`
`Current Medicinal Chemistry, 2008, 15, 1570-1585
`
`
`
`Structure –Activity Relationships of PDE5 Inhibitors
`
`D. Er(cid:5)s3, Cs. Szántai-Kis1,3, R. Kiss4, Gy. Kéri1,2,3, B. Hegymegi-Barakonyi1,3, I. Kövesdi3 and L. (cid:4)rfi*,1,3,4
`
`1Rational Drug-Design Laboratory Cooperative Research Centre, Semmelweis University, 1367 Budapest 5, POB131
`2Semmelweis University, Department of Medical Chemistry, Pathobiochemistry Research Group of the Hungarian Academy of Sci-
`ences, Puskin u. 9., Budapest, 1088, Hungary
`3Vichem Chemie Ltd., Herman Ottó u. 15., Budapest, 1022, Hungary
`4Semmelweis University, Department of Pharmaceutical Chemistry, H(cid:3)gyes E. u. 9., Budapest, 1092, Hungary
`
`Abstract: cGMP has a short-term effect on smooth muscle tone and a longer-term effect on responses to chronic drug treatment or pro-
`liferative signals. cGMP-Phosphodiesterase type 5 (PDE5) hydrolizes cGMP, and the result is smooth muscle contraction. PDE5 is a rela-
`tively novel therapeutic target of various diseases, such as erectile dysfunction and pulmonary hypertension. The most intensively exam-
`ined and marketed PDE5 inhibitor was sildenafil (Viagra) but recently vardenafil (Levitra) and tadalafil (Cialis) were launched with
`beneficial ADME parameters and PDE5 selectivity. The increasing interest in PDE5 inhibition made it reasonable to collect the available
`inhibitory data from the scientific literature and set up a structure-activity relationship study. Chemical structures of 438 compounds and
`their cGMP-PDE5 inhibitory data (IC50) were collected from recently published articles. In this paper physiology, regulation and inhibi-
`tion of PDE5 (and briefly other PDE-s) are discussed and inhibitors are tabulated by the core structures. Finally, a general QSAR model
`built from these data is presented. All data used in the QSAR study were summarized in a Supplement (for description please see the on-
`line version of the article).
`
`Keywords: PDE5, QSAR, virtual sceening.
`
`INTRODUCTION
`
`Phosphodiesterase 5 (PDE5) is a relatively new therapeutic tar-
`get in the treatment of erectile disfunction (ED), and there are many
`other disorders in which PDE5 inhibiton might have therapeutic
`value. Phosphodiesterases catalyze the degradation of cXMPs to
`XMPs, thus inhibition of PDEs will result in the prolongation of
`vasodilatator effects. The first PDE5 inhibitor, sildenafil (Viagra,
`Pfizer Pharmaceuticals) [1] was launched in 1998 and it revolution-
`ized ED management and market. In 1997 the ED market totalled
`US $137 million and increased to US $1.8 billion in 2003 [2]. Since
`then two other drugs were introduced: tadalafil (Cialis, Lilly ICOS
`LLC) [3] and vardenafil (Levitra, Bayer and GlaxoSmithKline) [4].
`The latter two drugs differ from sildenafil in their ADME parame-
`ters and PDE5 selectivity.
`
`In this paper we summarize up the recently published results of
`PDE research, especially of PDE5. Physiology, regulation and inhi-
`bition of PDE5 are discussed and the inhibitors collected from the
`literature are tabulated. Finally, a general QSAR model built from
`these data is described. The generalization ability of this model is
`high since many different cores were involved in the model build-
`ing (the experimental data spans a range of greater than 5 orders of
`magnitude – the values range from pIC50 = 4.59 – 9.7) and the Q2
`value of the external validation is acceptable (0.69), enabling us to
`use this model for in silico screening and forecasting PDE5 inhibi-
`tory side effects of new drug candidates.
`
`THE PDE SYSTEM
`
`Different factors such as nitric oxide (NO), atrial natriuretic
`peptide (ANP) and other endogenous vasodilatators stimulate
`adenylyl- and guanylyl cyclases, which transform ATP and GTP
`into cAMP and cGMP, respectively in smooth muscle cells
`(SMCs). NO is produced in vascular endothel cells by nitric oxide
`synthases, and as NO is a small molecule it can easily cross the cell
`membranes by diffusion and get into the SMCs. ANP is released
`into the circulation from the atria in response to hypervolaemia-
`induced stretch and acts as a hormone [5]. While NO activates the
`soluble NO-stimulated guanylyl cyclase (NOS-GC) [6], the other
`factors activate the membrane-bound cyclases [7].
`
`
`*Address correspondence to this author at the Semmelweis University, Department of
`Pharmaceutical Chemistry, H(cid:5)gyes E. u. 9., Budapest, 1092, Hungary; Tel: +36-20-
`825-9625; Fax: +361-217-0891; E-mail: orlasz@gytk.sote.hu
`
`
`
`0929-8673/08 $55.00+.00
`
`cAMP and cGMP are second messengers and relatively simple
`molecules. However, their physiological effects are very diverse
`and cell specific. There are many different cyclases and PDEs
`which can be expressed in cells which explains how the effects of
`these simple molecules are cell specific. About 10 different
`adenylyl cyclase genes and about 20 different PDE genes with var-
`ied regulation, physiological properties, etc. have been identified in
`mammalian species. A human cell can express 1-2 cyclases and 3-4
`PDEs, which means that the number of possible combinations is
`very large [8].
