A structural role for glycosylation

Background: Protein glycosylation, the covalent attachment of carbohydrates, is very common, but in many cases the biological function of glycosylation is not well understood. Recently, uorescence energy transfer experiments have shown that glycosylation can strongly change the global conformational distributions of peptides. We intend to show the physical mechanism behind this structural e ect using a theoretical model. Results: The framework of the hp-model of Dill and co-workers is used to describe peptides and their glycosylated counterparts. Conformations are completely enumerated and exact results are obtained for the e ect of glycosylation. On glycosylation the model peptides experience conformational changes similar to those seen in experiment. This e ect is highly speci c for the sequence of amino acids and also depends on the size of the glycan. Experimentally testable predictions are made for related peptides. Conclusions: Glycans can, by means of entropical contributions, modulate the free energy landscape of polypeptides and thereby specifically stabilize polypeptide conformations. With respect to glycoproteins the results suggest that the loss of chain entropy during protein folding is partly balanced by an increase in carbohydrate entropy. 2 Introduction Glycosylation, the covalent addition of a carbohydrate, is one of the most common modi cations of proteins (for reviews see e.g. [1, 2]). On the phenotype level defects in the glycosylation apparatus often manifest as severe and frequently fatal diseases. In many cases the molecular mechanisms behind these disorders are not understood. What seems certain by now is that glycosylation can have profound e ects on the structure and mobility of polypeptides. This insight comes mainly from NMR experiments with molecules corresponding to stretches of peptide around the glycosylation sites of some glycoproteins [3, 4, 5, 6, 7]. Due to the high exibility and the fast motions of the peptides, there are usually no clear nuclear Overhauser e ects. Therefore most of the information obtained from NMR experiments is about sequence local structure. Alternatively, CD spectroscopy has been used in cases where glycosylation is coupled to characteristic changes in secondary structure [8, 9]. Recently, Imperiali and Rickert have carried out uorescence energy transfer (FET) experiments on three glycopeptides and their non-glycosylated counterparts [10] to assess the structural e ect of glycosylation. From these FET experiments the distribution of distances between two uorophores at either end of the polypeptide can be calculated. This distribution gives a more global picture of the conformational distributions of the peptides. The investigation of Imperiali and Rickert shows that glycosylation can have quite diverse e ects on the global conformations of the peptides depending on the sequence of amino acids. The peptides in the FET experiments are N-glycosylated, i.e. the carbohydrate is attached via an amide bond to the 3 nitrogen of an asparagine sidechain. N-glycosylation is known to happen co-translationally and hence during or before the folding of the protein. The results of Imperiali and Rickert therefore suggest that glycosylation may have an impact on the folding process. The aim of this work is to theoretically elucidate the physical basis of the various e ects observed in the FET experiments, and to propose a biophysical role for glycosylation. Since the e ects observed in the FET experiments concern a coarse length scale, our theoretical model needs not to describe the molecule in atomic detail. Nevertheless to produce meaningful results the model should capture the essential features of the biopolymeric system. We use a slightly extended version of the hp-model put forward by Dill and others (for review see [11]). In this model biopolymers are represented as chains of point-like residues on a grid. There are only two types of residues: hydrophobic (h) residues and polar (p) residues (we use lower case letters for hp-model sequences to distinguish them from the amino acid sequences in the standard one letter code). Despite its simplicity the hp-model has been very successful in explaining a variety of kinetic and structural properties of proteins. A great advantage of the model is that for short chains all conformations can be enumerated. This complete enumeration yields exact results free of any statistical error. If we can show that the hp-model glycopeptides exhibit a behaviour similar to that seen in the experiments then, applying Occams razor, the essential physics captured in the simple hp-model