Kemian Nobel 2013. Jaakko Akola

Kemian Nobel 2013 Jaakko Akola Materiaalit ja molekyläärinen mallinnus Fysiikan laitos Tampereen teknillinen yliopisto & Laskennallisen nanotieteen huippuyksikkö (COMP) Aalto yliopisto

Sisällysluettelo Kemian Nobelin palkinnon historiaa Nobelistit Vuoden 2013 palkinnon perusteet Tietokonesimulaatiot, mallinnus, supertietokoneet Simulaatiomenetelmiä, esimerkkejä, multiskaalamallinnus QM/MM-menetelmä Esimerkkejä: Au-nanopartikkelit, liuotetut elektronit 2

Kemian Nobel Kemia oli Alfred Nobelin työn ja keksintöjen kannalta kaikista olennaisin tieteenala, ja se on mainittu toisena alana hänen testamentissaan. Palkinto jaettu v. 1901 lähtien 105 kertaa (poikkeuksena vuodet 1916, 1917, 1924, 1933, 1940, 1941, 1942) 166 palkinnon saajaa, 63 kertaa yhdelle henkilölle, vain 4 kertaa naiselle (M. Curie), 1 henkilö saanut kahdesti (F. Sanger) Keski-ikä 58 vuotta, nuorin 35 (F. Joliot) ja vanhin 85 (J.B. Fenn 2002) Artturi Ilmari Virtanen v. 1945, "tutkimuksistaan ja keksinnöistään maatalous- ja ravintokemian alalla, erityisesti hänen rehunsäilytysmenetelmästään à AIV-rehu Kvanttikemiallisista laskentamenetelmistä Nobel 1998, W. Kohn, J. Pople 3

Kemian Nobel 2013 for the development of multiscale models for complex chemical systems monimutkaisten kemiallisten systeemien multiskaalamallinnusmenetelmien kehittämisestä??? 4

Nobelistit: Martin Karplus Yhdysvaltain ja Itävallan kansalainen Syntynyt 1930 in Wienissä Ph.D. 1953, California Institute of Technology, CA, USA Professeur Conventionné, Université de Strasbourg, France; Theodore William Richards Professor of Chemistry, Emeritus, Harvard University, Cambridge, MA, USA 5

Nobelistit: Michael Levitt Yhdysvaltojen, Iso-Britannian ja Israelin kansalainen Syntynyt 1947 Pretoriassa, Etelä-Afrikassa Ph.D. 1971, University of Cambridge, UK Robert W. and Vivian K. Cahill Professor in Cancer Research, Stanford University School of Medicine, Stanford, CA, USA 6

Nobelistit: Arieh Warshel Yhdysvaltojen ja Israelin kansalainen Syntynyt 1940, Kibbutz Sde- Nahum, Israel Ph.D. 1969, Weizmann Institute of Science, Rehovot, Israel Distinguished Professor, University of Southern California, Los Angeles, CA, USA 7

Mitä on tietokonemallinnus? Molekyylien mallinnus perustui aikaisemmin fysikaalisiin rakennelmiin, malleihin (esim. pallot ja tikut) Nykyään mallinnus tehdään tietokonesimulaatioita käyttäen, mallinnus on siis siirtynyt ns. kyberavaruuteen Tämän vuoden nobelistit tekivät uraa uurtavaa työtä 1970-luvulla tehokkaiden tietokonemenetelmien kehittämisessä Erityisesti nobelistit ansioituivat klassisen fysiikan ja kvanttimekaniikkaan pohjautuvien menetelmien yhdistämisessä, jonka ansiosta pystymme nykypäivänä tutkimaan monimutkaisia kemiallisia prosesseja, esim. kasvien fotosynteesiä tai katalyysireaktioita autojen pakoputkistossa Tänä päivänä tietokone on kemistille yhtä tärkeä työkalu kuin koeputki. Simulaatioden tarkkuus mahdollistaa, että niiden avulla voidaan ennustaa ilmiöitä ennen kuin varsinaisia kokeita on edes tehty. 8

Luonto Teoreettinen malli Koe Simulaatio Teoria Kokeellinen mittaus Simuloitu ennustus Teoreettinen ennustus Vertailu Luonnonilmiön ymmärtäminen Teorian/mallin testaus 9