`
`The synthesized cXMPs activate complex pathways generating
`different biological effects. cXPMs bind to protein kinases (cAMP
`binds to the cAMP-dependent protein kinase (PKA) and cGMP
`binds to the cGMP-dependent protein kinase (PKG)) activating
`them. They also interact with ion-channels and PDEs. As a result,
`the intracellular calcium ion concentration decreases and the activ-
`ity of myosin phosphatase increases, therefore the sensitivity to
`calcium ions decreases. The short-term output is SMC relaxation
`(vasodilatation) and there are also longer-term responses. The first
`is that constant cyclase stimulation increases the expession of PDEs
`leading to be less responsive to cXMP. This is the major cause of
`the tolerance to NO-releasing drugs. In human cells only cXPMs
`induce PDE1C and its inhibition leads to suppression of SMC pro-
`liferation [8].
`
`The role of PDEs in this signal transduction process is hydro-
`lyzing the accumulated cXMPs to XMPs, thus PDEs decrease the
`signal. To date we know 11 PDE gene families [8]. Brief informa-
`tion on PDE families are summarized in Table 1 [8, 9].
`
`A highly conserved catalytic region is located in the C-terminal
`part of PDEs. According PDEs share an average of about 30% se-
`quence homology in the catalytic domain [9]. We confirmed this
`information by collecting primary sequences of four PDEs from the
`Brookhaven Protein Data Bank (PDB) [10] and calculating their
`sequence homology. The result is shown in Table 2. PDE5 differs
`most from the other examined PDEs.
`
`The N-terminal part where the regulation domain(s) are located
`in PDEs is pretty variable. From the above comparison of the pri-
`mary sequence of the catalytic domain of PDEs it can be recognized
`that the enzymes catalyze the same reactions and that they differ in
`the mode of their regulation.
`
`© 2008 Bentham Science Publishers Ltd.
`
`ATI 1008-0001
`
`ATI v. ICOS
`IPR2018-01183
`
`
`
`Structure –Activity Relationships of PDE5 Inhibitors
`
`Table 1.
`
`
`Human PDE Families
`
`Family
`
`Gene(s)
`
`N-terminal
`
`regulation domain(s)
`
`PDE1
`
`A, B, C
`
`PDE2
`
`A
`
`PDE3
`
`A, B
`
`CBD* (2)
`
`GAF# (2)
`
`PMAD%
`
`Current Medicinal Chemistry, 2008 Vol. 15, No. 16 1571
`
`Substrate
`
`Inhibitor(s)
`
`Target disease
`
`cAMP, cGMP
`
`Vinpocetine
`
`Urge incontinence, low compliance bladder, acute ischemic
`stroke
`
`cAMP, cGMP
`
`
`
`
`
`cAMP, cGMP
`
`
`
`Olprinone
`
`PDE4
`
`A, B, C, D
`
`UCR domain (2)
`
`cAMP
`
`PDE5
`
`A
`
`GAF (2)
`
`cGMP
`
`PDE6
`
`PDE7
`
`PDE8
`
`A, B, C
`
`A, B
`
`A, B
`
`GAF (2)
`
`-
`
`PAS domain
`
`cGMP
`
`cAMP
`
`cAMP
`
`Cilostazol
`
`Cilomilast
`
`Roflumilast
`
`Sildenafil
`
`Vardenafil
`
`Tadalafil
`
`Exisulind
`
`CP461
`
`
`
`
`
`
`
`Gastric intramuscular acidosis, systemic infalmmation after
`cardiopulmonary bypass
`
`Angiographic restenosis, intermittent claudication
`
`Asthma, chronic obstructive
`pulmonary disease, allergic rhinitis
`
`Pulmonary hypertension, female sexual dysfunction
`
`Erectile dysfunction
`
`
`
`Various cancers
`
`
`
`
`
`
`
`
`
`PDE9
`
`PDE10
`
`PDE11
`
`A
`
`A
`
`A
`
`-
`
`GAF (2)
`
`GAF (2)
`
`cGMB
`
`cAMP, cGMP
`
`cAMP, cGMP
`
`
`
`
`
`
`
`
`
`
`
`*Ca/Calmodulin binding domain.
`#GAF domain, binds cGMP and/or other proteins.
`%Putative membrane-association domain.
`
`
`
`Table 2.
`
`
`
`Sequence Homology (%) of the Catalytic Domain of Four
`PDEs
`
`
`
`PDE1B
`
`PDE3B
`
`PDE4B
`
`PDE5A
`
`PDE1B
`
`PDE3B
`
`PDE4B
`
`PDE5A
`
`100
`
`33.5
`
`36.6
`
`20.5
`
`
`
`100
`
`31.1
`
`16.5
`
`
`
`
`
`100
`
`18.5
`
`
`
`
`
`
`
`100
`
`PDE5
`
`As Table 1 indicates, the substrate molecule of PDE5 is cGMP.
`Furthermore, PDE5 is the major cGMP hydrolyzing PDE enzyme
`[8]. Three isoforms of PDE5 are known, A1, A2 and A3. The iso-
`forms vary in their N-terminal regulatory domain only, and the A1
`isoform is predominant. The A3 form is SMC specific, while the
`other two do not show cell specificity [11].
`
`PDE5 can be isolated from a large variety of tissues, for exam-
`ple it can be found in corpus cavernosum [12], platelets [13], lung
`[14], brain [15], kidneys [16], vaginal tissues [17], spleen, endothe-
`lial cells [18], Purkinje neurons [19], cerebellum, retina, thymus,
`heart, liver, esophagus, stomach, pancreas, small intestine, colon,
`prostate and urethra [9].
