Changes in purine specificity in tandem GAF chimeras
from cyanobacterial cyaB1 adenylate cyclase and rat
phosphodiesterase 2
Ju
¨
rgen U. Linder, Sandra Bruder, Anita Schultz and Joachim E. Schultz
Abteilung Pharmazeutische Biochemie, Fakulta
¨
tfu
¨
r Chemie und Pharmazie, Universita
¨
tTu
¨
bingen, Germany
The second messengers cAMP and cGMP mediate
many intracellular functions. Swift signal modulation
requires highly regulated biosynthesis and degradation.
In mammals, the latter is accomplished by a family of
11 phosphodiesterase (PDE) isoforms which possess
similar C-terminal catalytic domains (20–45% identity
[1]). Different regulatory features are imparted by the
N-terminal domains. Five PDE families, 2, 5, 6, 10
and 11, have an N-terminal tandem GAF domain of
about 500 amino acids, i.e. two GAF domains, termed
A and B, are sequentially connected by a short linker.
The acronym GAF is derived from the proteins of
initial identification: mammalian cGMP-binding PDEs,
Anabaena adenylate cyclases (ACs; EC 4.6.1.1), and
Escherichia coli transcription factor F hlA. Meanwhile,
GAF domains have been identified in more than 3000

December 2006, accepted 12 January 2007)
doi:10.1111/j.1742-4658.2007.05700.x
The C-terminal catalytic domains of the 11 mammalian phosphodiesterase
families (PDEs) are important drug targets. Five of the 11 PDE families
contain less well-characterized N-terminal GAF domains. cGMP is the lig-
and for the GAF domains in PDEs 2, 5, 6 and 11, and cAMP is the ligand
for PDE10. Structurally related tandem GAF domains signalling via cAMP
are present in the cyanobacterial adenylate cyclases cyaB1 and cyaB2.
Because current high-resolution crystal structures of the tandem GAF
domains of PDE2 and cyaB2 do not reveal how cNMP specificity is enco-
ded, we generated chimeras between the tandem GAF domains of cyaB1
and PDE2. Both bind the ligand in the GAF B subdomains. Segmental
replacements in the highly divergent b1–b3 region of the GAF B sub-
domain of cyaB1 by the corresponding PDE2 regions switched signalling
from cAMP to cGMP. Using 10 chimeric constructs, we demonstrated
that, for this switch in purine specificity, only 11% of the sequence of the
cyanobacterial GAF B needs to be replaced by PDE2 sequences. We were
unable, however, to switch the purine specificity of the PDE2 tandem GAF
domain from cGMP to cAMP in reverse constructs, i.e. by replacement of
PDE2 segments with those from the cyaB1 GAF tandem domain. The data
provide a novel view on the structure–function relationships underlying the
purine specificity of cNMP-binding GAF domains and indicate that, as
potential drug targets, they must be characterized structurally and bio-
chemically one by one.
Abbreviations
AC, adenylate cyclase; PDE, phosphodiesterase; mPDE2, mouse PDE2; rPDE2, rat PDE2.
1514 FEBS Journal 274 (2007) 1514–1523 ª 2007 The Authors Journal compilation ª 2007 FEBS
to the mammalian PDE GAF domains. cAMP is the
ligand for these cyanobacterial tandem GAF domains,
and cAMP binding results in stimulation of the

almost identical with the mouse sequence, can
functionally replace the cyanobacterial one in the
cyaB1 AC, converting it into a cGMP-activated
enzyme [7]. Thus, the cyanobacterial AC is a powerful
tool for biochemically characterizing the GAF domains
of PDEs because, in the chimeras, cyclic nucleotides are
not concomitantly substrate and allosteric activator as
in PDE assays and product inhibition of the AC is
absent [7].
A common feature of both GAF domain structures
is that the bound cAMP or cGMP molecules are bur-
ied, i.e. less than 10% of the surface of the bound cyc-
lic nucleotide is solvent exposed [4,9]. Therefore, for
entry and exit of the compounds, the structures must
relax. The amino acids that contact the ligands in
the crystal structures as determined by ligplots
(Fig. 1C–E [4,9]) may be involved in the selectivity
filter for cAMP and cGMP, respectively. We generated
chimeras between the rPDE2 and cyaB1 tandem GAF
domains to address the question of how cyclic nucleo-
tide specificity is encoded and obtained proteins in
which exclusively cGMP instead of cAMP activates
the associated AC. Thus we defined a rather small pro-
tein segment important in ligand specification.
Results
In the structure of the mPDE2A tandem GAF
domain, 12 amino acids have been identified as inter-
acting with bound cGMP [4]. The ribose phosphate
moiety interacts with Ile453, Ala454, Tyr476, Asp480,
T487 and Glu507 (rPDE numbering). The purine moi-

