Computational studies on nuclear receptors : mechanism of allosterism and design of novel ligands

(1)

DISSERTATIONS | TUOMAS ORAVILAHTI | COMPUTATIONAL STUDIES ON NUCLEAR RECEPTORS: MECHANISM OF ... | No 631

Dissertations in Health Sciences

PUBLICATIONS OF

THE UNIVERSITY OF EASTERN FINLAND

uef.fi

PUBLICATIONS OF

THE UNIVERSITY OF EASTERN FINLAND Dissertations in Health Sciences

ISBN 978-952-61-3818-3 ISSN 1798-5706

Nuclear receptors (NRs) are transcription factors, which regulate gene expression in

response to ligand binding. This feature makes them ideal drug targets. This thesis utilizes molecular dynamics simulations to study the molecular level mechanism of allosteric activation of a nuclear receptor heterodimer. In addition, new ligands for steroid hormone receptors are designed and

studied with molecular modelling methods.

TUOMAS ORAVILAHTI

COMPUTATIONAL STUDIES ON NUCLEAR RECEPTORS:

MECHANISM OF ALLOSTERISM

AND DESIGN OF NOVEL LIGANDS

(2)

(3)

COMPUTATIONAL STUDIES ON NUCLEAR RECEPTORS: MECHANISM OF ALLOSTERISM AND

DESIGN OF NOVEL LIGANDS

(4)

(5)

Tuomas Oravilahti

COMPUTATIONAL STUDIES ON NUCLEAR RECEPTORS: MECHANISM OF ALLOSTERISM AND

DESIGN OF NOVEL LIGANDS

To be presented by permission of the Faculty of Health Sciences, University of Eastern Finland

for public examination in MD100 Auditorium, Kuopio on June 17^th, 2021, at 12 o’clock noon

Publications of the University of Eastern Finland Dissertations in Health Sciences

No 631

Faculty of Health Sciences/School of Pharmacy University of Eastern Finland, Kuopio

2021

(6)

(7)

Series Editors

Professor Tomi Laitinen, M.D., Ph.D.

Institute of Clinical Medicine, Clinical Physiology and Nuclear Medicine Faculty of Health Sciences

Professor Tarja Kvist, Ph.D.

Department of Nursing Science Faculty of Health Sciences Professor Ville Leinonen, M.D., Ph.D.

Institute of Clinical Medicine, Neurosurgery Faculty of Health Sciences

Professor Tarja Malm, Ph.D.

A.I. Virtanen Institute for Molecular Sciences Faculty of Health Sciences

Lecturer Veli-Pekka Ranta, Ph.D.

School of Pharmacy Faculty of Health Sciences

Distributor:

University of Eastern Finland Kuopio Campus Library

P.O.Box 1627 FI-70211 Kuopio, Finland

www.uef.fi/kirjasto GRANO Helsinki, 2021

ISBN (print): 978-952-61-3818-3 ISBN (PDF): 978-952-61-3819-0

ISSNL: 1798-5706 ISSN: 1798-5706 ISSN (PDF): 1798-5714

(8)

(9)

Author’s address: Faculty of Health Sciences/School of Pharmacy University of Eastern Finland

KUOPIO, FINLAND

Doctoral programme: Doctoral Programme in Drug Research

Supervisors: Adjunct Professor Maija Lahtela-Kakkonen, Ph.D.

Faculty of Health Sciences/School of Pharmacy University of Eastern Finland

KUOPIO, FINLAND

Adjunct Professor Tuomo Laitinen, Ph.D.

KUOPIO, FINLAND

Docent Mikael Peräkylä, Ph.D.

KUOPIO, FINLAND

Reviewers: Professor Petr Pávek, Ph.D.

Department of Pharmacology and Toxicology and Centre for Drug Development

Charles University

HRADEC KRÁLOVÉ, CZECH REPUBLIC Markus Miettinen, Ph.D.

Max Planck Institute of Colloids and Interfaces POTSDAM, GERMANY

Opponent: Professor Björn Windshügel, Ph.D Fraunhofer ITMP

HAMBURG, GERMANY

(10)

(11)

Oravilahti, Tuomas

Computational studies on nuclear receptors: mechanism of allosterism and design of novel ligands.

Kuopio: University of Eastern Finland

Publications of the University of Eastern Finland Dissertations in Health Sciences 631. 2021, 74 p.

ISBN (print): 978-952-61-3818-3 ISSNL: 1798-5706

ISSN: 1798-5706,

ISBN: (PDF): 978-952-61-3819-0 ISSN: (PDF): : 1798-5714

ABSTRACT

Nuclear receptors (NRs) are transcription factors, which regulate gene expression in response to ligand binding, a feature which makes them ideal drug targets.

While many NRs like steroid hormone receptors function as homodimers, one group of NRs can form heterodimers with another NR called retinoid X receptor (RXR). Both partners of a heterodimer complex can bind a ligand, and different heterodimers respond to ligand binding differently. Some of these heterodimers can also be activated in an allosteric manner through ligand binding on RXR. This thesis utilizes molecular dynamics simulations to study the molecular level mechanism of allosteric activation of the peroxisome proliferator-activated receptor (PPARα)-RXRα heterodimer. It was found that allosteric activation may result from stabilization of the co-activator binding site upon ligand binding to the heterodimeric partner.

In addition, structural determinants of several new estrogen receptor (ER) binders were explored in docking studies. Single amino acid differences in the ligand binding pockets of ERα and ERβ were found to contribute to the binding orientation of the compounds and selectivity between these two ER subtypes.

A rational design approach was used to discover novel structures for androgen receptor (AR) binders, by combining the existing features of ER binders, AR binders and a novel structural feature. In experimental tests, these compounds were found to bind to AR with high affinity and to inhibit AR activation comparably or better than the existing AR antagonists used in clinical practice.

Keywords: Nuclear receptors, peroxisome proliferator-activated receptor, estrogen receptor, androgen receptor, drug design, allosterism

(12)

(13)

Oravilahti, Tuomas

Laskennallisia tumareseptoritutkimuksia: allosteriamekanismi ja uudet sitoutujat.

Kuopio: Itä-Suomen yliopisto

Publications of the University of Eastern Finland Dissertations in Health Sciences 631. 2021, 74 p.

ISBN (print): 978-952-61-3818-3 ISSNL: 1798-5706

ISSN: 1798-5706,

ISBN: (PDF): 978-952-61-3819-0 ISSN: (PDF): : 1798-5714

TIIVISTELMÄ

Tumareseptorit ovat transkriptiotekijöitä, jotka sääntelevät geenien ilmentymistä pienimolekyylisten ligandien sitoutumisen seurauksena. Tämä tumareseptorien ominaisuus tekee niistä ihanteellisia kohteita lääkeaineille. Osa tumareseptoreista muodostaa dimeereitä toisen tumareseptorin, RXR:n (retinoid X receptor) kanssa.

Kumpikin kompleksin osapuoli voi sitoa ligandia, ja eri heterodimeerit vastaavat ligandien sitoutumiseen eri tavoin. Jotkut näistä heterodimeereistä aktivoituvat allosteerisesti myös silloin, kun ligandi sitoutuu RXR:iin. Tässä väitöskirjatyössä tutkittiin molekyylidynamiikka-menetelmällä allosteerisen aktivoitumisen molekyylitason mekanismia peroxisome proliferator-activated receptor (PPARα)- RXRα- heterodimeerissä. Tutkimuksessa havaittiin, että allosteerinen

aktivoituminen RXRα:n ligandin sitoutuessa voi olla seurausta liikkeiden vähentymisestä koaktivaattoriproteiinien sitoutumiskohdassa PPARα:ssa.