should be su cient to explain the behaviour of real glycopeptides. We can then draw conclusions from our exact results which also should be valid for real glycopeptides. We will nd that a large part of the structural e ects seen in the experiments is probably 4 due to entropical contributions from the carbohydrate. Of course one cannot expect quantitative agreement between the coarse hp-model and experiment. But as we will show below the qualitative agreement is impressive and allows us to make some predictions which can be tested experimentally. Results and discussion Extended hp-model We use a slightly extended version of the hp-model of Dill and co-workers [11]. As in the orginial hp-model biopolymers are modeled as self-avoiding walks on a two dimensional square lattice. Each monomer occupies one lattice point and the lattice spacing corresponds to a typical distance between two neighbouring monomers, i.e. about 4 A. The low dimensionality of the lattice is no essential restriction since the short peptides considered here are either extended or have some turn conformation (Fig. 1); both groups of conformations can be modeled reasonably well in two dimensions. There are two types of residues: hydrophobic (h) and polar (p) residues. Interaction energies are -1 (in arbitrary units) for each nonbonded pair of h residues on neighbouring grid points, and zero for hp or pp pairs. Hence the conformational energy always equals the negative number of nonbonded hh pairs. We extend the original hp-model by two features in order to reconstruct the experimental situation as closely as possible within our coarse model. First, the biopolymers in our calculations are signi cantly branched: a GlcNAc (N-acetyl-D-glucosamine) dimer is attached to a p residue of the peptide chain corresponding to asparagine (right column of Fig. 1). The trivial second extension is that we not only consider amino acids but also two other monomers. First, the dansyl (Dns) chromophore with its hydrophobic naph5 thalene double ring is modeled as h residue. Second, each GlcNAc monomer is represented by two p residues: the AcHN group alone has the size of Gly, and the remaining glucose unit with six carbons and ve oxygens roughly has the size of a larger hydrophilic amino acid. Complete enumeration The hp-model on the two dimensional square lattice allows only for a few thousand conformations for the short (glyco-)peptides considered here. Hence we can completely enumerate all n possible conformations i and calculate exactly the thermodynamical average hxi of any observable x using [12] hxi = n Xi=1 xi exp( Ei=(kBT ))= n Xi=1 exp( Ei=(kBT )); (1) where xi is the value of x in conformation i, Ei is the energy of conformation i, kB is the Boltzmann constant and T is the absolute temperature. Ei is given by the hp-model and xi can be calculated directly from the conformation. The only free quantity is T , or more conveniently kBT in the same units as the hp-model energy. For very small values of kBT the conformational distribution is dominated by only a few conformations with the maximum number of hh-contacts. This is unrealistic because we know from experiment that these small peptides are highly exible, in contrast to proteins consisting of longer polypeptide chains. On the other hand for very high values of kBT we neglect all energetic contributions coming from hydrophobic contacts. In our calculations we choose a value of kBT = 0:6. This means e.g. that the conformations of lowest energy (Ei = 3) of the rst non-glycosylated peptide investigated below, which make up 30 out of 4067 conformations, have a weight of 28% in the partition function in Eq. 1. Hence, non-minimum energy conformations still prevail in the partition function and the peptide 6 will sample a large number of conformations as it does in the experiments. Fortunately, it turns out that the qualitative results are not sensitive to changes in temperature, as long as we prevent the conformational distribution from being dominated by a few low energy conformations by choosing very low temperatures. For comparison with the experimental results of Imperiali and Rickert we calculate the probability distribution of the distance between Trp ( uorescence donor), and the dansyl group ( uorescence acceptor). We obtain the distribution of donor-acceptor distances from Eq. 1 by setting xi equal to (rda;i; rj). The latter expression equals 1 if the donor-acceptor distance rda;i of conformation i is rj and 0 if this is not the case. The resulting thermodynamical average is the probability !j of nding conformations with a donor-acceptor distance of rj. Peptides and glycopeptides