Thr258, are conserved in the b1 sheet of the cyaB1
and cyaB2 GAF B subdomains. However, the b2–b3
elements of cyaB1 and cyaB2 are diverged to an extent
that precludes prediction of a binding mode for cAMP
in cyaB1.
Initially, we replaced the entire b1–b3 region, Ile250
to Ile283, of the cyaB1 GAF B subdomain with
Asn411 to Tyr443 of GAF B from rPDE2A (Fig. 1B,
construct I in Table 1). The purified recombinant pro-
tein was stimulated 10.5-fold by cGMP with an EC
50
of 360 lm, similar to wild-type cyaB1, and, surpris-
ingly, activation by cAMP was eliminated (Fig. 2A,
Table 1). Obviously, exchange of this region switched
the specificity of the cyaB1 tandem GAF domain from
cAMP to cGMP, confirming the above considerations.
Next we swapped the anterior part, Ile250 to Gly267,
in cyaB1 GAF B with Asn411 to Asn426 of rPDE2A
(construct II in Table 1). Construct II was stimulated
J. U. Linder et al. Cyclic nucleotide specificity in GAF domains
FEBS Journal 274 (2007) 1514–1523 ª 2007 The Authors Journal compilation ª 2007 FEBS 1515
by neither cAMP nor cGMP up to 10 mm (Fig. 2A).
We were unable to determine whether cNMP access
and binding was abrogated or whether intramolecular
signalling was impaired because of incorrect folding of
the GAF ensemble in spite of a robust basal activity
of the C-terminal AC. Swapping the C-terminal W270
to I283 section of cyaB1 for Val429 to Tyr443 from
rPDE2a (construct III in Table 1) yielded a protein
that was highly stimulated by both cAMP and cGMP,

identify the amino acids that impeded cAMP stimula-
CyaB2-GAFB-cAMP
rPDE2A-GAF-B-cGMP CyaB2-GAF-A-cAMP CyaB2-GAF-B-cAMP
A B
C D E
Fig. 1. Structural properties of the b1–b3 sheets and the interactions with cyclic nucleotide monophosphates. (A) GAF B–cAMP complex of
cyaB2. The regions that correspond to the parts of cyaB1 exchanged in constructs II and III are shown in blue (A283-E303) and red (L304-
P321). (B) Alignment of the respective sequence segments. (C) Interaction sites of cGMP in the structure of mPDE2 (amino acid numbering
corresponds to rPDE2). (D) Interaction sites for cAMP in the structure of the cyaB2 GAF A domain. (E) Interaction sites for cAMP in the
structure of the cyaB2 GAF B domain. Amino acids interacting with the purine moiety are colour-coded for clarity in (B–E). Hydrophobic inter-
actions are depicted by grey ovals in (C)E).
Cyclic nucleotide specificity in GAF domains J. U. Linder et al.
1516 FEBS Journal 274 (2007) 1514–1523 ª 2007 The Authors Journal compilation ª 2007 FEBS
Table 1. Stimulation of cyaB1-rPDE2 chimeras by cAMP and cGMP. Assays were performed with 75 lM ATP as described in Experimental procedures. cyaB1-WT, Wild-type cyaB1 AC
holoenzyme; rPDE2-cyaB1, chimera in which the cyaB1 AC tandem GAF domains (residues 50–385) are replaced by the rPDE2 tandem GAF domains. Sequences in bold indicate residues
from rPDE2. N ¼ 2–8; usually recombinant protein from two expressions were used. Basal activity is expressed as nmol cAMPÆmg
)1
Æmin
)1
.
Construct
EC
50
(cAMP)
(l
M)
EC
50
(cGMP)
(l