`
`As Table 1 shows, the N-terminal part of PDE5 contains two,
`so called GAF domains, GAF A and GAF B. Originally, GAF do-
`main was found to be present in cGMP-regulated phosphodi-
`esterases, adenyl cyclases and bacterial transcription factor called
`FhlA. The initial letters of these proteins gave the name to the GAF
`domain [20]. The GAF domain can bind cGMP, which directly
`activates the enzyme.
`
`PKG can phosphorylate PDE5 on serine 92 if the GAF domain
`binds cGMP. This phosphorylation does not directly enhance the
`
`activity of PDE5, but it increases the apparent affinity of PDE5 for
`cGMP binding [21]. It is suspected that the role of the phosphoryla-
`tion is stabilizing PDE5 in its cGMP-bound, active state. The exis-
`tence of two states (inactive and active) of PDE5 explains why the
`inhibitory effect of inhibitors depends on the cGMP concentration -
`the two states have different affinity for the same inhibitor [8]. The
`mechanism of action and regulation of PDE5 is depicted in Fig. (1).
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Fig. (1). The effects and regulation of PDE5.
`
`NO: nitric oxide, NOS: NO-stimulated Guanylyl Cyclase, MB-GC: Mem-
`brane-bound Gunaylyl Cyclase, ANP: Atrial Natriuretic Peptide. The dotted
`arrow represents the phosphorylation of PDE5 on Ser-92 by PKG.
`
`INHIBITION OF PDE5
`
`The main effects of PDE5 inhibition are cardiovascular. Silde-
`nafil was originally tested as an antianginal agent [22], but it
`
`ATI 1008-0002
`
`
`
`1572 Current Medicinal Chemistry, 2008 Vol. 15, No. 16
`
`Er(cid:3)s et al.
`
`quickly turned out that it only slightly decreases the systemic blood
`pressure [23]. It has also been demonstrated that administration of a
`PDE5 inhibitor in a therapeutic dose does not affect the electrocar-
`diogram [24] and does not have any inotropic effect [25].
`
`Pulmonary artery pressure, on the other hand, is significantly
`decreased by sildenafil suggesting that PDE5 inhibitors might be
`used for the treatment of pulmonary hypertension [14, 26].
`
`The effective PDE5 inhibitor is a possible therapeutic agent for
`the treatment of congestive heart failure [18]. In healthy heart, sild-
`enafil moderately caused relaxing effects on the coronary blood
`vessels. In contrast, this effect is significantly increased in the pres-
`ence of myocardial ischaemia [27].
`
`Admittedly the most important cardiovascular effect of PDE5
`inhibitors takes place in the corpus cavernosum. ED is the prevail-
`ing yet undertreated state in which the patients have difficulties in
`producing and/or maintaining erection.
`
`The risk factors of ED (hypertension, diabetes, smoking, lipid
`abnormality, obesity and lack of physical activity) are common in
`those with coronary artery disease, 75 % of men with chronic coro-
`nary artery disease also suffer from ED [28].
`
`Before developing Viagra, the treatment of ED was much less
`effective and it consisted of injection therapies, prosthetic implants,
`vacuum devices and less effective oral agents [2]. The introduction
`the PDE5 inhibitors will enhance the treatment of ED, which entails
`the improvement of mood and reduction of depression [28].
`
`The lack of sexual activity decreases testosteron concentration
`in blood [29]. Testosteron levels increase with PDE5 inhibition.
`
`Another sexual problem which can be related to PDE5 inhibi-
`tion is rapid ejaculation [30].
`
`As PDE5 is also present in the female genitals, it is supposed
`that it has an important role in the female sexual response as well
`[17].
`
`In kidney sildenafil stimulates renin secretion, this may explain
`why PDE5 inhibitors have only a moderate effect on the blood
`pressure [16].
`
`In the central nerveous system sildenafil activates the serotonin
`transformer (SERT) which stimulates serotonin uptake [31]. This
`can be major source of the CNS side effects of sildenafil.
`
`It has been shown that sildenafil induces neurogenesis and
`promotes functional recovery after stroke in rats [15].
`
`There are other diseases in which PDE5 inhibitors might be
`useful such as in respiratory system diseases (sildenafil has also
`antiinflammatory effects) [32] and benign prostatic hyperplasia
`associated urinary dysfunction [9]. PDE5 inhibitors may be poten-
`tial therapeutical agents in the management of preeclampsia, which
`is the major cause of perinatal and maternal morbidity [33]. In rat
`experiments zaprinast, which is another PDE5 inhibitor, has been
`shown to promote recovery from ischemic acute renal failure [9].
`
`PDE5 INHIBITORS AS DRUGS
`
`All clinically useful PDE5 inhibitor drugs have nanomolar IC50
`against PDE5. The agents differ in their selectivity towards other
`phosphodiesterases and in pharmacokinetics.
`
`The first PDE5 inhibitor on the market for the treatment of ED
`was sildenafil [1]. For example, it has been shown that Viagra, and
`any other PDE5 inhibitor drug, do not block the HERG channel
`[34], do not induce apoptosis [35] and do not contribute to the de-
`velopment of myocardial infarction or ischemia [23]. Despite of the
`success of Viagra there are a number of problems which inspired
`the research for further PDE5 inhibitory drugs.
`
`Research resulted in two more PDE5 inhibitor drugs, vardenafil
`(Levitra, Bayer and GlaxoSmithKline) [4] and tadalafil (Cialis,
`
`Lilly ICOS LLC) [36]. The selectivity of these drugs are better than
`that of sildenafil, but the side effect profile of them is similar to
`sildenafil, except the vision-related side effects. Contraindication
`for new drugs are similar to sildenafil, but vardenafil and tadalafil
`certainly have beneficial properties.