Lisäksi tutkittiin telakointimenetelmillä rakenteellisia tekijöitä, jotka määrittävät uusien yhdisteiden sitoutumista estrogeenireseptoriin (ER). Yksittäisten

aminohappojen erot ligandia sitovassa taskussa vaikuttavat ligandien sitoutumiskonformaatioihin ja selektiivisyyteen ER:n alatyyppien välillä.

Uusien androgeenireseptori (AR)-antagonistien suunnittelussa sovellettiin rationaalisen suunnittelun menetelmää yhdistämällä ER:iin ja AR:iin sitoutuvien yhdisteiden tunnettuja piirteitä uuteen rakenteelliseen ominaisuuteen.

Kokeellisessa testauksessa havaittiin, että nämä yhdisteet sitoutuvat AR:iin korkella affiniteetillä ja inhiboivat AR:a yhtä hyvin tai paremmin kuin kliinisessä käytössä olevat AR antagonistit.

Avainsanat: Tumareseptorit, peroxisome proliferator-activated receptor, estrogeenireseptori, androgeenireseptori, lääkeainesuunnittelu, allosteria

(14)

(15)

ACKNOWLEDGEMENTS

The studies behind this thesis were done mainly at the University of Kuopio, Department of Chemistry, over the years 2004–2010. I wish to thank my principal supervisor docent Mikael Peräkylä and professor Reino Laatikainen for their guidance to molecular modelling and all the support in conducting my research. In addition, I want to thank all the people conducting the experimental work, which was essential for the success of these computational studies, Dr. Juha Pulkkinen, Dr. Pekka Poutiainen and Dr. Ferdinand Molnár among others.

The actual writing of this thesis started ten years after the original publications, at the Department of Pharmacy, University of Eastern Finland. I wish to express my sincere gratitude to my supervisors, Adjunct Professors Maija Lahtela-Kakkonen and Tuomo Laitinen, who finally made it possible after all these years.

In addition, I would like to thank professor Chris Oostenbrink, who was the leader of the drug design group and my host during my research visits to Amsterdam and Vienna. During those visits, I learned to use new programs and tools for modelling and drug design. Thank you also for being my host at the lab, I always felt very welcomed.

Lastly, it must be noted that the writing process required a significant amount of time on weekends, late nights, and holidays. I want to express my deepest gratitude to my wife, Anniina, for keeping the home running and taking care of the children during my absence to help me to get to this point.

Kuopio, May 2021

(16)

(17)

LIST OF ORIGINAL PUBLICATIONS

This dissertation is based on the following original publications:

I Venäläinen, T.^a, Molnár, F., Oostenbrink, C., Carlberg, C., & Peräkylä, M.

(2010). Molecular mechanism of allosteric communication in the human PPARα‐RXRα heterodimer. Proteins: Structure, Function, and Bioinformatics, 78(4), 873-887.

II Poutiainen, P. K.,^b Venäläinen, T. A.^a,b, Peräkylä, M., Matilainen, J. M., Väisänen, S., Honkakoski, P., ... & Pulkkinen, J. T. (2010). Synthesis and biological evaluation of phenolic 4, 5-dihydroisoxazoles and 3-hydroxy ketones as estrogen receptor α and β agonists. Bioorganic & medicinal chemistry, 18(10), 3437-3447.

III Poutiainen, P. K., Oravilahti, T., Perakyla, M., Palvimo, J. J., Ihalainen, J. A., Laatikainen, R., & Pulkkinen, J. T. (2012). Design, synthesis, and biological evaluation of nonsteroidal cycloalkane [d] isoxazole-containing androgen receptor modulators. Journal of medicinal chemistry, 55(14), 6316-6327.

The publications were adapted with the permission of the copyright owners.

aTuomas Oravilahti, birth name Tuomas Venäläinen

bEqual contribution.

Overall, TO did the computational parts and the other authors performed the experimental parts of the publications. In publication I, TO designed and

conducted the molecular dynamics simulations and the analyses of clusters and fluctuations. The other main authors performed the in vitro studies and supportive computational analyses. In publication II, the two main authors contributed

equally: TO conducted docking studies and the related analyses, while the

syntheses and other experimental work were undertaken by the other authors. In publication III, TO designed the molecular structure of the new binders; the syntheses and biological evaluation of those compounds were done by the other investigators.

(18)

(19)

ABBREVIATIONS

9cRA 9-cis-retinoic acid ADT Androgen deprivation

therapy

AF-1 Activation function-1 AF-2 Activation function-2 AR Androgen receptor CAR Constitutive androstane

receptor

DAX Dosage-sensitive sex

reversal, adrenal hypoplasia critical region, on

chromosome X, gene 1 DBD DNA binding domain DHT Dihydrotestosterone E2 Estradiol

ER Estrogen receptor FXR Farnesoid X receptor GCNF Germ cell nuclear factor GR Glucocorticoid receptor

GW GW409544

LBD Ligand binding domain LBP Ligand binding pocket LRH-1 Liver receptor Homolog-1 MD Molecular dynamics MM-PBSA Molecular Mechanics

Poisson-Boltzmann Surface Area

N-CoR Nuclear corepressor

NGF1-B Nerve growth factor 1B NMR Nuclear magnetic resonance NOR-1 Neuron derived orphan

Receptor-1 NR Nuclear receptor NURR1 Nurr-related Factor-1 PPAR Peroxisome-proliferator-

activated receptor PR Progesterone receptor RAR Retinoid acid receptor RBA Relative binding affinity RE Response element

RID Receptor interacting domain RMSD Root mean square deviation RMSF Root mean square

fluctuations

RXR Retinoid X receptor

SCA Statistical coupling analysis SERM Selective estrogen receptor

modulator

SF-1 Steroidogenic Factor 1 SHP Small heterodimer partner SMRT Silencing mediator for

retinoid and thyroid hormone receptors

SRC Steroid receptor coactivator TI Thermodynamic integration

(22)

TR Thyroid hormone receptor VDR Vitamin D receptor

(23)

1 INTRODUCTION

Nuclear receptors (NRs) are transcription factors, which regulate gene expression in response to ligand binding and their ability to start and stop gene expression makes them ideal drug targets. The natural ligands of the NRs include molecules like steroid hormones, fatty acids and some of these receptors function as sensors for xenobiotics. Compounds binding to NRs have been used for a long time in a wide range of applications, ranging from cancer treatment to diabetes and even birth control. There are still many mysteries surrounding nuclear receptors, for example their interaction with each other, the ligands that bind to them, and other regulatory proteins, which participate in their activation. In recent years, the simplified view of ligand-induced activation of a nuclear receptor has been complemented with other players, e.g. potential impact of DNA binding. In

addition, the mystery of the so-called phantom ligand effect, i.e. a nuclear receptor acting as if it there was a ligand present in the binding site, still remains to be resolved.

The development of new NR binders has evident clinical potential and different computational approaches have been applied in investigations into NR activation and modelling of NR binders at the molecular level. The possibilities of utilizing computational methods in drug development have recently dramatically increased when more knowledge about these proteins has accumulated. In 2005, when the work for this thesis was started, the protein data bank contained less than 30 000 protein structures. In January 2021, there are now over 170 000 protein structures.

Today, computational chemistry is routinely applied to accelerate the process of drug discovery by filtering out compounds to streamline experimental testing and by rationalizing experimental findings to guide the synthesis of new compounds.

Molecular docking is one of the most popular structure-based methods, which has benefitted from the increase in the number of available protein structures.