AR NLSNAEICSVFLLDQ N EL WTKVAAAADTTQ-LI EIRIP
Construct III 3.3 ± 0.3 3.5 ± 0.3 0.94 363 ± 21 11.3 ± 0.2 12.7 ± 0.5 AR ILMQADRSTLFLYRKEMG EL VAKVFDGGVVDDESY EIRI
Construct IV 8.2 ± 0.4 13.7 ± 1.0 0.60 7.2 ± 0.1 10.9 ± 0.3 12.8 ± 0.5 AR ILMQADRSTLFLYRQ N EL VAKVFDGGVVDDESY EIRIP
Construct V 5.5 ± 0.5 6.8 ± 0.2 0.81 56.5 ± 7.3 15.4 ± 5.4 18.1 ± 6.3 AR ILMQADRSTLFLYRQEMN EL VAKVFDGGVVDDESY EIRIP
Construct VI 7.3 ± 2.4 19.9 ± 0.8 0.37 24.3 ± 3.3 46.1 ± 7.2 54.2 ± 6.6 AR ILMQADRSTLFLLDQ N EL VAKVFDGGVVDDESY EIRIP
Construct VII 78.9 ± 30.9 9.9 ± 0.3 7.97 7.8 ± 0.7 35.6 ± 2.3 394 ± 30 AR ILMQADRSSVFLLDQ N EL VAKVFDGGVVDDESY EIRIP
Construct VIII 56.9 ± 32.6 4.6 ± 0.6 12.4 9.9 ± 1.0 27.1 ± 4.8 96.4 ± 5.3 AR ILMQADRSSVFLYRKEMG EL VAKVFDGGVVDDESY EIRIP
Construct IX 1.4 ± 0.1 0.04 ± 0.01 35.0 79.7 ± 7.4 5.4 ± 0.3 7.1 ± 0.4 AR ILMQADRSSLFLYRKEMG EL VAKVFDGGVVDDESY EIRIP
Construct X 72.8 ± 21.8 814 ± 122 0.09 3.8 ± 0.2 34.3 ± 6.7 38.5 ± 7.6 AR ILMQADRSTVFLYRKEMG EL VAKVFDGGVVDDESY EIRIP
Construct XI 11.7 ± 0.5 > 3000 < 0.01 1.6 ± 0.1 48.9 ± 2.5 7.9
b
AR ILMQADRSSVFLYRKEMG EL WTKVAAAADTTQ-LI EIRIP
rPDE2-cyaB1
a
(> 3000) 3 NA 2.4 2
b
10
409
AR NLSNAEICSVFLLDQ N EL VAKVFDGGVVDDESY EIRIP
448
Construct XII (> 3000) (> 3000) NA 0.6 ± 0.1 1.1
b
1.3
b
AR ILMQADRSTLFLYRKEMG EL WTKVAAAADTTQ-LI EIRIP
Construct XIII (> 3000) (> 3000) NA 0.6 ± 0.1 1.4
b
1.2
b
AR ILMQADRSTLFLYRKEMG EL VAKVFDGGVVDDESY EIRIP

Next we examined whether a T258S ⁄ L259V double
mutation in the background of construct III was suffi-
cient to change cNMP specificity (Table 1, con-
structs VIII–XI). Corresponding assays demonstrated
that this was the case (Table 1, Fig. 3B). Properties of
constructs VII and VIII were comparable, i.e. cGMP
was 12-fold more potent in stimulating the associated
AC activity than cAMP (Table 1). A notable difference
between the exchange of the N-terminal region Ile250
to Gly267 (construct I) and the T258 ⁄ L259 couple
(construct VIII) was the complete abrogation of cAMP
stimulation in the former versus the remaining cAMP
response in the latter (compare Fig. 2A and Fig. 3B).
However, the loss of cAMP stimulation in construct I
was achieved at the expense of reduced efficacy for
cGMP (compare constructs I and VIII in Table 1).
Next, we asked whether the switch in cNMP specificity
was attributable to either T258S or L259V (con-
structs IX and X, Fig. 3C, Table 1). In construct IX
(T258S) the preference for cGMP over cAMP was
35-fold (Table 1), and the affinity for both cyclic nuc-
leotides was enhanced by almost two orders of mag-
nitude compared with construct VIII (Table 1). The
EC
50
for cGMP was 40 nm and that for cAMP
1.4 lm. Therefore construct IX differed from con-
struct I and VII by an increase in cGMP sensitivity
without affecting the EC
50