`
`Vardenafil is a potent PDE5 inhibitor (IC50 = 0.7 nM) and its
`effect develops quickly [37]. This increases its efficacy and reliabil-
`ity even in the case of the difficult to treat patient groups (diabetics,
`patients with severe ED, prostatectomized men, etc) [2]. Fatty foods
`do not influence the relative bioavailability, but delays the absorp-
`tion of vardenafil. An important warning concerning the administra-
`tion of vardenafil is that CYP3A4 inhibitors, for example ritonavir,
`can affect the hepatic metabolism of vardenafil [37]. Vardenafil is
`more selective than sildenafil, but it blocks PDE6. The two drugs
`have very similar chemical structure: the main difference is the
`position of one nitrogen atom in the heterocyclic core which causes
`the different selectivity of the compounds. The extra methylene
`group in Vardenafil do not interfere with binding, but can enhance
`the permeability.
`
`The effect of tadalafil lasts the longest of these three drugs, up
`to 36 hours (the effect of sildenafil and vardenafil lasts for 4-8
`hours). Its effect is not influenced by food, and develops very
`quickly (in 16-17 minutes) [2]. Tadalafil is a selective PDE5 inhibi-
`tor, but it also inhibits PDE11, an enzyme with unknown physio-
`logical functions [38].
`
`General side effects of the three drugs are vasodilatation (due to
`PDE1 inhibition), vision related disturbance (due to PDE6 inhibi-
`tion), increased heart rate and inhibition of platelet aggregation (due
`to PDE3 inhibition) [38], nasal congestion [37], headache, flushing,
`dyspepsia. Although these side effects are tolerable, a chronic use
`of a PDE5 inhibitor can enhance them [39]. Sildenafil can cause
`migrain [40] and it may have certain central side effects associated
`with its SERT stimulation such as dizziness, depression, insomnia,
`abnormal dreams, anxiety and aggressive behavior [31]. The ration-
`ally designed second generation PDE5 inhibitors will lack these
`side effects [41].
`
`Coadministration of PDE5 inhibitors with nitrates or NO gener-
`ating molecules is contraindicated because it causes significant
`vasodilatation and reduction of blood pressure. Alpha-adrenergic
`blockers can be administered with precaution with sildenafil and
`tadalafil, but are contraindicated with vardenafil [23].
`
`PDE5 INHIBITORS
`
`All PDE5 inhibitors published so far act by binding to the ac-
`tive site of PDE5. Theoretically it might be possible to find a new
`group of PDE5 inhibitors binding to the GAF domain. This idea
`could lead to the discovery of more selective PDE5 inhibitors [8].
`
`In Table 3 we listed the PDE5 inhibitory compound types found
`in the literature by June of 2007 (note that not all compounds were
`used in the QSAR modeling).
`
`The X-ray structure of PDE5A co-crystallized with GMP
`(1t9s.pdb [88]), sildenafil (1tbf.pdb [88]), tadalafil (1xoz.pdb [89])
`and vardenafil (1xp0.pdb [89]) has unraveled the binding mode of
`the inhibitors. Fig. (2) shows the four superimposed structures, we
`visualized the superimposition with Sybyl 7.0 software [90].
`
`QSAR MODELING
`
`The graphical 2D database of PDE5 inhibitors containing struc-
`tures of 438 molecules with their IC50 data was built in ISISBase
`software [91]. We have not found a QSAR study of PDE5 inhibiton
`from such a large dataset of diverse structures in the literature. The
`negative logarithm of IC50, i.e., the pIC50 values were used for
`modeling. 3D structures of the molecules were calculated by the
`CONCORD algorithm of the Tripos software package [92]. The
`
`ATI 1008-0003
`
`
`
`Structure –Activity Relationships of PDE5 Inhibitors
`
`Current Medicinal Chemistry, 2008 Vol. 15, No. 16 1573
`
`Table 3.
`
`
`The Structures of Published PDE5 Inhibitors
`
`Compound group
`
`IC50 (nM)
`
`Structure
`
`Ref.
`
`O
`
`R1
`
`N
`
`N
`
`N
`
`O
`
`R1
`
`N
`
`R3
`
`N
`
`N
`
`R2
`
`N
`
`N
`
`R2
`
`N
`
`N
`
`R4
`
`R3
`
`1
`
`2
`
`Polycyclic guanines
`
`0.3 -16000
`
`[42, 43]
`
`R1: CH3, C2H5, i-Pr
`
`R2: H, alkyl, (subst.)-aryl-CH3, -CH2CH2OH
`
`R3: H, alkyl, (subst.)-aryl-alkyl,alkyl-oxy, CH2OCH3,
`SO2C2H5,(subst.)-amino-alkyl, CF3, SC2H5, CN,
`COOCH3, COONH2, COONHCH3, NH2, triazino
`
`R4: H, CH2-Phe
`
`
`
`O
`
`N
`
`3
`
`R3
`
`R1
`
`N
`
`N
`
`R4
`
`R2
`
`R1
`
`R2
`
`N
`
`N
`
`R3
`
`R4
`
`R5
`
`[18]
`
`[44]
`
`[45, 46, 47]
`
`R9
`
`R8
`
`4
`
`R1, R2: H, Cl, CH3, H3CO, H3CCH2O
`R3: H, Cl
`R4,R5: COOCH3, CH2OH, NH2,-COOCH2-, CO(disubst.)N
`R6, R7: H, H3CO, Cl, -OCH2O-
`R8: H, H3CO, Cl, Br, NH2
`R9: H, CH3
`
`R6
`
`R7
`
`
`
`Cl
`
`O
`
`5
`
`NN
`
`HN
`
`R1
`
`R2
`
`R3
`R1: H, CN, NO2, CF3, Cl, COOHCONH2, CON(CH3)2
`R2: H, Cl
`R3: Cl, subst.-amino
`
`
`
`O
`
`HN
`
`R1
`
`N
`
`6
`R1: subst.-Ph
`R2: H, (subst.)-CH2-Phe
`
`R2
`
`N
`
`N
`
`
`
`Naphthalenes
`
`6.2 - >10000
`
`Aminophthalazines
`
`0.56 -600
`
`Imidazoquinazo-
`
`linones
`
`0.2 - 40
`
`4-Aryl-1-isoquinolinones
`
`1 - >10000
`
`
`
`R2
`
`R1
`
`O
`
`N
`
`R3
`
`R4
`
`O
`
`O
`
`R5
`
`7
`
`R1: H, alkyloxy R2: alkyloxy R3: H, alkyl,
`subst.-amino,(subst.)-aryl, CH2-Phe
`R4: COOH, COOCH3, COOC2H5, CONH2,
`CONHCH3, CON(CH3)2
`R5:CH3, OH, OCH3, Br
`
`[48]
`
`
`
`
`
`ATI 1008-0004
`
`
`
`1574 Current Medicinal Chemistry, 2008 Vol. 15, No. 16
`
`Compound group
`
`IC50 (nM)
`
`Structure
`
`Er(cid:3)s et al.