Supertietokoneet Tietokonesimulaatiot ovat nykyään rinnakkaistettuja, ja ne vaativat runsaasti laskentaresursseja CRAY XC30 (SISU), CSC Tieteen tietotekniikan keskus, Espoo Cray XC30 supertietokoneen ensimmäinen vaihe käsittää neljä kabinettia, joihin on asennettu 1472 kappaletta 2.6 GHz E5-2670 suoritinta, kussakin 8 laskentaydintä. Näistä 11776 laskentaytimestä saadaan yhteensä 244 TFlop/s teoreettinen maksimiteho. Kahden prosessorin (16 ytimen) laskentasolmut on yhdistetty Crayn nopealla Aries kytkentäverkolla, joka mahdollistaa koko koneen suuruiset rinnakkaistyöt. Sisun toinen vaihe asennetaan vuonna 2014, jolloin laskentakapasiteetti ylittää petaflopin sekunnissa. 10

Jatkoa IBM Blue Gene/Q (JUQUEEN) supertietokone Forschungszentrum Jülichissä, Saksassa Numero 1 Euroopassa, TOP10 maailmassa Prosessori: IBM PowerPC A2, 1.6 GHz, 16 ydintä per noodi 28 672 noodia, 458 752 laskentaydintä Huipputeho 5.9 Pflop/s Supertietokoneet kuluttavat valtavasti sähköä, ja niiden jäähdytys aiheuttaa rakennusteknisiä haasteita. Lisäksi ne vanhenevat jopa muutamassa vuodessa. Kemian Nobel 2013 8.1.2014 11

her traditional simulation methods thods Quantum mechanical methods -Solve the Schrödinger equation antum mechanical methods -Computationally Quantum mechanical very intensive methods - olve Can the Schrödinger equation -Solve be combined the Schrödinger with the equation other omputationally very intensive -Computationally methods very intensive an be combined with the other - Can be combined with the other ethods methods -Brownian/Langevin dynamics Stochastic Stochastic simulations simulations Mallinnus- / simulaatiomenetelmiä -Brownian/Langevin dynamics Kvanttikemialliset menetelmät random numbers to generate escale Monte may Carlo not be -menetelmät well-defined Stochastic simulations -No -Brownian/Langevin hydrodynamics dynamics -No hydrodynamics -No hydrodynamics Monte Carlo Coarse-grained models Monte Carlo nsitions Schrödingerin the system yhtälö, -Use random Monte elektronit numbers Carlo mukana, to generate raskas transitions putationally in Monte the fast system Carlo -Computationally se random numbers to generate -Use random numbers fast to generate ransitions -Timescale in the system transitions may in the not system be well-defined omputationally fast imescale may not be well-defined -Timescale may not be well-defined Atomitason simulaatiot - Integrate over secondary properti Coarse-grained models Molekyylidynamiikkaa (MD tai MM) - Integrate Coarse-grained atomitasolla, over secondary models Newtonin properties lait -Computationally Coarse-grained fast models - Integrate -Microscopic over secondary properties - Integrate Karkeutetut over information secondary mallit lostproperties -Microscopic information lost -Microscopic information lost -Microscopic information lost Satunnaisluvut määräävät systeemin muutokset odelling of bio- and nanosystems, TUT fall 2008 g of bio- and nanosystems, TUT fall 2008 lling of bio- and nanosystems, TUT fall 2008 ar modelling of bio- and nanosystems, TUT fall 2008 Atomitason kuvaus menetetty, laskennallisesti nopeita 12