Focusing on interactions of a ligand and the active site of a rigid,

crystallographic protein structure may well suffice for many applications, for example in screening of potential binders. However, this approach may not be adequate for handling of other kinds of problems like allostery. Allosteric

activation means that a protein is activated by an effector binding at an allosteric binding site, which is not located at the actual active site; an allosteric effect can be

(24)

mediated by changing the protein’s 3-D conformation or mobility. In the case of NRs, it relates to the effect of different parts of the complex other than the actual ligand binding site, for example, on the ways in which ligand binding to a

heterodimeric partner can exert effects on the entire molecular complex.

Molecular dynamics (MD) simulations is a powerful approach to obtain insights into protein movements or changes in the conformation, beyond the information that can be obtained by crystallographic studies. Thus, this method is suitable for examining the mechanisms of allostery at the molecular level.

(25)

2 REVIEW OF THE LITERATURE

2.1 NUCLEAR RECEPTORS

2.1.1 The nuclear receptor superfamily

The nuclear receptors are proteins that regulate gene expression in the cell nucleus. They form a large superfamily of proteins, which are thought to be evolutionary derived from a common ancestor (Figure 1). This superfamily has been further subdivided into subgroups using phylogenetic analysis.

Figure 1. Phylogenetic classification of nuclear receptors to subgroups I–IV. Figure:

modified from (Mazaira et al., 2018)

(26)

The first group of these receptors contains receptors like thyroid hormone receptor (TR), retinoid acid receptor (RAR), peroxisome-proliferator-activated receptor (PPAR), vitamin D receptor (VDR) and farnesoid X receptor (FXR). The most interesting member (from the point of view of this study) of the second subgroup of NRs is the retinoid X receptor, RXR. The third subfamily includes steroid

receptors like the estrogen receptor (ER), progesterone receptor (PR),

glucocorticoid receptor (GR) and androgen receptor (AR). (Weikum et al., 2018;

Mazaira et al., 2018; Aranda and Pascual, 2001; Auwerx et al., 1999; Laudet, 1997) The other three subgroups contain receptors for which the ligands are not known. Subgroup IV includes nerve growth factor 1B (NGF1-B), nurr-related Factor- 1 (NURR1) and neuron derived orphan receptor-1 (NOR-1). Subgroup V contains steroidogenic factor 1 (SF-1) and liver receptor homolog-1 (LRH-1). At present, the only member in subgroup VI is germ cell nuclear factor (GCNF). In addition, there is a subgroup 0, which contains two nuclear receptors (DAX and SHP) with atypical structures. (Weikum et al., 2018)

The natural ligands of the NRs are a wide group of hydrophobic molecules such as steroid hormones (Beato et al., 1995), fatty acids (Kersten et al., 2000;

Schoonjans et al., 1996) and xenobiotics (Kliewer et al., 2002). NRs for which the regulatory ligand is not known, are called “orphan receptors” (Germain et al., 2006). These can be re-classified as “adopted orphans” if the ligand is discovered.

NRs can also be classified based on their dimerization and DNA-binding properties. In this classification system, the steroid hormone receptors belong to class I. They form homodimers and bind to DNA at response elements organized as inverted (palindromic) repeats (Figure 2). The response elements are hexameric DNA sequences, which NRs recognize and to which they subsequently bind

(Khorasanizadeh and Rastinejad, 2001). Class II NRs form heterodimers with RXR and bind to direct repeats. Class III NRs form homodimers similarly as class I receptors, but they bind primarily to direct repeats. Finally, class IV receptors bind to DNA as monomers. Class I includes the steroid hormone receptors and class II contains all of the other known ligand-dependent receptors. Most orphan

receptors fall into classes III and IV. It should be noted that the general rule of homodimerization or heterodimerization of class I and class II receptors does not exclude the possibility for the formation of other, atypical heterodimers, e.g. ER-AR or GR-PPAR, which may have clinical importance. (De Bosscher et al., 2020;

Mangelsdorf et al., 1995)

The role and function of different members of the NR superfamily have been a focus of research and there are several extensive review articles (Glaría Percaz et

(27)

al., 2020; Liu, B. et al., 2019; Han, C., 2018; Oladimeji and Chen, 2018; Check, 2017;

Grimm et al., 2016; Yan and Xie, 2016; Yamazaki et al., 2007). This thesis mainly focuses on PPARα and RXRα, and steroid hormone receptors ER and AR.

Figure 2. NR binding to DNA. Class I receptors bind to response elements (RE) organized as inverted (palindromic) repeats of hexameric half sites (A) and class II receptors bind to direct repeats (B). Monomeric receptors bind to extended half sites, which contain an A/T rich region along with the hexameric half site. LBD:

Ligand binding domain, DBD: DNA binding domain, AF-1: Activation function-1.

Figure: modified from (Weikum et al., 2018)

2.1.2 Peroxisome proliferator-activated receptors (PPARs)

Peroxisomes are cell organelles which have multiple roles in different parts of the body, e.g. β-oxidation of fatty acids and defence against oxidative stress. Their dysfunction is involved in several serious diseases like cancer and metabolic and neurodegenerative diseases. Peroxisome proliferators are a group of structurally diverse compounds which cause an increase in both the size and number of peroxisomes. The first PPAR receptor (later named PPARα) was discovered in 1983 (Lalwani et al., 1987; Lalwani et al., 1983). Its expression was found to display the same tissue specificity as the effects of peroxisome proliferators, and thus, the receptor was thought to mediate the effects of these compounds. Later, more PPAR subtypes (PPARβ/δ and PPARγ) were discovered (Dreyer et al., 1992); their properties have been extensively reviewed (Mirza et al., 2019; Han, L., Shen, Bittner, Kraemer and Azhar, 2017a; Janani and Ranjitha Kumari, 2014; Aleshin et al., 2013; Berger and Wagner, 2002), but they are not a central focus of this thesis.

Compared to the stringent specificity of steroid hormone receptors, all PPARs have a large ligand binding pocket, which means that they can be activated by a variety of fatty acids and synthetic ligands. Their target genes are partly specific to the PPAR subtype and partly common to all subtypes. (Jo et al., 2020; Youssef and Badr, 2013)

PPARα is expressed largely in skeletal muscle, heart, liver, and brown adipose tissue; it is activated by natural ligands like fatty acids and their metabolites (Han,

(28)

L., Shen, Bittner, Kraemer and Azhar, 2017b). In liver, PPARα controls the

expression of the genes involved in many steps of fatty acid metabolism, including fatty acid uptake, intracellular transport, oxidation, and lipoprotein metabolism.

Dietary unsaturated fatty acids affect hepatic gene expression, and these effects are mostly mediated by PPARα (Sanderson et al., 2008). PPARα has a central role in ensuring energy availability in times of fasting and starvation (Contreras et al., 2013).

In addition to lipid metabolism, PPARα has other roles. It is involved in attenuating inflammatory responses in cooperation with GRα (Bougarne et al., 2019; Bougarne et al., 2018). It also has a protective role against

neurodegenerative diseases (Fidaleo et al., 2014). It also seems to mediate protection from oxidative stress (Sekulic-Jablanovic et al., 2017). PPARα is also expressed in the central nervous system, and it has been suggested as a potential target for treating Alzheimer’s disease (D'Orio et al., 2018).

The central role of PPARα in important metabolic pathways, and the fact that it can be activated by ligands, make PPARα an interesting drug target. The fibrates are a group of synthetic ligands which bind to PPARα and which are administered to lower high triglyceride levels in plasma and elevate HDL cholesterol levels, with the intention to reduce the risk of cardiovascular disease in individuals with diabetes (Contreras et al., 2013; Staels et al., 1997). The fibrate group of

compounds includes fenofibrate, gemfibrozil, bezafibrate, and ciprofibrate, which are in clinical use as lipid-lowering drugs (Youssef and Badr, 2013). PPARα also seems to have a role in mediating the effects of metformin in the therapy of diabetes (Kang et al., 2016).