`
`(Table 3). Contd…..
`
`Ref.
`
`[49, 52]
`
`[50 , 51, 52]
`
`[53, 54,
`
`55, 42]
`
`N
`
`O
`
`R1
`
`O
`
`N
`
`S
`
`O
`
`HN
`
`N
`
`N
`
`O
`
`N
`
`N
`
`R2
`
`8
`
`R1: H, Cl, subst.-CH2-Phe-amino
`R2: H, CH2-Phe-4-OH
`
`N
`
`N
`
`N
`
`O
`
`S
`
`N
`
`O
`
`HN
`
`R
`
`O
`
`N
`
`9
`
`O
`
`N
`
`N
`
`O
`
`
`
`N
`
`N
`
`R: subst.-Phe
`
`R1
`
`N
`
`O
`
`O
`
`N
`
`R2
`
`R3
`
`N
`
`N
`
`10
`
`9b
`
`Sildenafil
`
`
`
`R4
`R1: H, alkyl, (subst.)-amino-alkyl, (subst.)-aryl-
`alkyl, H3CCH2O,C2H5OH,
`Phe-(O,S)-(CH2)2
`R2: alkyl, subst.-CH2-Phe
`R3: H, (subst.)-CH2-Phe
`R4: (subst.)-aryl, 4-Br-CH2-Phe,c-alkyl-amino,
`H3COCH2
`
`
`
`N R2
`
`O
`
`NH
`
`O
`
`N
`
`R1
`
`Q
`
`N
`
`O
`
`O
`
`NH
`
`Pyrazolo-pyrido-pyrimidines
`
`0.31 - 13
`
`Pyrazolo-pyrimidinones
`
`0.05 - 40
`
`Xanthines
`
`0.3 - >10000
`
`Pyrrolo-quinolones
`
`0.019 - 148 (Ki)
`
`11
`
`12
`
`X
`
`O
`
`X
`
`O
`
`[18, 56, 57, 58, 59]
`
`X = CH2, O
`Q = CH2, CH2CH2
`R1 = CH3, CH2-Phe, 2-pyridyl-CH2
`R2 = subst.-aryl, CO-subst.-aryl,
`COO-CH2-Phe
`
`
`
`
`
`
`
`
`
`ATI 1008-0005
`
`
`
`Structure –Activity Relationships of PDE5 Inhibitors
`
`(Table 3). Contd…..
`
`Current Medicinal Chemistry, 2008 Vol. 15, No. 16 1575
`
`Compound group
`
`IC50 (nM)
`
`Structure
`
`Ref.
`
`Y
`
`X
`
`R2
`
`Q
`
`Y
`
`N
`
`R1
`
`R5
`
`R6
`
`N
`
`R4
`
`13
`
`N
`
`13a
`
`Tadalafil
`
`O
`
`N
`
`N
`
`O
`
`O
`
`O
`
`N R3
`
`[57, 60, 61, 62, 63, 64]
`
`Beta-carbolines
`
`0.7 - >10000
`
`R1
`
`NH
`
`Q: -, CH2, CH2CH2
`X: N, CH
`R1: (subst.-)aryl
`R2: H, alkyl, CH2-Phe, subst.-amino-ethyl, N-subst.-aza-cycloalkyl, Phe
`R3: CO-subst.-furyl, CH2-subst.-furyl, subst.-pyridyl,
`subst.-pyrimidyl, COOCH3, COCH=CH-(4-(CH3)2N-C2H4-O)-Phe
`R4: H, CH3; R5: H, OCH3; R6: H, OCH3
`Y: O, H2
`
`14
`
`Naphthyridine derivatives
`
`0.22 - 11
`
`Quinolines
`
`0.05 - 160
`
`Pyrazolo-pyrido-pyridazines
`
`0.03 - 0.3
`
`N
`
`O
`
`O
`
`O
`
`R2
`
`N
`
`O
`
`R1
`
`N
`
`O
`
`R2
`
`O
`
`N
`
`N
`
`O
`
`O
`
`O
`
`15
`
`O
`
`O
`
`16
`
`O
`
`R1: subst.-aryl-(CH2)
`R2: aryl-alkyl-oxy, aryl-alkyl-amino,
`N-CH3-piperazine, OH(CH2)2NH-
`
`O
`
`Cl
`
`R1: H, COOC2H5, subst.-CONH, CH2OH
`R2: H, CF3, CN
`R3: H, CF3
`R4: H, C2H5
`R5: H, 4-pyridyl
`
`HN
`
`N
`
`17
`
`R1
`
`R5
`
`R2
`
`R3
`
`O
`
`R4
`
`Cl
`
`
`
`
`
`R: 4-CONHCH3-1-imidazole, 4-OH-1-piperidine,
`4-COOH-1-piperidine, 4-pydidyl-CH2
`
`HN
`
`N
`
`N
`
`R
`
`N
`
`N
`
`N
`
`18
`
`
`
`
`
`[65]
`
`[66, 67]
`
`[68]
`
`ATI 1008-0006
`
`
`
`Er(cid:3)s et al.