Kvanttikemialliset simulaatiot: esimerkki molecular mechanics simulations provide additional information. 8122 J. Phys. Chem. B, Vol. 110, No. 15, 2006 Akola and Jones In section II we outline the methods of calculation that we have used, and the results are presented in section III. Our concluding remarks are given in section IV. ATP:n hydrolyysi (reaktio) aktiinissa II. Methods of Calculation 8128 J. Phys. Chem. B, Vol. 110, No. 15, 2006 Akola and Jones The Car-Parrinello molecular dynamics (CPMD) package 7 is based on DF theory and was used for most of the calculations in the companion paper. 8 The approximation of Perdew, Burke, ~200 atomia, and Ernzerhof (PBE) is used for the exchange-correlation elektronit energy, 9 and the mukana plane wave basis set has a kinetic energy cutoff of 80 Ry. The electron-ion interaction is described by nonlocal, norm-conserving, and separable pseudopotentials; 10 for the metal cations Na +, Mg 2+, and Ca 2+ Figure 11. Snapshot of ADP Pi-actin after 5 ns of molecular dynamics (blue ribbons) compared to the initial crystal structure of ATP-actin J. Akola J. Phys. & R. Chem. O. B, Vol. Jones, 110,, the contribution of semicore (redno. ribbons). 2006 15, The actin subdomains 8127 are labeled according to the standard notation. ADP electrons is included as discussed in ref 3- and H2PO4 8. The - are shown the licorice format, and the Mg cation is represented by a greeninteraction sphere (simulation). between (charged) replicas is avoided Comparisonby of ATP- solving and ADP P the i-actin Poisson shows that the catalytically important equation using the method of Hockney. 11 Gln137 and His161 side chains are accessible to the separated phosphate. Geometries The Gln137 side are chain the ATP state does not connect to the terminal phosphate at optimized using the simulated annealing any time, while theprocedure, hydroxyl group of H 2PO where the - 4 binds occasionally to the carbonyl oxygen of Gln137. Similarly, His161 does backbone atoms (N, C R, C) and not reach the terminal sidephosphate chainbefore terminating hydrolysis, but it forms frequent H-bonds with the separated phosphate via its protonated aliphatic carbons are fixed toimidazole preserve nitrogen (N ɛ). Gln137original and His161 are important actinfor hydrolysis and the final phosphate cleaving process, where conformation. H 2PO - 4 passes through the active site back door to yield ADP-actin. The initial protein coordinates were 35 Trajectories of γ-phosphate in contact with H taken from the X-ray 2O work. To accommodate a proton transfer, the protein has to show diminished mobility, and in only one case (WAT1, ATPactin) did a water molecule leave the nucleotide binding cleft Indications of environment-assisted deprotonation are observed modify the pk a of WAT1 via a conformational change. measurement of globular yeastwhile actin being replaced (G-actin) by another. bound to a gelsolin segment (PDB code 1YAG) with a metastable Mg 2+ in reaction 3 as the H-bond between the imidazole group of His161 (base) and WAT2 shortens during H - 3O + formation. It IV. Discussion and Concluding Remarks ATP complex inside. 5 has been proposed that in the reactive conformation His161 Our earlieratp work hydrolysis on at the ATP active site hydrolysis of actin has been studied should bind to WAT1, as in the X-ray measurement for Li in + - water 4 using a DF method. The active site is approximated by the ATP actin, and this is consistent with the large fluctuations in showed that ATP can reactive be triphosphate approximated tail of ATP complexed by methyl with Mg 2+, the the His161 side chain location found in our molecular mechanics surrounding water molecules, and the nearest actin residues. This simulations of ATP-actin. However, the mobility of Gln137 triphosphate (MTP), and the present model systemreactions includes the reactive are cavity based of fouron water is not enhanced, and it is an unlikely base for WAT1, which MTP 4- complexed with Mg 2+ molecules (WAT1-WAT4) and γ-phosphate, all triphosphate prefers to transfer a proton to WAT2 as n H(O w) is decreased. coordinating. The residues, active and asite full hydration of actin shell of Mgiswith The probable role of Gln137 in ATP hydrolysis is to place modeled by selecting MTP, 4- stabilizing Mg 2+ aspartate side chains. The large system size enables WAT1 in a preferable position and orientation. us to study, the important likely aspects in catalytic the ATP hydrolysis water (reaction The four reactions lead to different protonation and coordination of end products. The most energetically favorable situation (WAT709), 5 paths, catalytically important residues, proton wires, role of and the actin residues divalent and cations, water and coordination oxygens of endwithin products), and 5 it is achieved when a nonbridging O γ (not linked to Mg 2+ )is includes a significant portion of the electrostatic field imposed protonated resulting in H 2PO Å. This model contains 18 residues by the whole and protein. 10Thewater calculationsmolecules, - 4 and ADP 3-, which are coupled use an extended plane by a H-bond. Classical simulations (8 ns) of ADP P i-actin wave basis and are very demanding terms of CPU time and confirm the importance of the phosphate orientation where the Figurewhere 10. Optimized missing end products. hydrogens The perspective have memory. been is similar added to Figure using simple initial Figure Mg-O γ linkage 1. Simulated is preserved. The total active energy (-2.1 sitekcal/ of actin shown from two perspectives. 1a. Notice chemical that H 2 PO rules. 4- in The the middle water is molecules attachedfour toreaction the are other paths have placed phosphate been modeled by constraining certain mol) to saturatekemian the Nobel (a) is slightly Front below 2013 view: the starting The leveltriphosphate but less than typical tail 8.1.2014 of ATP, water molecules, 13 and coordinates, and the first gives the simplest picture of dissociative path as the bridging P γ-o s bond is stretched. A barrier of mol). residue side chains are plotted in licorice representation, and the Mg values for ATP hydrolysis in aqueous solution (-5 to-10 kcal/ via an H-bond and an O-Mg-O bridge. hydrogen bond (H-bond) network, 28.8and kcal/mol some is obtained residue at 2.8 Å, after side whichchains 15,36,37 We suspect that the complete cleavage of H 2PO - 4 the metaphos- can release additional energy. Comparison of ATP- and ADP cation is the magenta sphere. (b) Side view: The triphosphate and phate 2+ formed is approached by WAT1, which is in contact with transfers a proton to the phosphate via a proton wire (WAT2 and WAT3). This path supports the suggestion that WAT1 is the catalytic water, because its location enables an in-line nucleophilic attack on the terminal phosphorus. However, the local dipole moment of WAT1 (3.1 D) in the initial configuration differs little from that of bulk water. The other reactions employ two coordinates that control (a) the nucleophilic attack of the catalytic water as the β- and γ-phosphates are separated and (b) the protonation number of WAT1. The three paths display different orders of events and can be characterized as dissociative (reaction 2) or associative (reactions 3 and 4). The lowest-energy barrier is for reaction 2, where the terminal phosphate is cleaved (20.4 kcal/mol, TS2a), and the deprotonation of WAT1 follows (21.0 kcal/mol, TS2b). These two events compensate for each other energetically and lead to the flat plateau in total energy between TS2a and TS2b. In the structure TS2a the dipole moment of WAT1 is relatively small (3.2 D) and the interaction with the metaphosphate is noncovalent (P γ-o w distance 2.4 Å). The reaction is completed when a proton is transferred to a carboxylate group of Asp154, resulting in a cleaved HPO 4 2-. In the associative reaction 3, deprotonation of WAT1 results in a reactive OH - ion, which then makes a nucleophilic attack on P γ.thebarrierobtained(39.8kcal/mol,ts3)issignificantly higher than those for the previous reactions, but it agrees well with our study of ATP hydrolysis in water (39.1 kcal/mol) 4 and the results of Okimoto (42.0 kcal/mol) for ATP hydrolysis in an isolated active site of myosin. 31 The corresponding transition state contains a pentacoordinated phosphorus with P γ-o s and P γ-o w distances of 2.1 Å, and the ELF shows electron-gaslike localization of electron density that indicates chemical bonding. Reaction 4 displays features that are very similar to those of reaction 3, although nucleophilic attack is concurrent with the deprotonation of WAT1. The barrier is 37.6 kcal/mol. Associative reactions are energetically unfavorable, perhaps because deprotonation of WAT1 costs a significant amount of P -actin structures reveals few differences (especially at the