In 1992, PPAR was found to form heterodimers with RXR (Kliewer et al., 1992).

Heterodimerisation, and the differential responses to ligands of RXR and its partner, add complexity to the signalling of all class II NRs. RXR has also different isotypes, named RXRα, RXRβ and RXRγ, which display different expression profiles (Bookout et al., 2006). All these isotypes are highly expressed in the organs of the metabolic system (liver, kidney, adipose tissue and muscle). RXRβ and RXRγ are highly expressed in the central nervous system, whereas RXRα has only moderate expression in the brain.

RXRs are activated by retinoids and vitamin A (retinol) derivatives as well as some toxic environmental chemicals (Unsworth et al., 2017; le Maire et al., 2009).

RXR-selective retinoids are called rexinoids. (Li, Y. et al., 2007) 9-cis-retinoic acid (9cRA) is a retinoic acid derivative which can activate both RXR and RAR. Thus, 9cRA

(29)

can mediate RAR-specific effects by binding to RXR, but also enhance the activation of several other NRs, like PPARs, by binding to RXR.

2.2 THE STRUCTURE AND ACTIVATION OF THE NR

2.2.1 The structure and domain organization of the nuclear receptor

The NR’s structure is organized into five to six separate domains designated as A to F (Figure 3) from the N-terminal to the C-terminal (Germain et al., 2006; Aranda and Pascual, 2001). These domains have their own specific functions. The domains function autonomously and can be interchanged between other NRs without loss of function. The DNA-binding domain (DBD) and ligand binding domain (LBD) are the most conserved domains. The typical size of the DBD is 62–64 amino acid residues (Moore et al., 2006). The DNA-binding domain (DBD) recognizes a specific DNA sequence, a response element, in the nuclear DNA and binds to it. The N- terminal A/B region contains another activation function (AF-1), which can operate independently of ligand binding. This region is less conserved and less organized compared to LBD and DBD, and its size varies between the various NRs. The D region is poorly conserved; it forms a hinge region between DBD and LBD, allowing them more freedom to adopt different conformations. Some receptors also contain a C-terminal F-region, which may be related to receptor activity in some NRs (Begam et al., 2017).

Region:

N-

A/B C D E F

Function: AF-1 DBD LBD, AF-2 -C

Figure 3. General domain organization of nuclear receptors. LBD= ligand binding domain; DBD= DNA binding domain; AF= activation function.

LBD binds the ligand and contains a ligand-dependent activation function-2 (AF-2), which is a hydrophobic groove for co-activator proteins, composed of helices 3, 4 and 12. The typical size of the LBD is 222–284 amino acid residues (Moore et al., 2006). The helix 10 of the ligand binding domain forms a major part of the interface for homo- and hetero-dimerisation (Moras and Gronemeyer, 1998). In addition, helices 9, 7, and the loop connecting helices 7 and 8, contribute to dimer stability. (Weikum et al., 2018; Moore et al., 2006; Bourguet et al., 2000;

Moras and Gronemeyer, 1998)

(30)

2.2.2 The mechanism of ligand-dependent activation

The mechanism of activation is controlled in a NR by at least two parts of the receptor: AF-1 and AF-2. AF-2 is located in the LBD. Some NRs also have a ligand- independent AF-1 that is located in the N-terminal domain of the NR. The AF-2- dependent activity of an NR is determined by the conformation of the so-called helix 12, which is the C-terminal helix of the NR. (Nettles and Greene, 2005)

Binding of an agonist stabilizes the active conformation of helix 12, which becomes tightly packed against helices 3 and 4 (Moras and Gronemeyer, 1998), and this refolding is required for coactivator binding. NR coactivators are proteins or protein complexes that mediate the activation of the target genes (McKenna and O’Malley, 2002; Leo and Chen, 2000). The coactivators can have histone acetylase activity to open the DNA to allow reading, or they can directly interact with the RNA polymerase II to initiate transcription. The result of agonist binding to a NR is the formation of a protein–DNA complex that contains a DNA sequence recognizing part (NR DBD) and a ligand recognizing part (NR LBD). The complex functions to initiate transcription of an NR-specific target gene (Figure 4).

Figure 4. The mechanism of ligand dependent activation (above) and repression (below) of NRs.

(31)

The structures of the coactivators are not known, but a short α-helical part of common coactivators has been crystallized with several NRs (Zhou et al., 2010; Hur et al., 2004). This coactivator peptide is bound on the surface of the NR into a groove formed by helices 3, 4 and 12. A crystal structure (Figure 5) of a full length NR as a heterodimer in complex with a ligand, co-activator peptides and DNA exists (Chandra et al., 2008), but it is not known whether this represents a

physiological structure of the complex or whether all NRs form similar complexes.

Crystallization of the full length receptor has proved difficult because of its flexibility. LBDs and DBDs can be crystallized separately more easily, and the molecular modeling of NRs is mostly restricted to these individual domains.

Figure 5. The structure of the full length PPARγ-RXRα heterodimer in complex with DNA; PPARγshown in orange, RXRαin grey. The helix 12 of PPARγ is in the agonistic conformation and a coactivator peptide is bound to the complex. The DBD of RXR is bound on the β-sheet area of PPARγ.

The conformation of helix 12 is crucial for the ligand-dependent activation of a NR (Figure 4). Agonist binding to the NR leads to the stabilization of the agonistic

(32)

conformation of helix 12, which is needed for binding of the coactivator protein.

Antagonists block the correct docking of helix 12 against helices 3 and 11, and prevent the formation of the coactivator binding pocket. (Nettles and Greene, 2005; Nagy and Schwabe, 2004)

The crystal structure of ER in complex with the antagonist drug, raloxifene, demonstrates that the mechanism of the NR antagonists is to block the correct positioning of helix 12. In the complex with an antagonist, helix 12 is bound to the groove between the helices 3 and 5, thus competing with the binding of the coactivator. (Brzozowski et al., 1997)

Another helix-12-independent mechanism of ligand-dependent activation has also been described. Based on amide hydrogen/deuterium exchange studies, it has been suggested that partial agonists may not stabilize helix 12, but rather other parts of the ligand binding pocket (Bruning et al., 2007). These parts include helix 3 and the β-sheet area. In addition, it has been shown with a group of novel synthetic ligands that direct stabilization of helix 12 is not necessary to activate PPARα or PPARγ (Östberg et al., 2004).

2.2.3 Constitutively active receptors

Many of the NRs like retinoic acid receptor (RAR), vitamin D receptor (VDR), thyroid hormone receptor (TR) and steroid hormone receptors are not active without an agonist. However, some others, such as PPARs and constitutive androstane receptor (CAR) have significant constitutive activity, i.e. they are active without any ligand. A ligand is not needed, if the agonistic helix 12 conformation is sufficiently stable in the apo form of the receptor. Nonetheless, even in these cases, agonist binding can still increase the basal activity, and antagonist binding can reduce it.

The latter is called inverse agonism. (Molnár et al., 2005; Frank et al., 2003) 2.2.4 Permissive and non-permissive complexes

The class II receptors form heterodimers with RXR. The heterodimeric structure is needed to form a contact with the DNA response elements of the target genes.

Both partners of a heterodimer complex can bind a ligand, and different heterodimers respond differently to ligand binding. On the basis of the

heterodimer's response to ligands, NRs can be classified as either permissive or non-permissive receptors (Figure 6).