`
`(Table 3). Contd…..
`
`Ref.
`
`[69, 70]
`
`[71]
`
`[72]
`
`[73]
`
`[74]
`
`1576 Current Medicinal Chemistry, 2008 Vol. 15, No. 16
`
`Compound group
`
`IC50 (nM)
`
`Structure
`
`HO
`
`HO
`
`OH
`
`HO
`
`O
`
`R1
`
`R1, R3: H, OH
`R2, R4 : H, CH3
`R5: H, alkyl
`
`R3
`
`O
`
`R4
`
`O
`
`O
`
`19
`
`R5
`
`HO
`
`O
`
`
`
`R1
`
`O
`
`O
`
`20
`
`R2
`
`R3
`
`O
`
`R2
`
`N
`
`21
`
`R1
`
`N
`
`N
`
`R1: H, NHBu, NH-3-Cl-4-H3CO-CH2-Phe
`R2: OH, OCH2CH3, (aryl)-alkyl-NH
`R3: (aryl)-alkyl-NH
`
`
`
`R1: aryl, t-But
`R2: aryl, alkyl
`R3: (subst.)CH2-Phe, Phe
`
`
`
`R2
`
`E
`
`G
`
`D
`
`R1
`
`A B
`
`O
`
`N
`
`R1
`
`N
`
`N
`
`N
`
`R2
`
`O
`
`HN
`
`O
`
`22
`
`NN
`
`R3
`
`R3
`
`A, B: C, N
`G, D: CH, N
`E: CH, N, O
`
`23
`
`R1: alkyl
`R2: H, alkyl
`R3: H, SO2N(CH2CH2)2N-R
`
`
`
`R6
`
`R7
`
`N
`
`N
`
`R8
`
`R9
`
`O
`
`N
`
`N
`
`O
`
`R
`
`R6: H, CH3, Br
`R7: H, CH3
`R8: H, CH3, Br, CN
`R9: H, alkyl
`R: H, SO2N(CH2CH2)2N-R'
`
`24
`
`
`
`Flavonoids
`
`13 - >100000
`
`Pyrazolo-pyridines
`
`0.6 - 180
`
`Pyrimido-pyridazinones
`
`0.031 - 3.1
`
`Other pyrimidine condensed
`heterocycles
`
`1 – 500
`
`Pyrido-purinones
`
`9 – 4870
`
`
`
`
`ATI 1008-0007
`
`
`
`Structure –Activity Relationships of PDE5 Inhibitors
`
`(Table 3). Contd…..
`
`Compound group
`
`IC50 (nM)
`
`Structure
`
`Current Medicinal Chemistry, 2008 Vol. 15, No. 16 1577
`
`Ref.
`
`[52]
`
`
`
`[44]
`
`[37]
`
`[75]
`
`[76]
`
`
`
`
`
`N
`
`
`
`O
`
`O
`
`HN
`
`N
`
`25
`
`O
`
`N
`
`O
`
`N
`
`N
`
`HN
`
`O
`
`HN
`
`O
`
`HN
`
`26
`
`N
`
`N
`
`N
`
`O
`
`OH
`
`
`
`O
`
`N
`
`N
`
`O
`
`N
`
`
`
`HN
`
`O
`
`O
`
`29
`
`
`
`OH
`
`N
`
`N
`
`N
`
`N
`
`30
`
`OH
`
`
`
`27
`
`O
`
`S
`
`O N
`
`28
`
`N+
`
`N
`
`N
`
`Purine analogue
`
`10
`
`Vardenafil analogue
`
`5
`
`Quinazoline analogue
`
`3.7
`
`Vardenafil
`
`0.7
`
`Chelerythrine
`
`19000
`
`Cl
`
`N
`
`O
`
`O
`
`Dipyridamole
`
`690
`
`HO
`
`N
`
`N
`
`OH
`
`
`
`
`ATI 1008-0008
`
`
`
`Er(cid:3)s et al.
`
`(Table 3). Contd…..
`
`Ref.
`
`[70]
`
`[77]
`
`[77]
`
`[77]
`
`[77]
`
`[77]
`
`[78]
`
`
`
`1578 Current Medicinal Chemistry, 2008 Vol. 15, No. 16
`
`Compound group
`
`IC50 (nM)
`
`Structure
`
`N
`
`N
`
`NH
`
`
`
`F
`
`O
`
`O
`
`F
`
`O
`
`N
`
`HN
`
`O
`
`31
`
`
`
`
`
`
`
`O
`
`O
`
`O
`
`O
`
`O
`
`O
`
`F
`
`F
`
`
`
`O
`
`O
`
`
`
`N
`
`N
`
`32
`
`N
`
`NH
`
`O
`
`H2N
`
`O
`
`N
`
`O
`
`33
`
`N
`
`34
`
`O
`
`HO
`
`N
`
`Cl
`
`O
`
`NH
`
`35
`
`Cl
`
`N
`
`Cl
`
`O
`
`NH
`
`36
`
`Cl
`
`O
`
`HN
`
`O
`
`N
`
`O
`
`S
`
`HN
`
`O
`
`37
`
`
`
`Zaprinast
`
`3300
`
`Zardaverine
`
`81000
`
`Filaminast
`
`53000
`
`Cilomilast
`
`53000
`
`Roflumilast
`
`17000
`
`Piclamilast
`
`3500
`
`DMPPO
`
`3
`
`ATI 1008-0009
`
`
`
`Structure –Activity Relationships of PDE5 Inhibitors
`
`(Table 3). Contd…..
