1, Na–HAKN1 (a protein containing the N-terminal arm

of MATa2) or R55K–Na–HAKN1 (a protein with both modifications).

A

B

Fig 5 K54V–HAKN1 shows relaxed specificity Binding of K54V– HAKN1 (150 ng) to different oligonucleotides was analysed in an electrophoretic mobility shift assay (A) (B) Competition of K54V– HAKN1 binding to BS1 with a 25-fold molar excess of different oligonucleotides (depicted in Fig 1).

of other homeodomain–DNA complexes suggests the

possibility that Arg55 recognizes the second position

of TGAC Arg55 participates in binding to G residues

in other homeodomains, such as yeast MATa1

(GATG; [36]) or Drosophila extradenticle (TGAT;

[35]) Consistent with a role in DNA binding, an

Arg55 to Ala mutation completely disrupts the

inter-action of HAKN1 with DNA (not shown) To further

analyse its involvement in base-specific contacts, we

reasoned that a conservative substitution for Lys

would not affect nonspecific interactions (i.e

electro-static interactions with the phosphate backbone), but

would preclude the establishment of hydrogen bonds

with the guanine base of G The results shown in

Fig 6 indicate that R55K–HAKN1 is unable to bind

DNA, supporting the hypothesis that Arg55 is

involved in base-specific contacts, which are disrupted

upon mutation to Lys Another explanation would be

that this change disturbs the overall folding of the

homeodomain, but this seems unlikely because several

homeodomains, notably MATa2, contain Lys at

posi-tion 55 Assuming that the N-terminal arm of MATa2

and Arg55 may be incompatible as both portions may

interact with the same positions of the target site, we

also constructed a mutant in which the N-terminal

arm of MATa2 was inserted into the R55K mutant of

HAKN1 This protein was also ineffective in binding

to the HAKN1 target site or its variants (Fig 6)

A model for the HAKN1–DNA interaction

Based on the analysis of the binding of wild-type and

mutant HAKN1 homeodomains to different DNA

tar-get sites, we propose a model for the interaction of

HAKN1 with DNA A set of four amino acids,

located within helix III of the homeodomain, would

make base-specific contacts with defined nucleotides

within the tGACAg sequence Arg55 would establish a

pair of hydrogen bonds with positions O6 and N7 of

guanine in GACA As mentioned above, similar

inter-actions have been observed in complexes of other

homeodomains with DNA [35,36] Asn51 would

inter-act with the first adenine in GACA, also establishing

a pair of hydrogen bonds, as in most homeodomain–

DNA complexes The next position (C, or G in the

opposite strand) would be contacted by Lys54

Although there is no evidence in the literature about a

specific contact made by a lysine at this position,

resi-due 54 interacts with the nucleotide adjacent to that

bound by Asn51 in several homeodomains, for

exam-pleMATa2 (Arg54, TTAC; [5]), TTF1 (Tyr54, CAAG;

[37]), bicoid (Arg54, TAAT or TAAG; [38]) and Hahr1

(Thr54, TAAA, in this case in combination with

Phe47; [39]) Additionally, lysine determines a prefer-ence for C at an adjacent position when present at position 50 in bicoid and other mutant homeodomains (including HAKN1, see above), presumably by inter-acting with guanine bases through hydrogen bonds as observed in the Lys50–engrailed crystal structure [40] Finally, our results also indicate that Ile50 is involved in establishing a preference for A or T at the 3¢ side of the core Mutations of this residue to Ser or Lys were able to confer a new binding specificity to HAKN1, changing to a net preference for T or C, respectively Ile50 is present in MATa1, where it inter-acts with a TA dinucleotide adjacent to the position contacted by Met54 [36] Accordingly, Ile50 may also

be involved in contacts with an adjacent position, which is protected by HAKN1 in footprinting experi-ments and interferes with binding when modified by hydroxyl radical attack

To examine the consistence of the interactions des-cribed above, we have constructed a theoretical model

of the HAKN1–DNA complex using the program swiss-model [41] available in the ExPASy web server Different models for wild-type and mutant HAKN1 were obtained using the homeodomain–DNA com-plexes of extradenticle [35], MATa1 [36] and MATa2 [5] as templates Figure 7 shows the alignment of helix III of the HAKN1 homeodomain along the major groove of DNA (the MATa1 binding site in this case) Amino acids in red are those present in wild-type HAKN1 that putatively contact the GAC core Note that Arg55 and Asn51 establish hydrogen bonds with adjacent G and A, respectively Interestingly, Lys54 also appears making a hydrogen bond with the N7 of

an adjacent purine (adenine in this case) present in the complementary strand A similar contact could be made with a guanine complementary to C in GAC, further suggesting that Lys54 is likely to contact this position The position of Lys55 in the corresponding mutant is shown in yellow Clearly, the specific con-tacts made by Arg55 are lost and are replaced by an interaction with the phosphate backbone Val54 is shown in pink The shorter side chain and the loss of