The peptides used by Imperiali and Rickert in their FET experiments are derived from three naturally occurring sequences of amino acids from two proteins (Fig. 1). Imperiali and Rickert add two chromophore containing groups at either end of the peptides and measure the distribution of distances rda between these chromophores. We calculate rda distributions for the hpmodels of these three peptides and their N-glycosylated versions, and we also try some predictions for related peptides with mutated sequences and di erent glycan lengths. In the FET experiments the glycopeptides have a GlcNAc dimer attached to the Asn sidechain. As motivated above we model this disaccharide as tetramer `pppp' (Fig. 1) because of its bulkiness and its hydrophilic character. 7 Peptide 1 Peptide 1 (top of Fig. 1) is derived from amino acids A19{A26 of in uenza hemagglutinin [13]. It has the sequence (Dan)OAVPNGTWV, where O is ornithine and Dan is the dansyl chromophore attached to the sidechain of this residue. The peptide sequence translates into the hp-model sequence hphhhppphh. Fig. 2 shows the distribution of donor-acceptor distances for peptide 1 and for the corresponding glycopeptides with 0 to 4 GlcNAc monomers. As in the FET experiments the distribution for the non-glycosylated peptide shows two peaks, a higher one at smaller values of rda, corresponding to narrow turn structures, and a smaller peak at greater values of rda, corresponding to more extended structures. Glycosylation with two GlcNAc residues causes an increase in the probability ! of the narrow turns by 13% of its previous value or 5% in terms of total probability, at the expense of ! for larger values. Experiment[10] also shows a stabilization of narrow turns of peptide 1 due to glycosylation. What causes this shifting of probability from extended to turn structures? To clarify the physical reason in the framework of the hp-model we note that for the non-glycosylated peptide the curve in Fig. 2 is obtained from !j = Pj (rda;i; rj) exp( Ei kBT ) Pj exp( Ej kBT ) : (2) with the quantities as introduced at the end of the above section on complete enumeration. Using the same terminology the curve for the glycosylated peptide results from !(g) j = Pj ni (rda;i; rj) exp( Ei kBT ) Pj ni exp( Ej kBT ) : (3) 8 Here ni = 0; 1; 2; : : : is the number of carbohydrate conformations for a peptide in conformation i. Since we have modeled the sugars as p-monomers, they can only contribute to the partition function via the degeneracy factors ni. We can recast Eq. 3 by drawing the ni into the exponents and arrive at !(g) j = Pj (rda;i; rj) exp( (Ei kBT lnni) kBT ) Pj exp( (Ej kBT lnni) kBT ) : (4) A comparison of Eqs. 2 and 4 shows that in the framework of the hp-model the carbohydrate only contributes entropical terms kBT lnni which modulate the free energy surface of the peptide. The \enthalpic" hh-contacts and the entropic contributions from the carbohydrate stabilize the conformations in the highest peak in Fig. 2 in a weakly cooperative manner. The reason for this cooperativity is that 1. the best way for the peptide to prevent interference with the glycan is to pair both ends, and 2. this pairing is also stabilized by relatively many hh-contacts. Hence, there is a mutual ampli cation of glycan entropy and hh-pairs between the ends of the peptide. In terms of Eq. 4 this means that structures with low values of Ei also have high values of ni. The shift to more turn-like low-energy structures upon glycosylation for peptide 1 is illustrated in Fig. 3. In order to quantify this cooperativity we consider the e ects of glycosylation and hh-contacts on the height of the turn peak separately. First, to quantify the e ect of the glycan alone we mutate all h-residues to p-resides. The height of peak of narrow turns due to the glycosylation increases slightly by 0.014 from 0.098 to 0.112 for this all-p peptide. Second, the e ect of hh-pairs is obtained by a comparison of the non-glycosylated all-p peptide and the nonglycosylated version of the original peptide 1 with sequence hphhhppphh. Here the height of the peak corresponding to narrow turns rises by 0.268 9 from 0.098 to 0.366. If the two e ects would be additive one would expect a combined increase of 0.014+0.268=0.282 to a total height of the peak of 0.380. Instead, the narrow turns have a probability of 0.414 for the glycosylated peptide 1, which means that we have in fact a weakly cooperative stabilization of turns due to glycosylation and hh-contacts. An important question is how the size of the glycan in uences the stability of the turn-like structures. Although disaccharides are not uncommon, glycoproteins