`
`Current Medicinal Chemistry, 2008 Vol. 15, No. 16 1579
`
`Compound group
`
`IC50 (nM)
`
`Structure
`
`Ref.
`
`O
`
`HN
`
`O
`
`N
`
`SO
`
`O
`
`38
`
`N
`
`N
`
`HO
`
`O
`
`HN
`
`O
`
`N
`
`NH
`
`O
`
`
`
`O
`
`O
`
`N
`
`39
`
`O
`
`HN
`
`O
`
`N
`
`NH2
`
`40
`
`O
`
`HN
`
`S
`
`O
`
`N
`
`
`
`S
`
`N
`
`NH
`
`41
`
`O
`
`HN
`
`O
`
`N
`
`N
`
`N
`
`O
`
`42
`
`
`
`
`
`
`
`Quinazolone analogue
`
`6.5
`
`Benzo-furo-pyrimidone
`analogue
`
`18
`
`Benzothiophene analogue
`
`1.7
`
` Benzothiophene analogue
`
`3.5
`
`Imidazo-pyridine analogue
`
`10
`
`
`
`
`
`
`[79]
`
`[79]
`
`[79]
`
`[79]
`
`[80]
`
`
`
`ATI 1008-0010
`
`
`
`Er(cid:3)s et al.
`
`(Table 3). Contd…..
`
`Ref.
`
`[81]
`
`[82]
`
`[83]
`
`[84]
`
`[85]
`
`[86]
`
`N
`
`OH
`
`
`
`Cl
`
`O
`
`
`
`
`
`N
`
`O
`
`O
`
`O
`
`Cl
`
`O
`
`OH
`
`
`
`O
`
`Structure
`
`O
`
`O
`
`S
`
`N
`
`O
`
`HN
`
`N
`
`NN
`
`N
`
`O
`
`N
`
`43
`
`S
`
`HN
`
`HN
`
`N
`
`N
`
`44
`
`O
`
`N
`
`N
`
`N
`
`N
`
`45
`
`HN
`
`O
`
`NH
`
`O
`
`46
`
`N
`
`N
`
`N
`
`N
`
`N
`
`Cl
`
`O
`
`47
`
`OH
`
`
`
`R3
`
`N
`
`O
`
`N
`
`R1
`
`R2
`
`N
`
`48
`
`R1: (subst.) aryl, heteroaryl
`R2: alkylamino, benzylamin
`R3: alkyl, CH2-Phe
`
`
`
`1580 Current Medicinal Chemistry, 2008 Vol. 15, No. 16
`
`Compound group
`
`IC50 (nM)
`
`Triazolo-purinone analogue
`
`0.34
`
`Thia-triaza-fluorene
`
`analogue
`
`2.6
`
`Pyrimidine analogue
`
`14
`
`Anthranylamine analogue
`
`0.4
`
`Imidazo-pyridine analogue
`
`10
`
`pyrido[2,3-b]pyrazin-
`3(4H)ones
`
`0.4 - >10000
`
`ATI 1008-0011
`
`
`
`Structure –Activity Relationships of PDE5 Inhibitors
`
`(Table 3). Contd…..
`
`Compound group
`
`IC50 (nM)
`
`Structure
`
`Current Medicinal Chemistry, 2008 Vol. 15, No. 16 1581
`
`Ref.
`
`[87]
`
`N
`
`N
`
`N
`
`R2
`
`X
`
`R1
`
`N
`
`NH
`
`N
`
`N
`
`N
`
`R2
`
`X
`
`R1
`
`O
`
`49
`
`O
`
`50
`
`X: N, C
`R1: (subst.) aryl, heteroaryl
`R2: H, CH2-Phe
`
`
`
`Quinazolones / Purines
`
`2000 - 27000
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Fig. (2). GMP and PDE5 inhibitors co-crystallized with PDE5A.
`yellow: GMP, blue: sildenafil, orange: vardenafil, green: tadalafil.
`
`structures of the molecules and the inhibitory data were exported in
`a structure definition (sdf) file, which was imported into the Dragon
`[93] software. Dragon is a descriptor calculation program which
`can calculate 1612 type of 0, 1, 2 and 3 dimensional descriptors
`from the molecular structure. After the Dragon descriptors had been
`calculated, they were imported into our in-house developed auto-
`matic model generator program, 3DNET4W. Using the 1D descrip-
`tor pre-selection features of 3DNET4W the constant descriptors and
`one descriptor from each pair of highly correlated descriptors were
`removed from the dataset. As a result of this pre-selection, the
`number of starting descriptors decreased to 307. At this point we
`had to exclude 18 molecules from the calculations, because they
`appeared to be heavy outliers. They were identified by building a
`fast MLR model (with 'scout scan' sequential variable subset selec-
`tion) including all molecules. Compounds with residue error greater
`than 1.5 pIC50 unit were identified as outliers and excluded from
`the calculations. Table 4 shows the number of excluded molecules
`from each compound group.