(33)

Figure 6. Activation of permissive, conditionally permissive and non-permissive heterodimers. RXR= retinoid X receptor; LXR= liver X receptor; RAR= Retinoid acid receptor; VDR= vitamin D receptor; RLU: relative light units; ETOH= ethanol. Figure:

(Shulman et al., 2004)

Permissive receptors, such as PPARs, can be activated with PPAR ligands but also with a RXR ligand and a synergistic effect can be obtained with two ligands.

Non-permissive receptors do not respond to a RXR ligand but exhibit a full response with their own ligand. Conditionally permissive receptors, such as RAR, display synergy, or increased activity with two ligands, but they cannot be activated with RXR ligand alone. (Nettles and Greene, 2005; Shulman et al., 2004)

2.2.5 Coactivators and corepressors

Transcriptional coregulators are a group of proteins that regulate gene expression by interacting with and modulating the activity of NRs and other transcription factors. More than 400 coregulators have been identified and they have been classified in two major classes based on their activity: coactivators and

corepressors (see Figure 4). Coactivators induce gene expression as a response to

(34)

agonist binding to NR. Corepressors in turn repress gene expression when NR is not binding a ligand or when it is binding an antagonist. (Dasgupta et al., 2014)

The family of steroid receptor coactivators (SRCs) was the first NR coactivator family identified. It consists of three members named SRC-1, SRC-2, and SRC-3, which bind to a broad range of NRs, including PR, ER, TR, RXR, GR and PPAR. The receptor interacting domain (RID) of the SRC contains three conserved amino acid sequences, which are important for the coactivator–NR interaction. This sequence is called the NR box and it contains a distinctive LXXLL motif, in which L is leucine and X is any amino acid residue. (Dasgupta et al., 2014; Leo and Chen, 2000)

The structure of the coactivators is not known, and it is not clear how it interacts with a NR heterodimer. Crystal structures have indicated that NR heterodimers can bind two peptides with the LXXLL motif simultaneously, one on the AF-2 for each of the receptors, and there is some evidence that all three LXXLL motifs of the coactivator can be involved with the interaction at the same time.

However, it is not known where the third NR box would bind. (de Vera, Ian Mitchelle S et al., 2017)

In the absence of an agonist, corepressors can bind to the NR (Figure 4). This complex can be stabilized by antagonists. Corepressors, such as nuclear

corepressor (N-CoR) and silencing mediator for retinoid and thyroid hormone receptors (SMRT), also bind to PPARα and repress the transcriptional activity.

Corepressors have similar helical binding motifs as the coactivators, which also bind to the same area. However, for corepressor binding, the helix 12 adopts a different position than in the case of coactivator binding. Thus, ligand binding on NR controls coregulator recruitment using helix 12 position as a mediator. (Nagy and Schwabe, 2004; Xu et al., 2002)

2.3 ALLOSTERIC ACTIVATIONS OF NR COMPLEXES

2.3.1 Models for allostery

The term ‘allostery’ refers to the regulation of the functional activity of a protein via effector binding to an allosteric binding site, which is distinct from the active site.

(Lu et al., 2014) Oxygen binding to hemoglobin has been an interesting topic of allostery research for a long time. Linus Pauling suggested a model for cooperative oxygen binding to hemoglobin in 1935 (Pauling 1935). Later, the same

phenomenon has been examined by several others and various explanations for cooperative binding have been postulated. The term ‘allostery’ was used first by Monod and Jacob in 1961 (Monod and Jacob, 1961).

(35)

Since these early days, different suggestions for the model describing allostery have been proposed. In 1965, Monod, Wyman and Changeux published a model for allostery that was later named the MWC model (Monod et al., 1965). This model describes how oligomeric proteins react to ligand binding by a

simultaneous change in the conformation of all subunits (Figure 7). The difference between the MWC model and another early allostery model called KNF is that in the latter, the subunits are not necessarily adopting the same conformation (Koshland et al., 1966). Both models assume that the protein is an oligomer with identical, rigid subunits. (Cui 2008)

Figure 7. Allosteric models. The long arrows indicate which side is favored. The MWC model assumes equilibrium between high (squares) and low (circles) reactivity (or affinity) conformations. Ligand binding changes the balance of conformations. In contrast, the KNF model assumes that only the ligand binding subunit changes conformation, but it interacts with other subunits to move the equilibrium towards the active (liganded) state. In Population Shift model ligand binding makes the bioactive conformation (*) energetically more favorable.

Modified from (Lu et al., 2014)

The Population Shift model develops these concepts further by stating that the different protein conformations would be in equiliblium and the allosteric event (e.g. ligand binding) would shift the equilibrium towards the bioactive

conformation (Cui 2008). However, none of these models provide an atomistic- level description on how the conformation change is actually induced by ligand binding, and what actually happens at the molecular level.

C NF

P

(36)

2.3.2 Ligand effects inducing coactivator binding

In the context of the NRs, the question related to allostery has not been the affinity of the ligands, but rather the activation of the NR complex. As described above (see Figure 6), the permissive (class II) heterodimers can be activated also with ligands binding to RXR, and they gain their full activity only with ligands bound to both partners of the complex (Nettles and Greene, 2005; Shulman et al., 2004). The classical route of activation of the NR complex requires that a coactivator binds to the NR complex. Unfortunately, the full structure of the co-activator proteins or the NR-coactivator complex is not known. Thus, it is not known why the permissive heterodimers can be activated with ligand binding to RXR, and why they are further activated with a RXR ligand when the partnering NR is binding a ligand.

Sequence analysis has revealed a network of amino acids that seem to play a role in allosteric activation of NR heterodimers. This statistical coupling analysis (SCA) method detects the coevolution of amino acids in a protein, suggesting that these amino acids are connected in some manner. The network of amino acids links together the heterodimerization interface, the co-activator binding surface, the helix 12, and the ligand binding pocket. It was also observed that a mutation in a single amino acid in this network could disrupt the permissivity. (Shulman et al., 2004)

In 1997, Shulman et al. described the so-called ‘phantom ligand effect’, or activation of RAR by a ligand specific to its heterodimeric partner, RXR (Schulman et al., 1997). The effect was so named because RAR seemed to act as if there was a ligand bound to its ligand binding pocket, even though the ligand used in the experiment was RXR specific. Mutations on the AF-2 region of RXR had only a minor effect on the induction by the RXR ligand, but similar inactivating mutants of RAR made the heterodimer unresponsive to the RXR ligand. These results suggest that the RAR was trans-activated by a ligand binding to RXR. The authors proposed that the mechanism for trans-activation must be some type of conformational change in the LBD of RXR. These results were questioned later by Sato et al.; these investigators revealed that the ligand binding to RXR has no effect on the affinity of the SRC-1 NR2 peptide (a coactivator peptide) on RAR (Sato et al., 2010). It was also suggested that the earlier findings of Shulman et al. (1997) could be a result of binding of the RXR antagonist directly to RAR. However, it is unlikely that this explanation would be true to all permissive NRs with different ligand specifities.

Later it was concluded that in the case of the PPARγ–RXRα heterodimer, only one coactivator protein is bound to the complex, and it utilizes either two out of its three LXXLL motifs to occupy the binding sites on each heterodimeric partner or all

(37)

three motifs (de Vera, Ian Mitchelle S et al., 2017; Osz, Pethoukhov et al., 2012).

These experimental results, however, did not rule out other possibilities of coactivator binding, as stated by the authors.

Other investigators have suggested that RXR–RAR, PPAR–RXR and RXR–VDR heterodimers all bind one coactivator protein, and it is bound to RXR (Rochel et al., 2011) while others have indicated that even though some parts of RXR are affected by coactivator binding either by direct contact or by a distant allosteric effect, RXR- AF-2 is not needed for coactivator binding (Belorusova Anna et al., 2020). RAR seems to behave differently when it exists as a homodimer and binds two coactivator proteins, on both RARs (Osz, Brélivet et al., 2012). CAR in turn seems to recruit one or two coactivator proteins, depending on whether only CAR or both partners are binding a ligand (Pavlin et al., 2014).