Atomitason simulaatiot: esimerkki 1 Cholesterol modulates glycolipid conformation and receptor activity. Nature Chemical Biology 7, 260-262 (2011). Prof. Ilpo Vattulaisen ryhmä, TTY, Fysiikan laitos 14

irst the ain ms. attaeal em acitful o- isof ot en the ing in,, it of ith ny ldof ed ng perature.] In each of the two equilibrium simulations, FiP35 underwent multiple folding and unfolding transitions, for a total of 15 events in the two simulations (Fig. 2A). We found the folding time, as calculated from the average waiting time in the unfolded state, to be 10 T 3 ms, in relatively close agreement with the experimental folding time [14 ms (12)]. The population of the folded state in the simulations was 60%. In our simulations, a well-defined sequence of events leads from the disordered, unfolded (26, 27), as have computational studies in the context of the diffusion-collision model for protein folding (28). Our simulations show that both hairpins also form with similar rate constants in the context of the full WW domain. The first hairpin forms with atimeconstantof5t 1 ms, slower than in isolation by a factor of 4; such a slowdown is greater than expected for a simple hydrodynamic drag. To determine its origin, we performed simulations in which we systematically added residues Atomitason simulaatiot: esimerkki Fig. 1. Folding proteins at x-ray resolution, showing comparison of x-ray structures (blue) (15, 24) and last frame of MD simulation (red): (A) simulation of villin at 300 K, (B)simulationofFiP35at 337 K. Simulations were initiated from completely extended structures. Villin and FiP35 folded to their native states after 68 ms and 38 ms, respectively, A B and simulations were continued for an additional 20 msafterthefoldingevent toverify thestability ofthe native fold. Atomic-Level Characterization of the Structural Dynamics of Proteins Science 330, 341 (2010). David E. Shaw n ryhmä (D.E. Shaw Research, New York) Kaksi proteiinia, jotka laskostuvat 68 µs ja 37 µs kuluessa simulaatiossa Sininen: simulaatio; Punainen: kokeellinen rakenne TOBER 2010 VOL 330 SCIENCE www.sciencemag.org 15