`
`The experimentally determined activity of these outlier mole-
`cules might be erroneous or the calculated descriptors are not suffi-
`cient to describe their activity. The relatively high number of out-
`liers of the naphthalenes might be the result of certain differences in
`the used PDE5 enzyme or the applied assay protocol compared to
`the other literature sources. The source of enzyme was beagle in
`studies on naphthalenes [18] whereas in other cases, rabbit, pig,
`guinea pig, rat, bovine or human PDE5 were used.
`
`After decreasing the number of descriptors and removing 18
`outliers we started the calculations with a dataset containing 307
`
`descriptors and 420 molecules. The dataset was splitted into two
`main parts, the first part was the so called Model Building Set
`(MBS) and the second part was the External Validation Set (EVS).
`The MBS (300 molecules) was used to build the models and the
`EVS (120 molecules) to validate them. In the cyclic-iterative model
`optimization process, the MBS was further split randomly and re-
`peatedly (six times) in two halves yielding the training (TS) and
`monitoring (MS) sets. The former was used to fit the current model
`and the latter to control the predictive ability of that model.
`
`Table 4.
`
`Number of Outliers in Each Class of Compounds Exhibit-
`ing the Residual Error >1.5 pIC50 Unit as Evaluated in a
`Fast MLR Model
`
`
`
`
`
`Compound group
`
`Number of outliers
`
`Naphthalenes
`
`Polycyclic guanine inhibitors
`
`Quinolines
`
`Aminophthalazines
`
`4-Aryl-1-isoquinolinones
`
`Flavonoids
`
`Xanthines
`
`7
`
`3
`
`2
`
`2
`
`2
`
`1
`
`1
`
`Three types of models were built: multiple linear regression
`(MLR), partial least squares (PLS) and artificial neural network
`(ANN). Variable subset selection (VSS) was performed with the
`genetic algorithm (GA) and sequential selection algorithms (SSA).
`
`ATI 1008-0012
`
`
`
`1582 Current Medicinal Chemistry, 2008 Vol. 15, No. 16
`
`Er(cid:3)s et al.
`
`The aim of the optimization was to improve repeated evaluation
`statistics (standard error of prediction, SEP) of the predictions and
`effective descriptor scoring functions were used to facilitate quick
`generation of MLR and PLS models with optimal predictive ability.
`ANN models were generated using the MLR and PLS models as
`starting points. In this case only the network architecture was opti-
`mized. The optimizations were stopped when any change in the
`descriptor set of the given model increased the average SEP of the
`six cross validations for that model. During the descriptor selection
`process, models having SEP values lower than a pre-defined
`threshold were collected in the "model bank". When the automatic
`process stopped, the best model was selected from the model bank
`by using bootstrap validation (random split half 512 times). In the
`case of the bootstrap validation the crossvalidated squared correla-
`tion coefficient (Q2) was the selection criterion in order to obtain a
`final model with balanced Q2 and SEP values. This way the pro-
`
`gram selected a PLS model with 47 variables and 19 PLS compo-
`nents. The descriptors are listed in Table 5. The predictive ability of
`this model was checked by calculating the activity values of the
`molecules in the EVS. Obtained parameters, Q2 = 0.69, R2 = 0.69,
`SEP = 0.82, (correlation between the experimental and predicted
`values is depicted in Fig. 3) validated the final model, enabling this
`model for in silico screening of large compound libraries.
`
`After the successful external validation of the final model, we
`wanted to ensure that the model was not only a ‘poor chance’ corre-
`lation, so we made a ‘Scrambled-Y’ test. In this test, the biological
`responses are randomly shuffled and a new QSAR model is devel-
`oped using the original independent variables. If the new models
`developed from the dataset with randomized responses have sig-
`nificantly weaker statistical parameter(s) (Q2
`LHO, SEPLHO, etc.) than
`the original model, then the proposed model is not the result of a
`chance correlation [94, 95].
`
`Descriptors Used in the Best Model
`
`Table 5.
`
`
`Descriptor
`
`Eigenvector coefficient sum from polarizability / van der Waals weighted distance matrix (2 descriptors)
`
`Radial Distribution Function - 3.0 and 14.0 / weighted by atomic masses (2 descriptors)
`
`Broto-Moreau autocorrelation of a topological structure - lag 2 and 3 / weighted by atomic masses (2 descriptors)
`
`Eigenvalue 09 from edge adj. matrix weighted by dipole moments
`
`H total index / weighted by atomic van der Waals volumes
`
`Geary autocorrelation - lag 1 and 6 / weighted by atomic Sanderson electronegativities (2 descriptors)
`
`R autocorrelation of lag 1 and 3 / weighted by atomic masses (2 descriptors)
`
`Highest eigenvalue n. 3 and 5 of Burden matrix / weighted by atomic masses (2 descriptors)
`
`Balaban X and V indices (2 descriptors)
`
`Valence connectivity index chi-1
`
`Span R
`
`Total information content index (neighborhood symmetry of 1-order)
`
`T total size index / weighted by atomic Sanderson electronegativities
`
`Leverage-weighted total index / weighted by atomic polarizabilities
`
`Moran autocorrelation - lag 6 / weighted by atomic polarizabilities
`
`Radial Distribution Function - 11.5 / weighted by atomic polarizabilities
`
`Average shape profile index of order 2
`
`Randic shape profile no. 1, 2, 4, 17 and 18 (5 descriptors)
`
`Total information index of atomic composition
`
`Solvation connectivity index chi-2
`
`Kier symmetry index
`
`Highest eigenvalue n. 8 of Burden m