In contrast to earlier observations on the ability of NR to activate the

heterodimeric partner, Kojetin et al. described the mechanism of silencing on the non-permissive TRβ–RXR. This group applied NMR measurements and observed that the apo-form of TRβ destabilized parts of the agonist-bound RXRα, such that it resembled an apo–RXRα. However, ligand binding to TRβ re-stabilized these regions, in a similar manner as occurred with agonist binding to a RXRα

homodimer. This allosteric signalling used the SCA-detected network of residues described above and, in addition, the RXRα helix 5 was found to play a central role.

(Kojetin et al., 2015)

Despite these efforts, many aspects of the overall picture of the multiple forms of co-activator binding on different NR heterodimers are still unclear. The reason for this uncertainty lies mainly in the difficulties of crystallizing the flexible complex consisting of the NR LBDs and DBDs among other domains, ligands, DNA and coactivators. There are also no crystal structures available of full-length coactivator proteins. In addition, the crystal structures may be different than the actual, functional solution structures, this being especially true with complexes consisting of multiple domains which can move quite flexibly in relation to each other. Finally, ligand binding, co-activator binding, DNA binding and heterodimerization are all equilibrium reactions. The conditions needed for the experiments may change the outcome of the complex formation, thus giving a biased view of the biological phenomenon being investigated.

2.3.3 Allostery on PPAR–RXR complexes

Solving the crystal structure of the full length NR first time by Chandra et al. in 2008 was a major event in NR research, since it provided new perspectives of the

(38)

interactions of NR LBD with DBD, ligand, coactivators and the DNA (Figure 5). All these parts play a role in the complex phenomenon of NR activation. The crystal structure revealed a previously unknown heterodimerization interface between the PPARγ LBD and the RXRα DBD. However, these new findings could not really explain the phantom ligand effect or permissivity. Instead, two individual

coactivator peptides were seen to be bound to PPARγ and RXRα coactivator binding sites. (Chandra et al., 2008)

The overall architecture of the PPARγ-RXRα-DNA complex seen in the above- mentioned crystal structure has been also questioned. In solution, a different domain organization with no direct contact between LBD and DBD, has been observed (Rochel et al., 2011).

Later it was shown by de Vera et al. (2017) that binding of NR on DNA stabilized the PPARγ–RXRα dimerization surface, by influencing the affinity of coregulator binding. Similar findings have been also made with other NRs (Bernardes et al., 2012; Zhang et al., 2011; Meijsing et al., 2009; Hall et al., 2002). Differences in the DNA sequence, in addition to affecting the affinity of the heterodimer to DNA, also influence the affinity of the heterodimer for the co-activators. The authors stated that these DNA-binding-induced effects were compatible with both the more compact crystal structure and the elongated solution structure of the PPARγ–

RXRα–DNA complex. (de Vera, Ian Mitchelle S et al., 2017)

Ricci et al. used the first crystal structure of the full-length PPARγ–RXRα

heterodimer to study allosteric effects in the complex (Ricci et al., 2016). Their MD simulations emphasized the role of a flexible area between helices 2 and 3.

Furthermore, they concluded that the allosteric pathways utilized so-called ‘highly frustrated’ (Ferreiro et al., 2007; Bryngelson and Wolynes, 1987) contacts on the solvent-exposed loops and helices, instead of the more rigid areas in the core of the protein complex. The ‘frustrated fit’ has been described by others (Clark et al., 2016; Johnson et al., 2015), suggesting that when RXR is bound to its natural ligand 9cRA it is less ordered than in its apo form, and this “frustration” can then mediate allosteric effects.

2.4 STEROID RECEPTORS AND THEIR LIGANDS

2.4.1 Androgen receptor in prostate cancer

The androgen receptor (AR) is a member of the NR superfamily, expressed in a diverse range on tissues (Davey and Grossmann, 2016). Its natural ligand

(39)

dihydrotestosterone (DHT) is a steroid hormone that regulates the development of male reproductive system and the secondary sexual characteristics.

The AR has a similar overall structure and domain structure as the other NRs.

However, AR has an exceptional C-terminal extension that is found to be folded on the surface of the LBD when the helix 12 adopts the agonistic conformation. Like many steroid hormone receptors, AR functions as a homodimer. The C-terminal extension blocks the formation of the type of dimer that is seen in the crystal structures of RXR heterodimers by folding onto the dimerization surface (Bledshoe et al. 2002). The structure of the homodimer is different from many other

structures, e.g PPAR heterodimers with RXR, but it does resemble the structure of the GR homodimer (Nadal et al., 2017).

Prostate cancer is common in men over 50 years of age. Genetic alterations commonly occur in signalling pathways that promote growth as well as those blocking apoptosis in prostate cancer cells. Some of these pathways are regulated by AR, and some are able to alter the structure of the AR itself. This, along with the fact that the activity of AR can be regulated with small molecules, makes the AR an important target for the treatment of prostate cancer. (Shafi et al., 2013)

Prostate cancer is often treated with androgen deprivation therapy (ADT), which means deprivation of the AR of testosterone. This is achieved by surgical or medical orchidectomy. AR antagonists can be used in addition to achieve complete androgen blockade. (Davey and Grossmann, 2016)

After an initial response, most tumors become refractory to ADT. This state is called hormone refractory or castration resistant prostate cancer (CRPC). There are multiple mechanisms leading to the loss of response, including an increase in AR expression, AR point mutations and other variants, AR function modulation by changes in cell signalling pathways, and changes in coregulator proteins or androgen biosynthesis. (Fujita and Nonomura, 2019; Shafi et al., 2013) 2.4.2 Antiandrogens and AR mutations

Compounds that compete with androgens for binding to AR and inhibit its activation are called antiadrogens. Steroidal antiandrogens like cyproterone acetate were used to treat advanced prostate cancer because of their ability to bind to AR and block DHT and testosterone binding (Figure 8). However, steroidal antiandrogens had side effects which limited their usability. Steroidal

antiandrogens were commonly replaced by nonsteroidal compounds in prostate cancer therapy because of their more favourable safety profile. First generation nonsteroidal antiandrogens flutamide and nilutamide and second generation

(40)

compound bicalutamide all have similar, relatively low affinities for the AR (Figure 9). This means that DHT can still activate AR to stimulate the growth of prostate cancer cells. (Crawford et al., 2018)

Figure 8. Structures of steroidal AR agonists dihydrotestosterone and R1881 and antagonist cyproterone acetate.

Enzalutamide (Figure 9) is a third generation antiandrogen which, compared with bicalutamide, has five to eight fold higher binding affinity for the AR. Apalutamide has a similar structure as enzalutamide. Darolutamide is structurally different than the other two third generation antiandrogens. When compared to other two, its advantage is its lower penetration across the blood–brain barrier. It is also active against some of the common AR mutants, including the F876L mutation that makes the tumor resistant to enzalutamide and apalutamide. (Crawford et al., 2018; Tran et al., 2009)

(41)

Figure 9. Structures nosteroidal of AR antagonists.