undegraded proteins.
Cyclic nucleotide specificity in GAF domains J. U. Linder et al.
1518 FEBS Journal 274 (2007) 1514–1523 ª 2007 The Authors Journal compilation ª 2007 FEBS
important contributions to functionally important
features.
Next we examined whether the effect of the double
mutation T258S ⁄ L259V would also materialize in the
cyaB1 AC parent protein, i.e. a reduction in cAMP
efficacy and a persistent discrimination against cGMP.
In the T258S ⁄ L259V-cyaB1 AC (construct XI, Fig. 4,
Table 1), the EC
50
for cAMP was 12 lm and maximal
stimulation was about 50-fold. The response to cGMP
was greatly diminished (Fig. 4). The data corresponded
to the high preference for cAMP over cGMP of the
parent cyaB1 protein [7], with a 10-fold increased EC
50
for cAMP. Further the specific activities of basal and
1mm cAMP-stimulated T258S ⁄ L259V-cyaB1 AC were
reduced approximately 45-fold (1.4 and 84 (basal),
67.5 and 2244 nmol cAMPÆmg
)1
Æmin
)1
(activated with
1mm cAMP)). We conclude that to some extent
Thr258 ⁄ Leu259 in cyaB1 govern cAMP signalling;
however, they do not alone define purine specificity of
the regulatory domain. Only together with other regio-

)1
Æmin
)1
.
Addition of cAMP or cGMP up to 10 mm slightly and
dose-dependently activated AC activity (maximal acti-
vation 1.4-fold, Table 1). Clearly, we could not repeat
the regional exchange of the b1–b3 region with the
rPDE2 tandem GAF domain serving as the receiver
domain. Apart from the problem of poor expression
and perhaps partial misfolding of constructs XII–XIV,
it is obvious that the data with the cyaB1 GAF
domain cannot be generalized, i.e. a free exchange of
corresponding GAF domain regions between different
cNMP-binding tandem GAF domains is obviously
impossible. This means that these GAF domains, in
spite of the high similarity between them, have prob-
ably retained considerable individuality in ligand
binding and intramolecular signalling.
Discussion
So far, the structures of two tandem GAF domains
are known. The mPDE2 one contains two cGMP moi-
eties bound to each GAF B in a parallel dimer [4]. The
other structurally elucidated tandem GAF domain is
from the cyanobacterial AC cyaB2 [9]. It signals via
cAMP which binds to GAF A and ⁄ or GAF B. Hence,
the structure contains four cAMP molecules in an anti-
parallel dimer. For construction of chimeras, we used
a rat PDE2, which has an amino-acid sequence identi-
cal with the mouse isoform in this region, and the

GAF A, a F205A mutation (Phe433 in rPDE2) abro-
gated cGMP binding [13], and, in PDE6, both of the
amino acids corresponding to Phe433 ⁄ Asp434 in
rPDE2A are required for cGMP binding [14].
Our initial mutations in cyaB1 GAF B also involved
individual replacement of the residues positionally cor-
responding to Phe433 and Asp434 in rPDE2. Neither an
A275D mutation [7] nor a
274
AAA
276
to FDG triple
exchange affected cAMP specificity of the cyaB1 tandem
GAF domain (T. Kanacher, unpublished results). Nei-
ther was cAMP stimulation lost nor cGMP activation
gained.
While these experiments were in progress, the crystal
structure of the cAMP-binding GAF tandem domain
of cyaB2 became available [9]. It showed that the
backbone conformations of the b1–b3 regions are par-
ticularly divergent, as is the primary sequence. This
was not surprising because the corresponding segments
A250 to I283 of cyaB1 and N411 to Y443 in the
rPDE2 tandem GAF domains are only 24% identical.
Therefore, we hypothesized that regional chimeras
instead of point mutants might yield clues about which
structural elements are involved in coding purine spe-
cificity. The expectation of this hypothesis was fulfilled
by construct I. A cyaB1 tandem GAF domain with
replacement of this segment by the rPDE2 region,

between members of this tandem GAF domain family.
The structural differences affect coding for purine spe-
cificity and possibly also signalling to the associated
catalytic output. Furthermore, considering that cAMP
in the cyaB2 and cGMP in the mPDE2 GAF tandem
domain structures are less than 10% exposed to sol-
vent, amino acids additional to those identified as
binding to and interacting with cNMPs in the crystal
structures may be required. For cNMPs to access the
binding pocket, the structures undoubtedly must open
up. Therefore, a structure without bound ligand would
be helpful in this context to elucidate the structural
transitions upon cNMP binding.
When one compares the sequences of the cAMP-
binding GAF tandem domain of PDE10, the PDE
from Trypanosoma brucei [15] and cyaB1 and cyaB2
ACs with the cGMP-binding tandem GAF domains
from mammalian PDEs in conjunction with the avail-
able biochemical and structural data, it is impossible
to predict which amino acids or sequences encode pur-
ine specificity in each case. Our data contribute to the
complex structure–function relationships underlying
purine specificity of cNMP binding GAF domains. In
addition, they highlight that, in order to potentially
use the tandem GAF domains as potential drug tar-
gets, we must characterize GAF domains structurally
and biochemically one by one.
Experimental procedures
Recombinant DNAs
The cyaB1 holoenzyme and a chimera consisting of the