often carry much larger and more branched carbohydrate trees [2]. Since the stabilization is of entropical nature and the entropy of the glycan increases with size one could expect that the stabilizing e ect shows the same tendency and that by only considering disaccharides we underestimate its magnitude. Indeed, we nd that as the glycan is elongated from none to four GlcNAc monomers, turn structures are preferred more and more with turn probabilities of 0.366, 0.379, 0.414, 0.429, and 0.437 (Fig. 2). The addition of the second monomer obviously provides the largest contribution. Interestingly, there is evidence from recent NMR experiments on a di erent set of glycopeptides that the addition of a second monomer can change the conformational distribution of glycopeptides signi cantly [6]. Peptide 2 Peptide 2 (middle of Fig. 1) is also taken from a turn in in uenza hemagglutinin. It corresponds to residues A282{A288 and its sequence after attachment of the chromophore at the N-terminus reads (Dan)OITPNGTWA. This translates into the hp-model sequence hphphppphh. This sequence differs from that of peptide 1 only in fourth position where the latter has a hydrophobic residue. Since this single mutation does not a ect the ability 10 to form hh-contacts between the ends, the donor-acceptor distance distribution is qualitatively similar for both peptides. This similarity is in good agreement with the experimental ndings [10]. As in peptide 1 narrow turns are increasingly stabilized with growing length of the glycan, and the second sugar monomer also contributes most to this stabilization (left side of Fig. 4). What happens if we mutate one of the h-residues at either end of peptide 2 which is critical for the turn formation, e.g. h10 at the C-terminus? This would correspond to a mutation like A10D of the real peptide. The model mutant h10p is no longer able to form hh-contacts and the conformational distribution changes drastically (right side of Fig. 4). Turns are here much less populated than structures with intermediate values of rda. Glycosylation of the mutant still strengthens somewhat turn structures but much less than in peptide 1 or in the wildtype of peptide 2. The comparison of the two parts of Fig. 4 again demonstrates the weak cooperativity of hh-contacts and glycosylation with respect to turn stabilization, since the turn-stabilization due to glycosylation is stronger for the original peptide with its ability to form hh-contacts than for the mutant. Peptide 3 Peptide 3 (bottom of Fig. 1) corresponds to residues H152{H161 of an immunoglobulin [14] (PDB code 2fbj). The modi ed peptide used in the FET experiments has the sequence (Dns)SGTMNVTWGK with the dansyl group appended at the N-terminus. The corresponding hp-model sequence is hppphphphpp. This sequence does not permit the formation of hh-contacts, hence the model peptide prefers more extended conformations than peptides 1 and 2. Interestingly, this is in analogy to the structures of the original pep11 tides in the their native protein environment where peptides 1 and 2 form turns whereas peptide 3 is part of an extended -strand (Fig. 1). The calculated rda distribution (not shown) is almost indistinguishable from that of the h10p mutant of peptide 2 (left side of Fig. 4) which is not surprising because both sequences cannot form hh-contacts, have the same positions of donor and acceptor, and di er in length by only one monomer. Although the detailed shape of the experimental rda distribution is not reproduced in the calculations, there are remarkable parallels between both distributions and their relations to those of peptides 1 and 2. As in the FET experiments the distribution obtained for peptide 3 is qualitatively di erent to that of peptides 1 and 2. For peptide 3 there is no clear shift of probability towards turns due to glycosylation. Instead, both experiment and calculation show a polarization of probabilities with small gains for turns and very extended conformations at the expense of conformations with intermediate values of rda. Since peptide 3 is very similar to the mutant h10p of peptide 2 it may be possible to apply a reverse mutation to peptide 3 and transform it to a peptide with a rda distribution similar to that of the wildtype of peptide 2. This can indeed be done by applying mutation p2h to peptide 3 (Fig. 5). The mutant p2h is able to form hh-contacts that stabilize more closed turn conformations. It clearly shows also the turn promoting e ect of glycosylation seen for peptides 1 and 2. This computational result suggests that experimentally a mutation like G2L applied to the original sequence of peptide 3 should lead to a measured rda distributions more similar to that seen for peptides 1 and 2. 