The use of antiandrogen commonly leads to resistance, by mechanisms including AR overexpression and AR mutations. Point mutations are found in approximately 15–30 % of patients with CRPC, most commonly in the ligand binding pocket (Fujita and Nonomura, 2019). For example, many patients respond well at first to

treatment with enzalutamide, but later they develop resistance. One mechanism for the resistance is the appearance of the F876L mutation in the AR ligand binding pocket, which allows the AR to use enzalutamide as an agonist to drive tumour growth. However, the AR with the F876L mutation still responds to bicalutamide. In contrast, the W741C mutation turns bicalutamide into an agonist, but this is not the case for enzalutamide. Mutation T877A confers AR resistance to

hydroxyflutamide (Figure 9) and cyproterone acetate (Figure 8), and AR with T878A can be activated with flutamide, bicalutamide and exalutamide, but also with progesterone and estrogen. Because of these kinds of differences in activity profiles of antiandrogens in AR mutants, it has been suggested that the use of structurally distinct compounds in combination or in series might represent an

(42)

efficient strategy for combating the resistance caused by AR mutations. (Fujita and Nonomura, 2019; Romanel et al., 2015; Bambury and Scher, 2015)

2.4.3 Rational design of AR ligands

Because the resistance to a certain structure is a common problem in the use of antiandrogens, much work has been done in developing new types of

antiandrogens. The resistance to the antiandrogens flutamide and bicalutamide was associated with over-expression of the AR, suggesting that a more potent AR inhibitor without partial agonist activity could be used in these cases. This was the starting point for the development of enzalutamide. The new structure was found by developing a library of compounds based on a known AR binder and testing them in prostate cancer cell lines. (Bhattacharya et al., 2015)

Another approach was taken by Bassetto et al. Their work started with the observation that one of the most common mutants of the AR, W741L, converted bicalutamide into an agonist. The reason for this phenomenon lies in the flexible structure of bicalutamide. Mutation of W741 to a smaller residue creates more space within the ligand binding pocket of AR, and bicalutamide can be twisted to fill that space, thus leading to the correct positioning of the AR helix 12 for activation. Bassetto and coworkers aimed to design a structure that would resemble bicalutamide, but would be more rigid to avoid the loss of antagonistic activity caused by mutations. For this purpose, the structural features of

bicalutamide were combined with another, more rigid, AR binder to create a new compound (Figure 10). (Bassetto et al., 2017)

Figure 10. Rational design of new antiandrogens. Figure redrawn from (Bassetto

(43)

Prekovic et al. used docking to examine the effects of AR mutations on ligand responses (Prekovic et al., 2016). For example, the F877L mutation expands the ligand binding pocket of AR and allows enzalutamide to bind and behave like an agonist. Similar conclusions had been made earlier based on MD simulations (Balbas et al., 2013). The authors suggested that these observations should be taken into account when designing the next generation antiandrogens, which could maintain their antagonist activity even in mutated receptors.

2.4.4 The estrogen receptor

The estrogen receptor (ER) is one of the steroid receptors belonging to the NR superfamily. It has two subtypes, named ERα and ERβ. The subtypes are highly homologous in the DBD, but only 56 % in the LBD. In the ligand binding pocket, however, there are only two amino acids that are different between ERα and ERβ.

Human ERα has 595 amino acids (66 kDa) and ERβ contains 530 amino acids (54 kDa). ERs have a C-terminal F-domain (see Figure 3), which affects the activity of the ERs and may contribute to the selective activation of specific target genes. The natural ligand for both isotypes is 17β-estradiol (E2). ERs function as homodimers.

(Liu, J., 2020; Begam et al., 2017; Moon et al., 2009; Dahlman-Wright et al., 2006) In some organs, both isotypes are expressed at similar levels, but ERα or ERβ may predominate in some tissues or different cell types in the same tissue. ERα is mainly expressed in uterus, testes, bone, breast, liver and white adipose tissue while ERβ tends to be expressed in colon, testes and vascular endothelium. In ovary, prostate and brain, both isotypes are expressed but in different regions. ERs have a similar domain structure as the other receptors in the NR superfamily.

(Begam et al., 2017; Moon et al., 2009)

ERα and ERβ have their own roles and functions in the immune, skeletal, cardiovascular and central nervous systems. Their activation can even have opposite effects, and the final outcome depends on the balance between ERα and ERβ signalling. Despite the high similarity of the ligand binding pockets, it has been possible to develop ligands that are selective for either ERα or ERβ. (Begam et al., 2017; Moon et al., 2009)

2.4.5 Selective estrogen receptor modulators

Compounds which have high affinity for ERs but not for other steroid hormone receptors, are called selective estrogen receptor modulators, or SERMs. The interesting feature of these compounds is that they can display tissue specific agonistic or antagonistic activity, that is, behave as agonists or antagonists

(44)

depending on the tissue type. This kind of feature is of major interest in drug development because it offers a possibility to avoid undesired side effects outside of the target tissue. A good example is tamoxifen, which acts as an antagonist in the breast but as an agonist in bone, so it can be used simultaneously to treat osteoporosis and prevent breast cancer. These opposite outcomes are a result from the cell surroundings in different tissues, varying levels of expression of coactivator and corepressor proteins and phosphorylation of ER. A ligand binding to ER can change the shape of ER and in this way, also the interactions it can have with its coregulators. However, e.g. high level of expression of coactivators may convert the equilibrium from antiestrogenic behaviour in one cell type to

estrogenic behaviour in another type of cells. (Maximov et al., 2013; Dutertre and Smith, 2000)

Docking and MD have been used in attempts to predict the agonistic or

antagonistic properties ER binders. (Cotterill et al., 2019; Puranik et al., 2019; Ng et al., 2014) Docking into the agonistic and antagonistic crystal structures of ER can be used to distinguish agonistic and antagonistic binders. However, it is unlikely that these methods can accurately predict the very subtle differences in agonistic and antagonistic behavior, especially in the case of tissue specific effects.

Figure 11. Structures of estradiol and selective estrogen receptor modulators. All

(45)

2.4.6 Clinical use of ER antagonists

The first generation SERMs, tamoxifen and toremifene (Figure 11), are still in clinical use in the treatment of breast cancer. In addition to their antagonist activity in breast and agonistic activity in bone, tamoxifene also has agonistic activity in uterus; with long term use, it may increase the risk of endometrium cancer. Toremifene resembles tamoxifene in terms of both structure and

biological activity. A second generation compound, raloxifene, is indicated for the treatment and prevention of osteoporosis. It does not have a similar agonistic effect in uterus i.e. it differs from tamoxifen and toremifene. (Liu, J., 2020; Dutertre and Smith, 2000)

Basedoxifene is a third generation SERM which has a strong antagonistic effecs in breast and endometrium. It is used to treat vasomotor symptoms in

menopausal women. Ospemifene is structurally similar to tamoxifen and toremifene with similar biological properties. Lasofoxifene, another third generation SERM, is used in the treatment of osteoporosis and it is being

evaluated to treat breast cancer. Arzoxifene failed in phase III trials and has been discontinued. (Li, H. et al., 2020; Liu, J., 2020; Dutertre and Smith, 2000)

(46)

(47)

3 CO PUTATIONAL ETHODS

3.1 MD SIMULATIONS

3.1.1 Basic concepts

Structural information of biomolecules from crystallographic or nuclear magnetic resonance (NMR) measurements is needed for MD simulations. When the

positions of the atoms in the system are known, it is possible to calculate the forces exerted on each atom by other atoms in the system. The forces are used to calculate the new positions of the atoms after a short period of time, and the new forces exerted on the atoms. By repeating this sequence of calculations, a

trajectory of atoms moving over time is produced. Thus, it is possible to calculate the movements of the molecule. The bonds between atoms cannot form or break, so it is not possible to model chemical reactions with a classical MD simulation.