Multiskaalamallinnus Regimes of different simulation methods Other traditional simulation methods Time (s) Aika (s) 1 10-3 10 Other -6 traditional simulation methods 10-9 10-12 10-15 Molecular modelling of bio- and nanosystems, TUT fall 2008 Quantum mechanical methods -Solve the Schrödinger equation -Computationally very intensive - Can be combined with the other methods QM Classical MD QM Monte Carlo, Coarse- Graining -Brownian/Langevin MM dynamics karkeutetut Other traditional -No hydrodynamics simulation methods -Solve the Schrödinger equation -Computationally very intensive - Can be combined with the other methods Monte Carlo -Use random numbers to generate transitions in the system -Timescale may not be well-defined Quantum mechanical methods Molecular modelling of bio- and nanosystems, TUT fall 2008 -Use random numbers to generate transitions in the system -Timescale may not be well-defined Quantum mechanical methods -Solve the Schrödinger equation -Computationally very intensive Monte Carlo -Use random numbers to generate transitions in the system Stochastic simulations Finite Element Methods Linear size (m) Continuum Models 10-12 10-9 -Microscopic information 10-6 -6 lost 10-3 -3 1 Stochastic simulations - Can be combined with the other Monte Carlo Coarse-grained models methods - Integrate over secondary properties -Microscopic information lost -Timescale may not be well-defined Stochastic simulations -Brownian/Langevin dynamics -No hydrodynamics Monte Coarse-grained models - Integrate over secondary properties Carlo -Microscopic information lost -Brownian/Langevin dynamics -No hydrodynamics Coarse-grained models - Integrate over secondary properties Finite elements = hilamalli Pituusskaala (m) Jatkumomallit Molecular modelling of bio- and nanosystems, TUT fall 2008 Molecular modelling of of bio- and nanosystems, TUT fall 2008 16

Lisää atomiryhmiä, molekyylejä Aika elektronit mukana simulaatioita atomitasolla = jatkumo = karkeutus = MD tai MM, molekyylidynamiikka tai -mekaniikka = kvanttikemia, QM Pituus 17

QM/MM-menetelmä Yhdistetään kvanttikemian (QM) ja molekyylidynamiikan (MM) menetelmät Kaksi aluetta, joissa käytetään eri menetelmiä + vuorovaikutus näiden alueiden välillä (vaikein osuus) Ainoastaan tärkein osa kuvataan kvanttikemiallisesti QM: 1-100 atomia rajapinta MM: 1000-1000000 atomia 18

Esimerkkejä QM/MM-simulaatioista Kasvien yhteyttäminen, fotosynteesi L. Guidoni, King s College, Lontoo 19

Kultananopartikkeli vedessä Kvanttimekaniikka: Kultananopartikkeli (klusteri), jossa on 25 kulta-atomia (vihreät pallot) 18 orgaanista sivuryhmää (lakritsi) Klassinen fysiikka: Suolaa (natrium), vesimolekyylejä (paljon) Kemian Nobel 2013 8.1.2014 20

Proteiini ja Au-nanopartikkeli Kemian Nobel 2013 8.1.2014 21

Reaktio atomitasolla Pyrimme linkittämään nanopartikkelin proteiiniin à muodostetaan niiden välille Au-S sidos Tämän seurauksena nanopartikkelin pinnalla yksi Au-S rikkoutuu, ja lopulta yksi pinnan molekyyleistä irtautuu Kemian Nobel 2013 8.1.2014 22

Ylimääräinen elektroni vedessä Ennen Liuotettu elektroni Muodostuu säteilytyksen tuloksena, (esim. Olkiluoto) Pariutuu lopulta protonin H+ kanssa, ja muodostaa vetyatomin, H à H2 kaasua Elektroni ja ympäröivä vesi (33 kpl) à QM-simulaatio Jälkeen Muu vesi kuuluu MMsimulaatioon Kemian Nobel 2013 8.1.2014 23