12 E ect in terms of free energy We can express the e ect of glycosylation also in terms of free energy. The complete enumeration yields for the free energy of a subset sub with respect to the set of remaining conformations [12] Asub = kBT lnQsub=(Q Qsub); (5) where Q = Pi exp( Ei=(kBT )) and Qsub = Psub exp( Esub=kBT ) are the total and partial partition functions, respectively. A natural subset for proteins is the native state and the remaining conformations correspond to the unfolded state. For the short peptides investigated here which have no native state we have to make another physically sensible choice, e.g. the group of turn structures with rda 2 as \native" state. The stabilization of turns due to glycosylation is given by Arda 2 = Aglycosylated rda 2 Anot glycosylated rda 2 : (6) In the peptides investigated above we nd values of A of the order of 0.1 in units of hh-contact energy. Assuming a realistic value for hydrophobic contact energies of a few kJ/mol, glycosylation leads to a turn stabilization of the order of 1 kJ/mol. This e ect is small but measurable. In fact recent measurements for a number of glycoproteins suggest that the stabilization measured in terms of increase of melting temperature Tm is of the order of a few Kelvin at most [15, 16]. Implications for proteins Despite the simplicity of our model we have found qualitative agreement with experiment. The conclusion from our calculations for peptides is that glycosylation can amplify the preference for certain peptide conformations by 13 selectively increasing the entropy of these conformations. Short peptides likethose considered here and in the FET experiments are very exible and donot collapse to a single conformation. For a protein the situation is di erentbecause it condenses into a single native structure, sitting at the bottom ofa funnel shaped energy landscape (for a review see e.g. ref. [17]). Neverthe-less, the observations made for the peptides have also some implications forproteins.For the particular peptides considered here, according to both experimentand calculation, glycosylation promotes conformations which are similar tothose adopted by the peptides in their native protein environment. Thistendency is particularly clear for the turn-forming peptides 1 and 2, and lesspronounced but also present for the more extended peptide 3.In general it is found that carbohydrate chains are attached to the surfaceof the protein and are not incorporated into the bulk of the native structure.There may be hydrogen bonds or other contacts between glycan and aminoacids but overall the position of the carbohydrate can be described as pro-truding from the native polypeptide into the solvent [18], where it can moverelatively freely. However, this is probably not true for the unfolded, non-native polypeptide. For the unfolded chain both the conformational freedomof the glycan and of the polypeptide are restricted as a consequence of theexcluded volume of the carbohydrate. Hence, during the process of proteinfolding the carbohydrate may gain entropy whereas the polypeptide looseschain entropy. We think that one of the roles of glycosylation in protein fold-ing is to balance to some extent this loss of chain entropy experienced by thefolding polypeptide. The mechanism responsible for this e ect is the sameas in the peptides investigated above. Interestingly, DeKoster and Robert-14 son [16] also support the notion of an entropic stabilization since in theirdi erential calorimetry measurements they observe an increase in Tm uponglycosylation but no increase in enthalpy of unfolding.AcknowledgmentWe thank Oliver Eulenstein and Ulrich Essmann for helpful discussions. D.H.gratefully acknowledges support by the CompChem project of the High Per-formance Computing Center (HLRZ) at GMD-SCAI.References1. Lis, H. & Sharon, N. (1993). Protein glycosylation. Eur. J. Biochem.218, 1{27.2. Dwek, R. A. (1996). Glycobiology: Toward understanding the functionof sugars. Chem. Rev. 96, 683{720.3. Wormald, M. R., et al., & Dwek, R. A. (1991). The conformational e ectsof N-glycosylation on the tailpiece from serum IgM. Eur. J. Biochem.198, 131{139.4. Davis, J. T., Hirani, S., Bartlett, C. & Reid, B. R. (1994). 