N-terminal rPDE2A tandem GAF domain, which carried
corresponding inserts of the cyaB1 tandem GAF domain
(constructs XII–XIV) were generated by introduction of a
silent BssHII at A409 ⁄ R410 (numbering of rPDE2A), a SacI
site at E427 ⁄ L428, and an EcoRI restriction site at R446 ⁄
I447 ⁄ P448, and employing the same cloning strategy as for
constructs I–III.
Constructs I–XI were expressed from pQE30. For expres-
sion of the inverse chimeras (constructs XII to XIV), a
pET16b vector with the multiple cloning site of pQE30
(bp267–431 of pET16b were replaced by bp199–90 of
pQE30) was used retaining the N-terminal MRGSH
6
GS
tag. The fidelity of all constructs was verified by double-
stranded sequencing. Primer sequences and additional clo-
ning details are available on request.
Expression and purification of bacterially
expressed proteins
Chimeric constructs based on the cyanobacterial GAF tan-
dem domain were transformed into E. coli BL21(DE3)-
[pRep4] and those based on the rPDE2 GAF tandem
domain into E. coli BL21(DE3)[pLysS]. Cultures were
grown in Lennox L broth at 30 °C containing 100 mgÆL
)1
ampicillin and 50 mgÆL
)1
kanamycin. Expression was
induced at an A
600

35% glycerol) and stored at )20 °C.
Purity of recombinant proteins
The purity and integrity of expression products was exam-
ined by SDS ⁄ PAGE and western blotting. For western
blots, proteins were blotted on to poly(vinylidene difluo-
ride) membranes and sequentially probed with an antibody
to RGSH
4
(Qiagen) and with a peroxidase-conjugated goat
anti-(mouse IgG-F
c
) secondary antibody (Dianova, Ham-
burg, Germany). Detection was carried out with the ECL
Plus kit (GE Health Care, Munich, Germany). Western
blots are included in the figures to demonstrate the absence
of proteolytic degradation products. Because fragments that
lost the His tag do not bind to the Ni
2+
resin used for
purification, they do not constitute a purity problem.
AC assay
Enzyme activity was assayed for 4 min (cyaB1-based
constructs) or 10 min (rPDE2a-GAF-based constructs) at
37 °C in 100 lL containing 22% glycerol, 50 mm Tris ⁄ HCl,
pH 7.5, 10 mm MgCl
2
,10lg BSA and 75 lm [a-
32
P]ATP
(25 kBq). [2,8-

ase 11. J Biol Chem 281, 2841–2846.
6 McAllister-Lucas LM, Haik TL, Colbran JL, Sonnen-
burg WK, Seger D, Turko IV, Beavo JA, Francis SH &
Corbin JD (1995) An essential aspartic acid at each of
two allosteric cGMP-binding sites of a cGMP-specific
phosphodiesterase. J Biol Chem 270, 30671–30679.
7 Kanacher T, Schultz A, Linder JU & Schultz JE (2002)
A GAF-domain-regulated adenylyl cyclase from Ana-
baena is a self-activating cAMP switch. EMBO J 21,
3672–3680.
8 Bruder S, Linder JU, Martinez SE, Zheng N, Beavo JA
& Schultz JE (2005) The cyanobacterial tandem GAF
domains from the cyaB2 adenylyl cyclase signal via both
cAMP-binding sites. Proc Natl Acad Sci USA 102,
3088–3092.
9 Martinez SE, Bruder S, Schultz A, Zheng N, Schultz
JE, Beavo JA & Linder JU (2005) Crystal structure of
the tandem GAF domains from a cyanobacterial adeny-
lyl cyclase: modes of ligand binding and dimerization.
Proc Natl Acad Sci USA 102, 3082–3087.
10 Zoraghi R, Corbin JD & Francis SH (2004) Properties
and functions of GAF domains in cyclic nucleotide
phosphodiesterases and other proteins. Mol Pharmacol
65, 267–278.
11 Bruder S, Schultz A & Schultz JE (2006) Characteri-
zation of the tandem GAF domain of human phos-
phodiesterase 5 using a cyanobacterial adenylyl cyclase
as a reporter enzyme. J Biol Chem 281, 19969–19976.
12 Wu AY, Tang XB, Martinez SE, Ikeda K & Beavo JA
(2004) Molecular determinants for cyclic nucleotide