(Hollingsworth and Dror, 2018; Durrant and McCammon, 2011)

The forces described above are calculated using a molecular mechanical force field. The term ‘force field’ refers to a mathematical equation and its parameters, which are used to calculate the potential energy of the system. Typically, the force field contains definitions for different atom types, parameters for bonds, angles and torsions as well as for non-bonded interactions. An example of a force field equation for calculating the forces in the system is shown in equation (1).

(Monticelli and Tieleman, 2013)

𝐸𝑡𝑜𝑡𝑎𝑙= ∑ 𝑘_𝑑

2 (𝑑 − 𝑑₀)²+

𝑏𝑜𝑛𝑑𝑠

∑ 𝑘_𝜃

2 (𝜃 − 𝜃₀)²+

𝑎𝑛𝑔𝑙𝑒𝑠

∑ 𝑘_𝜑

2 (𝜑 − 𝜑₀)²+

𝑖𝑚𝑝𝑟𝑜𝑝𝑒𝑟 𝑑𝑖ℎ𝑒𝑑𝑟𝑎𝑙𝑠

∑ 𝑘∅

2 [1 + cos(𝑛∅ − ∅₀)] + ∑ 4𝜀_𝑖𝑗[(𝜎_𝑖𝑗 𝑟_𝑖𝑗)

12

− (𝜎_𝑖𝑗 𝑟_𝑖𝑗)

6

]

𝑖<𝑗 𝑑𝑖ℎ𝑒𝑑𝑟𝑎𝑙𝑠

+ ∑𝑞_𝑖𝑞_𝑗 𝜖𝑅_𝑖𝑗

𝑖<𝑗

(1)

The first term in the equation represents the bond stretching with a harmonic potential. kd represents the force constant and d0 the equilibrium bond length.

Angle bending and improper dihedrals are described analogously. The improper dihedrals term is used for out-of-plane bending of planar groups.

Dihedral-term is applied on torsion angles for atoms separated by three bonds.

In this term, k∅ is a force constant related to the barrier height. The number of

(48)

minima in the energy function is determined by n. The phase factor ∅0 determines the positions of the minima.

The last two terms represent non-bonded interactions. The non-bonded interactions are usually calculated for atoms which are not connected with bonds or which are separated by at least three bonds, but a scaling factor can be applied on non-bonded interactions, which are also covered by torsion terms. The second- last term is called Lennard-Jones 12-6 potential, and it represents the van der Waals component of the potential energy. In this term, σ is the distance defined for each pair of atom types, in which the potential between the two atoms in question is zero. In shorter distances there is a repulsive force, and in longer distances there is an attractive force, the strength of which is determined by ε. The last term in the equation represents the electrostatic interactions between atoms with (partial) charges (q). ϵ is the dielectric constant.

The force field parameters are developed to reproduce quantum mechanical or experimentally observed geometries and energies in chosen simulation

conditions. A number of force fields are currently available for simulation of biological macromolecules, e.g. Amber (Cornell et al., 1995), Gromos (Oostenbrink et al., 2005), OPLS (Harder et al., 2016) and CHARMM (Mackerell, 2004). The major force fields applied in MD simulations have also parameters for various polar and apolar solvents, common organic molecules, proteins, carbohydrates, nucleic acids and some common phospholipids. (Monticelli and Tieleman, 2013)

The force field is used to calculate forces on each atom, and the Newtonian equation of motion is then used to calculate the accelerations and new positions of the atoms after a short time step. Verlet (Verlet, 1967) and leap-frog (Hockney, 1970) algorithms are common methods to numerically integrate the equation in a MD simulation. (Allen, 2004; Leach, 2001)

The MD simulations are usually performed in a certain temperature and pressure to mimic in vivo or in vitro conditions. The temperature in the MD simulation is calculated from the kinetic energy of the system. Controlling the temperature of the system can be done by scaling the velocities of the atoms in the system. Commonly used algorithms for temperature control in MD simulations are Berendsen (Berendsen et al., 1984) and Nosé-Hoover (Hoover, 1985; Nosé, 1984) algorithms. Constant pressure conditions are maintained by changing the volume of the system. Berendsen (Berendsen et al., 1984) and Parrinello–Rahman (Parrinello and Rahman, 1981; Parrinello and Rahman, 1980) are commonly used algorithms for pressure control. (Leach, 2001)

(49)

3.1.2 Challenges and limitations of MD simulations

MD simulations are used to study dynamic properties of molecules. This kind of knowledge is difficult to get with other methods. However, a series of calculations for a system with many atoms requires a lot of computational capacity, and this task is currently not possible without a simplification of the description of the system. MD simulation uses a force field to approximate the quantum mechanical wavefunction, reducing the computational task from quantum mechanics to classical mechanics, ignoring the electronic motions. (Vanommeslaeghe et al., 2014; Guvench and MacKerell, 2008; Leach, 2001)

The outcome of the MD simulation is crucially dependent on the success of the parametrization of the force fields, which are constantly developed futher for better accuracy. Different force fields may have a different approach on their parametrization and the equations may contain different terms. The performances of the force fields may also differ, when the geometries they produce are

compared to e.g. NMR measurements. A force field parametrized for a certain specific purpose will perform better in that purpose, compared to a more general-purpose force field. In addition, a classical MD force field uses partial charges assigned on each atom before the simulation starts (Jorgensen, 2007).

Ignoring electronic motion means that electronic polarization on response to surroundings of the atoms is not captured. Polarizable force fields have been developed which improve accuracy, but with the cost of increasing computational time (Cieplak et al., 2009). (Rahman et al., 2020; Henriques et al., 2015; Beauchamp et al., 2012; Durrant and McCammon, 2011)

Interesting biological processes, such as ligand binding and conformational changes of protein, happen on timescales from nanosecond to microsecond or longer. On the other hand, the length of the time steps in a classical all-atom MD simulation must be short, typically 1–2 femtoseconds, to avoid instability, so a huge number of time steps are required for a long timescale MD simulation. Thus, the computational capacity currently available also limits the length of the

simulations. (Hollingsworth and Dror, 2018; Kmiecik et al., 2018)

3.2 FREE ENERGY CALCULATIONS

Free energy calculations can be used to provide estimates of free energy

differences between two states of a system, for example, a protein with a bound and unbound ligand. Binding of a drug to its target protein is required for its therapeutic effects, and thus it is important in drug design to be able predict

Computational studies on nuclear receptors : mechanism of allosterism and design of novel ligands

Dissertations in Health Sciences

uef.fi

TUOMAS ORAVILAHTI

TUOMAS ORAVILAHTI

COMPUTATIONAL STUDIES ON NUCLEAR RECEPTORS:

MECHANISM OF ALLOSTERISM

AND DESIGN OF NOVEL LIGANDS

COMPUTATIONAL STUDIES ON NUCLEAR RECEPTORS: MECHANISM OF ALLOSTERISM AND

DESIGN OF NOVEL LIGANDS

Tuomas Oravilahti

COMPUTATIONAL STUDIES ON NUCLEAR RECEPTORS: MECHANISM OF ALLOSTERISM AND

DESIGN OF NOVEL LIGANDS

ABSTRACT

TIIVISTELMÄ

ACKNOWLEDGEMENTS

LIST OF ORIGINAL PUBLICATIONS

CONTENTS

ABBREVIATIONS

1 INTRODUCTION

2 REVIEW OF THE LITERATURE

2.1 NUCLEAR RECEPTORS

2.2 THE STRUCTURE AND ACTIVATION OF THE NR

2.3 ALLOSTERIC ACTIVATIONS OF NR COMPLEXES

2.4 STEROID RECEPTORS AND THEIR LIGANDS

3 CO PUTATIONAL ETHODS

3.1 MD SIMULATIONS

3.2 FREE ENERGY CALCULATIONS