1H NMRstudies on an Asn-linked glycopeptide. J. Biol. Chem. 269, 3331{3338.5. Andreotti, A. H. & Kahne, D. (1993). E ects of Glycosylation on PeptideBackbone Conformation. J. Am. Chem. Soc. 115, 3352{3353.6. Liang, R., Andreotti, A. H. & Kahne, D. (1995). Sensitivity of Glycopep-tide Conformation to Carbohydrate Chain Length. J. Am. Chem. Soc.117, 10395{10396.15 7. Live, D. H., Kumar, R. A., Beebe, X. & Danishefsky, S. J. (1996). Con-formational in uences of glycosylation of a peptide: A possible modelfor the e ect of glycosylation on the rate of protein folding. Proc. Natl.Acad. Sci. USA 93, 12759{12761.8. Aubert, J. P., Helbecque, N. & Loucheux-Lefebvre, M. H. (1981). Circu-lar dichroism studies of synthetic Asn-X-Ser/Thr-containing: Structure-glycosylation relationship. Arch. Biochem. Biophys. 208, 20{29.9. Otvos, L., Thurin, J., Kollat, E., Urge, L., Mantsch, H. H. & Hollosi, M.(1991). Glycosylation of synthetic peptides breaks helices. Phosphoryla-tion results in distorted structure. Int. J. Pept. Protein Res. 38, 476{482.10. Imperiali, B. & Rickert, K. W. (1995). Conformational implications ofasparagine-linked glycosylation. Proc. Natl. Acad. Sci. USA 92, 97{101.11. Dill, K. A., et al., & Chan, H. S. (1995). Principles of protein folding {A perspective from simple exact models. Protein Science 4, 561{602.12. McQuarrie, D. A. (1976). Statistical Mechanics. Harper & Row, NewYork.13. Wilson, I. A., Skehel, J. J. & Wiley, D. C. (1981). Structure of thehemagglutinin membrane glycoprotein of in uenza virus at 3 A resolu-tion. Nature 289, 366{373.14. Suh, S. W., et al., & Davies, D. R. (1986). The galactan-binding im-munoglobulin Fab J539. An X-ray di raction study at 2.6 A resolution.Proteins: struc. funct. genet. 1, 74{80.16 15. Wang, C., Eufemi, M., Turano, C. & Giartosio, A. (1996). In uence ofthe carbohydrate moiety on the stability of glycoproteins. Biochemistry35, 7299{7307.16. DeKoster, G. T. & Robertson, A. D. (1997). Thermodynamics of Unfold-ing for Kazal-Type Serine Protease Inhibitors: Entropic Stabilization ofOvomucoid First Domain by Glycosylation. Biochemistry 36, 2323{2331.17. Wolynes, P. G., Luthey-Schulten, Z. & Onuchic, J. N. (1996). Fast-foldingexperiments and the topography of protein folding energy landscapes.Chemistry & Biology 3, 425{432.18. Imberty, A. & Perez, S. (1995). Stereochemistry of N-glycosylation sitesin glycoproteins. Protein Engineering 8, 699{709.19. Bernstein, F. C., et al., & Tasumi, M. (1977). The Protein Data Bank: acomputer-based archival le for macromolecular structures. J. Mol. Biol.112, 535{542.20. Kraulis, P. (1991). MOLSCRIPT: a program to produce both detailedand schematic plots of protein structures. J. Appl. Cryst. 24, 946{950.17 Figure captionsFigure 1: Conformations of original peptides (not glycosylated) taken fromprotein x-ray structures (left column) and hp-models of modi ed peptidesas used in the FET experiments [10] (right column). Labels indicate aminoacids in standard one letter code, and GlcNAc sugar moiety. In the rightcolumn black spheres are hydrophobic (h) residues, white and grey spheresare polar (p) residues. Peptides 1 (top row) and 2 (middle row) are takenfrom PDB [19] entry 1hgd of in uenza hemagglutinin [13]. Peptide 3 (bottomrow) is taken from PDB entry 2fbj of an immunoglobulin [14]. The pictureswere prepared using Molscript [20].Figure 2: Histogram of probability of donor-acceptor distance for peptide 1with glycosylations consisting of 0 to 4 GlcNAc monomers. Donor-acceptordistance is given in units of lattice spacings.Figure 3: Change of probability distribution of peptide 1 upon glycosylation.Probability di erences are given in percent, energy E is in units of hh-contactenergies, donor-acceptor distance rda is in units of lattice spacings. Probabil-ity di erence contours have been smoothed for clarity. Glycosylation causesa shift of probabilities from more extended high energy areas (minimum atrda = 3 and E = 1) to low energy turn structures (maximum at rda = 1and E = 3).18 Figure 4: Histogram of probability of donor-acceptor distance rda for peptide2 (left) and its mutant h10p (right). Dotted lines: non-glycosylated peptides,solid lines: glycopeptides. Axes as in Fig 2.Figure 5: Histogram of probability of rda for mutant p2h of peptide 3. Wild-type of peptide 3 has a distribution (not shown) very similar to mutant h10pof peptide 2 (right half of Fig. 4). Axes and lines as in Fig 4.19 Fig. 1 of Ho mann and Florke P NLGVA TVVODnsWAVTGP NGlcNAcGlcNAc PGNI STITP N Dns OIAGlcNAcGlcNAcGTW GWTVSMGNKTGlcNAcGlcNAcTVWKGGMTNDnsS Fig. 2 of Ho mann and Florke 02468donor−acceptor distance0.000.100.200.300.400.50 probability4 GlcNAc3 GlcNAc2 GlcNAc1 GlcNAcno GlcNAc Fig. 3 of Ho mann and Florke 21.510.50-0.5-1-1.5-20 1 2 3 4 5 6 7 8-3-2-10 donor-acceptor distanceenergy Fig. 4 of Ho mann and Florke 02468donor−acceptor distance0.000.100.200.300.40 probability 02468donor−acceptor distanceh10p Fig. 5 of Ho mann and Florke 02468donor−acceptor distance0.000.100.200.30 probability