a hydrogen bond explain the decrease in affinity and relaxed specificity Finally, the variants at position 50 (Ile in orange, Ser in green and Lys in blue) are also represented All these residues are oriented towards the 3¢ end of the core and probably establish contacts with the complementary strand It should be emphasized that the mutagenesis experiments described here do not prove that certain amino acids make base-specific contacts, especially when a new specificity at a defined position was not achieved The combination of these experiments with the footprinting results and previous

knowledge, however, are highly indicative that this is

the case Determination of the three-dimensional

struc-ture of the complex will be required to evaluate the

accuracy of the DNA–protein contacts proposed by

this model

Discussion

In this study, we investigated the interaction of the

homeodomain of the KNOX protein HAKN1 with

DNA As no structural studies on the interaction of

any KNOX protein with DNA have been reported,

ours constitutes a first approach to understand these

interactions at the molecular level Electrophoretic

mobility shift assays, footprinting analyses and missing

nucleoside experiments using different binding sites

and mutated proteins allowed us to establish a model

for HAKN1–DNA interaction This model postulates

that HAKN1 binds to a TGACNN core primarily

through interactions of certain helix III amino acids

(Ile50, Asn51, Lys54 and Arg55) with DNA This

particular combination of amino acids is present only

in KNOX proteins, indicating that they may have been selected through evolution to generate a defined specif-icity Among them, the incorporation of Ile50 and Arg55 must have been particularly important Other homeodomains that contain Ile50 and Arg55 are those

of the TGIF, Meis and Bell families [22] and yeast MATa1 [36] TGIF and Meis proteins bind the sequence TGTCA (TGACA on the complementary strand [42,43]), which is identical to that recognized by HAKN1 They possess Arg at position 54, suggesting that Lys54 in HAKN1 may not be the only means of recognizing the TGAC core Accordingly, the Bell protein ATH1, which contains Val54, also selects the sequence TGACA from a random population (I Viola, unpublished results) The presence of Val54 produces, however, a relaxed specificity at the fourth position and reduced affinity within the context of both the HAKN1 (this study) and the ATH1 homeo-domain (I Viola, unpublished results) MATa1, in turn, binds a completely different sequence (GATGT⁄ ACATC [44]), indicating that other factors apart from these residues also influence specificity The GA dinu-cleotide in GATGT is recognized by Arg55 and Asn51

of MATa1, as proposed here for the GA dinucleotide

in TGAC The GT dinucleotide is, in turn, contacted

by Met54 and Ile50 through interactions with the com-plementary strand [36] This means that, in MATa1, positions contacted by residues 55⁄ 51 and 54 ⁄ 50 are separated by one additional base pair This may be originated by the presence of Val47, which binds the nucleotide adjacent to the A recognized by Asn51 in many homeodomains The above-mentioned TGAC binding proteins (including KNOX proteins) contain Asn47, which may not establish specific contacts with DNA These differences may also originate changes in the relative orientation of DNA contacting amino acids

The model presented here can also be compared with the structures determined for the TALE proteins extradenticle and PBX1 bound to DNA [35,45] These proteins bind the sequence TGAT, with Arg55 and Asn51 establishing hydrogen bonds with the GA dinu-cleotide, as proposed here for HAKN1 The initial T is also contacted by Arg55 through van der Waals inter-actions in the PBX1 complex [45] The second T makes van der Waals contacts with Asn47 As PBX1 and extradenticle contain Ile54, this situation may resemble the binding behaviour of K54V–HAKN1, which shows relaxed specificity at this position

Our results with HAKN1 clearly support the idea that there is a general recognition code for homeo-domains Accordingly, recognition at the left side of

A

B

Fig 7 A model for the interaction of HAKN1 with DNA (A)

Dia-gram of the HAKN1 DNA binding site with the residues putatively

involved in binding each position (B) Spatial model of the

interac-tion of helix III of wild-type and mutant HAKN1 homeodomains

with DNA The model was constructed with the program SWISS

-MODEL [41] using the structures of the DNA complexes of MATa1

(1YRN), extradenticle (1B8I) and MATa2 (1APL) as templates.

Amino acids in red are those present in wild-type HAKN1 that

puta-tively contact the GAC core Residues at position 50 are: Ile in

orange, Ser in green and Lys in blue Val54 and Lys55, present in

the mutants, are shown in pink and yellow, respectively.

the conserved A that is contacted by the universally

present Asn51 is determined by the N-terminal arm

and⁄ or Arg55 The presence of Arg55 determines a G

5¢ to the conserved A, while the N-terminal arm seems

to determine a preference for A⁄ T base pairs The set

of residues present at positions 47, 50 and 54 influence

binding preferences at the right side

The putative DNA-contacting amino acids of

HAKN1 are also present in all described KNOX

pro-teins, indicating that they may all recognize identical

or similar sequences This raises the question of how

the specificity of interaction is achieved in vivo, because

different KNOX proteins have different functions

A similar paradox has been noted for animal

homeo-domains, for which current evidence suggests that

spe-cificity arises from the interaction of homeodomain

proteins with other factors that somehow influence

their DNA binding properties [46] Plant KNOX

pro-teins interact with propro-teins from the Bell family, which

also belong to the TALE superclass [47,48] and bind

similar sequences [49] Chen et al [49] have shown that

potato KNOX and Bell proteins bind two tandem

cop-ies of a TGAC motif separated by one additional

nuc-leotide As both types of homedomains seem to

establish similar contacts with their target sites

(I Viola, unpublished results), this indicates that the

respective recognition helices must lie in an antiparallel

orientation within the major groove at opposite sides

of the DNA According to the footprinting data

pre-sented here, the central nucleotide pair and the first

two pairs of the second TGAC would be contacted by

both proteins This may indicate that some

rearrange-ments may occur upon complex formation by KNOX

and Bell proteins, either before or after binding to

DNA In the complexes formed by PBX1 and

extra-denticle, the presence of Gly50, which does not contact

DNA, may allow the binding of an additional

homeo-domain in tandem immediately following TGAT

[35,43,45] The presence of Ile50, that interacts with

nucleotides located at the 3¢ side of the core, may

explain the requirement of a larger distance between

both binding sites in the complexes formed by KNOX

and Bell proteins

Sequences outside the homeodomain may also

influ-ence the binding properties of the protein Indeed, a

stretch of 16 amino acids located immediately

C-ter-minal to the homeodomain forms an a-helix that has

been shown to influence the DNA binding affinity of

the PBX1 homeodomain [50,51] As the protein used

in our assays includes a C-terminal portion, we have

analysed the structure of the region immediately

fol-lowing the HAKN1 homeodomain using several

secon-dary structure prediction programs We have only

observed a short region (five to eight amino acids depending on the program) that has a propensity to form an a-helix Therefore, we consider it unlikely that

an effect of the C-terminal tail, similar to that observed with PBX1, occurs in HAKN1 or other KNOX proteins

In summary, the results presented here constitute a framework to understand at the molecular level how KNOX proteins interact with DNA and how these interactions contribute to the establishment of active transcription complexes that influence defined develop-mental pathways within plant cells

Experimental procedures

Cloning, expression and purification of recombinant proteins

HAKN1 homeodomain and C-terminal sequences were amplified and cloned in-frame into the EcoRI and PstI sites

of the expression vector pMAL-c2 (New England Biolabs, Beverly, MA, USA) Amplifications were performed using PfuDNA polymerase and oligonucleotides MALN1: 5¢-GC GGAATTCAAAAAGAGAAAGAAAGGG-3¢ and MALC: 5¢-GGCCTGCAGCTAGAGAAGTGAAACATC-3¢ with HAKN1 cDNA [28] as the template

An I50S mutant was constructed using complementary oligonucleotides I50SF and I50SR (5¢-CAACTGGTTC AGCAACCAAAGGAA-3¢ and 5¢-TTCCTTTGGTTGC TGAACCAGTTG-3¢; introduced mutations underlined) together with primers MALC and MALN1, respectively, to amplify partially overlapping N-terminal and C-terminal HAKN1 fragments The resulting products were mixed in buffer containing 50 mm Tris⁄ HCl (pH 7.2), 10 mm MgSO4, and 0.1 mm dithiothreitol, incubated at 95
C dur-ing 5 min, and annealed by allowdur-ing the solution to cool to

24
C in approximately 1 h After this, 0.5 mm of each dNTP and 5 units of the Klenow fragment of E coli DNA polymerase I were added, and incubation was followed for

1 h at 37
C An aliquot of this reaction was used directly

to amplify the annealed fragments using primers MALN1 and MALC Mutants I50K, K54V, R55A and R55K were constructed in a similar way, using oligonucleotides I50KF (5¢-CAACTGGTTCAAAAACCAAAGGAA-3¢), I50KR TTCCTTTGGTTTTTGAACCAGTTG-3¢), K54VF TAAACCARAGGGTGCGGCAYTGGA-3¢), K54VR (5¢-TCCARTGCCGCACCCTYTGGTTTA-3¢), R55AF (5¢-CA AAGGAAGGCGCACTGGAA-3¢), R55AR (5¢-TTCCAG TGCGCCTTCCTTTG-3¢), R55KF (5¢-CAAAGGAAGAA GCACTGGAA-3¢) and R55KR (5¢-TTCCAGTGCTTCT TCCTTTG-3¢) to introduce the mutations The N-terminal arm of the MATa2 homeodomain (amino acids 1–9) was introduced into the HAKN1 homeodomain using two successive rounds of amplification with oligonucleotides